/* Copyright 2006-2007 Niels Provos
 * Copyright 2007-2012 Nick Mathewson and Niels Provos
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. The name of the author may not be used to endorse or promote products
 *    derived from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

/* Based on software by Adam Langly. Adam's original message:
 *
 * Async DNS Library
 * Adam Langley <agl@imperialviolet.org>
 * Public Domain code
 *
 * This software is Public Domain. To view a copy of the public domain dedication,
 * visit http://creativecommons.org/licenses/publicdomain/ or send a letter to
 * Creative Commons, 559 Nathan Abbott Way, Stanford, California 94305, USA.
 *
 * I ask and expect, but do not require, that all derivative works contain an
 * attribution similar to:
 *	Parts developed by Adam Langley <agl@imperialviolet.org>
 *
 * You may wish to replace the word "Parts" with something else depending on
 * the amount of original code.
 *
 * (Derivative works does not include programs which link against, run or include
 * the source verbatim in their source distributions)
 *
 * Version: 0.1b
 */

#include "event2/event-config.h"
#include "evconfig-private.h"

#include <sys/types.h>

#ifndef _FORTIFY_SOURCE
#define _FORTIFY_SOURCE 3
#endif

#include <string.h>
#include <fcntl.h>
#ifdef EVENT__HAVE_SYS_TIME_H
#include <sys/time.h>
#endif
#ifdef EVENT__HAVE_STDINT_H
#include <stdint.h>
#endif
#include <stdlib.h>
#include <string.h>
#include <errno.h>
#ifdef EVENT__HAVE_UNISTD_H
#include <unistd.h>
#endif
#include <limits.h>
#include <sys/stat.h>
#include <stdio.h>
#include <stdarg.h>
#ifdef _WIN32
#include <winsock2.h>
#include <winerror.h>
#include <ws2tcpip.h>
#ifndef _WIN32_IE
#define _WIN32_IE 0x400
#endif
#include <shlobj.h>
#endif

#include "event2/buffer.h"
#include "event2/bufferevent.h"
#include "event2/dns.h"
#include "event2/dns_struct.h"
#include "event2/dns_compat.h"
#include "event2/util.h"
#include "event2/event.h"
#include "event2/event_struct.h"
#include "event2/listener.h"
#include "event2/thread.h"

#include "defer-internal.h"
#include "log-internal.h"
#include "mm-internal.h"
#include "strlcpy-internal.h"
#include "ipv6-internal.h"
#include "util-internal.h"
#include "evthread-internal.h"
#ifdef _WIN32
#include <ctype.h>
#include <winsock2.h>
#include <windows.h>
#include <iphlpapi.h>
#include <io.h>
#else
#include <sys/socket.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#endif

#ifdef EVENT__HAVE_NETINET_IN6_H
#include <netinet/in6.h>
#endif

#define EVDNS_LOG_DEBUG EVENT_LOG_DEBUG
#define EVDNS_LOG_WARN EVENT_LOG_WARN
#define EVDNS_LOG_MSG EVENT_LOG_MSG

#ifndef EVDNS_NAME_MAX
#define EVDNS_NAME_MAX 255
#endif

#include <stdio.h>

#undef MIN
#undef MAX
#define MIN(a,b) ((a)<(b)?(a):(b))
#define MAX(a,b) ((a)>(b)?(a):(b))

#define ASSERT_VALID_REQUEST(req) \
	EVUTIL_ASSERT((req)->handle && (req)->handle->current_req == (req))

#define u64 ev_uint64_t
#define u32 ev_uint32_t
#define u16 ev_uint16_t
#define u8  ev_uint8_t

/* maximum number of addresses from a single packet */
/* that we bother recording */
#define MAX_V4_ADDRS 32
#define MAX_V6_ADDRS 32

/* Maximum allowable size of a DNS message over UDP without EDNS.*/
#define DNS_MAX_UDP_SIZE 512
/* Maximum allowable size of a DNS message over UDP with EDNS.*/
#define EDNS_MAX_UDP_SIZE 65535

#define EDNS_ENABLED(base) \
	(((base)->global_max_udp_size) > DNS_MAX_UDP_SIZE)

#define TYPE_A	       EVDNS_TYPE_A
#define TYPE_CNAME     5
#define TYPE_PTR       EVDNS_TYPE_PTR
#define TYPE_SOA       EVDNS_TYPE_SOA
#define TYPE_AAAA      EVDNS_TYPE_AAAA
#define TYPE_OPT       41

#define CLASS_INET     EVDNS_CLASS_INET

/* Timeout in seconds for idle TCP connections that server keeps alive. */
#define SERVER_IDLE_CONN_TIMEOUT 10
/* Timeout in seconds for idle TCP connections that client keeps alive. */
#define CLIENT_IDLE_CONN_TIMEOUT 5
/* Default maximum number of simultaneous TCP client connections that DNS server can hold. */
#define MAX_CLIENT_CONNECTIONS 10

/* Persistent handle.  We keep this separate from 'struct request' since we
 * need some object to last for as long as an evdns_request is outstanding so
 * that it can be canceled, whereas a search request can lead to multiple
 * 'struct request' instances being created over its lifetime. */
struct evdns_request {
	struct request *current_req;
	struct evdns_base *base;

	int pending_cb; /* Waiting for its callback to be invoked; not
			 * owned by event base any more. */

	/* elements used by the searching code */
	int search_index;
	struct search_state *search_state;
	char *search_origname;	/* needs to be free()ed */
	int search_flags;
	u16 tcp_flags;
};

struct request {
	u8 *request;  /* the dns packet data */
	u16 request_size; /* size of memory block stored in request field */
	u8 request_type; /* TYPE_PTR or TYPE_A or TYPE_AAAA */
	unsigned int request_len;
	int reissue_count;
	int tx_count;  /* the number of times that this packet has been sent */
	void *user_pointer;  /* the pointer given to us for this request */
	evdns_callback_type user_callback;
	struct nameserver *ns;	/* the server which we last sent it */

	/* these objects are kept in a circular list */
	/* XXX We could turn this into a CIRCLEQ. */
	struct request *next, *prev;

	struct event timeout_event;

	u16 trans_id;  /* the transaction id */
	unsigned request_appended :1;	/* true if the request pointer is data which follows this struct */
	unsigned transmit_me :1;  /* needs to be transmitted */
	unsigned need_cname :1;   /* make a separate callback for CNAME */

	/* XXXX This is a horrible hack. */
	char **put_cname_in_ptr; /* store the cname here if we get one. */

	struct evdns_base *base;

	struct evdns_request *handle;
};

struct reply {
	unsigned int type;
	unsigned int have_answer : 1;
	u32 rr_count;
	union {
		u32 *a;
		struct in6_addr *aaaa;
		char *ptr_name;
		void *raw;
	} data;
	char *cname;
};

enum tcp_state {
	TS_DISCONNECTED,
	TS_CONNECTING,
	TS_CONNECTED
};

struct tcp_connection {
	struct bufferevent *bev;
	enum tcp_state state;
	u16 awaiting_packet_size;
};

struct evdns_server_port;

struct client_tcp_connection {
	LIST_ENTRY(client_tcp_connection) next;
	struct tcp_connection connection;
	struct evdns_server_port *port;
};

struct nameserver {
	evutil_socket_t socket;	 /* a connected UDP socket */
	struct tcp_connection *connection; /* intended for TCP support */
	struct sockaddr_storage address;
	ev_socklen_t addrlen;
	int failed_times;  /* number of times which we have given this server a chance */
	int timedout;  /* number of times in a row a request has timed out */
	struct event event;
	/* these objects are kept in a circular list */
	struct nameserver *next, *prev;
	struct event timeout_event;  /* used to keep the timeout for */
				     /* when we next probe this server. */
				     /* Valid if state == 0 */
	/* Outstanding probe request for this nameserver, if any */
	struct evdns_request *probe_request;
	char state;  /* zero if we think that this server is down */
	char choked;  /* true if we have an EAGAIN from this server's socket */
	char write_waiting;  /* true if we are waiting for EV_WRITE events */
	struct evdns_base *base;

	/* Number of currently inflight requests: used
	 * to track when we should add/del the event. */
	int requests_inflight;
};


/* Represents a local port where we're listening for DNS requests. */
struct evdns_server_port {
	evutil_socket_t socket; /* socket we use to read queries and write replies. */
	int refcnt; /* reference count. */
	char choked; /* Are we currently blocked from writing? */
	char closing; /* Are we trying to close this port, pending writes? */
	evdns_request_callback_fn_type user_callback; /* Fn to handle requests */
	void *user_data; /* Opaque pointer passed to user_callback */
	struct event event; /* Read/write event */
	/* circular list of replies that we want to write. */
	struct server_request *pending_replies;
	struct event_base *event_base;

	/* Structures for tcp support */
	struct evconnlistener *listener;
	LIST_HEAD(client_list, client_tcp_connection) client_connections;
	unsigned client_connections_count;
	unsigned max_client_connections;
	struct timeval tcp_idle_timeout;

#ifndef EVENT__DISABLE_THREAD_SUPPORT
	void *lock;
#endif
};

/* Represents part of a reply being built.	(That is, a single RR.) */
struct server_reply_item {
	struct server_reply_item *next; /* next item in sequence. */
	char *name; /* name part of the RR */
	u16 type; /* The RR type */
	u16 class; /* The RR class (usually CLASS_INET) */
	u32 ttl; /* The RR TTL */
	char is_name; /* True iff data is a label */
	u16 datalen; /* Length of data; -1 if data is a label */
	void *data; /* The contents of the RR */
};

/* Represents a request that we've received as a DNS server, and holds */
/* the components of the reply as we're constructing it. */
struct server_request {
	/* Pointers to the next and previous entries on the list of replies */
	/* that we're waiting to write.	 Only set if we have tried to respond */
	/* and gotten EAGAIN. */
	struct server_request *next_pending;
	struct server_request *prev_pending;

	u16 trans_id; /* Transaction id. */
	struct evdns_server_port *port; /* Which port received this request on? */
	struct client_tcp_connection *client; /* Equal to NULL in case of UDP connection. */
	struct sockaddr_storage addr; /* Where to send the response in case of UDP. Equal to NULL in case of TCP connection.*/
	ev_socklen_t addrlen; /* length of addr */
	u16 max_udp_reply_size; /* Maximum size of udp reply that client can handle. */

	int n_answer; /* how many answer RRs have been set? */
	int n_authority; /* how many authority RRs have been set? */
	int n_additional; /* how many additional RRs have been set? */

	struct server_reply_item *answer; /* linked list of answer RRs */
	struct server_reply_item *authority; /* linked list of authority RRs */
	struct server_reply_item *additional; /* linked list of additional RRs */

	/* Constructed response.  Only set once we're ready to send a reply. */
	/* Once this is set, the RR fields are cleared, and no more should be set. */
	char *response;
	size_t response_len;

	/* Caller-visible fields: flags, questions. */
	struct evdns_server_request base;
};

struct evdns_base {
	/* An array of n_req_heads circular lists for inflight requests.
	 * Each inflight request req is in req_heads[req->trans_id % n_req_heads].
	 */
	struct request **req_heads;
	/* A circular list of requests that we're waiting to send, but haven't
	 * sent yet because there are too many requests inflight */
	struct request *req_waiting_head;
	/* A circular list of nameservers. */
	struct nameserver *server_head;
	int n_req_heads;

	struct event_base *event_base;

	/* The number of good nameservers that we have */
	int global_good_nameservers;

	/* inflight requests are contained in the req_head list */
	/* and are actually going out across the network */
	int global_requests_inflight;
	/* requests which aren't inflight are in the waiting list */
	/* and are counted here */
	int global_requests_waiting;

	int global_max_requests_inflight;

	struct timeval global_timeout;	/* 5 seconds by default */
	int global_max_reissues;  /* a reissue occurs when we get some errors from the server */
	int global_max_retransmits;  /* number of times we'll retransmit a request which timed out */
	/* number of timeouts in a row before we consider this server to be down */
	int global_max_nameserver_timeout;
	/* true iff we will use the 0x20 hack to prevent poisoning attacks. */
	int global_randomize_case;
	/* Maximum size of a UDP DNS packet. */
	u16 global_max_udp_size;

	/* The first time that a nameserver fails, how long do we wait before
	 * probing to see if it has returned?  */
	struct timeval global_nameserver_probe_initial_timeout;

	/* Combination of DNS_QUERY_USEVC, DNS_QUERY_IGNTC flags
	 * to control requests via TCP. */
	u16 global_tcp_flags;
	/* Idle timeout for outgoing TCP connections. */
	struct timeval global_tcp_idle_timeout;

	/** Port to bind to for outgoing DNS packets. */
	struct sockaddr_storage global_outgoing_address;
	/** ev_socklen_t for global_outgoing_address. 0 if it isn't set. */
	ev_socklen_t global_outgoing_addrlen;

	struct timeval global_getaddrinfo_allow_skew;

	int so_rcvbuf;
	int so_sndbuf;

	int getaddrinfo_ipv4_timeouts;
	int getaddrinfo_ipv6_timeouts;
	int getaddrinfo_ipv4_answered;
	int getaddrinfo_ipv6_answered;

	struct search_state *global_search_state;

	TAILQ_HEAD(hosts_list, hosts_entry) hostsdb;

#ifndef EVENT__DISABLE_THREAD_SUPPORT
	void *lock;
#endif

	int disable_when_inactive;

	/* Maximum timeout between two probe packets
	 * will change `global_nameserver_probe_initial_timeout`
	 * when this value is smaller */
	int ns_max_probe_timeout;
	/* Backoff factor of probe timeout */
	int ns_timeout_backoff_factor;
};

struct hosts_entry {
	TAILQ_ENTRY(hosts_entry) next;
	union {
		struct sockaddr sa;
		struct sockaddr_in sin;
		struct sockaddr_in6 sin6;
	} addr;
	int addrlen;
	char hostname[1];
};

static struct evdns_base *current_base = NULL;

struct evdns_base *
evdns_get_global_base(void)
{
	return current_base;
}

/* Given a pointer to an evdns_server_request, get the corresponding */
/* server_request. */
#define TO_SERVER_REQUEST(base_ptr)					\
	((struct server_request*)					\
	  (((char*)(base_ptr) - evutil_offsetof(struct server_request, base))))

#define REQ_HEAD(base, id) ((base)->req_heads[id % (base)->n_req_heads])

static struct nameserver *nameserver_pick(struct evdns_base *base);
static void evdns_request_insert(struct request *req, struct request **head);
static void evdns_request_remove(struct request *req, struct request **head);
static void nameserver_ready_callback(evutil_socket_t fd, short events, void *arg);
static int evdns_transmit(struct evdns_base *base);
static int evdns_request_transmit(struct request *req);
static void nameserver_send_probe(struct nameserver *const ns);
static void search_request_finished(struct evdns_request *const);
static int search_try_next(struct evdns_request *const req);
static struct request *search_request_new(struct evdns_base *base, struct evdns_request *handle, int type, const char *const name, int flags, evdns_callback_type user_callback, void *user_arg);
static void evdns_requests_pump_waiting_queue(struct evdns_base *base);
static u16 transaction_id_pick(struct evdns_base *base);
static struct request *request_new(struct evdns_base *base, struct evdns_request *handle, int type, const char *name, int flags, evdns_callback_type callback, void *ptr);
static struct request *request_clone(struct evdns_base *base, struct request* current);
static void request_submit(struct request *const req);

static int server_request_free(struct server_request *req);
static void server_request_free_answers(struct server_request *req);
static void server_port_free(struct evdns_server_port *port);
static void server_port_ready_callback(evutil_socket_t fd, short events, void *arg);
static int evdns_base_resolv_conf_parse_impl(struct evdns_base *base, int flags, const char *const filename);
static int evdns_base_set_option_impl(struct evdns_base *base,
    const char *option, const char *val, int flags);
static void evdns_base_free_and_unlock(struct evdns_base *base, int fail_requests);
static void evdns_request_timeout_callback(evutil_socket_t fd, short events, void *arg);
static int evdns_server_request_format_response(struct server_request *req, int err);
static void incoming_conn_cb(struct evconnlistener *listener, evutil_socket_t fd,
    struct sockaddr *address, int socklen, void *arg);

static int strtoint(const char *const str);

#ifdef EVENT__DISABLE_THREAD_SUPPORT
#define EVDNS_LOCK(base)  EVUTIL_NIL_STMT_
#define EVDNS_UNLOCK(base) EVUTIL_NIL_STMT_
#define ASSERT_LOCKED(base) EVUTIL_NIL_STMT_
#else
#define EVDNS_LOCK(base)			\
	EVLOCK_LOCK((base)->lock, 0)
#define EVDNS_UNLOCK(base)			\
	EVLOCK_UNLOCK((base)->lock, 0)
#define ASSERT_LOCKED(base)			\
	EVLOCK_ASSERT_LOCKED((base)->lock)
#endif

static evdns_debug_log_fn_type evdns_log_fn = NULL;

void
evdns_set_log_fn(evdns_debug_log_fn_type fn)
{
	evdns_log_fn = fn;
}

#ifdef __GNUC__
#define EVDNS_LOG_CHECK	 __attribute__ ((format(printf, 2, 3)))
#else
#define EVDNS_LOG_CHECK
#endif

static void evdns_log_(int severity, const char *fmt, ...) EVDNS_LOG_CHECK;
static void
evdns_log_(int severity, const char *fmt, ...)
{
	va_list args;
	va_start(args,fmt);
	if (evdns_log_fn) {
		char buf[512];
		int is_warn = (severity == EVDNS_LOG_WARN);
		evutil_vsnprintf(buf, sizeof(buf), fmt, args);
		evdns_log_fn(is_warn, buf);
	} else {
		event_logv_(severity, NULL, fmt, args);
	}
	va_end(args);
}

#define log evdns_log_

/* Initialize tcp_connection structure. */
static void
init_tcp_connection(struct tcp_connection *conn, struct bufferevent *bev)
{
	memset(conn, 0, sizeof(*conn));
	conn->state = TS_DISCONNECTED;
	conn->bev = bev;
	conn->awaiting_packet_size = 0;
}

/* Disconnect tcp connection. */
static void
evdns_tcp_disconnect(struct tcp_connection *conn)
{
	if (!conn)
		return;
	conn->state = TS_DISCONNECTED;
	conn->awaiting_packet_size = 0;
	if (conn->bev) {
		bufferevent_free(conn->bev);
		conn->bev = NULL;
	}
}

/* Add new tcp client to the list of TCP clients in the TCP DNS server. */
static struct client_tcp_connection*
evdns_add_tcp_client(struct evdns_server_port *port, struct bufferevent *bev)
{
	struct client_tcp_connection *client;
	EVUTIL_ASSERT(port && bev);
	if (port->max_client_connections == port->client_connections_count)
		goto error;

	client = mm_calloc(1, sizeof(*client));
	if (!client)
		goto error;
	init_tcp_connection(&client->connection, bev);
	client->port = port;
	LIST_INSERT_HEAD(&port->client_connections, client, next);

	++port->client_connections_count;
	/* we need to hold evdns_server_port as long as one connection at least stays alive */
	++port->refcnt;
	return client;
error:
	return NULL;
}

/* Remove tcp client and free all associated data from the TCP DNS server. */
static int
evdns_remove_tcp_client(struct evdns_server_port *port, struct client_tcp_connection *client)
{
	if (!port || !client)
		goto error;

	evdns_tcp_disconnect(&client->connection);
	LIST_REMOVE(client, next);
	mm_free(client);
	--port->client_connections_count;
	--port->refcnt;
	return 0;
error:
	return -1;
}

/* Remove all tcp clients and free all associated data from the TCP DNS server. */
static void
evdns_remove_all_tcp_clients(struct evdns_server_port *port)
{
	struct client_tcp_connection *client;
	while ((client = LIST_FIRST(&port->client_connections))) {
		evdns_remove_tcp_client(port, client);
	}
}

/* Create new tcp connection structure for DNS client. */
static struct tcp_connection *
new_tcp_connecton(struct bufferevent *bev)
{
	struct tcp_connection *conn;
	if (!bev)
		return NULL;

	conn = mm_calloc(1, sizeof(*conn));
	if (!conn)
		return NULL;
	init_tcp_connection(conn, bev);
	return conn;
}

/* Disconnect and free all associated data for the tcp connection in DNS client. */
static void
disconnect_and_free_connection(struct tcp_connection *conn)
{
	if (!conn)
		return;
	evdns_tcp_disconnect(conn);
	mm_free(conn);
}

/* This walks the list of inflight requests to find the */
/* one with a matching transaction id. Returns NULL on */
/* failure */
static struct request *
request_find_from_trans_id(struct evdns_base *base, u16 trans_id) {
	struct request *req = REQ_HEAD(base, trans_id);
	struct request *const started_at = req;

	ASSERT_LOCKED(base);

	if (req) {
		do {
			if (req->trans_id == trans_id) return req;
			req = req->next;
		} while (req != started_at);
	}

	return NULL;
}

/* a libevent callback function which is called when a nameserver */
/* has gone down and we want to test if it has came back to life yet */
static void
nameserver_prod_callback(evutil_socket_t fd, short events, void *arg) {
	struct nameserver *const ns = (struct nameserver *) arg;
	(void)fd;
	(void)events;

	EVDNS_LOCK(ns->base);
	nameserver_send_probe(ns);
	EVDNS_UNLOCK(ns->base);
}

/* a libevent callback which is called when a nameserver probe (to see if */
/* it has come back to life) times out. We increment the count of failed_times */
/* and wait longer to send the next probe packet. */
static void
nameserver_probe_failed(struct nameserver *const ns) {
	struct timeval timeout;
	int i;

	ASSERT_LOCKED(ns->base);
	(void) evtimer_del(&ns->timeout_event);
	if (ns->state == 1) {
		/* This can happen if the nameserver acts in a way which makes us mark */
		/* it as bad and then starts sending good replies. */
		return;
	}

	memcpy(&timeout, &ns->base->global_nameserver_probe_initial_timeout,
	    sizeof(struct timeval));
	for (i = ns->failed_times; i > 0 && timeout.tv_sec < ns->base->ns_max_probe_timeout; --i) {
		timeout.tv_sec *= ns->base->ns_timeout_backoff_factor;
		timeout.tv_usec *= ns->base->ns_timeout_backoff_factor;
		if (timeout.tv_usec > 1000000) {
			timeout.tv_sec += timeout.tv_usec / 1000000;
			timeout.tv_usec %= 1000000;
		}
	}
	if (timeout.tv_sec > ns->base->ns_max_probe_timeout) {
		timeout.tv_sec = ns->base->ns_max_probe_timeout;
		timeout.tv_usec = 0;
	}

	ns->failed_times++;

	if (evtimer_add(&ns->timeout_event, &timeout) < 0) {
		char addrbuf[128];
		log(EVDNS_LOG_WARN,
		    "Error from libevent when adding timer event for %s",
		    evutil_format_sockaddr_port_(
			    (struct sockaddr *)&ns->address,
			    addrbuf, sizeof(addrbuf)));
	}
}

static void
request_swap_ns(struct request *req, struct nameserver *ns) {
	if (ns && req->ns != ns) {
		EVUTIL_ASSERT(req->ns->requests_inflight > 0);
		req->ns->requests_inflight--;
		ns->requests_inflight++;

		req->ns = ns;
	}
}

/* called when a nameserver has been deemed to have failed. For example, too */
/* many packets have timed out etc */
static void
nameserver_failed(struct nameserver *const ns, const char *msg, int err) {
	struct request *req, *started_at;
	struct evdns_base *base = ns->base;
	int i;
	char addrbuf[128];

	ASSERT_LOCKED(base);
	/* if this nameserver has already been marked as failed */
	/* then don't do anything */
	if (!ns->state) return;

	log(EVDNS_LOG_MSG, "Nameserver %s has failed: %s",
	    evutil_format_sockaddr_port_(
		    (struct sockaddr *)&ns->address,
		    addrbuf, sizeof(addrbuf)),
	    msg);

	base->global_good_nameservers--;
	EVUTIL_ASSERT(base->global_good_nameservers >= 0);
	if (base->global_good_nameservers == 0) {
		log(EVDNS_LOG_MSG, "All nameservers have failed");
	}

	ns->state = 0;
	ns->failed_times = 1;

	if (ns->connection) {
		disconnect_and_free_connection(ns->connection);
		ns->connection = NULL;
	} else if (err == ENOTCONN) {
		/* XXX: If recvfrom results in ENOTCONN, the socket remains readable
		 * which triggers another recvfrom. The observed behavior is 100% CPU use.
		 * This occurs on iOS (kqueue) after the process has been backgrounded
		 * for a long time (~300 seconds) and then resumed.
		 * All sockets, TCP and UDP, seem to get ENOTCONN and must be closed.
		 * https://github.com/libevent/libevent/issues/265 */
		const struct sockaddr *address = (const struct sockaddr *)&ns->address;
		evutil_closesocket(ns->socket);
		ns->socket = evutil_socket_(address->sa_family,
			SOCK_DGRAM | EVUTIL_SOCK_NONBLOCK | EVUTIL_SOCK_CLOEXEC, 0);

		if (base->global_outgoing_addrlen &&
			!evutil_sockaddr_is_loopback_(address)) {
			if (bind(ns->socket,
					(struct sockaddr *)&base->global_outgoing_address,
					base->global_outgoing_addrlen) < 0) {
				log(EVDNS_LOG_WARN, "Couldn't bind to outgoing address");
			}
		}

		event_del(&ns->event);
		event_assign(&ns->event, ns->base->event_base, ns->socket,
			EV_READ | (ns->write_waiting ? EV_WRITE : 0) | EV_PERSIST,
			nameserver_ready_callback, ns);
		if (!base->disable_when_inactive && event_add(&ns->event, NULL) < 0) {
			log(EVDNS_LOG_WARN, "Couldn't add %s event",
				ns->write_waiting ? "rw": "read");
		}
	}
	if (evtimer_add(&ns->timeout_event,
		&base->global_nameserver_probe_initial_timeout) < 0) {
		log(EVDNS_LOG_WARN,
		    "Error from libevent when adding timer event for %s",
		    evutil_format_sockaddr_port_(
			    (struct sockaddr *)&ns->address,
			    addrbuf, sizeof(addrbuf)));
		/* ???? Do more? */
	}

	/* walk the list of inflight requests to see if any can be reassigned to */
	/* a different server. Requests in the waiting queue don't have a */
	/* nameserver assigned yet */

	/* if we don't have *any* good nameservers then there's no point */
	/* trying to reassign requests to one */
	if (!base->global_good_nameservers) return;

	for (i = 0; i < base->n_req_heads; ++i) {
		req = started_at = base->req_heads[i];
		if (req) {
			do {
				if (req->tx_count == 0 && req->ns == ns) {
					/* still waiting to go out, can be moved */
					/* to another server */
					request_swap_ns(req, nameserver_pick(base));
				}
				req = req->next;
			} while (req != started_at);
		}
	}
}

static void
nameserver_up(struct nameserver *const ns)
{
	char addrbuf[128];
	ASSERT_LOCKED(ns->base);
	if (ns->state) return;
	log(EVDNS_LOG_MSG, "Nameserver %s is back up",
	    evutil_format_sockaddr_port_(
		    (struct sockaddr *)&ns->address,
		    addrbuf, sizeof(addrbuf)));
	evtimer_del(&ns->timeout_event);
	if (ns->probe_request) {
		evdns_cancel_request(ns->base, ns->probe_request);
		ns->probe_request = NULL;
	}
	ns->state = 1;
	ns->failed_times = 0;
	ns->timedout = 0;
	ns->base->global_good_nameservers++;
}

static void
request_trans_id_set(struct request *const req, const u16 trans_id) {
	req->trans_id = trans_id;
	*((u16 *) req->request) = htons(trans_id);
}

/* Called to remove a request from a list and dealloc it. */
/* head is a pointer to the head of the list it should be */
/* removed from or NULL if the request isn't in a list. */
/* when free_handle is one, free the handle as well. */
static void
request_finished(struct request *const req, struct request **head, int free_handle) {
	struct evdns_base *base = req->base;
	int was_inflight = (head != &base->req_waiting_head);
	EVDNS_LOCK(base);
	ASSERT_VALID_REQUEST(req);

	if (head)
		evdns_request_remove(req, head);

	log(EVDNS_LOG_DEBUG, "Removing timeout for request %p", (void *)req);
	if (was_inflight) {
		evtimer_del(&req->timeout_event);
		base->global_requests_inflight--;
		req->ns->requests_inflight--;
	} else {
		base->global_requests_waiting--;
	}
	/* it was initialized during request_new / evtimer_assign */
	event_debug_unassign(&req->timeout_event);

	if (req->ns &&
	    req->ns->requests_inflight == 0 &&
	    req->base->disable_when_inactive) {
		event_del(&req->ns->event);
		evtimer_del(&req->ns->timeout_event);
	}

	if (!req->request_appended) {
		/* need to free the request data on it's own */
		mm_free(req->request);
	} else {
		/* the request data is appended onto the header */
		/* so everything gets free()ed when we: */
	}

	if (req->handle) {
		EVUTIL_ASSERT(req->handle->current_req == req);

		if (free_handle) {
			search_request_finished(req->handle);
			req->handle->current_req = NULL;
			if (! req->handle->pending_cb) {
				/* If we're planning to run the callback,
				 * don't free the handle until later. */
				mm_free(req->handle);
			}
			req->handle = NULL; /* If we have a bug, let's crash
					     * early */
		} else {
			req->handle->current_req = NULL;
		}
	}

	mm_free(req);

	evdns_requests_pump_waiting_queue(base);
	EVDNS_UNLOCK(base);
}

/* This is called when a server returns a funny error code. */
/* We try the request again with another server. */
/* */
/* return: */
/*   0 ok */
/*   1 failed/reissue is pointless */
static int
request_reissue(struct request *req) {
	const struct nameserver *const last_ns = req->ns;
	ASSERT_LOCKED(req->base);
	ASSERT_VALID_REQUEST(req);
	/* the last nameserver should have been marked as failing */
	/* by the caller of this function, therefore pick will try */
	/* not to return it */
	request_swap_ns(req, nameserver_pick(req->base));
	if (req->ns == last_ns) {
		/* ... but pick did return it */
		/* not a lot of point in trying again with the */
		/* same server */
		return 1;
	}

	req->reissue_count++;
	req->tx_count = 0;
	req->transmit_me = 1;

	return 0;
}

/* this function looks for space on the inflight queue and promotes */
/* requests from the waiting queue if it can. */
/* */
/* TODO: */
/* add return code, see at nameserver_pick() and other functions. */
static void
evdns_requests_pump_waiting_queue(struct evdns_base *base) {
	ASSERT_LOCKED(base);
	while (base->global_requests_inflight < base->global_max_requests_inflight &&
		   base->global_requests_waiting) {
		struct request *req;

		EVUTIL_ASSERT(base->req_waiting_head);
		req = base->req_waiting_head;

		req->ns = nameserver_pick(base);
		if (!req->ns)
			return;

		/* move a request from the waiting queue to the inflight queue */
		req->ns->requests_inflight++;

		evdns_request_remove(req, &base->req_waiting_head);

		base->global_requests_waiting--;
		base->global_requests_inflight++;

		request_trans_id_set(req, transaction_id_pick(base));

		evdns_request_insert(req, &REQ_HEAD(base, req->trans_id));
		evdns_request_transmit(req);
		evdns_transmit(base);
	}
}

/* TODO(nickm) document */
struct deferred_reply_callback {
	struct event_callback deferred;
	struct evdns_request *handle;
	u8 request_type;
	u8 have_reply;
	u32 ttl;
	u32 err;
	evdns_callback_type user_callback;
	struct reply reply;
};

static void
reply_run_callback(struct event_callback *d, void *user_pointer)
{
	struct deferred_reply_callback *cb =
	    EVUTIL_UPCAST(d, struct deferred_reply_callback, deferred);

	switch (cb->request_type) {
	case TYPE_A:
		if (cb->have_reply) {
			cb->user_callback(DNS_ERR_NONE, DNS_IPv4_A,
			    cb->reply.rr_count, cb->ttl,
			    cb->reply.data.a,
			    user_pointer);
			if (cb->reply.cname)
				cb->user_callback(DNS_ERR_NONE, DNS_CNAME, 1,
				    cb->ttl, cb->reply.cname, user_pointer);
		} else
			cb->user_callback(cb->err, 0, 0, cb->ttl, NULL, user_pointer);
		break;
	case TYPE_PTR:
		if (cb->have_reply) {
			char *name = cb->reply.data.ptr_name;
			cb->user_callback(DNS_ERR_NONE, DNS_PTR, 1, cb->ttl,
			    &name, user_pointer);
		} else {
			cb->user_callback(cb->err, 0, 0, cb->ttl, NULL, user_pointer);
		}
		break;
	case TYPE_AAAA:
		if (cb->have_reply) {
			cb->user_callback(DNS_ERR_NONE, DNS_IPv6_AAAA,
			    cb->reply.rr_count, cb->ttl,
			    cb->reply.data.aaaa,
			    user_pointer);
			if (cb->reply.cname)
				cb->user_callback(DNS_ERR_NONE, DNS_CNAME, 1,
				    cb->ttl, cb->reply.cname, user_pointer);
		} else
			cb->user_callback(cb->err, 0, 0, cb->ttl, NULL, user_pointer);
		break;
	default:
		EVUTIL_ASSERT(0);
	}

	if (cb->handle && cb->handle->pending_cb) {
		mm_free(cb->handle);
	}

	if (cb->reply.data.raw) {
		mm_free(cb->reply.data.raw);
	}

	if (cb->reply.cname) {
		mm_free(cb->reply.cname);
	}

	mm_free(cb);
}

static void
reply_schedule_callback(struct request *const req, u32 ttl, u32 err, struct reply *reply)
{
	struct deferred_reply_callback *d = mm_calloc(1, sizeof(*d));

	if (!d) {
		event_warn("%s: Couldn't allocate space for deferred callback.",
		    __func__);
		return;
	}

	ASSERT_LOCKED(req->base);

	d->request_type = req->request_type;
	d->user_callback = req->user_callback;
	d->ttl = ttl;
	d->err = err;
	if (reply) {
		d->have_reply = 1;
		memcpy(&d->reply, reply, sizeof(struct reply));
		/* We've taken ownership of the data. */
		reply->data.raw = NULL;
	}

	if (req->handle) {
		req->handle->pending_cb = 1;
		d->handle = req->handle;
	}

	event_deferred_cb_init_(
	    &d->deferred,
	    event_get_priority(&req->timeout_event),
	    reply_run_callback,
	    req->user_pointer);
	event_deferred_cb_schedule_(
		req->base->event_base,
		&d->deferred);
}

static int
client_retransmit_through_tcp(struct evdns_request *handle)
{
	struct request *req = handle->current_req;
	struct evdns_base *base = req->base;
	struct request *newreq = request_clone(base, req);
	ASSERT_LOCKED(base);
	if (!newreq)
		return 1;
	request_finished(req, &REQ_HEAD(req->base, req->trans_id), 0);
	handle->current_req = newreq;
	newreq->handle = handle;
	request_submit(newreq);
	return 0;
}

#define _QR_MASK    0x8000U
#define _OP_MASK    0x7800U
#define _AA_MASK    0x0400U
#define _TC_MASK    0x0200U
#define _RD_MASK    0x0100U
#define _RA_MASK    0x0080U
#define _Z_MASK     0x0040U
#define _AD_MASK    0x0020U
#define _CD_MASK    0x0010U
#define _RCODE_MASK 0x000fU
#define _Z_MASK_DEPRECATED 0x0070U

/* this processes a parsed reply packet */
static void
reply_handle(struct request *const req, u16 flags, u32 ttl, struct reply *reply) {
	int error;
	char addrbuf[128];
	int retransmit_via_tcp = 0;
	static const int error_codes[] = {
		DNS_ERR_FORMAT, DNS_ERR_SERVERFAILED, DNS_ERR_NOTEXIST,
		DNS_ERR_NOTIMPL, DNS_ERR_REFUSED
	};

	ASSERT_LOCKED(req->base);
	ASSERT_VALID_REQUEST(req);

	if (flags & (_RCODE_MASK | _TC_MASK) || !reply || !reply->have_answer) {
		/* there was an error */
		if (flags & _TC_MASK) {
			error = DNS_ERR_TRUNCATED;
			retransmit_via_tcp = (req->handle->tcp_flags & (DNS_QUERY_IGNTC | DNS_QUERY_USEVC)) == 0;
		} else if (flags & _RCODE_MASK) {
			u16 error_code = (flags & _RCODE_MASK) - 1;
			if (error_code > 4) {
				error = DNS_ERR_UNKNOWN;
			} else {
				error = error_codes[error_code];
			}
		} else if (reply && !reply->have_answer) {
			error = DNS_ERR_NODATA;
		} else {
			error = DNS_ERR_UNKNOWN;
		}

		switch (error) {
		case DNS_ERR_NOTIMPL:
		case DNS_ERR_REFUSED:
			/* we regard these errors as marking a bad nameserver */
			if (req->reissue_count < req->base->global_max_reissues) {
				char msg[64];
				evutil_snprintf(msg, sizeof(msg), "Bad response %d (%s)",
					 error, evdns_err_to_string(error));
				nameserver_failed(req->ns, msg, 0);
				if (!request_reissue(req)) return;
			}
			break;
		case DNS_ERR_SERVERFAILED:
			/* rcode 2 (servfailed) sometimes means "we
			 * are broken" and sometimes (with some binds)
			 * means "that request was very confusing."
			 * Treat this as a timeout, not a failure.
			 */
			log(EVDNS_LOG_DEBUG, "Got a SERVERFAILED from nameserver"
				"at %s; will allow the request to time out.",
			    evutil_format_sockaddr_port_(
				    (struct sockaddr *)&req->ns->address,
				    addrbuf, sizeof(addrbuf)));
			/* Call the timeout function */
			evdns_request_timeout_callback(0, 0, req);
			return;
		default:
			/* we got a good reply from the nameserver: it is up. */
			if (req->handle == req->ns->probe_request) {
				/* Avoid double-free */
				req->ns->probe_request = NULL;
			}

			nameserver_up(req->ns);
		}

		if (retransmit_via_tcp) {
			log(EVDNS_LOG_DEBUG, "Recieved truncated reply(flags 0x%x, transanc ID: %d). Retransmiting via TCP.",
				req->handle->tcp_flags, req->trans_id);
			req->handle->tcp_flags |= DNS_QUERY_USEVC;
			client_retransmit_through_tcp(req->handle);
			return;
		}

		if (req->handle->search_state &&
		    req->request_type != TYPE_PTR) {
			/* if we have a list of domains to search in,
			 * try the next one */
			if (!search_try_next(req->handle)) {
				/* a new request was issued so this
				 * request is finished and */
				/* the user callback will be made when
				 * that request (or a */
				/* child of it) finishes. */
				return;
			}
		}

		/* all else failed. Pass the failure up */
		reply_schedule_callback(req, ttl, error, NULL);
		request_finished(req, &REQ_HEAD(req->base, req->trans_id), 1);
	} else {
		/* all ok, tell the user */
		reply_schedule_callback(req, ttl, 0, reply);
		if (req->handle == req->ns->probe_request)
			req->ns->probe_request = NULL; /* Avoid double-free */
		nameserver_up(req->ns);
		request_finished(req, &REQ_HEAD(req->base, req->trans_id), 1);
	}
}

static int
name_parse(u8 *packet, int length, int *idx, char *name_out, int name_out_len) {
	int name_end = -1;
	int j = *idx;
	int ptr_count = 0;
#define GET32(x) do { if (j + 4 > length) goto err; memcpy(&t32_, packet + j, 4); j += 4; x = ntohl(t32_); } while (0)
#define GET16(x) do { if (j + 2 > length) goto err; memcpy(&t_, packet + j, 2); j += 2; x = ntohs(t_); } while (0)
#define GET8(x) do { if (j >= length) goto err; x = packet[j++]; } while (0)

	char *cp = name_out;
	const char *const end = name_out + name_out_len;

	/* Normally, names are a series of length prefixed strings terminated */
	/* with a length of 0 (the lengths are u8's < 63). */
	/* However, the length can start with a pair of 1 bits and that */
	/* means that the next 14 bits are a pointer within the current */
	/* packet. */

	for (;;) {
		u8 label_len;
		GET8(label_len);
		if (!label_len) break;
		if (label_len & 0xc0) {
			u8 ptr_low;
			GET8(ptr_low);
			if (name_end < 0) name_end = j;
			j = (((int)label_len & 0x3f) << 8) + ptr_low;
			/* Make sure that the target offset is in-bounds. */
			if (j < 0 || j >= length) return -1;
			/* If we've jumped more times than there are characters in the
			 * message, we must have a loop. */
			if (++ptr_count > length) return -1;
			continue;
		}
		if (label_len > 63) return -1;
		if (cp != name_out) {
			if (cp + 1 >= end) return -1;
			*cp++ = '.';
		}
		if (cp + label_len >= end) return -1;
		if (j + label_len > length) return -1;
		memcpy(cp, packet + j, label_len);
		cp += label_len;
		j += label_len;
	}
	if (cp >= end) return -1;
	*cp = '\0';
	if (name_end < 0)
		*idx = j;
	else
		*idx = name_end;
	return 0;
 err:
	return -1;
}

/* parses a raw request from a nameserver */
static int
reply_parse(struct evdns_base *base, u8 *packet, int length)
{
	int j = 0, k = 0;  /* index into packet */
	u16 t_;	 /* used by the macros */
	u32 t32_;  /* used by the macros */
	char tmp_name[256], cmp_name[256]; /* used by the macros */
	int name_matches = 0;

	u16 trans_id, questions, answers, authority, additional, datalength;
	u16 flags = 0;
	u32 ttl, ttl_r = 0xffffffff;
	struct reply reply;
	struct request *req = NULL;
	unsigned int i, buf_size;

	ASSERT_LOCKED(base);

	GET16(trans_id);
	GET16(flags);
	GET16(questions);
	GET16(answers);
	GET16(authority);
	GET16(additional);
	(void) authority; /* suppress "unused variable" warnings. */
	(void) additional; /* suppress "unused variable" warnings. */

	req = request_find_from_trans_id(base, trans_id);
	if (!req) return -1;
	EVUTIL_ASSERT(req->base == base);

	memset(&reply, 0, sizeof(reply));

	/* If it's not an answer, it doesn't correspond to any request. */
	if (!(flags & _QR_MASK)) return -1;  /* must be an answer */
	if ((flags & (_RCODE_MASK|_TC_MASK)) && (flags & (_RCODE_MASK|_TC_MASK)) != DNS_ERR_NOTEXIST) {
		/* there was an error and it's not NXDOMAIN */
		goto err;
	}
	/* if (!answers) return; */  /* must have an answer of some form */

	/* This macro skips a name in the DNS reply. */
#define SKIP_NAME						\
	do { tmp_name[0] = '\0';				\
		if (name_parse(packet, length, &j, tmp_name,	\
			sizeof(tmp_name))<0)			\
			goto err;				\
	} while (0)

	reply.type = req->request_type;

	/* skip over each question in the reply */
	for (i = 0; i < questions; ++i) {
		/* the question looks like
		 *   <label:name><u16:type><u16:class>
		 */
		tmp_name[0] = '\0';
		cmp_name[0] = '\0';
		k = j;
		if (name_parse(packet, length, &j, tmp_name, sizeof(tmp_name)) < 0)
			goto err;
		if (name_parse(req->request, req->request_len, &k,
			cmp_name, sizeof(cmp_name))<0)
			goto err;
		if (!base->global_randomize_case) {
			if (strcmp(tmp_name, cmp_name) == 0)
				name_matches = 1;
		} else {
			if (evutil_ascii_strcasecmp(tmp_name, cmp_name) == 0)
				name_matches = 1;
		}

		j += 4;
		if (j > length)
			goto err;
	}

	if (!name_matches)
		goto err;

	/* We can allocate less for the reply data, but to do it we'll have
	 * to parse the response. To simplify things let's just allocate
	 * a little bit more to avoid complex evaluations.
	 */
	buf_size = MAX(length - j, EVDNS_NAME_MAX);
	reply.data.raw = mm_malloc(buf_size);

	/* now we have the answer section which looks like
	 * <label:name><u16:type><u16:class><u32:ttl><u16:len><data...>
	 */

	for (i = 0; i < answers; ++i) {
		u16 type, class;

		SKIP_NAME;
		GET16(type);
		GET16(class);
		GET32(ttl);
		GET16(datalength);

		if (type == TYPE_A && class == CLASS_INET) {
			int addrcount;
			if (req->request_type != TYPE_A) {
				j += datalength; continue;
			}
			if ((datalength & 3) != 0) /* not an even number of As. */
			    goto err;
			addrcount = datalength >> 2;

			ttl_r = MIN(ttl_r, ttl);
			/* we only bother with the first four addresses. */
			if (j + 4*addrcount > length) goto err;
			memcpy(&reply.data.a[reply.rr_count],
				   packet + j, 4*addrcount);
			j += 4*addrcount;
			reply.rr_count += addrcount;
			reply.have_answer = 1;
		} else if (type == TYPE_PTR && class == CLASS_INET) {
			if (req->request_type != TYPE_PTR) {
				j += datalength; continue;
			}
			if (name_parse(packet, length, &j, reply.data.ptr_name,
						   buf_size)<0)
				goto err;
			ttl_r = MIN(ttl_r, ttl);
			reply.have_answer = 1;
			break;
		} else if (type == TYPE_CNAME) {
			char cname[EVDNS_NAME_MAX];
			if (name_parse(packet, length, &j, cname,
				sizeof(cname))<0)
				goto err;
			if (req->need_cname)
				reply.cname = mm_strdup(cname);
			if (req->put_cname_in_ptr && !*req->put_cname_in_ptr)
				*req->put_cname_in_ptr = mm_strdup(cname);
		} else if (type == TYPE_AAAA && class == CLASS_INET) {
			int addrcount;
			if (req->request_type != TYPE_AAAA) {
				j += datalength; continue;
			}
			if ((datalength & 15) != 0) /* not an even number of AAAAs. */
				goto err;
			addrcount = datalength >> 4;  /* each address is 16 bytes long */
			ttl_r = MIN(ttl_r, ttl);

			/* we only bother with the first four addresses. */
			if (j + 16*addrcount > length) goto err;
			memcpy(&reply.data.aaaa[reply.rr_count],
				   packet + j, 16*addrcount);
			reply.rr_count += addrcount;
			j += 16*addrcount;
			reply.have_answer = 1;
		} else {
			/* skip over any other type of resource */
			j += datalength;
		}
	}

	if (!reply.have_answer) {
		for (i = 0; i < authority; ++i) {
			u16 type, class;
			SKIP_NAME;
			GET16(type);
			GET16(class);
			GET32(ttl);
			GET16(datalength);
			if (type == TYPE_SOA && class == CLASS_INET) {
				u32 serial, refresh, retry, expire, minimum;
				SKIP_NAME;
				SKIP_NAME;
				GET32(serial);
				GET32(refresh);
				GET32(retry);
				GET32(expire);
				GET32(minimum);
				(void)expire;
				(void)retry;
				(void)refresh;
				(void)serial;
				ttl_r = MIN(ttl_r, ttl);
				ttl_r = MIN(ttl_r, minimum);
			} else {
				/* skip over any other type of resource */
				j += datalength;
			}
		}
	}

	if (ttl_r == 0xffffffff)
		ttl_r = 0;

	reply_handle(req, flags, ttl_r, &reply);
	if (reply.data.raw)
		mm_free(reply.data.raw);
	return 0;
 err:
	if (req)
		reply_handle(req, flags, 0, NULL);
	if (reply.data.raw)
		mm_free(reply.data.raw);
	return -1;
}

/* Parse a raw request (packet,length) sent to a nameserver port (port) from */
/* a DNS client (addr,addrlen), and if it's well-formed, call the corresponding */
/* callback. */
static int
request_parse(u8 *packet, int length, struct evdns_server_port *port,
				struct sockaddr *addr, ev_socklen_t addrlen, struct client_tcp_connection *client)
{
	int j = 0;	/* index into packet */
	u16 t_;	 /* used by the macros */
	u32 t32_;  /* used by the macros */
	char tmp_name[256]; /* used by the macros */

	int i;
	u16 trans_id, flags, questions, answers, authority, additional;
	struct server_request *server_req = NULL;
	u32 ttl;
	u16 type, class, rdlen;

	ASSERT_LOCKED(port);

	/* Get the header fields */
	GET16(trans_id);
	GET16(flags);
	GET16(questions);
	GET16(answers);
	GET16(authority);
	GET16(additional);

	if (flags & _QR_MASK) return -1; /* Must not be an answer. */
	flags &= (_RD_MASK|_CD_MASK); /* Only RD and CD get preserved. */

	server_req = mm_malloc(sizeof(struct server_request));
	if (server_req == NULL) return -1;
	memset(server_req, 0, sizeof(struct server_request));

	server_req->trans_id = trans_id;
	if (addr) {
		memcpy(&server_req->addr, addr, addrlen);
		server_req->addrlen = addrlen;
	}

	server_req->port = port;
	server_req->client = client;
	server_req->base.flags = flags;
	server_req->base.nquestions = 0;
	server_req->base.questions = mm_calloc(sizeof(struct evdns_server_question *), questions);
	if (server_req->base.questions == NULL)
		goto err;

	for (i = 0; i < questions; ++i) {
		u16 type, class;
		struct evdns_server_question *q;
		int namelen;
		if (name_parse(packet, length, &j, tmp_name, sizeof(tmp_name))<0)
			goto err;
		GET16(type);
		GET16(class);
		namelen = (int)strlen(tmp_name);
		q = mm_malloc(sizeof(struct evdns_server_question) + namelen);
		if (!q)
			goto err;
		q->type = type;
		q->dns_question_class = class;
		memcpy(q->name, tmp_name, namelen+1);
		server_req->base.questions[server_req->base.nquestions++] = q;
	}

#define SKIP_RR \
	do { \
		SKIP_NAME; \
		j += 2 /* type */ + 2 /* class */ + 4 /* ttl */; \
		GET16(rdlen); \
		j += rdlen; \
	} while (0)

	for (i = 0; i < answers; ++i) {
		SKIP_RR;
	}

	for (i = 0; i < authority; ++i) {
		SKIP_RR;
	}

	server_req->max_udp_reply_size = DNS_MAX_UDP_SIZE;
	for (i = 0; i < additional; ++i) {
		SKIP_NAME;
		GET16(type);
		GET16(class);
		GET32(ttl);
		GET16(rdlen);
		(void)ttl;
		j += rdlen;
		if (type == TYPE_OPT) {
			/* In case of OPT pseudo-RR `class` field is treated
			 * as a requestor's UDP payload size. */
			server_req->max_udp_reply_size = MAX(class, DNS_MAX_UDP_SIZE);
			evdns_server_request_add_reply(&(server_req->base),
				EVDNS_ADDITIONAL_SECTION,
				"", /* name */
				TYPE_OPT, /* type */
				DNS_MAX_UDP_SIZE, /* class */
				0, /* ttl */
				0, /* datalen */
				0, /* is_name */
				NULL /* data */
			);
			break;
		}
	}

	port->refcnt++;

	/* Only standard queries are supported. */
	if (flags & _OP_MASK) {
		evdns_server_request_respond(&(server_req->base), DNS_ERR_NOTIMPL);
		return -1;
	}

	port->user_callback(&(server_req->base), port->user_data);

	return 0;
err:
	if (server_req) {
		if (server_req->base.questions) {
			for (i = 0; i < server_req->base.nquestions; ++i)
				mm_free(server_req->base.questions[i]);
			mm_free(server_req->base.questions);
		}
		mm_free(server_req);
	}
	return -1;

#undef SKIP_RR
#undef SKIP_NAME
#undef GET32
#undef GET16
#undef GET8
}

/* Try to choose a strong transaction id which isn't already in flight */
static u16
transaction_id_pick(struct evdns_base *base) {
	ASSERT_LOCKED(base);
	for (;;) {
		u16 trans_id;
		evutil_secure_rng_get_bytes(&trans_id, sizeof(trans_id));

		if (trans_id == 0xffff) continue;
		/* now check to see if that id is already inflight */
		if (request_find_from_trans_id(base, trans_id) == NULL)
			return trans_id;
	}
}

/* choose a namesever to use. This function will try to ignore */
/* nameservers which we think are down and load balance across the rest */
/* by updating the server_head global each time. */
static struct nameserver *
nameserver_pick(struct evdns_base *base) {
	struct nameserver *started_at = base->server_head, *picked;
	ASSERT_LOCKED(base);
	if (!base->server_head) return NULL;

	/* if we don't have any good nameservers then there's no */
	/* point in trying to find one. */
	if (!base->global_good_nameservers) {
		base->server_head = base->server_head->next;
		return base->server_head;
	}

	/* remember that nameservers are in a circular list */
	for (;;) {
		if (base->server_head->state) {
			/* we think this server is currently good */
			picked = base->server_head;
			base->server_head = base->server_head->next;
			return picked;
		}

		base->server_head = base->server_head->next;
		if (base->server_head == started_at) {
			/* all the nameservers seem to be down */
			/* so we just return this one and hope for the */
			/* best */
			EVUTIL_ASSERT(base->global_good_nameservers == 0);
			picked = base->server_head;
			base->server_head = base->server_head->next;
			return picked;
		}
	}
}

/* this is called when a namesever socket is ready for reading */
static void
nameserver_read(struct nameserver *ns) {
	struct sockaddr_storage ss;
	ev_socklen_t addrlen = sizeof(ss);
	char addrbuf[128];
	const size_t max_packet_size = ns->base->global_max_udp_size;
	u8 *packet = mm_malloc(max_packet_size);
	ASSERT_LOCKED(ns->base);

	if (!packet) {
		nameserver_failed(ns, "not enough memory", 0);
		return;
	}

	for (;;) {
		const int r = recvfrom(ns->socket, (void*)packet,
		    max_packet_size, 0,
		    (struct sockaddr*)&ss, &addrlen);
		if (r < 0) {
			int err = evutil_socket_geterror(ns->socket);
			if (EVUTIL_ERR_RW_RETRIABLE(err))
				goto done;
			nameserver_failed(ns,
			    evutil_socket_error_to_string(err), err);
			goto done;
		}
		if (evutil_sockaddr_cmp((struct sockaddr*)&ss,
			(struct sockaddr*)&ns->address, 0)) {
			log(EVDNS_LOG_WARN, "Address mismatch on received "
			    "DNS packet.  Apparent source was %s",
			    evutil_format_sockaddr_port_(
				    (struct sockaddr *)&ss,
				    addrbuf, sizeof(addrbuf)));
			goto done;
		}

		ns->timedout = 0;
		reply_parse(ns->base, packet, r);
	}
done:
	mm_free(packet);
}

/* Read a packet from a DNS client on a server port s, parse it, and */
/* act accordingly. */
static void
server_udp_port_read(struct evdns_server_port *s) {
	u8 packet[1500];
	struct sockaddr_storage addr;
	ev_socklen_t addrlen;
	int r;
	ASSERT_LOCKED(s);

	for (;;) {
		addrlen = sizeof(struct sockaddr_storage);
		r = recvfrom(s->socket, (void*)packet, sizeof(packet), 0,
					 (struct sockaddr*) &addr, &addrlen);
		if (r < 0) {
			int err = evutil_socket_geterror(s->socket);
			if (EVUTIL_ERR_RW_RETRIABLE(err))
				return;
			log(EVDNS_LOG_WARN,
			    "Error %s (%d) while reading request.",
			    evutil_socket_error_to_string(err), err);
			return;
		}
		request_parse(packet, r, s, (struct sockaddr*) &addr, addrlen, NULL);
	}
}

static int
server_send_response(struct evdns_server_port *port, struct server_request *req)
{
	u16 packet_size = 0;
	struct bufferevent *bev = NULL;
	if (req->client) {
		bev = req->client->connection.bev;
		EVUTIL_ASSERT(bev);
		EVUTIL_ASSERT(req->response_len <= 65535);
		packet_size = htons((u16)req->response_len);
		if (bufferevent_write(bev, &packet_size, sizeof(packet_size)))
			goto beferevent_error;
		if (bufferevent_write(bev, (void*)req->response, req->response_len))
			goto beferevent_error;
		return (int)req->response_len;
	} else {
		int r = sendto(port->socket, req->response, (int)req->response_len, 0,
					(struct sockaddr*) &req->addr, (ev_socklen_t)req->addrlen);
		return r;
	}

beferevent_error:
	log(EVDNS_LOG_WARN, "Failed to send reply to request %p for client %p", (void *)req, (void *)req->client);
	/* disconnect if we got bufferevent error */
	evdns_remove_tcp_client(port, req->client);
	return -1;
}

/* Try to write all pending replies on a given DNS server port. */
static void
server_port_flush(struct evdns_server_port *port)
{
	struct server_request *req = port->pending_replies;
	ASSERT_LOCKED(port);
	while (req) {
		int r = server_send_response(port, req);
		if (r < 0) {
			int err = evutil_socket_geterror(port->socket);
			if (EVUTIL_ERR_RW_RETRIABLE(err))
				return;
			log(EVDNS_LOG_WARN, "Error %s (%d) while writing response to port; dropping", evutil_socket_error_to_string(err), err);
		}
		if (server_request_free(req)) {
			/* we released the last reference to req->port. */
			return;
		} else {
			EVUTIL_ASSERT(req != port->pending_replies);
			req = port->pending_replies;
		}
	}

	/* We have no more pending requests; stop listening for 'writeable' events. */
	(void) event_del(&port->event);
	event_assign(&port->event, port->event_base,
				 port->socket, EV_READ | EV_PERSIST,
				 server_port_ready_callback, port);

	if (event_add(&port->event, NULL) < 0) {
		log(EVDNS_LOG_WARN, "Error from libevent when adding event for DNS server.");
		/* ???? Do more? */
	}
}

/* set if we are waiting for the ability to write to this server. */
/* if waiting is true then we ask libevent for EV_WRITE events, otherwise */
/* we stop these events. */
static void
nameserver_write_waiting(struct nameserver *ns, char waiting) {
	ASSERT_LOCKED(ns->base);
	if (ns->write_waiting == waiting) return;

	ns->write_waiting = waiting;
	(void) event_del(&ns->event);
	event_assign(&ns->event, ns->base->event_base,
	    ns->socket, EV_READ | (waiting ? EV_WRITE : 0) | EV_PERSIST,
	    nameserver_ready_callback, ns);
	if (event_add(&ns->event, NULL) < 0) {
		char addrbuf[128];
		log(EVDNS_LOG_WARN, "Error from libevent when adding event for %s",
		    evutil_format_sockaddr_port_(
			    (struct sockaddr *)&ns->address,
			    addrbuf, sizeof(addrbuf)));
		/* ???? Do more? */
	}
}

/* a callback function. Called by libevent when the kernel says that */
/* a nameserver socket is ready for writing or reading */
static void
nameserver_ready_callback(evutil_socket_t fd, short events, void *arg) {
	struct nameserver *ns = (struct nameserver *) arg;
	(void)fd;

	EVDNS_LOCK(ns->base);
	if (events & EV_WRITE) {
		ns->choked = 0;
		if (!evdns_transmit(ns->base)) {
			nameserver_write_waiting(ns, 0);
		}
	}
	if (events & EV_READ) {
		nameserver_read(ns);
	}
	EVDNS_UNLOCK(ns->base);
}

/* a callback function. Called by libevent when the kernel says that */
/* a server socket is ready for writing or reading. */
static void
server_port_ready_callback(evutil_socket_t fd, short events, void *arg) {
	struct evdns_server_port *port = (struct evdns_server_port *) arg;
	(void) fd;

	EVDNS_LOCK(port);
	if (events & EV_WRITE) {
		port->choked = 0;
		server_port_flush(port);
	}
	if (events & EV_READ) {
		server_udp_port_read(port);
	}
	EVDNS_UNLOCK(port);
}

/* This is an inefficient representation; only use it via the dnslabel_table_*
 * functions, so that is can be safely replaced with something smarter later. */
#define MAX_LABELS 128
/* Structures used to implement name compression */
struct dnslabel_entry { char *v; off_t pos; };
struct dnslabel_table {
	int n_labels; /* number of current entries */
	/* map from name to position in message */
	struct dnslabel_entry labels[MAX_LABELS];
};

/* Initialize dnslabel_table. */
static void
dnslabel_table_init(struct dnslabel_table *table)
{
	table->n_labels = 0;
}

/* Free all storage held by table, but not the table itself. */
static void
dnslabel_clear(struct dnslabel_table *table)
{
	int i;
	for (i = 0; i < table->n_labels; ++i)
		mm_free(table->labels[i].v);
	table->n_labels = 0;
}

/* return the position of the label in the current message, or -1 if the label */
/* hasn't been used yet. */
static int
dnslabel_table_get_pos(const struct dnslabel_table *table, const char *label)
{
	int i;
	for (i = 0; i < table->n_labels; ++i) {
		if (!strcmp(label, table->labels[i].v))
			return table->labels[i].pos;
	}
	return -1;
}

/* remember that we've used the label at position pos */
static int
dnslabel_table_add(struct dnslabel_table *table, const char *label, off_t pos)
{
	char *v;
	int p;
	if (table->n_labels == MAX_LABELS)
		return (-1);
	v = mm_strdup(label);
	if (v == NULL)
		return (-1);
	p = table->n_labels++;
	table->labels[p].v = v;
	table->labels[p].pos = pos;

	return (0);
}

/* Converts a string to a length-prefixed set of DNS labels, starting */
/* at buf[j]. name and buf must not overlap. name_len should be the length */
/* of name.	 table is optional, and is used for compression. */
/* */
/* Input: abc.def */
/* Output: <3>abc<3>def<0> */
/* */
/* Returns the first index after the encoded name, or negative on error. */
/*	 -1	 label was > 63 bytes */
/*	 -2	 name too long to fit in buffer. */
/* */
static off_t
dnsname_to_labels(u8 *const buf, size_t buf_len, off_t j,
				  const char *name, const size_t name_len,
				  struct dnslabel_table *table) {
	const char *end = name + name_len;
	int ref = 0;
	u16 t_;

#define APPEND16(x) do {						\
		if (j + 2 > (off_t)buf_len)				\
			goto overflow;					\
		t_ = htons(x);						\
		memcpy(buf + j, &t_, 2);				\
		j += 2;							\
	} while (0)
#define APPEND32(x) do {						\
		if (j + 4 > (off_t)buf_len)				\
			goto overflow;					\
		t32_ = htonl(x);					\
		memcpy(buf + j, &t32_, 4);				\
		j += 4;							\
	} while (0)

	if (name_len > 255) return -2;

	for (;;) {
		const char *const start = name;
		if (table && (ref = dnslabel_table_get_pos(table, name)) >= 0) {
			APPEND16(ref | 0xc000);
			return j;
		}
		name = strchr(name, '.');
		if (!name) {
			const size_t label_len = end - start;
			if (label_len > 63) return -1;
			if ((size_t)(j+label_len+1) > buf_len) return -2;
			if (table) dnslabel_table_add(table, start, j);
			buf[j++] = (ev_uint8_t)label_len;

			memcpy(buf + j, start, label_len);
			j += (int) label_len;
			break;
		} else {
			/* append length of the label. */
			const size_t label_len = name - start;
			if (label_len > 63) return -1;
			if ((size_t)(j+label_len+1) > buf_len) return -2;
			if (table) dnslabel_table_add(table, start, j);
			buf[j++] = (ev_uint8_t)label_len;

			memcpy(buf + j, start, label_len);
			j += (int) label_len;
			/* hop over the '.' */
			name++;
		}
	}

	/* the labels must be terminated by a 0. */
	/* It's possible that the name ended in a . */
	/* in which case the zero is already there */
	if (!j || buf[j-1]) buf[j++] = 0;
	return j;
 overflow:
	return (-2);
}

/* Finds the length of a dns request for a DNS name of the given */
/* length. The actual request may be smaller than the value returned */
/* here */
static size_t
evdns_request_len(const struct evdns_base *base, const size_t name_len)
{
	int addional_section_len = 0;
	if (EDNS_ENABLED(base)) {
		addional_section_len = 1 + /* length of domain name string, always 0 */
			2 + /* space for resource type */
			2 + /* space for UDP payload size */
			4 + /* space for extended RCODE flags */
			2;  /* space for length of RDATA, always 0 */
	}
	return 96 + /* length of the DNS standard header */
		name_len + 2 +
		4 /* space for the resource type */ +
		addional_section_len;
}

/* build a dns request packet into buf. buf should be at least as long */
/* as evdns_request_len told you it should be. */
/* */
/* Returns the amount of space used. Negative on error. */
static int
evdns_request_data_build(const struct evdns_base *base,
	const char *const name, const size_t name_len,
	const u16 trans_id, const u16 type, const u16 class, u8 *const buf,
	size_t buf_len)
{
	off_t j = 0;  /* current offset into buf */
	u16 t_;	 /* used by the macros */
	u32 t32_;  /* used by the macros */

	APPEND16(trans_id);
	APPEND16(0x0100);  /* standard query, recusion needed */
	APPEND16(1);  /* one question */
	APPEND16(0);  /* no answers */
	APPEND16(0);  /* no authority */
	APPEND16(EDNS_ENABLED(base) ? 1 : 0); /* additional */

	j = dnsname_to_labels(buf, buf_len, j, name, name_len, NULL);
	if (j < 0) {
		return (int)j;
	}

	APPEND16(type);
	APPEND16(class);

	if (EDNS_ENABLED(base)) {
		/* The OPT pseudo-RR format 
		 * (https://tools.ietf.org/html/rfc6891#section-6.1.2)
		 * +------------+--------------+------------------------------+
		 * | Field Name | Field Type   | Description                  |
		 * +------------+--------------+------------------------------+
		 * | NAME       | domain name  | MUST be 0 (root domain)      |
		 * | TYPE       | u_int16_t    | OPT (41)                     |
		 * | CLASS      | u_int16_t    | requestor's UDP payload size |
		 * | TTL        | u_int32_t    | extended RCODE and flags     |
		 * | RDLEN      | u_int16_t    | length of all RDATA          |
		 * | RDATA      | octet stream | {attribute,value} pairs      |
		 * +------------+--------------+------------------------------+ */
		buf[j++] = 0;  /* NAME, always 0 */
		APPEND16(TYPE_OPT);  /* OPT type */
		APPEND16(base->global_max_udp_size);  /* max UDP payload size */
		APPEND32(0);  /* No extended RCODE flags set */
		APPEND16(0);  /* length of RDATA is 0 */
	}

	return (int)j;
 overflow:
	return (-1);
}

/* exported function */
struct evdns_server_port *
evdns_add_server_port_with_base(struct event_base *base, evutil_socket_t socket, int flags, evdns_request_callback_fn_type cb, void *user_data)
{
	struct evdns_server_port *port;
	if (flags)
		return NULL; /* flags not yet implemented */
	if (!(port = mm_malloc(sizeof(struct evdns_server_port))))
		return NULL;
	memset(port, 0, sizeof(struct evdns_server_port));


	port->socket = socket;
	port->refcnt = 1;
	port->choked = 0;
	port->closing = 0;
	port->user_callback = cb;
	port->user_data = user_data;
	port->pending_replies = NULL;
	port->event_base = base;
	port->max_client_connections = MAX_CLIENT_CONNECTIONS;
	port->tcp_idle_timeout.tv_sec = SERVER_IDLE_CONN_TIMEOUT;
	port->tcp_idle_timeout.tv_usec = 0;
	port->client_connections_count = 0;
	LIST_INIT(&port->client_connections);
	event_assign(&port->event, port->event_base,
				 port->socket, EV_READ | EV_PERSIST,
				 server_port_ready_callback, port);
	if (event_add(&port->event, NULL) < 0) {
		mm_free(port);
		return NULL;
	}
	EVTHREAD_ALLOC_LOCK(port->lock, EVTHREAD_LOCKTYPE_RECURSIVE);
	return port;
}

/* exported function */
struct evdns_server_port *
evdns_add_server_port_with_listener(struct event_base *base, struct evconnlistener *listener, int flags, evdns_request_callback_fn_type cb, void *user_data)
{
	struct evdns_server_port *port;
	if (!listener)
		return NULL;
	if (flags)
		return NULL; /* flags not yet implemented */

	if (!(port = mm_calloc(1, sizeof(struct evdns_server_port))))
		return NULL;
	port->socket = -1;
	port->refcnt = 1;
	port->choked = 0;
	port->closing = 0;
	port->user_callback = cb;
	port->user_data = user_data;
	port->pending_replies = NULL;
	port->event_base = base;
	port->max_client_connections = MAX_CLIENT_CONNECTIONS;
	port->client_connections_count = 0;
	LIST_INIT(&port->client_connections);
	port->listener = listener;
	evconnlistener_set_cb(port->listener, incoming_conn_cb, port);

	EVTHREAD_ALLOC_LOCK(port->lock, EVTHREAD_LOCKTYPE_RECURSIVE);
	return port;
}

static void
server_tcp_event_cb(struct bufferevent *bev, short events, void *ctx);

static int
tcp_read_message(struct tcp_connection *conn, u8 **msg, int *msg_len)
{
	struct bufferevent *bev = conn->bev;
	struct evbuffer *input = bufferevent_get_input(bev);
	u8 *packet = NULL;
	int r = 0;

	EVUTIL_ASSERT(conn);
	EVUTIL_ASSERT(conn->state == TS_CONNECTED);

	/* reading new packet size */
	if (!conn->awaiting_packet_size) {
		if (evbuffer_get_length(input) < sizeof(ev_uint16_t))
			goto awaiting_next;

		bufferevent_read(bev, (void*)&conn->awaiting_packet_size,
			sizeof(conn->awaiting_packet_size));
		conn->awaiting_packet_size = ntohs(conn->awaiting_packet_size);
		if (conn->awaiting_packet_size <= 0)
			goto fail;
	}

	/* reading new packet content */
	if (evbuffer_get_length(input) < conn->awaiting_packet_size)
		goto awaiting_next;

	packet = mm_malloc(conn->awaiting_packet_size);
	if (!packet)
		goto fail;

	r = (int)bufferevent_read(bev, (void*)packet, conn->awaiting_packet_size);
	if (r != conn->awaiting_packet_size) {
		mm_free(packet);
		packet = NULL;
		goto fail;
	}

	*msg = packet;
	*msg_len = r;
awaiting_next:
	return 0;
fail:
	return 1;
}

static void
server_tcp_read_packet_cb(struct bufferevent *bev, void *ctx)
{
	u8 *msg = NULL;
	int msg_len = 0;
	int rc;
	struct client_tcp_connection *client = (struct client_tcp_connection *)ctx;
	struct evdns_server_port *port = client->port;
	struct tcp_connection *conn = &client->connection;
	EVUTIL_ASSERT(port && bev);
	EVDNS_LOCK(port);

	while (1) {
		if (tcp_read_message(conn, &msg, &msg_len)) {
			log(EVDNS_LOG_MSG, "Closing client connection %p due to error", (void *)bev);
			evdns_remove_tcp_client(port, client);
			rc = port->refcnt;
			EVDNS_UNLOCK(port);
			if (!rc)
				server_port_free(port);
			return;
		}

		/* Only part of the message was recieved. */
		if (!msg)
			break;

		request_parse(msg, msg_len, port, NULL, 0, client);
		mm_free(msg);
		msg = NULL;
		conn->awaiting_packet_size = 0;
	}

	bufferevent_setwatermark(bev, EV_READ,
			conn->awaiting_packet_size ? conn->awaiting_packet_size : sizeof(ev_uint16_t), 0);
	bufferevent_setcb(bev, server_tcp_read_packet_cb, NULL, server_tcp_event_cb, ctx);
	EVDNS_UNLOCK(port);
}

static void
server_tcp_event_cb(struct bufferevent *bev, short events, void *ctx)
{
	struct client_tcp_connection *client = (struct client_tcp_connection *)ctx;
	struct evdns_server_port *port = client->port;
	int rc;
	EVUTIL_ASSERT(port && bev);
	EVDNS_LOCK(port);
	if (events & (BEV_EVENT_EOF | BEV_EVENT_ERROR | BEV_EVENT_TIMEOUT)) {
		log(EVDNS_LOG_DEBUG, "Closing connection %p", (void *)bev);
		evdns_remove_tcp_client(port, client);
	}
	rc = port->refcnt;
	EVDNS_UNLOCK(port);
	if (!rc)
		server_port_free(port);
}

static void
incoming_conn_cb(struct evconnlistener *listener, evutil_socket_t fd,
				  struct sockaddr *address, int socklen, void *arg)
{
	struct evdns_server_port *port = (struct evdns_server_port*)arg;
	struct bufferevent *bev = bufferevent_socket_new(port->event_base, fd, BEV_OPT_CLOSE_ON_FREE);
	struct client_tcp_connection *client = NULL;
	struct tcp_connection *cd = NULL;

	if (!bev)
		goto error;
	log(EVDNS_LOG_DEBUG, "New incoming client connection %p", (void *)bev);

	bufferevent_set_timeouts(bev, &port->tcp_idle_timeout, &port->tcp_idle_timeout);

	client = evdns_add_tcp_client(port, bev);
	if (!client)
		goto error;
	cd = &client->connection;

	cd->state = TS_CONNECTED;
	bufferevent_setwatermark(bev, EV_READ, sizeof(ev_uint16_t), 0);
	bufferevent_setcb(bev, server_tcp_read_packet_cb, NULL, server_tcp_event_cb, (void *)client);
	bufferevent_enable(bev, EV_READ);

	return;
error:
	if (bev)
		bufferevent_free(bev);
	return;
}

struct evdns_server_port *
evdns_add_server_port(evutil_socket_t socket, int flags, evdns_request_callback_fn_type cb, void *user_data)
{
	return evdns_add_server_port_with_base(NULL, socket, flags, cb, user_data);
}

/* exported function */
void
evdns_close_server_port(struct evdns_server_port *port)
{
	EVDNS_LOCK(port);
	evdns_remove_all_tcp_clients(port);
	if (--port->refcnt == 0) {
		EVDNS_UNLOCK(port);
		server_port_free(port);
	} else {
		port->closing = 1;
		EVDNS_UNLOCK(port);
	}
}

/* exported function */
int
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)
{
	struct server_request *req = TO_SERVER_REQUEST(req_);
	struct server_reply_item **itemp, *item;
	int *countp;
	int result = -1;

	EVDNS_LOCK(req->port);
	if (req->response) /* have we already answered? */
		goto done;

	switch (section) {
	case EVDNS_ANSWER_SECTION:
		itemp = &req->answer;
		countp = &req->n_answer;
		break;
	case EVDNS_AUTHORITY_SECTION:
		itemp = &req->authority;
		countp = &req->n_authority;
		break;
	case EVDNS_ADDITIONAL_SECTION:
		itemp = &req->additional;
		countp = &req->n_additional;
		break;
	default:
		goto done;
	}
	while (*itemp) {
		itemp = &((*itemp)->next);
	}
	item = mm_malloc(sizeof(struct server_reply_item));
	if (!item)
		goto done;
	item->next = NULL;
	if (!(item->name = mm_strdup(name))) {
		mm_free(item);
		goto done;
	}
	item->type = type;
	item->dns_question_class = class;
	item->ttl = ttl;
	item->is_name = is_name != 0;
	item->datalen = 0;
	item->data = NULL;
	if (data) {
		if (item->is_name) {
			if (!(item->data = mm_strdup(data))) {
				mm_free(item->name);
				mm_free(item);
				goto done;
			}
			item->datalen = (u16)-1;
		} else {
			if (!(item->data = mm_malloc(datalen))) {
				mm_free(item->name);
				mm_free(item);
				goto done;
			}
			item->datalen = datalen;
			memcpy(item->data, data, datalen);
		}
	}

	*itemp = item;
	++(*countp);
	result = 0;
done:
	EVDNS_UNLOCK(req->port);
	return result;
}

/* exported function */
int
evdns_server_request_add_a_reply(struct evdns_server_request *req, const char *name, int n, const void *addrs, int ttl)
{
	return evdns_server_request_add_reply(
		  req, EVDNS_ANSWER_SECTION, name, TYPE_A, CLASS_INET,
		  ttl, n*4, 0, addrs);
}

/* exported function */
int
evdns_server_request_add_aaaa_reply(struct evdns_server_request *req, const char *name, int n, const void *addrs, int ttl)
{
	return evdns_server_request_add_reply(
		  req, EVDNS_ANSWER_SECTION, name, TYPE_AAAA, CLASS_INET,
		  ttl, n*16, 0, addrs);
}

/* exported function */
int
evdns_server_request_add_ptr_reply(struct evdns_server_request *req, struct in_addr *in, const char *inaddr_name, const char *hostname, int ttl)
{
	u32 a;
	char buf[32];
	if (in && inaddr_name)
		return -1;
	else if (!in && !inaddr_name)
		return -1;
	if (in) {
		a = ntohl(in->s_addr);
		evutil_snprintf(buf, sizeof(buf), "%d.%d.%d.%d.in-addr.arpa",
				(int)(u8)((a	)&0xff),
				(int)(u8)((a>>8 )&0xff),
				(int)(u8)((a>>16)&0xff),
				(int)(u8)((a>>24)&0xff));
		inaddr_name = buf;
	}
	return evdns_server_request_add_reply(
		  req, EVDNS_ANSWER_SECTION, inaddr_name, TYPE_PTR, CLASS_INET,
		  ttl, -1, 1, hostname);
}

/* exported function */
int
evdns_server_request_add_cname_reply(struct evdns_server_request *req, const char *name, const char *cname, int ttl)
{
	return evdns_server_request_add_reply(
		  req, EVDNS_ANSWER_SECTION, name, TYPE_CNAME, CLASS_INET,
		  ttl, -1, 1, cname);
}

/* exported function */
void
evdns_server_request_set_flags(struct evdns_server_request *exreq, int flags)
{
	struct server_request *req = TO_SERVER_REQUEST(exreq);
	req->base.flags &= ~(EVDNS_FLAGS_AA|EVDNS_FLAGS_RD);
	req->base.flags |= flags;
}

static int
evdns_server_request_format_response(struct server_request *req, int err)
{
	unsigned char buf[1024 * 64];
	size_t buf_len = sizeof(buf);
	off_t j = 0, r;
	u16 t_;
	u32 t32_;
	int i;
	u16 flags;
	struct dnslabel_table table;

	if (err < 0 || err > 15) return -1;

	/* Set response bit and error code; copy OPCODE and RD fields from
	 * question; copy RA and AA if set by caller. */
	flags = req->base.flags;
	flags |= (_QR_MASK | err);

	dnslabel_table_init(&table);
	APPEND16(req->trans_id);
	APPEND16(flags);
	APPEND16(req->base.nquestions);
	APPEND16(req->n_answer);
	APPEND16(req->n_authority);
	APPEND16(req->n_additional);

	/* Add questions. */
	for (i=0; i < req->base.nquestions; ++i) {
		const char *s = req->base.questions[i]->name;
		j = dnsname_to_labels(buf, buf_len, j, s, strlen(s), &table);
		if (j < 0) {
			dnslabel_clear(&table);
			return (int) j;
		}
		APPEND16(req->base.questions[i]->type);
		APPEND16(req->base.questions[i]->dns_question_class);
	}

	/* Add answer, authority, and additional sections. */
	for (i=0; i<3; ++i) {
		struct server_reply_item *item;
		if (i==0)
			item = req->answer;
		else if (i==1)
			item = req->authority;
		else
			item = req->additional;
		while (item) {
			r = dnsname_to_labels(buf, buf_len, j, item->name, strlen(item->name), &table);
			if (r < 0)
				goto overflow;
			j = r;

			APPEND16(item->type);
			APPEND16(item->dns_question_class);
			APPEND32(item->ttl);
			if (item->is_name) {
				off_t len_idx = j, name_start;
				j += 2;
				name_start = j;
				r = dnsname_to_labels(buf, buf_len, j, item->data, strlen(item->data), &table);
				if (r < 0)
					goto overflow;
				j = r;
				t_ = htons( (short) (j-name_start) );
				memcpy(buf+len_idx, &t_, 2);
			} else {
				APPEND16(item->datalen);
				if (j+item->datalen > (off_t)buf_len)
					goto overflow;
				if (item->data) {
					memcpy(buf+j, item->data, item->datalen);
					j += item->datalen;
				} else {
					EVUTIL_ASSERT(item->datalen == 0);
				}
			}
			item = item->next;
		}
	}

	if (j > req->max_udp_reply_size && !req->client) {
overflow:
		j = req->max_udp_reply_size;
		buf[2] |= 0x02; /* set the truncated bit. */
	}

	req->response_len = j;

	if (!(req->response = mm_malloc(req->response_len))) {
		server_request_free_answers(req);
		dnslabel_clear(&table);
		return (-1);
	}
	memcpy(req->response, buf, req->response_len);
	server_request_free_answers(req);
	dnslabel_clear(&table);
	return (0);
}

/* exported function */
int
evdns_server_request_respond(struct evdns_server_request *req_, int err)
{
	struct server_request *req = TO_SERVER_REQUEST(req_);
	struct evdns_server_port *port = req->port;
	int r = -1;

	EVDNS_LOCK(port);
	if (!req->response) {
		if ((r = evdns_server_request_format_response(req, err))<0)
			goto done;
	}

	r = server_send_response(port, req);
	if (r < 0 && req->client) {
		int sock_err = evutil_socket_geterror(port->socket);
		if (EVUTIL_ERR_RW_RETRIABLE(sock_err))
			goto done;

		if (port->pending_replies) {
			req->prev_pending = port->pending_replies->prev_pending;
			req->next_pending = port->pending_replies;
			req->prev_pending->next_pending =
				req->next_pending->prev_pending = req;
		} else {
			req->prev_pending = req->next_pending = req;
			port->pending_replies = req;
			port->choked = 1;

			(void) event_del(&port->event);
			event_assign(&port->event, port->event_base, port->socket, (port->closing?0:EV_READ) | EV_WRITE | EV_PERSIST, server_port_ready_callback, port);

			if (event_add(&port->event, NULL) < 0) {
				log(EVDNS_LOG_WARN, "Error from libevent when adding event for DNS server");
			}

		}

		r = 1;
		goto done;
	}
	if (server_request_free(req)) {
		r = 0;
		goto done;
	}

	if (port->pending_replies)
		server_port_flush(port);

	r = 0;
done:
	EVDNS_UNLOCK(port);
	return r;
}

/* Free all storage held by RRs in req. */
static void
server_request_free_answers(struct server_request *req)
{
	struct server_reply_item *victim, *next, **list;
	int i;
	for (i = 0; i < 3; ++i) {
		if (i==0)
			list = &req->answer;
		else if (i==1)
			list = &req->authority;
		else
			list = &req->additional;

		victim = *list;
		while (victim) {
			next = victim->next;
			mm_free(victim->name);
			victim->name = NULL;
			if (victim->data) {
				mm_free(victim->data);
				victim->data = NULL;
			}
			mm_free(victim);
			victim = next;
		}
		*list = NULL;
	}
}

/* Free all storage held by req, and remove links to it. */
/* return true iff we just wound up freeing the server_port. */
static int
server_request_free(struct server_request *req)
{
	int i, rc=1, lock=0;
	if (req->base.questions) {
		for (i = 0; i < req->base.nquestions; ++i) {
			mm_free(req->base.questions[i]);
			req->base.questions[i] = NULL;
		}
		mm_free(req->base.questions);
		req->base.questions = NULL;
	}

	if (req->port) {
		EVDNS_LOCK(req->port);
		lock=1;
		if (req->port->pending_replies == req) {
			if (req->next_pending && req->next_pending != req)
				req->port->pending_replies = req->next_pending;
			else
				req->port->pending_replies = NULL;
		}
		rc = --req->port->refcnt;
	}

	if (req->response) {
		mm_free(req->response);
		req->response = NULL;
	}

	server_request_free_answers(req);

	if (req->next_pending && req->next_pending != req) {
		req->next_pending->prev_pending = req->prev_pending;
		req->prev_pending->next_pending = req->next_pending;
	}

	if (rc == 0) {
		EVDNS_UNLOCK(req->port); /* ????? nickm */
		server_port_free(req->port);
		mm_free(req);
		return (1);
	}
	if (lock)
		EVDNS_UNLOCK(req->port);
	mm_free(req);
	return (0);
}

/* Free all storage held by an evdns_server_port.  Only called when  */
static void
server_port_free(struct evdns_server_port *port)
{
	EVUTIL_ASSERT(port);
	EVUTIL_ASSERT(!port->refcnt);
	EVUTIL_ASSERT(!port->pending_replies);
	if (port->socket > 0) {
		evutil_closesocket(port->socket);
		port->socket = -1;
	}

	/* if tcp server */
	if (port->listener) {
		evconnlistener_free(port->listener);
	} else {
		(void) event_del(&port->event);
		event_debug_unassign(&port->event);
	}

	EVTHREAD_FREE_LOCK(port->lock, EVTHREAD_LOCKTYPE_RECURSIVE);
	mm_free(port);
}

/* exported function */
int
evdns_server_request_drop(struct evdns_server_request *req_)
{
	struct server_request *req = TO_SERVER_REQUEST(req_);
	server_request_free(req);
	return 0;
}

/* exported function */
int
evdns_server_request_get_requesting_addr(struct evdns_server_request *req_, struct sockaddr *sa, int addr_len)
{
	struct server_request *req = TO_SERVER_REQUEST(req_);
	if (addr_len < (int)req->addrlen)
		return -1;
	memcpy(sa, &(req->addr), req->addrlen);
	return req->addrlen;
}

static void
retransmit_all_tcp_requests_for(struct nameserver *server)
{
	int i = 0;
	for (i = 0; i < server->base->n_req_heads; ++i) {
		struct request *started_at = server->base->req_heads[i];
		struct request *req = started_at;
		if (!req)
			continue;

		do {
			if (req->ns == server && (req->handle->tcp_flags & DNS_QUERY_USEVC)) {
				if (req->tx_count >= req->base->global_max_retransmits) {
					log(EVDNS_LOG_DEBUG, "Giving up on request %p; tx_count==%d",
						(void *)req, req->tx_count);
					reply_schedule_callback(req, 0, DNS_ERR_TIMEOUT, NULL);
					request_finished(req, &REQ_HEAD(req->base, req->trans_id), 1);
				} else {
					(void) evtimer_del(&req->timeout_event);
					evdns_request_transmit(req);
				}
			}
			req = req->next;
		} while (req != started_at);
	}
}

/* this is a libevent callback function which is called when a request */
/* has timed out. */
static void
evdns_request_timeout_callback(evutil_socket_t fd, short events, void *arg) {
	struct request *const req = (struct request *) arg;
	struct evdns_base *base = req->base;

	(void) fd;
	(void) events;

	log(EVDNS_LOG_DEBUG, "Request %p timed out", arg);
	EVDNS_LOCK(base);

	if (req->tx_count >= req->base->global_max_retransmits) {
		struct nameserver *ns = req->ns;
		/* this request has failed */
		log(EVDNS_LOG_DEBUG, "Giving up on request %p; tx_count==%d",
		    arg, req->tx_count);
		reply_schedule_callback(req, 0, DNS_ERR_TIMEOUT, NULL);

		request_finished(req, &REQ_HEAD(req->base, req->trans_id), 1);
		nameserver_failed(ns, "request timed out.", 0);
	} else {
		/* if request is using tcp connection, so tear connection */
		if (req->handle->tcp_flags & DNS_QUERY_USEVC) {
			disconnect_and_free_connection(req->ns->connection);
			req->ns->connection = NULL;

			/* client can have the only connection to DNS server */
			retransmit_all_tcp_requests_for(req->ns);
		} else {
			/* retransmit it */
			log(EVDNS_LOG_DEBUG, "Retransmitting request %p; tx_count==%d by udp", arg, req->tx_count);
			(void) evtimer_del(&req->timeout_event);
			request_swap_ns(req, nameserver_pick(base));
			evdns_request_transmit(req);

			req->ns->timedout++;
			if (req->ns->timedout > req->base->global_max_nameserver_timeout) {
				req->ns->timedout = 0;
				nameserver_failed(req->ns, "request timed out.", 0);
			}
		}
	}

	EVDNS_UNLOCK(base);
}

/* try to send a request to a given server. */
/* */
/* return: */
/*   0 ok */
/*   1 temporary failure */
/*   2 other failure */
static int
evdns_request_transmit_to(struct request *req, struct nameserver *server) {
	int r;
	ASSERT_LOCKED(req->base);
	ASSERT_VALID_REQUEST(req);

	if (server->requests_inflight == 1 &&
		req->base->disable_when_inactive &&
		event_add(&server->event, NULL) < 0) {
		return 1;
	}

	r = sendto(server->socket, (void*)req->request, req->request_len, 0,
	    (struct sockaddr *)&server->address, server->addrlen);
	if (r < 0) {
		int err = evutil_socket_geterror(server->socket);
		if (EVUTIL_ERR_RW_RETRIABLE(err))
			return 1;
		nameserver_failed(req->ns, evutil_socket_error_to_string(err), err);
		return 2;
	} else if (r != (int)req->request_len) {
		return 1;  /* short write */
	} else {
		return 0;
	}
}

/* try to connect to a given server. */
/* */
/* return: */
/*   0 ok */
/*   1 temporary failure */
/*   2 other failure */
static int
evdns_tcp_connect_if_disconnected(struct nameserver *server)
{
	struct tcp_connection *conn = server->connection;
	struct timeval *timeout = &server->base->global_tcp_idle_timeout;
	if (conn && conn->state != TS_DISCONNECTED && conn->bev != NULL)
		return 0;

	disconnect_and_free_connection(conn);
	conn = new_tcp_connecton(bufferevent_socket_new(server->base->event_base, -1, BEV_OPT_CLOSE_ON_FREE));
	if (!conn)
		return 2;
	server->connection = conn;

	if (bufferevent_set_timeouts(conn->bev, timeout, timeout))
		return 1;

	EVUTIL_ASSERT(conn->state == TS_DISCONNECTED);
	if (bufferevent_socket_connect(conn->bev, (struct sockaddr *)&server->address, server->addrlen))
		return 1;

	conn->state = TS_CONNECTING;
	log(EVDNS_LOG_DEBUG, "New tcp connection %p created", (void *)conn);
	return 0;
}

static void
client_tcp_event_cb(struct bufferevent *bev, short events, void *ctx);


static void
client_tcp_read_packet_cb(struct bufferevent *bev, void *ctx)
{
	u8 *msg = NULL;
	int msg_len = 0;
	struct nameserver *server = (struct nameserver*)ctx;
	struct tcp_connection *conn = server->connection;
	EVUTIL_ASSERT(server && bev);
	EVDNS_LOCK(server->base);

	while (1) {
		if (tcp_read_message(conn, &msg, &msg_len)) {
			disconnect_and_free_connection(server->connection);
			server->connection = NULL;
			EVDNS_UNLOCK(server->base);
			return;
		}

		/* Only part of the message was recieved. */
		if (!msg)
			break;

		reply_parse(server->base, msg, msg_len);
		mm_free(msg);
		msg = NULL;
		conn->awaiting_packet_size = 0;
	}

	bufferevent_setwatermark(bev, EV_READ,
		conn->awaiting_packet_size ? conn->awaiting_packet_size : sizeof(ev_uint16_t), 0);
	bufferevent_setcb(bev, client_tcp_read_packet_cb, NULL, client_tcp_event_cb, ctx);
	EVDNS_UNLOCK(server->base);
}

static void
client_tcp_event_cb(struct bufferevent *bev, short events, void *ctx) {
	struct nameserver *server = (struct nameserver*)ctx;
	struct tcp_connection *conn = server->connection;
	EVUTIL_ASSERT(server);
	EVDNS_LOCK(server->base);
	EVUTIL_ASSERT(conn && conn->bev == bev && bev);

	log(EVDNS_LOG_DEBUG, "Event %d on connection %p", events, (void *)conn);

	if (events & (BEV_EVENT_TIMEOUT)) {
		disconnect_and_free_connection(server->connection);
		server->connection = NULL;
	} else if (events & (BEV_EVENT_EOF | BEV_EVENT_ERROR)) {
		disconnect_and_free_connection(server->connection);
		server->connection = NULL;
	} else if (events & BEV_EVENT_CONNECTED) {
		EVUTIL_ASSERT (conn->state == TS_CONNECTING);
		conn->state = TS_CONNECTED;
		evutil_make_socket_nonblocking(bufferevent_getfd(bev));
		bufferevent_setcb(bev, client_tcp_read_packet_cb, NULL, client_tcp_event_cb, server);
		bufferevent_setwatermark(bev, EV_READ, sizeof(ev_uint16_t), 0);
	}
	EVDNS_UNLOCK(server->base);
}

/* try to send a request to a given server. */
/* */
/* return: */
/*   0 ok */
/*   1 temporary failure */
/*   2 other failure */
static int
evdns_request_transmit_through_tcp(struct request *req, struct nameserver *server) {
	uint16_t packet_size;
	struct tcp_connection *conn = NULL;
	int r;
	ASSERT_LOCKED(req->base);
	ASSERT_VALID_REQUEST(req);

	if ((r = evdns_tcp_connect_if_disconnected(server)))
		return r;

	conn = server->connection;
	bufferevent_setcb(conn->bev, client_tcp_read_packet_cb, NULL, client_tcp_event_cb, server);

	log(EVDNS_LOG_DEBUG, "Sending request %p via tcp connection %p", (void *)req, (void *)conn);
	packet_size = htons(req->request_len);
	if (bufferevent_write(conn->bev, &packet_size, sizeof(packet_size)) )
		goto fail;
	if (bufferevent_write(conn->bev, (void*)req->request, req->request_len) )
		goto fail;
	if (bufferevent_enable(conn->bev, EV_READ))
		goto fail;
	if (evtimer_add(&req->timeout_event, &req->base->global_timeout) < 0)
		goto fail;

	return 0;
fail:
	log(EVDNS_LOG_WARN, "Failed to send request %p via tcp connection %p", (void *)req, (void *)conn);
	disconnect_and_free_connection(server->connection);
	server->connection = NULL;
	return 2;
}

/* try to send a request, updating the fields of the request */
/* as needed */
/* */
/* return: */
/*   0 ok */
/*   1 failed */
static int
evdns_request_transmit(struct request *req) {
	int retcode = 0, r;

	ASSERT_LOCKED(req->base);
	ASSERT_VALID_REQUEST(req);
	/* if we fail to send this packet then this flag marks it */
	/* for evdns_transmit */
	req->transmit_me = 1;
	EVUTIL_ASSERT(req->trans_id != 0xffff);

	if (!req->ns)
	{
		/* unable to transmit request if no nameservers */
		return 1;
	}

	if (req->ns->choked) {
		/* don't bother trying to write to a socket */
		/* which we have had EAGAIN from */
		return 1;
	}

	if (req->handle->tcp_flags & DNS_QUERY_USEVC) {
		r = evdns_request_transmit_through_tcp(req, req->ns);
		/*
		If connection didn't initiated now, so report about temporary problems.
		We don't mark name server as chocked so udp packets possibly have no
		problems during transmit. Simply we will retry attempt later */
		if (r == 1) {
			return r;
		}
	} else {
		r = evdns_request_transmit_to(req, req->ns);
	}
	switch (r) {
	case 1:
		/* temp failure */
		req->ns->choked = 1;
		nameserver_write_waiting(req->ns, 1);
		return 1;
	case 2:
		/* failed to transmit the request entirely. we can fallthrough since
		 * we'll set a timeout, which will time out, and make us retransmit the
		 * request anyway. */
		retcode = 1;
		EVUTIL_FALLTHROUGH;
	default:
		/* all ok */
		log(EVDNS_LOG_DEBUG,
		    "Setting timeout for request %p, sent to nameserver %p", (void *)req, (void *)req->ns);
		if (evtimer_add(&req->timeout_event, &req->base->global_timeout) < 0) {
			log(EVDNS_LOG_WARN,
		      "Error from libevent when adding timer for request %p",
			    (void *)req);
			/* ???? Do more? */
		}
		req->tx_count++;
		req->transmit_me = 0;
		return retcode;
	}
}

static void
nameserver_probe_callback(int result, char type, int count, int ttl, void *addresses, void *arg) {
	struct nameserver *const ns = (struct nameserver *) arg;
	(void) type;
	(void) count;
	(void) ttl;
	(void) addresses;

	if (result == DNS_ERR_CANCEL) {
		/* We canceled this request because the nameserver came up
		 * for some other reason.  Do not change our opinion about
		 * the nameserver. */
		return;
	}

	EVDNS_LOCK(ns->base);
	ns->probe_request = NULL;
	if (result == DNS_ERR_NONE || result == DNS_ERR_NOTEXIST) {
		/* this is a good reply */
		nameserver_up(ns);
	} else {
		nameserver_probe_failed(ns);
	}
	EVDNS_UNLOCK(ns->base);
}

static void
nameserver_send_probe(struct nameserver *const ns) {
	struct evdns_request *handle;
	struct request *req;
	char addrbuf[128];
	/* here we need to send a probe to a given nameserver */
	/* in the hope that it is up now. */

	ASSERT_LOCKED(ns->base);
	log(EVDNS_LOG_DEBUG, "Sending probe to %s",
	    evutil_format_sockaddr_port_(
		    (struct sockaddr *)&ns->address,
		    addrbuf, sizeof(addrbuf)));
	handle = mm_calloc(1, sizeof(*handle));
	if (!handle) return;
	req = request_new(ns->base, handle, TYPE_A, "google.com", DNS_QUERY_NO_SEARCH, nameserver_probe_callback, ns);
	if (!req) {
		mm_free(handle);
		return;
	}
	ns->probe_request = handle;
	/* we force this into the inflight queue no matter what */
	request_trans_id_set(req, transaction_id_pick(ns->base));
	req->ns = ns;
	request_submit(req);
}

/* returns: */
/*   0 didn't try to transmit anything */
/*   1 tried to transmit something */
static int
evdns_transmit(struct evdns_base *base) {
	char did_try_to_transmit = 0;
	int i;

	ASSERT_LOCKED(base);
	for (i = 0; i < base->n_req_heads; ++i) {
		if (base->req_heads[i]) {
			struct request *const started_at = base->req_heads[i], *req = started_at;
			/* first transmit all the requests which are currently waiting */
			do {
				if (req->transmit_me) {
					did_try_to_transmit = 1;
					evdns_request_transmit(req);
				}

				req = req->next;
			} while (req != started_at);
		}
	}

	return did_try_to_transmit;
}

/* exported function */
int
evdns_base_count_nameservers(struct evdns_base *base)
{
	const struct nameserver *server;
	int n = 0;

	EVDNS_LOCK(base);
	server = base->server_head;
	if (!server)
		goto done;
	do {
		++n;
		server = server->next;
	} while (server != base->server_head);
done:
	EVDNS_UNLOCK(base);
	return n;
}

int
evdns_count_nameservers(void)
{
	return evdns_base_count_nameservers(current_base);
}

/* exported function */
int
evdns_base_clear_nameservers_and_suspend(struct evdns_base *base)
{
	struct nameserver *server, *started_at;
	int i;

	EVDNS_LOCK(base);
	server = base->server_head;
	started_at = base->server_head;
	if (!server) {
		EVDNS_UNLOCK(base);
		return 0;
	}
	while (1) {
		struct nameserver *next = server->next;
		disconnect_and_free_connection(server->connection);
		server->connection = NULL;
		(void) event_del(&server->event);
		if (evtimer_initialized(&server->timeout_event))
			(void) evtimer_del(&server->timeout_event);
		if (server->probe_request) {
			evdns_cancel_request(server->base, server->probe_request);
			server->probe_request = NULL;
		}
		if (server->socket >= 0)
			evutil_closesocket(server->socket);
		mm_free(server);
		if (next == started_at)
			break;
		server = next;
	}
	base->server_head = NULL;
	base->global_good_nameservers = 0;

	for (i = 0; i < base->n_req_heads; ++i) {
		struct request *req, *req_started_at;
		req = req_started_at = base->req_heads[i];
		while (req) {
			struct request *next = req->next;
			req->tx_count = req->reissue_count = 0;
			req->ns = NULL;
			/* ???? What to do about searches? */
			(void) evtimer_del(&req->timeout_event);
			req->trans_id = 0;
			req->transmit_me = 0;

			base->global_requests_waiting++;
			evdns_request_insert(req, &base->req_waiting_head);
			/* We want to insert these suspended elements at the front of
			 * the waiting queue, since they were pending before any of
			 * the waiting entries were added.  This is a circular list,
			 * so we can just shift the start back by one.*/
			base->req_waiting_head = base->req_waiting_head->prev;

			if (next == req_started_at)
				break;
			req = next;
		}
		base->req_heads[i] = NULL;
	}

	base->global_requests_inflight = 0;

	EVDNS_UNLOCK(base);
	return 0;
}

int
evdns_clear_nameservers_and_suspend(void)
{
	return evdns_base_clear_nameservers_and_suspend(current_base);
}


/* exported function */
int
evdns_base_resume(struct evdns_base *base)
{
	EVDNS_LOCK(base);
	evdns_requests_pump_waiting_queue(base);
	EVDNS_UNLOCK(base);

	return 0;
}

int
evdns_resume(void)
{
	return evdns_base_resume(current_base);
}

static int
evdns_nameserver_add_impl_(struct evdns_base *base, const struct sockaddr *address, int addrlen) {
	/* first check to see if we already have this nameserver */

	const struct nameserver *server = base->server_head, *const started_at = base->server_head;
	struct nameserver *ns;
	int err = 0;
	char addrbuf[128];

	ASSERT_LOCKED(base);
	if (server) {
		do {
			if (!evutil_sockaddr_cmp((struct sockaddr*)&server->address, address, 1)) return 3;
			server = server->next;
		} while (server != started_at);
	}
	if (addrlen > (int)sizeof(ns->address)) {
		log(EVDNS_LOG_DEBUG, "Addrlen %d too long.", (int)addrlen);
		return 2;
	}

	ns = (struct nameserver *) mm_malloc(sizeof(struct nameserver));
	if (!ns) return -1;

	memset(ns, 0, sizeof(struct nameserver));
	ns->base = base;

	evtimer_assign(&ns->timeout_event, ns->base->event_base, nameserver_prod_callback, ns);

	ns->socket = evutil_socket_(address->sa_family,
	    SOCK_DGRAM|EVUTIL_SOCK_NONBLOCK|EVUTIL_SOCK_CLOEXEC, 0);
	if (ns->socket < 0) { err = 1; goto out1; }

	if (base->global_outgoing_addrlen &&
	    !evutil_sockaddr_is_loopback_(address)) {
		if (bind(ns->socket,
			(struct sockaddr*)&base->global_outgoing_address,
			base->global_outgoing_addrlen) < 0) {
			log(EVDNS_LOG_WARN,"Couldn't bind to outgoing address");
			err = 2;
			goto out2;
		}
	}

	if (base->so_rcvbuf) {
		if (setsockopt(ns->socket, SOL_SOCKET, SO_RCVBUF,
		    (void *)&base->so_rcvbuf, sizeof(base->so_rcvbuf))) {
			log(EVDNS_LOG_WARN, "Couldn't set SO_RCVBUF to %i", base->so_rcvbuf);
			err = -SO_RCVBUF;
			goto out2;
		}
	}
	if (base->so_sndbuf) {
		if (setsockopt(ns->socket, SOL_SOCKET, SO_SNDBUF,
		    (void *)&base->so_sndbuf, sizeof(base->so_sndbuf))) {
			log(EVDNS_LOG_WARN, "Couldn't set SO_SNDBUF to %i", base->so_sndbuf);
			err = -SO_SNDBUF;
			goto out2;
		}
	}

	memcpy(&ns->address, address, addrlen);
	ns->addrlen = addrlen;
	ns->state = 1;
	ns->connection = NULL;
	event_assign(&ns->event, ns->base->event_base, ns->socket,
				 EV_READ | EV_PERSIST, nameserver_ready_callback, ns);
	if (!base->disable_when_inactive && event_add(&ns->event, NULL) < 0) {
		err = 2;
		goto out2;
	}

	log(EVDNS_LOG_DEBUG, "Added nameserver %s as %p",
	    evutil_format_sockaddr_port_(address, addrbuf, sizeof(addrbuf)), (void *)ns);

	/* insert this nameserver into the list of them */
	if (!base->server_head) {
		ns->next = ns->prev = ns;
		base->server_head = ns;
	} else {
		ns->next = base->server_head->next;
		ns->prev = base->server_head;
		base->server_head->next = ns;
		ns->next->prev = ns;
	}

	base->global_good_nameservers++;

	return 0;

out2:
	evutil_closesocket(ns->socket);
out1:
	event_debug_unassign(&ns->event);
	mm_free(ns);
	log(EVDNS_LOG_WARN, "Unable to add nameserver %s: error %d",
	    evutil_format_sockaddr_port_(address, addrbuf, sizeof(addrbuf)), err);
	return err;
}

/* exported function */
int
evdns_base_nameserver_add(struct evdns_base *base, unsigned long int address)
{
	struct sockaddr_in sin;
	int res;
	memset(&sin, 0, sizeof(sin));
	sin.sin_addr.s_addr = address;
	sin.sin_port = htons(53);
	sin.sin_family = AF_INET;
#ifdef EVENT__HAVE_STRUCT_SOCKADDR_IN_SIN_LEN
	sin.sin_len = sizeof(sin);
#endif
	EVDNS_LOCK(base);
	res = evdns_nameserver_add_impl_(base, (struct sockaddr*)&sin, sizeof(sin));
	EVDNS_UNLOCK(base);
	return res;
}

int
evdns_nameserver_add(unsigned long int address) {
	if (!current_base)
		current_base = evdns_base_new(NULL, 0);
	return evdns_base_nameserver_add(current_base, address);
}

static void
sockaddr_setport(struct sockaddr *sa, ev_uint16_t port)
{
	if (sa->sa_family == AF_INET) {
		((struct sockaddr_in *)sa)->sin_port = htons(port);
	} else if (sa->sa_family == AF_INET6) {
		((struct sockaddr_in6 *)sa)->sin6_port = htons(port);
	}
}

static ev_uint16_t
sockaddr_getport(struct sockaddr *sa)
{
	if (sa->sa_family == AF_INET) {
		return ntohs(((struct sockaddr_in *)sa)->sin_port);
	} else if (sa->sa_family == AF_INET6) {
		return ntohs(((struct sockaddr_in6 *)sa)->sin6_port);
	} else {
		return 0;
	}
}

/* exported function */
int
evdns_base_nameserver_ip_add(struct evdns_base *base, const char *ip_as_string) {
	struct sockaddr_storage ss;
	struct sockaddr *sa;
	int len = sizeof(ss);
	int res;
	if (evutil_parse_sockaddr_port(ip_as_string, (struct sockaddr *)&ss,
		&len)) {
		log(EVDNS_LOG_WARN, "Unable to parse nameserver address %s",
			ip_as_string);
		return 4;
	}
	sa = (struct sockaddr *) &ss;
	if (sockaddr_getport(sa) == 0)
		sockaddr_setport(sa, 53);

	EVDNS_LOCK(base);
	res = evdns_nameserver_add_impl_(base, sa, len);
	EVDNS_UNLOCK(base);
	return res;
}

int
evdns_nameserver_ip_add(const char *ip_as_string) {
	if (!current_base)
		current_base = evdns_base_new(NULL, 0);
	return evdns_base_nameserver_ip_add(current_base, ip_as_string);
}

int
evdns_base_nameserver_sockaddr_add(struct evdns_base *base,
    const struct sockaddr *sa, ev_socklen_t len, unsigned flags)
{
	int res;
	EVUTIL_ASSERT(base);
	EVDNS_LOCK(base);
	res = evdns_nameserver_add_impl_(base, sa, len);
	EVDNS_UNLOCK(base);
	return res;
}

int
evdns_base_get_nameserver_addr(struct evdns_base *base, int idx,
    struct sockaddr *sa, ev_socklen_t len)
{
	int result = -1;
	int i;
	struct nameserver *server;
	EVDNS_LOCK(base);
	server = base->server_head;
	for (i = 0; i < idx && server; ++i, server = server->next) {
		if (server->next == base->server_head)
			goto done;
	}
	if (! server)
		goto done;

	if (server->addrlen > len) {
		result = (int) server->addrlen;
		goto done;
	}

	memcpy(sa, &server->address, server->addrlen);
	result = (int) server->addrlen;
done:
	EVDNS_UNLOCK(base);
	return result;
}

int
evdns_base_get_nameserver_fd(struct evdns_base *base, int idx)
{
	int result = -1;
	int i;
	struct nameserver *server;
	EVDNS_LOCK(base);
	server = base->server_head;
	for (i = 0; i < idx && server; ++i, server = server->next) {
		if (server->next == base->server_head)
			goto done;
	}
	if (! server)
		goto done;
	result = server->socket;
done:
	EVDNS_UNLOCK(base);
	return result;
}


/* remove from the queue */
static void
evdns_request_remove(struct request *req, struct request **head)
{
	ASSERT_LOCKED(req->base);
	ASSERT_VALID_REQUEST(req);

#if 0
	{
		struct request *ptr;
		int found = 0;
		EVUTIL_ASSERT(*head != NULL);

		ptr = *head;
		do {
			if (ptr == req) {
				found = 1;
				break;
			}
			ptr = ptr->next;
		} while (ptr != *head);
		EVUTIL_ASSERT(found);

		EVUTIL_ASSERT(req->next);
	}
#endif

	if (req->next == req) {
		/* only item in the list */
		*head = NULL;
	} else {
		req->next->prev = req->prev;
		req->prev->next = req->next;
		if (*head == req) *head = req->next;
	}
	req->next = req->prev = NULL;
}

/* insert into the tail of the queue */
static void
evdns_request_insert(struct request *req, struct request **head) {
	ASSERT_LOCKED(req->base);
	ASSERT_VALID_REQUEST(req);
	if (!*head) {
		*head = req;
		req->next = req->prev = req;
		return;
	}

	req->prev = (*head)->prev;
	req->prev->next = req;
	req->next = *head;
	(*head)->prev = req;
}

static int
string_num_dots(const char *s) {
	int count = 0;
	while ((s = strchr(s, '.'))) {
		s++;
		count++;
	}
	return count;
}

static struct request *
request_new(struct evdns_base *base, struct evdns_request *handle, int type,
	    const char *name, int flags, evdns_callback_type callback,
	    void *user_ptr) {

	const char issuing_now =
	    (base->global_requests_inflight < base->global_max_requests_inflight) ? 1 : 0;

	const size_t name_len = strlen(name);
	const size_t request_max_len = evdns_request_len(base, name_len);
	const u16 trans_id = issuing_now ? transaction_id_pick(base) : 0xffff;
	/* the request data is alloced in a single block with the header */
	struct request *const req =
	    mm_malloc(sizeof(struct request) + request_max_len);
	int rlen;
	char namebuf[256];
	(void) flags;

	ASSERT_LOCKED(base);

	if (!req) return NULL;

	if (name_len >= sizeof(namebuf)) {
		mm_free(req);
		return NULL;
	}

	memset(req, 0, sizeof(struct request));
	req->request_size = (u16)(sizeof(struct request) + request_max_len);
	req->base = base;

	evtimer_assign(&req->timeout_event, req->base->event_base, evdns_request_timeout_callback, req);

	if (base->global_randomize_case) {
		unsigned i;
		char randbits[(sizeof(namebuf)+7)/8];
		strlcpy(namebuf, name, sizeof(namebuf));
		evutil_secure_rng_get_bytes(randbits, (name_len+7)/8);
		for (i = 0; i < name_len; ++i) {
			if (EVUTIL_ISALPHA_(namebuf[i])) {
				if ((randbits[i >> 3] & (1<<(i & 7))))
					namebuf[i] |= 0x20;
				else
					namebuf[i] &= ~0x20;
			}
		}
		name = namebuf;
	}

	/* request data lives just after the header */
	req->request = ((u8 *) req) + sizeof(struct request);
	/* denotes that the request data shouldn't be free()ed */
	req->request_appended = 1;
	rlen = evdns_request_data_build(base, name, name_len, trans_id,
	    type, CLASS_INET, req->request, request_max_len);
	if (rlen < 0)
		goto err1;

	req->request_len = rlen;
	req->trans_id = trans_id;
	req->tx_count = 0;
	req->request_type = type;
	req->user_pointer = user_ptr;
	req->user_callback = callback;
	req->ns = issuing_now ? nameserver_pick(base) : NULL;
	req->next = req->prev = NULL;
	req->handle = handle;
	if (handle) {
		handle->current_req = req;
		handle->base = base;
	}

	if (flags & DNS_CNAME_CALLBACK)
		req->need_cname = 1;

	return req;
err1:
	mm_free(req);
	return NULL;
}

static struct request *
request_clone(struct evdns_base *base, struct request* current)
{
	const char issuing_now =
	    (base->global_requests_inflight < base->global_max_requests_inflight) ? 1 : 0;
	const u16 trans_id = issuing_now ? transaction_id_pick(base) : 0xffff;
	/* the request data is alloced in a single block with the header */
	struct request *const req = mm_malloc(current->request_size);
	EVUTIL_ASSERT(current && base);
	ASSERT_LOCKED(base);

	if (!req)
		return NULL;
	memcpy(req, current, current->request_size);

	evtimer_assign(&req->timeout_event, req->base->event_base, evdns_request_timeout_callback, req);

	/* request data lives just after the header */
	req->request = ((u8 *) req) + sizeof(struct request);
	/* We need to replace transact id */
	request_trans_id_set(req, trans_id);

	req->tx_count = 0;
	req->ns = issuing_now ? nameserver_pick(base) : NULL;
	req->next = req->prev = NULL;
	req->handle = NULL;
	log(EVDNS_LOG_DEBUG, "Clone new request TID %d from TID %d", req->trans_id, current->trans_id);

	return req;
}

static void
request_submit(struct request *const req) {
	struct evdns_base *base = req->base;
	ASSERT_LOCKED(base);
	ASSERT_VALID_REQUEST(req);
	if (req->ns) {
		/* if it has a nameserver assigned then this is going */
		/* straight into the inflight queue */
		evdns_request_insert(req, &REQ_HEAD(base, req->trans_id));

		base->global_requests_inflight++;
		req->ns->requests_inflight++;

		evdns_request_transmit(req);
	} else {
		evdns_request_insert(req, &base->req_waiting_head);
		base->global_requests_waiting++;
	}
}

/* exported function */
void
evdns_cancel_request(struct evdns_base *base, struct evdns_request *handle)
{
	struct request *req;

	if (!handle->current_req)
		return;

	if (!base) {
		/* This redundancy is silly; can we fix it? (Not for 2.0) XXXX */
		base = handle->base;
		if (!base)
			base = handle->current_req->base;
	}

	EVDNS_LOCK(base);
	if (handle->pending_cb) {
		EVDNS_UNLOCK(base);
		return;
	}

	req = handle->current_req;
	ASSERT_VALID_REQUEST(req);

	reply_schedule_callback(req, 0, DNS_ERR_CANCEL, NULL);
	if (req->ns) {
		/* remove from inflight queue */
		request_finished(req, &REQ_HEAD(base, req->trans_id), 1);
	} else {
		/* remove from global_waiting head */
		request_finished(req, &base->req_waiting_head, 1);
	}
	EVDNS_UNLOCK(base);
}

/* exported function */
struct evdns_request *
evdns_base_resolve_ipv4(struct evdns_base *base, const char *name, int flags,
    evdns_callback_type callback, void *ptr) {
	struct evdns_request *handle;
	struct request *req;
	log(EVDNS_LOG_DEBUG, "Resolve requested for %s", name);
	handle = mm_calloc(1, sizeof(*handle));
	if (handle == NULL)
		return NULL;
	EVDNS_LOCK(base);
	handle->tcp_flags = base->global_tcp_flags;
	handle->tcp_flags |= flags & (DNS_QUERY_USEVC | DNS_QUERY_IGNTC);
	if (flags & DNS_QUERY_NO_SEARCH) {
		req =
			request_new(base, handle, TYPE_A, name, flags,
				    callback, ptr);
		if (req)
			request_submit(req);
	} else {
		search_request_new(base, handle, TYPE_A, name, flags,
		    callback, ptr);
	}
	if (handle->current_req == NULL) {
		mm_free(handle);
		handle = NULL;
	}
	EVDNS_UNLOCK(base);
	return handle;
}

int evdns_resolve_ipv4(const char *name, int flags,
					   evdns_callback_type callback, void *ptr)
{
	return evdns_base_resolve_ipv4(current_base, name, flags, callback, ptr)
		? 0 : -1;
}


/* exported function */
struct evdns_request *
evdns_base_resolve_ipv6(struct evdns_base *base,
    const char *name, int flags,
    evdns_callback_type callback, void *ptr)
{
	struct evdns_request *handle;
	struct request *req;
	log(EVDNS_LOG_DEBUG, "Resolve requested for %s", name);
	handle = mm_calloc(1, sizeof(*handle));
	if (handle == NULL)
		return NULL;
	EVDNS_LOCK(base);
	handle->tcp_flags = base->global_tcp_flags;
	handle->tcp_flags |= flags & (DNS_QUERY_USEVC | DNS_QUERY_IGNTC);
	if (flags & DNS_QUERY_NO_SEARCH) {
		req = request_new(base, handle, TYPE_AAAA, name, flags,
				  callback, ptr);
		if (req)
			request_submit(req);
	} else {
		search_request_new(base, handle, TYPE_AAAA, name, flags,
		    callback, ptr);
	}
	if (handle->current_req == NULL) {
		mm_free(handle);
		handle = NULL;
	}
	EVDNS_UNLOCK(base);
	return handle;
}

int evdns_resolve_ipv6(const char *name, int flags,
    evdns_callback_type callback, void *ptr) {
	return evdns_base_resolve_ipv6(current_base, name, flags, callback, ptr)
		? 0 : -1;
}

struct evdns_request *
evdns_base_resolve_reverse(struct evdns_base *base, const struct in_addr *in, int flags, evdns_callback_type callback, void *ptr) {
	char buf[32];
	struct evdns_request *handle;
	struct request *req;
	u32 a;
	EVUTIL_ASSERT(in);
	a = ntohl(in->s_addr);
	evutil_snprintf(buf, sizeof(buf), "%d.%d.%d.%d.in-addr.arpa",
			(int)(u8)((a	)&0xff),
			(int)(u8)((a>>8 )&0xff),
			(int)(u8)((a>>16)&0xff),
			(int)(u8)((a>>24)&0xff));
	handle = mm_calloc(1, sizeof(*handle));
	if (handle == NULL)
		return NULL;
	log(EVDNS_LOG_DEBUG, "Resolve requested for %s (reverse)", buf);
	EVDNS_LOCK(base);
	handle->tcp_flags = base->global_tcp_flags;
	handle->tcp_flags |= flags & (DNS_QUERY_USEVC | DNS_QUERY_IGNTC);
	req = request_new(base, handle, TYPE_PTR, buf, flags, callback, ptr);
	if (req)
		request_submit(req);
	if (handle->current_req == NULL) {
		mm_free(handle);
		handle = NULL;
	}
	EVDNS_UNLOCK(base);
	return (handle);
}

int evdns_resolve_reverse(const struct in_addr *in, int flags, evdns_callback_type callback, void *ptr) {
	return evdns_base_resolve_reverse(current_base, in, flags, callback, ptr)
		? 0 : -1;
}

struct evdns_request *
evdns_base_resolve_reverse_ipv6(struct evdns_base *base, const struct in6_addr *in, int flags, evdns_callback_type callback, void *ptr) {
	/* 32 nybbles, 32 periods, "ip6.arpa", NUL. */
	char buf[73];
	char *cp;
	struct evdns_request *handle;
	struct request *req;
	int i;
	EVUTIL_ASSERT(in);
	cp = buf;
	for (i=15; i >= 0; --i) {
		u8 byte = in->s6_addr[i];
		*cp++ = "0123456789abcdef"[byte & 0x0f];
		*cp++ = '.';
		*cp++ = "0123456789abcdef"[byte >> 4];
		*cp++ = '.';
	}
	EVUTIL_ASSERT(cp + strlen("ip6.arpa") < buf+sizeof(buf));
	memcpy(cp, "ip6.arpa", strlen("ip6.arpa")+1);
	handle = mm_calloc(1, sizeof(*handle));
	if (handle == NULL)
		return NULL;
	log(EVDNS_LOG_DEBUG, "Resolve requested for %s (reverse)", buf);
	EVDNS_LOCK(base);
	handle->tcp_flags = base->global_tcp_flags;
	handle->tcp_flags |= flags & (DNS_QUERY_USEVC | DNS_QUERY_IGNTC);
	req = request_new(base, handle, TYPE_PTR, buf, flags, callback, ptr);
	if (req)
		request_submit(req);
	if (handle->current_req == NULL) {
		mm_free(handle);
		handle = NULL;
	}
	EVDNS_UNLOCK(base);
	return (handle);
}

int evdns_resolve_reverse_ipv6(const struct in6_addr *in, int flags, evdns_callback_type callback, void *ptr) {
	return evdns_base_resolve_reverse_ipv6(current_base, in, flags, callback, ptr)
		? 0 : -1;
}

/* ================================================================= */
/* Search support */
/* */
/* the libc resolver has support for searching a number of domains */
/* to find a name. If nothing else then it takes the single domain */
/* from the gethostname() call. */
/* */
/* It can also be configured via the domain and search options in a */
/* resolv.conf. */
/* */
/* The ndots option controls how many dots it takes for the resolver */
/* to decide that a name is non-local and so try a raw lookup first. */

struct search_domain {
	int len;
	struct search_domain *next;
	/* the text string is appended to this structure */
};

struct search_state {
	int refcount;
	int ndots;
	int num_domains;
	struct search_domain *head;
};

static void
search_state_decref(struct search_state *const state) {
	if (!state) return;
	state->refcount--;
	if (!state->refcount) {
		struct search_domain *next, *dom;
		for (dom = state->head; dom; dom = next) {
			next = dom->next;
			mm_free(dom);
		}
		mm_free(state);
	}
}

static struct search_state *
search_state_new(void) {
	struct search_state *state = (struct search_state *) mm_malloc(sizeof(struct search_state));
	if (!state) return NULL;
	memset(state, 0, sizeof(struct search_state));
	state->refcount = 1;
	state->ndots = 1;

	return state;
}

static void
search_postfix_clear(struct evdns_base *base) {
	search_state_decref(base->global_search_state);

	base->global_search_state = search_state_new();
}

/* exported function */
void
evdns_base_search_clear(struct evdns_base *base)
{
	EVDNS_LOCK(base);
	search_postfix_clear(base);
	EVDNS_UNLOCK(base);
}

void
evdns_search_clear(void) {
	evdns_base_search_clear(current_base);
}

static void
search_postfix_add(struct evdns_base *base, const char *domain) {
	size_t domain_len;
	struct search_domain *sdomain;
	while (domain[0] == '.') domain++;
	domain_len = strlen(domain);

	ASSERT_LOCKED(base);
	if (!base->global_search_state) base->global_search_state = search_state_new();
	if (!base->global_search_state) return;
	base->global_search_state->num_domains++;

	sdomain = (struct search_domain *) mm_malloc(sizeof(struct search_domain) + domain_len);
	if (!sdomain) return;
	memcpy( ((u8 *) sdomain) + sizeof(struct search_domain), domain, domain_len);
	sdomain->next = base->global_search_state->head;
	sdomain->len = (int) domain_len;

	base->global_search_state->head = sdomain;
}

/* reverse the order of members in the postfix list. This is needed because, */
/* when parsing resolv.conf we push elements in the wrong order */
static void
search_reverse(struct evdns_base *base) {
	struct search_domain *cur, *prev = NULL, *next;
	ASSERT_LOCKED(base);
	cur = base->global_search_state->head;
	while (cur) {
		next = cur->next;
		cur->next = prev;
		prev = cur;
		cur = next;
	}

	base->global_search_state->head = prev;
}

/* exported function */
void
evdns_base_search_add(struct evdns_base *base, const char *domain) {
	EVDNS_LOCK(base);
	search_postfix_add(base, domain);
	EVDNS_UNLOCK(base);
}
void
evdns_search_add(const char *domain) {
	evdns_base_search_add(current_base, domain);
}

/* exported function */
void
evdns_base_search_ndots_set(struct evdns_base *base, const int ndots) {
	EVDNS_LOCK(base);
	if (!base->global_search_state) base->global_search_state = search_state_new();
	if (base->global_search_state)
		base->global_search_state->ndots = ndots;
	EVDNS_UNLOCK(base);
}
void
evdns_search_ndots_set(const int ndots) {
	evdns_base_search_ndots_set(current_base, ndots);
}

static void
search_set_from_hostname(struct evdns_base *base) {
	char hostname[EVDNS_NAME_MAX + 1], *domainname;

	ASSERT_LOCKED(base);
	search_postfix_clear(base);
	if (gethostname(hostname, sizeof(hostname))) return;
	domainname = strchr(hostname, '.');
	if (!domainname) return;
	search_postfix_add(base, domainname);
}

/* warning: returns malloced string */
static char *
search_make_new(const struct search_state *const state, int n, const char *const base_name) {
	const size_t base_len = strlen(base_name);
	char need_to_append_dot;
	struct search_domain *dom;

	if (!base_len) return NULL;
	need_to_append_dot = base_name[base_len - 1] == '.' ? 0 : 1;

	for (dom = state->head; dom; dom = dom->next) {
		if (!n--) {
			/* this is the postfix we want */
			/* the actual postfix string is kept at the end of the structure */
			const u8 *const postfix = ((u8 *) dom) + sizeof(struct search_domain);
			const int postfix_len = dom->len;
			char *const newname = (char *) mm_malloc(base_len + need_to_append_dot + postfix_len + 1);
			if (!newname) return NULL;
			memcpy(newname, base_name, base_len);
			if (need_to_append_dot) newname[base_len] = '.';
			memcpy(newname + base_len + need_to_append_dot, postfix, postfix_len);
			newname[base_len + need_to_append_dot + postfix_len] = 0;
			return newname;
		}
	}

	/* we ran off the end of the list and still didn't find the requested string */
	EVUTIL_ASSERT(0);
	return NULL; /* unreachable; stops warnings in some compilers. */
}

static struct request *
search_request_new(struct evdns_base *base, struct evdns_request *handle,
		   int type, const char *const name, int flags,
		   evdns_callback_type user_callback, void *user_arg) {
	ASSERT_LOCKED(base);
	EVUTIL_ASSERT(type == TYPE_A || type == TYPE_AAAA);
	EVUTIL_ASSERT(handle->current_req == NULL);
	if ( ((flags & DNS_QUERY_NO_SEARCH) == 0) &&
	     base->global_search_state &&
		 base->global_search_state->num_domains) {
		/* we have some domains to search */
		struct request *req;
		if (string_num_dots(name) >= base->global_search_state->ndots) {
			req = request_new(base, handle, type, name, flags, user_callback, user_arg);
			if (!req) return NULL;
			handle->search_index = -1;
		} else {
			char *const new_name = search_make_new(base->global_search_state, 0, name);
			if (!new_name) return NULL;
			req = request_new(base, handle, type, new_name, flags, user_callback, user_arg);
			mm_free(new_name);
			if (!req) return NULL;
			handle->search_index = 0;
		}
		EVUTIL_ASSERT(handle->search_origname == NULL);
		handle->search_origname = mm_strdup(name);
		if (handle->search_origname == NULL) {
			/* XXX Should we dealloc req? If yes, how? */
			if (req)
				mm_free(req);
			return NULL;
		}
		handle->search_state = base->global_search_state;
		handle->search_flags = flags;
		base->global_search_state->refcount++;
		request_submit(req);
		return req;
	} else {
		struct request *const req = request_new(base, handle, type, name, flags, user_callback, user_arg);
		if (!req) return NULL;
		request_submit(req);
		return req;
	}
}

/* this is called when a request has failed to find a name. We need to check */
/* if it is part of a search and, if so, try the next name in the list */
/* returns: */
/*   0 another request has been submitted */
/*   1 no more requests needed */
static int
search_try_next(struct evdns_request *const handle) {
	struct request *req = handle->current_req;
	struct evdns_base *base = req->base;
	struct request *newreq;
	ASSERT_LOCKED(base);
	if (handle->search_state) {
		/* it is part of a search */
		char *new_name;
		handle->search_index++;
		if (handle->search_index >= handle->search_state->num_domains) {
			/* no more postfixes to try, however we may need to try */
			/* this name without a postfix */
			if (string_num_dots(handle->search_origname) < handle->search_state->ndots) {
				/* yep, we need to try it raw */
				newreq = request_new(base, NULL, req->request_type, handle->search_origname, handle->search_flags, req->user_callback, req->user_pointer);
				log(EVDNS_LOG_DEBUG, "Search: trying raw query %s", handle->search_origname);
				if (newreq) {
					search_request_finished(handle);
					goto submit_next;
				}
			}
			return 1;
		}

		new_name = search_make_new(handle->search_state, handle->search_index, handle->search_origname);
		if (!new_name) return 1;
		log(EVDNS_LOG_DEBUG, "Search: now trying %s (%d)", new_name, handle->search_index);
		newreq = request_new(base, NULL, req->request_type, new_name, handle->search_flags, req->user_callback, req->user_pointer);
		mm_free(new_name);
		if (!newreq) return 1;
		goto submit_next;
	}
	return 1;

submit_next:
	request_finished(req, &REQ_HEAD(req->base, req->trans_id), 0);
	handle->current_req = newreq;
	newreq->handle = handle;
	request_submit(newreq);
	return 0;
}

static void
search_request_finished(struct evdns_request *const handle) {
	ASSERT_LOCKED(handle->current_req->base);
	if (handle->search_state) {
		search_state_decref(handle->search_state);
		handle->search_state = NULL;
	}
	if (handle->search_origname) {
		mm_free(handle->search_origname);
		handle->search_origname = NULL;
	}
}

/* ================================================================= */
/* Parsing resolv.conf files */

static void
evdns_resolv_set_defaults(struct evdns_base *base, int flags) {
	int add_default = flags & DNS_OPTION_NAMESERVERS;
	if (flags & DNS_OPTION_NAMESERVERS_NO_DEFAULT)
		add_default = 0;

	/* if the file isn't found then we assume a local resolver */
	ASSERT_LOCKED(base);
	if (flags & DNS_OPTION_SEARCH)
		search_set_from_hostname(base);
	if (add_default)
		evdns_base_nameserver_ip_add(base, "127.0.0.1");
}

#ifndef EVENT__HAVE_STRTOK_R
static char *
strtok_r(char *s, const char *delim, char **state) {
	char *cp, *start;
	start = cp = s ? s : *state;
	if (!cp)
		return NULL;
	while (*cp && !strchr(delim, *cp))
		++cp;
	if (!*cp) {
		if (cp == start)
			return NULL;
		*state = NULL;
		return start;
	} else {
		*cp++ = '\0';
		*state = cp;
		return start;
	}
}
#endif

/* helper version of atoi which returns -1 on error */
static int
strtoint(const char *const str)
{
	char *endptr;
	const int r = strtol(str, &endptr, 10);
	if (*endptr) return -1;
	return r;
}

/* Parse a number of seconds into a timeval; return -1 on error. */
static int
evdns_strtotimeval(const char *const str, struct timeval *out)
{
	double d;
	char *endptr;
	d = strtod(str, &endptr);
	if (*endptr) return -1;
	if (d < 0) return -1;
	out->tv_sec = (int) d;
	out->tv_usec = (int) ((d - (int) d)*1000000);
	if (out->tv_sec == 0 && out->tv_usec < 1000) /* less than 1 msec */
		return -1;
	return 0;
}

/* helper version of atoi that returns -1 on error and clips to bounds. */
static int
strtoint_clipped(const char *const str, int min, int max)
{
	int r = strtoint(str);
	if (r == -1)
		return r;
	else if (r<min)
		return min;
	else if (r>max)
		return max;
	else
		return r;
}

static int
evdns_base_set_max_requests_inflight(struct evdns_base *base, int maxinflight)
{
	int old_n_heads = base->n_req_heads, n_heads;
	struct request **old_heads = base->req_heads, **new_heads, *req;
	int i;

	ASSERT_LOCKED(base);
	if (maxinflight < 1)
		maxinflight = 1;
	n_heads = (maxinflight+4) / 5;
	EVUTIL_ASSERT(n_heads > 0);
	new_heads = mm_calloc(n_heads, sizeof(struct request*));
	if (!new_heads)
		return (-1);
	if (old_heads) {
		for (i = 0; i < old_n_heads; ++i) {
			while (old_heads[i]) {
				req = old_heads[i];
				evdns_request_remove(req, &old_heads[i]);
				evdns_request_insert(req, &new_heads[req->trans_id % n_heads]);
			}
		}
		mm_free(old_heads);
	}
	base->req_heads = new_heads;
	base->n_req_heads = n_heads;
	base->global_max_requests_inflight = maxinflight;
	return (0);
}

/* exported function */
int
evdns_base_set_option(struct evdns_base *base,
    const char *option, const char *val)
{
	int res;
	EVDNS_LOCK(base);
	res = evdns_base_set_option_impl(base, option, val, DNS_OPTIONS_ALL);
	EVDNS_UNLOCK(base);
	return res;
}

static inline int
str_matches_option(const char *s1, const char *optionname)
{
	/* Option names are given as "option:" We accept either 'option' in
	 * s1, or 'option:randomjunk'.  The latter form is to implement the
	 * resolv.conf parser. */
	size_t optlen = strlen(optionname);
	size_t slen = strlen(s1);
	if (slen == optlen || slen == optlen - 1)
		return !strncmp(s1, optionname, slen);
	else if (slen > optlen)
		return !strncmp(s1, optionname, optlen);
	else
		return 0;
}

/* exported function */
int
evdns_server_port_set_option(struct evdns_server_port *port,
	enum evdns_server_option option, size_t value)
{
	int res = 0;
	EVDNS_LOCK(port);
	switch (option) {
	case EVDNS_SOPT_TCP_MAX_CLIENTS:
		if (!port->listener) {
			log(EVDNS_LOG_WARN, "EVDNS_SOPT_TCP_MAX_CLIENTS option can be set only on TCP server");
			res = -1;
			goto end;
		}
		port->max_client_connections = value;
		log(EVDNS_LOG_DEBUG, "Setting EVDNS_SOPT_TCP_MAX_CLIENTS to %u", port->max_client_connections);
		break;
	case EVDNS_SOPT_TCP_IDLE_TIMEOUT:
		if (!port->listener) {
			log(EVDNS_LOG_WARN, "EVDNS_SOPT_TCP_IDLE_TIMEOUT option can be set only on TCP server");
			res = -1;
			goto end;
		}
		port->tcp_idle_timeout.tv_sec = value;
		port->tcp_idle_timeout.tv_usec = 0;
		log(EVDNS_LOG_DEBUG, "Setting EVDNS_SOPT_TCP_IDLE_TIMEOUT to %u seconds",
			(unsigned)port->tcp_idle_timeout.tv_sec);
		break;
	default:
		log(EVDNS_LOG_WARN, "Invalid DNS server option %d", (int)option);
		res = -1;
		break;
	}
end:
	EVDNS_UNLOCK(port);
	return res;
}

static int
evdns_base_set_option_impl(struct evdns_base *base,
    const char *option, const char *val, int flags)
{
	ASSERT_LOCKED(base);
	if (str_matches_option(option, "ndots:")) {
		const int ndots = strtoint(val);
		if (ndots == -1) return -1;
		if (!(flags & DNS_OPTION_SEARCH)) return 0;
		log(EVDNS_LOG_DEBUG, "Setting ndots to %d", ndots);
		if (!base->global_search_state) base->global_search_state = search_state_new();
		if (!base->global_search_state) return -1;
		base->global_search_state->ndots = ndots;
	} else if (str_matches_option(option, "timeout:")) {
		struct timeval tv;
		if (evdns_strtotimeval(val, &tv) == -1) return -1;
		if (!(flags & DNS_OPTION_MISC)) return 0;
		log(EVDNS_LOG_DEBUG, "Setting timeout to %s", val);
		memcpy(&base->global_timeout, &tv, sizeof(struct timeval));
	} else if (str_matches_option(option, "getaddrinfo-allow-skew:")) {
		struct timeval tv;
		if (evdns_strtotimeval(val, &tv) == -1) return -1;
		if (!(flags & DNS_OPTION_MISC)) return 0;
		log(EVDNS_LOG_DEBUG, "Setting getaddrinfo-allow-skew to %s",
		    val);
		memcpy(&base->global_getaddrinfo_allow_skew, &tv,
		    sizeof(struct timeval));
	} else if (str_matches_option(option, "max-timeouts:")) {
		const int maxtimeout = strtoint_clipped(val, 1, 255);
		if (maxtimeout == -1) return -1;
		if (!(flags & DNS_OPTION_MISC)) return 0;
		log(EVDNS_LOG_DEBUG, "Setting maximum allowed timeouts to %d",
			maxtimeout);
		base->global_max_nameserver_timeout = maxtimeout;
	} else if (str_matches_option(option, "max-inflight:")) {
		const int maxinflight = strtoint_clipped(val, 1, 65000);
		if (maxinflight == -1) return -1;
		if (!(flags & DNS_OPTION_MISC)) return 0;
		log(EVDNS_LOG_DEBUG, "Setting maximum inflight requests to %d",
			maxinflight);
		evdns_base_set_max_requests_inflight(base, maxinflight);
	} else if (str_matches_option(option, "attempts:")) {
		int retries = strtoint(val);
		if (retries == -1) return -1;
		if (retries > 255) retries = 255;
		if (!(flags & DNS_OPTION_MISC)) return 0;
		log(EVDNS_LOG_DEBUG, "Setting retries to %d", retries);
		base->global_max_retransmits = retries;
	} else if (str_matches_option(option, "randomize-case:")) {
		int randcase = strtoint(val);
		if (randcase == -1) return -1;
		if (!(flags & DNS_OPTION_MISC)) return 0;
		base->global_randomize_case = randcase;
	} else if (str_matches_option(option, "bind-to:")) {
		/* XXX This only applies to successive nameservers, not
		 * to already-configured ones.	We might want to fix that. */
		int len = sizeof(base->global_outgoing_address);
		if (!(flags & DNS_OPTION_NAMESERVERS)) return 0;
		if (evutil_parse_sockaddr_port(val,
			(struct sockaddr*)&base->global_outgoing_address, &len))
			return -1;
		base->global_outgoing_addrlen = len;
	} else if (str_matches_option(option, "initial-probe-timeout:")) {
		struct timeval tv;
		if (evdns_strtotimeval(val, &tv) == -1) return -1;
		if (tv.tv_sec > 3600)
			tv.tv_sec = 3600;
		if (!(flags & DNS_OPTION_MISC)) return 0;
		log(EVDNS_LOG_DEBUG, "Setting initial probe timeout to %s",
		    val);
		memcpy(&base->global_nameserver_probe_initial_timeout, &tv,
		    sizeof(tv));
	} else if (str_matches_option(option, "max-probe-timeout:")) {
		const int max_probe_timeout = strtoint_clipped(val, 1, 3600);
		if (max_probe_timeout == -1) return -1;
		if (!(flags & DNS_OPTION_MISC)) return 0;
		log(EVDNS_LOG_DEBUG, "Setting maximum probe timeout to %d",
			max_probe_timeout);
		base->ns_max_probe_timeout = max_probe_timeout;
		if (base->global_nameserver_probe_initial_timeout.tv_sec > max_probe_timeout) {
			base->global_nameserver_probe_initial_timeout.tv_sec = max_probe_timeout;
			base->global_nameserver_probe_initial_timeout.tv_usec = 0;
			log(EVDNS_LOG_DEBUG, "Setting initial probe timeout to %s",
				val);
		}
	} else if (str_matches_option(option, "probe-backoff-factor:")) {
		const int backoff_backtor = strtoint_clipped(val, 1, 10);
		if (backoff_backtor == -1) return -1;
		if (!(flags & DNS_OPTION_MISC)) return 0;
		log(EVDNS_LOG_DEBUG, "Setting probe timeout backoff factor to %d",
			backoff_backtor);
		base->ns_timeout_backoff_factor = backoff_backtor;
	} else if (str_matches_option(option, "so-rcvbuf:")) {
		int buf = strtoint(val);
		if (buf == -1) return -1;
		if (!(flags & DNS_OPTION_MISC)) return 0;
		log(EVDNS_LOG_DEBUG, "Setting SO_RCVBUF to %s", val);
		base->so_rcvbuf = buf;
	} else if (str_matches_option(option, "so-sndbuf:")) {
		int buf = strtoint(val);
		if (buf == -1) return -1;
		if (!(flags & DNS_OPTION_MISC)) return 0;
		log(EVDNS_LOG_DEBUG, "Setting SO_SNDBUF to %s", val);
		base->so_sndbuf = buf;
	} else if (str_matches_option(option, "tcp-idle-timeout:")) {
		struct timeval tv;
		if (evdns_strtotimeval(val, &tv) == -1) return -1;
		if (!(flags & DNS_OPTION_MISC)) return 0;
		log(EVDNS_LOG_DEBUG, "Setting tcp idle timeout to %s", val);
		memcpy(&base->global_tcp_idle_timeout, &tv, sizeof(tv));
	} else if (str_matches_option(option, "use-vc:")) {
		if (!(flags & DNS_OPTION_MISC)) return 0;
		if (val && strlen(val)) return -1;
		log(EVDNS_LOG_DEBUG, "Setting use-vc option");
		base->global_tcp_flags |= DNS_QUERY_USEVC;
	} else if (str_matches_option(option, "ignore-tc:")) {
		if (!(flags & DNS_OPTION_MISC)) return 0;
		if (val && strlen(val)) return -1;
		log(EVDNS_LOG_DEBUG, "Setting ignore-tc option");
		base->global_tcp_flags |= DNS_QUERY_IGNTC;
	} else if (str_matches_option(option, "edns-udp-size:")) {
		const int sz = strtoint_clipped(val, DNS_MAX_UDP_SIZE, EDNS_MAX_UDP_SIZE);
		if (sz == -1) return -1;
		if (!(flags & DNS_OPTION_MISC)) return 0;
		log(EVDNS_LOG_DEBUG, "Setting edns-udp-size to %d", sz);
		base->global_max_udp_size = sz;
	}
	return 0;
}

int
evdns_set_option(const char *option, const char *val, int flags)
{
	if (!current_base)
		current_base = evdns_base_new(NULL, 0);
	return evdns_base_set_option(current_base, option, val);
}

static void
resolv_conf_parse_line(struct evdns_base *base, char *const start, int flags) {
	char *strtok_state;
	static const char *const delims = " \t";
#define NEXT_TOKEN strtok_r(NULL, delims, &strtok_state)


	char *const first_token = strtok_r(start, delims, &strtok_state);
	ASSERT_LOCKED(base);
	if (!first_token) return;

	if (!strcmp(first_token, "nameserver") && (flags & DNS_OPTION_NAMESERVERS)) {
		const char *const nameserver = NEXT_TOKEN;

		if (nameserver)
			evdns_base_nameserver_ip_add(base, nameserver);
	} else if (!strcmp(first_token, "domain") && (flags & DNS_OPTION_SEARCH)) {
		const char *const domain = NEXT_TOKEN;
		if (domain) {
			search_postfix_clear(base);
			search_postfix_add(base, domain);
		}
	} else if (!strcmp(first_token, "search") && (flags & DNS_OPTION_SEARCH)) {
		const char *domain;
		search_postfix_clear(base);

		while ((domain = NEXT_TOKEN)) {
			search_postfix_add(base, domain);
		}
		search_reverse(base);
	} else if (!strcmp(first_token, "options")) {
		const char *option;
		while ((option = NEXT_TOKEN)) {
			const char *val = strchr(option, ':');
			evdns_base_set_option_impl(base, option, val ? val+1 : "", flags);
		}
	}
#undef NEXT_TOKEN
}

/* exported function */
/* returns: */
/*   0 no errors */
/*   1 failed to open file */
/*   2 failed to stat file */
/*   3 file too large */
/*   4 out of memory */
/*   5 short read from file */
int
evdns_base_resolv_conf_parse(struct evdns_base *base, int flags, const char *const filename) {
	int res;
	EVDNS_LOCK(base);
	res = evdns_base_resolv_conf_parse_impl(base, flags, filename);
	EVDNS_UNLOCK(base);
	return res;
}

static char *
evdns_get_default_hosts_filename(void)
{
#ifdef _WIN32
	/* Windows is a little coy about where it puts its configuration
	 * files.  Sure, they're _usually_ in C:\windows\system32, but
	 * there's no reason in principle they couldn't be in
	 * W:\hoboken chicken emergency\
	 */
	char path[MAX_PATH+1];
	static const char hostfile[] = "\\drivers\\etc\\hosts";
	char *path_out;
	size_t len_out;

	if (! SHGetSpecialFolderPathA(NULL, path, CSIDL_SYSTEM, 0))
		return NULL;
	len_out = strlen(path)+strlen(hostfile)+1;
	path_out = mm_malloc(len_out);
	evutil_snprintf(path_out, len_out, "%s%s", path, hostfile);
	return path_out;
#else
	return mm_strdup("/etc/hosts");
#endif
}

static int
evdns_base_resolv_conf_parse_impl(struct evdns_base *base, int flags, const char *const filename) {
	size_t n;
	char *resolv;
	char *start;
	int err = 0;
	int add_default;

	log(EVDNS_LOG_DEBUG, "Parsing resolv.conf file %s", filename);

	add_default = flags & DNS_OPTION_NAMESERVERS;
	if (flags & DNS_OPTION_NAMESERVERS_NO_DEFAULT)
		add_default = 0;

	if (flags & DNS_OPTION_HOSTSFILE) {
		char *fname = evdns_get_default_hosts_filename();
		evdns_base_load_hosts(base, fname);
		if (fname)
			mm_free(fname);
	}

	if (!filename) {
		evdns_resolv_set_defaults(base, flags);
		return 1;
	}

	if ((err = evutil_read_file_(filename, &resolv, &n, 0)) < 0) {
		if (err == -1) {
			/* No file. */
			evdns_resolv_set_defaults(base, flags);
			return 1;
		} else {
			return 2;
		}
	}

	start = resolv;
	for (;;) {
		char *const newline = strchr(start, '\n');
		if (!newline) {
			resolv_conf_parse_line(base, start, flags);
			break;
		} else {
			*newline = 0;
			resolv_conf_parse_line(base, start, flags);
			start = newline + 1;
		}
	}

	if (!base->server_head && add_default) {
		/* no nameservers were configured. */
		evdns_base_nameserver_ip_add(base, "127.0.0.1");
		err = 6;
	}
	if (flags & DNS_OPTION_SEARCH && (!base->global_search_state || base->global_search_state->num_domains == 0)) {
		search_set_from_hostname(base);
	}

	mm_free(resolv);
	return err;
}

int
evdns_resolv_conf_parse(int flags, const char *const filename) {
	if (!current_base)
		current_base = evdns_base_new(NULL, 0);
	return evdns_base_resolv_conf_parse(current_base, flags, filename);
}


#ifdef _WIN32
/* Add multiple nameservers from a space-or-comma-separated list. */
static int
evdns_nameserver_ip_add_line(struct evdns_base *base, const char *ips) {
	const char *addr;
	char *buf;
	int r;
	ASSERT_LOCKED(base);
	while (*ips) {
		while (isspace(*ips) || *ips == ',' || *ips == '\t')
			++ips;
		addr = ips;
		while (isdigit(*ips) || *ips == '.' || *ips == ':' ||
		    *ips=='[' || *ips==']')
			++ips;
		buf = mm_malloc(ips-addr+1);
		if (!buf) return 4;
		memcpy(buf, addr, ips-addr);
		buf[ips-addr] = '\0';
		r = evdns_base_nameserver_ip_add(base, buf);
		mm_free(buf);
		if (r) return r;
	}
	return 0;
}

typedef DWORD(WINAPI *GetNetworkParams_fn_t)(FIXED_INFO *, DWORD*);

/* Use the windows GetNetworkParams interface in iphlpapi.dll to */
/* figure out what our nameservers are. */
static int
load_nameservers_with_getnetworkparams(struct evdns_base *base)
{
	/* Based on MSDN examples and inspection of  c-ares code. */
	FIXED_INFO *fixed;
	HMODULE handle = 0;
	ULONG size = sizeof(FIXED_INFO);
	void *buf = NULL;
	int status = 0, r, added_any;
	IP_ADDR_STRING *ns;
	GetNetworkParams_fn_t fn;

	ASSERT_LOCKED(base);
	if (!(handle = evutil_load_windows_system_library_(
			TEXT("iphlpapi.dll")))) {
		log(EVDNS_LOG_WARN, "Could not open iphlpapi.dll");
		status = -1;
		goto done;
	}
	if (!(fn = (GetNetworkParams_fn_t) GetProcAddress(handle, "GetNetworkParams"))) {
		log(EVDNS_LOG_WARN, "Could not get address of function.");
		status = -1;
		goto done;
	}

	buf = mm_malloc(size);
	if (!buf) { status = 4; goto done; }
	fixed = buf;
	r = fn(fixed, &size);
	if (r != ERROR_SUCCESS && r != ERROR_BUFFER_OVERFLOW) {
		status = -1;
		goto done;
	}
	if (r != ERROR_SUCCESS) {
		mm_free(buf);
		buf = mm_malloc(size);
		if (!buf) { status = 4; goto done; }
		fixed = buf;
		r = fn(fixed, &size);
		if (r != ERROR_SUCCESS) {
			log(EVDNS_LOG_DEBUG, "fn() failed.");
			status = -1;
			goto done;
		}
	}

	EVUTIL_ASSERT(fixed);
	added_any = 0;
	ns = &(fixed->DnsServerList);
	while (ns) {
		r = evdns_nameserver_ip_add_line(base, ns->IpAddress.String);
		if (r) {
			log(EVDNS_LOG_DEBUG,"Could not add nameserver %s to list,error: %d",
				(ns->IpAddress.String),(int)GetLastError());
			status = r;
		} else {
			++added_any;
			log(EVDNS_LOG_DEBUG,"Successfully added %s as nameserver",ns->IpAddress.String);
		}

		ns = ns->Next;
	}

	if (!added_any) {
		log(EVDNS_LOG_DEBUG, "No nameservers added.");
		if (status == 0)
			status = -1;
	} else {
		status = 0;
	}

 done:
	if (buf)
		mm_free(buf);
	if (handle)
		FreeLibrary(handle);
	return status;
}

static int
config_nameserver_from_reg_key(struct evdns_base *base, HKEY key, const TCHAR *subkey)
{
	char *buf;
	DWORD bufsz = 0, type = 0;
	int status = 0;

	ASSERT_LOCKED(base);
	if (RegQueryValueEx(key, subkey, 0, &type, NULL, &bufsz)
	    != ERROR_MORE_DATA)
		return -1;
	if (!(buf = mm_malloc(bufsz)))
		return -1;

	if (RegQueryValueEx(key, subkey, 0, &type, (LPBYTE)buf, &bufsz)
	    == ERROR_SUCCESS && bufsz > 1) {
		status = evdns_nameserver_ip_add_line(base,buf);
	}

	mm_free(buf);
	return status;
}

#define SERVICES_KEY TEXT("System\\CurrentControlSet\\Services\\")
#define WIN_NS_9X_KEY  SERVICES_KEY TEXT("VxD\\MSTCP")
#define WIN_NS_NT_KEY  SERVICES_KEY TEXT("Tcpip\\Parameters")

static int
load_nameservers_from_registry(struct evdns_base *base)
{
	int found = 0;
	int r;
#define TRY(k, name) \
	if (!found && config_nameserver_from_reg_key(base,k,TEXT(name)) == 0) { \
		log(EVDNS_LOG_DEBUG,"Found nameservers in %s/%s",#k,name); \
		found = 1;						\
	} else if (!found) {						\
		log(EVDNS_LOG_DEBUG,"Didn't find nameservers in %s/%s", \
		    #k,#name);						\
	}

	ASSERT_LOCKED(base);

	if (((int)GetVersion()) > 0) { /* NT */
		HKEY nt_key = 0, interfaces_key = 0;

		if (RegOpenKeyEx(HKEY_LOCAL_MACHINE, WIN_NS_NT_KEY, 0,
				 KEY_READ, &nt_key) != ERROR_SUCCESS) {
			log(EVDNS_LOG_DEBUG,"Couldn't open nt key, %d",(int)GetLastError());
			return -1;
		}
		r = RegOpenKeyEx(nt_key, TEXT("Interfaces"), 0,
			     KEY_QUERY_VALUE|KEY_ENUMERATE_SUB_KEYS,
			     &interfaces_key);
		if (r != ERROR_SUCCESS) {
			log(EVDNS_LOG_DEBUG,"Couldn't open interfaces key, %d",(int)GetLastError());
			return -1;
		}
		TRY(nt_key, "NameServer");
		TRY(nt_key, "DhcpNameServer");
		TRY(interfaces_key, "NameServer");
		TRY(interfaces_key, "DhcpNameServer");
		RegCloseKey(interfaces_key);
		RegCloseKey(nt_key);
	} else {
		HKEY win_key = 0;
		if (RegOpenKeyEx(HKEY_LOCAL_MACHINE, WIN_NS_9X_KEY, 0,
				 KEY_READ, &win_key) != ERROR_SUCCESS) {
			log(EVDNS_LOG_DEBUG, "Couldn't open registry key, %d", (int)GetLastError());
			return -1;
		}
		TRY(win_key, "NameServer");
		RegCloseKey(win_key);
	}

	if (found == 0) {
		log(EVDNS_LOG_WARN,"Didn't find any nameservers.");
	}

	return found ? 0 : -1;
#undef TRY
}

int
evdns_base_config_windows_nameservers(struct evdns_base *base)
{
	int r;
	char *fname;
	if (base == NULL)
		base = current_base;
	if (base == NULL)
		return -1;
	EVDNS_LOCK(base);
	fname = evdns_get_default_hosts_filename();
	log(EVDNS_LOG_DEBUG, "Loading hosts entries from %s", fname);
	evdns_base_load_hosts(base, fname);
	if (fname)
		mm_free(fname);

	if (load_nameservers_with_getnetworkparams(base) == 0) {
		EVDNS_UNLOCK(base);
		return 0;
	}
	r = load_nameservers_from_registry(base);

	EVDNS_UNLOCK(base);
	return r;
}

int
evdns_config_windows_nameservers(void)
{
	if (!current_base) {
		current_base = evdns_base_new(NULL, 1);
		return current_base == NULL ? -1 : 0;
	} else {
		return evdns_base_config_windows_nameservers(current_base);
	}
}
#endif

struct evdns_base *
evdns_base_new(struct event_base *event_base, int flags)
{
	struct evdns_base *base;

	if (evutil_secure_rng_init() < 0) {
		log(EVDNS_LOG_WARN, "Unable to seed random number generator; "
		    "DNS can't run.");
		return NULL;
	}

	/* Give the evutil library a hook into its evdns-enabled
	 * functionality.  We can't just call evdns_getaddrinfo directly or
	 * else libevent-core will depend on libevent-extras. */
	evutil_set_evdns_getaddrinfo_fn_(evdns_getaddrinfo);
	evutil_set_evdns_getaddrinfo_cancel_fn_(evdns_getaddrinfo_cancel);

	base = mm_malloc(sizeof(struct evdns_base));
	if (base == NULL)
		return (NULL);
	memset(base, 0, sizeof(struct evdns_base));
	base->req_waiting_head = NULL;

	EVTHREAD_ALLOC_LOCK(base->lock, EVTHREAD_LOCKTYPE_RECURSIVE);
	EVDNS_LOCK(base);

	/* Set max requests inflight and allocate req_heads. */
	base->req_heads = NULL;

	evdns_base_set_max_requests_inflight(base, 64);

	base->server_head = NULL;
	base->event_base = event_base;
	base->global_good_nameservers = base->global_requests_inflight =
		base->global_requests_waiting = 0;

	base->global_timeout.tv_sec = 5;
	base->global_timeout.tv_usec = 0;
	base->global_max_reissues = 1;
	base->global_max_retransmits = 3;
	base->global_max_nameserver_timeout = 3;
	base->global_search_state = NULL;
	base->global_randomize_case = 1;
	base->global_max_udp_size = DNS_MAX_UDP_SIZE;
	base->global_getaddrinfo_allow_skew.tv_sec = 3;
	base->global_getaddrinfo_allow_skew.tv_usec = 0;
	base->global_nameserver_probe_initial_timeout.tv_sec = 10;
	base->global_nameserver_probe_initial_timeout.tv_usec = 0;
	base->ns_max_probe_timeout = 3600;
	base->ns_timeout_backoff_factor = 3;
	base->global_tcp_idle_timeout.tv_sec = CLIENT_IDLE_CONN_TIMEOUT;

	TAILQ_INIT(&base->hostsdb);

#define EVDNS_BASE_ALL_FLAGS ( \
	EVDNS_BASE_INITIALIZE_NAMESERVERS | \
	EVDNS_BASE_DISABLE_WHEN_INACTIVE  | \
	EVDNS_BASE_NAMESERVERS_NO_DEFAULT | \
	0)

	if (flags & ~EVDNS_BASE_ALL_FLAGS) {
		flags = EVDNS_BASE_INITIALIZE_NAMESERVERS;
		log(EVDNS_LOG_WARN,
		    "Unrecognized flag passed to evdns_base_new(). Assuming "
		    "you meant EVDNS_BASE_INITIALIZE_NAMESERVERS.");
	}
#undef EVDNS_BASE_ALL_FLAGS

	if (flags & EVDNS_BASE_INITIALIZE_NAMESERVERS) {
		int r;
		int opts = DNS_OPTIONS_ALL;
		if (flags & EVDNS_BASE_NAMESERVERS_NO_DEFAULT) {
			opts |= DNS_OPTION_NAMESERVERS_NO_DEFAULT;
		}

#ifdef _WIN32
		r = evdns_base_config_windows_nameservers(base);
#else
		r = evdns_base_resolv_conf_parse(base, opts, "/etc/resolv.conf");
#endif
		if (r) {
			evdns_base_free_and_unlock(base, 0);
			return NULL;
		}
	}
	if (flags & EVDNS_BASE_DISABLE_WHEN_INACTIVE) {
		base->disable_when_inactive = 1;
	}

	EVDNS_UNLOCK(base);
	return base;
}

int
evdns_init(void)
{
	struct evdns_base *base = evdns_base_new(NULL, 1);
	if (base) {
		current_base = base;
		return 0;
	} else {
		return -1;
	}
}

const char *
evdns_err_to_string(int err)
{
    switch (err) {
	case DNS_ERR_NONE: return "no error";
	case DNS_ERR_FORMAT: return "misformatted query";
	case DNS_ERR_SERVERFAILED: return "server failed";
	case DNS_ERR_NOTEXIST: return "name does not exist";
	case DNS_ERR_NOTIMPL: return "query not implemented";
	case DNS_ERR_REFUSED: return "refused";

	case DNS_ERR_TRUNCATED: return "reply truncated or ill-formed";
	case DNS_ERR_UNKNOWN: return "unknown";
	case DNS_ERR_TIMEOUT: return "request timed out";
	case DNS_ERR_SHUTDOWN: return "dns subsystem shut down";
	case DNS_ERR_CANCEL: return "dns request canceled";
	case DNS_ERR_NODATA: return "no records in the reply";
	default: return "[Unknown error code]";
    }
}

static void
evdns_nameserver_free(struct nameserver *server)
{
	if (server->socket >= 0)
		evutil_closesocket(server->socket);
	(void) event_del(&server->event);
	event_debug_unassign(&server->event);
	if (server->state == 0)
		(void) event_del(&server->timeout_event);
	if (server->probe_request) {
		evdns_cancel_request(server->base, server->probe_request);
		server->probe_request = NULL;
	}
	event_debug_unassign(&server->timeout_event);
	disconnect_and_free_connection(server->connection);
	mm_free(server);
}

static void
evdns_base_free_and_unlock(struct evdns_base *base, int fail_requests)
{
	struct nameserver *server, *server_next;
	struct search_domain *dom, *dom_next;
	int i;

	/* Requires that we hold the lock. */

	/* TODO(nickm) we might need to refcount here. */

	while (base->req_waiting_head) {
		if (fail_requests)
			reply_schedule_callback(base->req_waiting_head, 0, DNS_ERR_SHUTDOWN, NULL);
		request_finished(base->req_waiting_head, &base->req_waiting_head, 1);
	}
	for (i = 0; i < base->n_req_heads; ++i) {
		while (base->req_heads[i]) {
			if (fail_requests)
				reply_schedule_callback(base->req_heads[i], 0, DNS_ERR_SHUTDOWN, NULL);
			request_finished(base->req_heads[i], &REQ_HEAD(base, base->req_heads[i]->trans_id), 1);
		}
	}
	base->global_requests_inflight = base->global_requests_waiting = 0;

	for (server = base->server_head; server; server = server_next) {
		server_next = server->next;
		/** already done something before */
		server->probe_request = NULL;
		evdns_nameserver_free(server);
		if (server_next == base->server_head)
			break;
	}
	base->server_head = NULL;
	base->global_good_nameservers = 0;

	if (base->global_search_state) {
		for (dom = base->global_search_state->head; dom; dom = dom_next) {
			dom_next = dom->next;
			mm_free(dom);
		}
		mm_free(base->global_search_state);
		base->global_search_state = NULL;
	}

	{
		struct hosts_entry *victim;
		while ((victim = TAILQ_FIRST(&base->hostsdb))) {
			TAILQ_REMOVE(&base->hostsdb, victim, next);
			mm_free(victim);
		}
	}

	mm_free(base->req_heads);

	EVDNS_UNLOCK(base);
	EVTHREAD_FREE_LOCK(base->lock, EVTHREAD_LOCKTYPE_RECURSIVE);

	mm_free(base);
}

void
evdns_base_free(struct evdns_base *base, int fail_requests)
{
	EVDNS_LOCK(base);
	evdns_base_free_and_unlock(base, fail_requests);
}

void
evdns_base_clear_host_addresses(struct evdns_base *base)
{
	struct hosts_entry *victim;
	EVDNS_LOCK(base);
	while ((victim = TAILQ_FIRST(&base->hostsdb))) {
		TAILQ_REMOVE(&base->hostsdb, victim, next);
		mm_free(victim);
	}
	EVDNS_UNLOCK(base);
}

void
evdns_shutdown(int fail_requests)
{
	if (current_base) {
		struct evdns_base *b = current_base;
		current_base = NULL;
		evdns_base_free(b, fail_requests);
	}
	evdns_log_fn = NULL;
}

static int
evdns_base_parse_hosts_line(struct evdns_base *base, char *line)
{
	char *strtok_state;
	static const char *const delims = " \t";
	char *const addr = strtok_r(line, delims, &strtok_state);
	char *hostname, *hash;
	struct sockaddr_storage ss;
	int socklen = sizeof(ss);
	ASSERT_LOCKED(base);

#define NEXT_TOKEN strtok_r(NULL, delims, &strtok_state)

	if (!addr || *addr == '#')
		return 0;

	memset(&ss, 0, sizeof(ss));
	if (evutil_parse_sockaddr_port(addr, (struct sockaddr*)&ss, &socklen)<0)
		return -1;
	if (socklen > (int)sizeof(struct sockaddr_in6))
		return -1;

	if (sockaddr_getport((struct sockaddr*)&ss))
		return -1;

	while ((hostname = NEXT_TOKEN)) {
		struct hosts_entry *he;
		size_t namelen;
		if ((hash = strchr(hostname, '#'))) {
			if (hash == hostname)
				return 0;
			*hash = '\0';
		}

		namelen = strlen(hostname);

		he = mm_calloc(1, sizeof(struct hosts_entry)+namelen);
		if (!he)
			return -1;
		EVUTIL_ASSERT(socklen <= (int)sizeof(he->addr));
		memcpy(&he->addr, &ss, socklen);
		memcpy(he->hostname, hostname, namelen+1);
		he->addrlen = socklen;

		TAILQ_INSERT_TAIL(&base->hostsdb, he, next);

		if (hash)
			return 0;
	}

	return 0;
#undef NEXT_TOKEN
}

static int
evdns_base_load_hosts_impl(struct evdns_base *base, const char *hosts_fname)
{
	char *str=NULL, *cp, *eol;
	size_t len;
	int err=0;

	ASSERT_LOCKED(base);

	if (hosts_fname == NULL ||
	    (err = evutil_read_file_(hosts_fname, &str, &len, 0)) < 0) {
		char tmp[64];
		strlcpy(tmp, "127.0.0.1   localhost", sizeof(tmp));
		evdns_base_parse_hosts_line(base, tmp);
		strlcpy(tmp, "::1   localhost", sizeof(tmp));
		evdns_base_parse_hosts_line(base, tmp);
		return err ? -1 : 0;
	}

	/* This will break early if there is a NUL in the hosts file.
	 * Probably not a problem.*/
	cp = str;
	for (;;) {
		eol = strchr(cp, '\n');

		if (eol) {
			*eol = '\0';
			evdns_base_parse_hosts_line(base, cp);
			cp = eol+1;
		} else {
			evdns_base_parse_hosts_line(base, cp);
			break;
		}
	}

	mm_free(str);
	return 0;
}

int
evdns_base_load_hosts(struct evdns_base *base, const char *hosts_fname)
{
	int res;
	if (!base)
		base = current_base;
	EVDNS_LOCK(base);
	res = evdns_base_load_hosts_impl(base, hosts_fname);
	EVDNS_UNLOCK(base);
	return res;
}

/* A single request for a getaddrinfo, either v4 or v6. */
struct getaddrinfo_subrequest {
	struct evdns_request *r;
	ev_uint32_t type;
};

/* State data used to implement an in-progress getaddrinfo. */
struct evdns_getaddrinfo_request {
	struct evdns_base *evdns_base;
	/* Copy of the modified 'hints' data that we'll use to build
	 * answers. */
	struct evutil_addrinfo hints;
	/* The callback to invoke when we're done */
	evdns_getaddrinfo_cb user_cb;
	/* User-supplied data to give to the callback. */
	void *user_data;
	/* The port to use when building sockaddrs. */
	ev_uint16_t port;
	/* The sub_request for an A record (if any) */
	struct getaddrinfo_subrequest ipv4_request;
	/* The sub_request for an AAAA record (if any) */
	struct getaddrinfo_subrequest ipv6_request;

	/* The cname result that we were told (if any) */
	char *cname_result;

	/* If we have one request answered and one request still inflight,
	 * then this field holds the answer from the first request... */
	struct evutil_addrinfo *pending_result;
	/* And this event is a timeout that will tell us to cancel the second
	 * request if it's taking a long time. */
	struct event timeout;

	/* And this field holds the error code from the first request... */
	int pending_error;
	/* If this is set, the user canceled this request. */
	unsigned user_canceled : 1;
	/* If this is set, the user can no longer cancel this request; we're
	 * just waiting for the free. */
	unsigned request_done : 1;
};

/* Convert an evdns errors to the equivalent getaddrinfo error. */
static int
evdns_err_to_getaddrinfo_err(int e1)
{
	/* XXX Do this better! */
	if (e1 == DNS_ERR_NONE)
		return 0;
	else if (e1 == DNS_ERR_NOTEXIST)
		return EVUTIL_EAI_NONAME;
	else
		return EVUTIL_EAI_FAIL;
}

/* Return the more informative of two getaddrinfo errors. */
static int
getaddrinfo_merge_err(int e1, int e2)
{
	/* XXXX be cleverer here. */
	if (e1 == 0)
		return e2;
	else
		return e1;
}

static void
free_getaddrinfo_request(struct evdns_getaddrinfo_request *data)
{
	/* DO NOT CALL this if either of the requests is pending.  Only once
	 * both callbacks have been invoked is it safe to free the request */
	if (data->pending_result)
		evutil_freeaddrinfo(data->pending_result);
	if (data->cname_result)
		mm_free(data->cname_result);
	event_del(&data->timeout);
	mm_free(data);
	return;
}

static void
add_cname_to_reply(struct evdns_getaddrinfo_request *data,
    struct evutil_addrinfo *ai)
{
	if (data->cname_result && ai) {
		ai->ai_canonname = data->cname_result;
		data->cname_result = NULL;
	}
}

/* Callback: invoked when one request in a mixed-format A/AAAA getaddrinfo
 * request has finished, but the other one took too long to answer. Pass
 * along the answer we got, and cancel the other request.
 */
static void
evdns_getaddrinfo_timeout_cb(evutil_socket_t fd, short what, void *ptr)
{
	int v4_timedout = 0, v6_timedout = 0;
	struct evdns_getaddrinfo_request *data = ptr;

	/* Cancel any pending requests, and note which one */
	if (data->ipv4_request.r) {
		/* XXXX This does nothing if the request's callback is already
		 * running (pending_cb is set). */
		evdns_cancel_request(NULL, data->ipv4_request.r);
		v4_timedout = 1;
		EVDNS_LOCK(data->evdns_base);
		++data->evdns_base->getaddrinfo_ipv4_timeouts;
		EVDNS_UNLOCK(data->evdns_base);
	}
	if (data->ipv6_request.r) {
		/* XXXX This does nothing if the request's callback is already
		 * running (pending_cb is set). */
		evdns_cancel_request(NULL, data->ipv6_request.r);
		v6_timedout = 1;
		EVDNS_LOCK(data->evdns_base);
		++data->evdns_base->getaddrinfo_ipv6_timeouts;
		EVDNS_UNLOCK(data->evdns_base);
	}

	/* We only use this timeout callback when we have an answer for
	 * one address. */
	EVUTIL_ASSERT(!v4_timedout || !v6_timedout);

	/* Report the outcome of the other request that didn't time out. */
	if (data->pending_result) {
		add_cname_to_reply(data, data->pending_result);
		data->user_cb(0, data->pending_result, data->user_data);
		data->pending_result = NULL;
	} else {
		int e = data->pending_error;
		if (!e)
			e = EVUTIL_EAI_AGAIN;
		data->user_cb(e, NULL, data->user_data);
	}

	data->user_cb = NULL; /* prevent double-call if evdns callbacks are
			       * in-progress. XXXX It would be better if this
			       * weren't necessary. */

	if (!v4_timedout && !v6_timedout) {
		/* should be impossible? XXXX */
		free_getaddrinfo_request(data);
	}
}

static int
evdns_getaddrinfo_set_timeout(struct evdns_base *evdns_base,
    struct evdns_getaddrinfo_request *data)
{
	return event_add(&data->timeout, &evdns_base->global_getaddrinfo_allow_skew);
}

static inline int
evdns_result_is_answer(int result)
{
	return (result != DNS_ERR_NOTIMPL && result != DNS_ERR_REFUSED &&
	    result != DNS_ERR_SERVERFAILED && result != DNS_ERR_CANCEL);
}

static void
evdns_getaddrinfo_gotresolve(int result, char type, int count,
    int ttl, void *addresses, void *arg)
{
	int i;
	struct getaddrinfo_subrequest *req = arg;
	struct getaddrinfo_subrequest *other_req;
	struct evdns_getaddrinfo_request *data;

	struct evutil_addrinfo *res;

	struct sockaddr_in sin;
	struct sockaddr_in6 sin6;
	struct sockaddr *sa;
	int socklen, addrlen;
	void *addrp;
	int err;
	int user_canceled;

	EVUTIL_ASSERT(req->type == DNS_IPv4_A || req->type == DNS_IPv6_AAAA);
	if (req->type == DNS_IPv4_A) {
		data = EVUTIL_UPCAST(req, struct evdns_getaddrinfo_request, ipv4_request);
		other_req = &data->ipv6_request;
	} else {
		data = EVUTIL_UPCAST(req, struct evdns_getaddrinfo_request, ipv6_request);
		other_req = &data->ipv4_request;
	}

	/** Called from evdns_base_free() with @fail_requests == 1 */
	if (result != DNS_ERR_SHUTDOWN) {
		EVDNS_LOCK(data->evdns_base);
		if (evdns_result_is_answer(result)) {
			if (req->type == DNS_IPv4_A)
				++data->evdns_base->getaddrinfo_ipv4_answered;
			else
				++data->evdns_base->getaddrinfo_ipv6_answered;
		}
		user_canceled = data->user_canceled;
		if (other_req->r == NULL)
			data->request_done = 1;
		EVDNS_UNLOCK(data->evdns_base);
	} else {
		data->evdns_base = NULL;
		user_canceled = data->user_canceled;
	}

	req->r = NULL;

	if (result == DNS_ERR_CANCEL && ! user_canceled) {
		/* Internal cancel request from timeout or internal error.
		 * we already answered the user. */
		if (other_req->r == NULL)
			free_getaddrinfo_request(data);
		return;
	}

	if (data->user_cb == NULL) {
		/* We already answered.  XXXX This shouldn't be needed; see
		 * comments in evdns_getaddrinfo_timeout_cb */
		free_getaddrinfo_request(data);
		return;
	}

	if (result == DNS_ERR_NONE) {
		if (count == 0)
			err = EVUTIL_EAI_NODATA;
		else
			err = 0;
	} else {
		err = evdns_err_to_getaddrinfo_err(result);
	}

	if (err) {
		/* Looks like we got an error. */
		if (other_req->r) {
			/* The other request is still working; maybe it will
			 * succeed. */
			/* XXXX handle failure from set_timeout */
			if (result != DNS_ERR_SHUTDOWN) {
				evdns_getaddrinfo_set_timeout(data->evdns_base, data);
			}
			data->pending_error = err;
			return;
		}

		if (user_canceled) {
			data->user_cb(EVUTIL_EAI_CANCEL, NULL, data->user_data);
		} else if (data->pending_result) {
			/* If we have an answer waiting, and we weren't
			 * canceled, ignore this error. */
			add_cname_to_reply(data, data->pending_result);
			data->user_cb(0, data->pending_result, data->user_data);
			data->pending_result = NULL;
		} else {
			if (data->pending_error)
				err = getaddrinfo_merge_err(err,
				    data->pending_error);
			data->user_cb(err, NULL, data->user_data);
		}
		free_getaddrinfo_request(data);
		return;
	} else if (user_canceled) {
		if (other_req->r) {
			/* The other request is still working; let it hit this
			 * callback with EVUTIL_EAI_CANCEL callback and report
			 * the failure. */
			return;
		}
		data->user_cb(EVUTIL_EAI_CANCEL, NULL, data->user_data);
		free_getaddrinfo_request(data);
		return;
	}

	/* Looks like we got some answers. We should turn them into addrinfos
	 * and then either queue those or return them all. */
	EVUTIL_ASSERT(type == DNS_IPv4_A || type == DNS_IPv6_AAAA);

	if (type == DNS_IPv4_A) {
		memset(&sin, 0, sizeof(sin));
		sin.sin_family = AF_INET;
		sin.sin_port = htons(data->port);
#ifdef EVENT__HAVE_STRUCT_SOCKADDR_IN_SIN_LEN
		sin.sin_len = sizeof(sin);
#endif

		sa = (struct sockaddr *)&sin;
		socklen = sizeof(sin);
		addrlen = 4;
		addrp = &sin.sin_addr.s_addr;
	} else {
		memset(&sin6, 0, sizeof(sin6));
		sin6.sin6_family = AF_INET6;
		sin6.sin6_port = htons(data->port);
#ifdef EVENT__HAVE_STRUCT_SOCKADDR_IN6_SIN6_LEN
		sin6.sin6_len = sizeof(sin6);
#endif

		sa = (struct sockaddr *)&sin6;
		socklen = sizeof(sin6);
		addrlen = 16;
		addrp = &sin6.sin6_addr.s6_addr;
	}

	res = NULL;
	for (i=0; i < count; ++i) {
		struct evutil_addrinfo *ai;
		memcpy(addrp, ((char*)addresses)+i*addrlen, addrlen);
		ai = evutil_new_addrinfo_(sa, socklen, &data->hints);
		if (!ai) {
			if (other_req->r) {
				evdns_cancel_request(NULL, other_req->r);
			}
			data->user_cb(EVUTIL_EAI_MEMORY, NULL, data->user_data);
			if (res)
				evutil_freeaddrinfo(res);

			if (other_req->r == NULL)
				free_getaddrinfo_request(data);
			return;
		}
		res = evutil_addrinfo_append_(res, ai);
	}

	if (other_req->r) {
		/* The other request is still in progress; wait for it */
		/* XXXX handle failure from set_timeout */
		evdns_getaddrinfo_set_timeout(data->evdns_base, data);
		data->pending_result = res;
		return;
	} else {
		/* The other request is done or never started; append its
		 * results (if any) and return them. */
		if (data->pending_result) {
			if (req->type == DNS_IPv4_A)
				res = evutil_addrinfo_append_(res,
				    data->pending_result);
			else
				res = evutil_addrinfo_append_(
				    data->pending_result, res);
			data->pending_result = NULL;
		}

		/* Call the user callback. */
		add_cname_to_reply(data, res);
		data->user_cb(0, res, data->user_data);

		/* Free data. */
		free_getaddrinfo_request(data);
	}
}

static struct hosts_entry *
find_hosts_entry(struct evdns_base *base, const char *hostname,
    struct hosts_entry *find_after)
{
	struct hosts_entry *e;

	if (find_after)
		e = TAILQ_NEXT(find_after, next);
	else
		e = TAILQ_FIRST(&base->hostsdb);

	for (; e; e = TAILQ_NEXT(e, next)) {
		if (!evutil_ascii_strcasecmp(e->hostname, hostname))
			return e;
	}
	return NULL;
}

static int
evdns_getaddrinfo_fromhosts(struct evdns_base *base,
    const char *nodename, struct evutil_addrinfo *hints, ev_uint16_t port,
    struct evutil_addrinfo **res)
{
	int n_found = 0;
	struct hosts_entry *e;
	struct evutil_addrinfo *ai=NULL;
	int f = hints->ai_family;

	EVDNS_LOCK(base);
	for (e = find_hosts_entry(base, nodename, NULL); e;
	    e = find_hosts_entry(base, nodename, e)) {
		struct evutil_addrinfo *ai_new;
		++n_found;
		if ((e->addr.sa.sa_family == AF_INET && f == PF_INET6) ||
		    (e->addr.sa.sa_family == AF_INET6 && f == PF_INET))
			continue;
		ai_new = evutil_new_addrinfo_(&e->addr.sa, e->addrlen, hints);
		if (!ai_new) {
			n_found = 0;
			goto out;
		}
		sockaddr_setport(ai_new->ai_addr, port);
		ai = evutil_addrinfo_append_(ai, ai_new);
	}
	EVDNS_UNLOCK(base);
out:
	if (n_found) {
		/* Note that we return an empty answer if we found entries for
		 * this hostname but none were of the right address type. */
		*res = ai;
		return 0;
	} else {
		if (ai)
			evutil_freeaddrinfo(ai);
		return -1;
	}
}

struct evdns_getaddrinfo_request *
evdns_getaddrinfo(struct evdns_base *dns_base,
    const char *nodename, const char *servname,
    const struct evutil_addrinfo *hints_in,
    evdns_getaddrinfo_cb cb, void *arg)
{
	struct evdns_getaddrinfo_request *data;
	struct evutil_addrinfo hints;
	struct evutil_addrinfo *res = NULL;
	int err;
	int port = 0;
	int want_cname = 0;
	int started = 0;

	if (!dns_base) {
		dns_base = current_base;
		if (!dns_base) {
			log(EVDNS_LOG_WARN,
			    "Call to getaddrinfo_async with no "
			    "evdns_base configured.");
			cb(EVUTIL_EAI_FAIL, NULL, arg); /* ??? better error? */
			return NULL;
		}
	}

	/* If we _must_ answer this immediately, do so. */
	if ((hints_in && (hints_in->ai_flags & EVUTIL_AI_NUMERICHOST))) {
		res = NULL;
		err = evutil_getaddrinfo(nodename, servname, hints_in, &res);
		cb(err, res, arg);
		return NULL;
	}

	if (hints_in) {
		memcpy(&hints, hints_in, sizeof(hints));
	} else {
		memset(&hints, 0, sizeof(hints));
		hints.ai_family = PF_UNSPEC;
	}

	evutil_adjust_hints_for_addrconfig_(&hints);

	/* Now try to see if we _can_ answer immediately. */
	/* (It would be nice to do this by calling getaddrinfo directly, with
	 * AI_NUMERICHOST, on plaforms that have it, but we can't: there isn't
	 * a reliable way to distinguish the "that wasn't a numeric host!" case
	 * from any other EAI_NONAME cases.) */
	err = evutil_getaddrinfo_common_(nodename, servname, &hints, &res, &port);
	if (err != EVUTIL_EAI_NEED_RESOLVE) {
		cb(err, res, arg);
		return NULL;
	}

	/* If there is an entry in the hosts file, we should give it now. */
	if (!evdns_getaddrinfo_fromhosts(dns_base, nodename, &hints, port, &res)) {
		cb(0, res, arg);
		return NULL;
	}

	/* Okay, things are serious now. We're going to need to actually
	 * launch a request.
	 */
	data = mm_calloc(1,sizeof(struct evdns_getaddrinfo_request));
	if (!data) {
		cb(EVUTIL_EAI_MEMORY, NULL, arg);
		return NULL;
	}

	memcpy(&data->hints, &hints, sizeof(data->hints));
	data->port = (ev_uint16_t)port;
	data->ipv4_request.type = DNS_IPv4_A;
	data->ipv6_request.type = DNS_IPv6_AAAA;
	data->user_cb = cb;
	data->user_data = arg;
	data->evdns_base = dns_base;

	want_cname = (hints.ai_flags & EVUTIL_AI_CANONNAME);

	/* If we are asked for a PF_UNSPEC address, we launch two requests in
	 * parallel: one for an A address and one for an AAAA address.  We
	 * can't send just one request, since many servers only answer one
	 * question per DNS request.
	 *
	 * Once we have the answer to one request, we allow for a short
	 * timeout before we report it, to see if the other one arrives.  If
	 * they both show up in time, then we report both the answers.
	 *
	 * If too many addresses of one type time out or fail, we should stop
	 * launching those requests. (XXX we don't do that yet.)
	 */

	EVDNS_LOCK(dns_base);

	if (hints.ai_family != PF_INET6) {
		log(EVDNS_LOG_DEBUG, "Sending request for %s on ipv4 as %p",
		    nodename, (void *)&data->ipv4_request);

		data->ipv4_request.r = evdns_base_resolve_ipv4(dns_base,
		    nodename, 0, evdns_getaddrinfo_gotresolve,
		    &data->ipv4_request);
		if (want_cname && data->ipv4_request.r)
			data->ipv4_request.r->current_req->put_cname_in_ptr =
			    &data->cname_result;
	}
	if (hints.ai_family != PF_INET) {
		log(EVDNS_LOG_DEBUG, "Sending request for %s on ipv6 as %p",
		    nodename, (void *)&data->ipv6_request);

		data->ipv6_request.r = evdns_base_resolve_ipv6(dns_base,
		    nodename, 0, evdns_getaddrinfo_gotresolve,
		    &data->ipv6_request);
		if (want_cname && data->ipv6_request.r)
			data->ipv6_request.r->current_req->put_cname_in_ptr =
			    &data->cname_result;
	}

	evtimer_assign(&data->timeout, dns_base->event_base,
	    evdns_getaddrinfo_timeout_cb, data);

	started = (data->ipv4_request.r || data->ipv6_request.r);

	EVDNS_UNLOCK(dns_base);

	if (started) {
		return data;
	} else {
		mm_free(data);
		cb(EVUTIL_EAI_FAIL, NULL, arg);
		return NULL;
	}
}

void
evdns_getaddrinfo_cancel(struct evdns_getaddrinfo_request *data)
{
	EVDNS_LOCK(data->evdns_base);
	if (data->request_done) {
		EVDNS_UNLOCK(data->evdns_base);
		return;
	}
	event_del(&data->timeout);
	data->user_canceled = 1;
	if (data->ipv4_request.r)
		evdns_cancel_request(data->evdns_base, data->ipv4_request.r);
	if (data->ipv6_request.r)
		evdns_cancel_request(data->evdns_base, data->ipv6_request.r);
	EVDNS_UNLOCK(data->evdns_base);
}