path: root/debian/src/geoip-asn-csv-to-dat.cpp

/* geoip-csv-to-dat - convert a country database from CSV to GeoIP binary format
 *
 * Copyright (c) 2009 Kalle Olavi Niemitalo.
 * Copyright (c) 2011 Patrick Matthäi
 *
 * Permission to use, copy, modify, and/or distribute this software for any
 * purpose with or without fee is hereby granted, provided that the above
 * copyright notice and this permission notice appear in all copies.
 *
 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
 */

#define _GNU_SOURCE 1
#include <algorithm>
#include <arpa/inet.h>
#include <cerrno>
#include <cstdio>
#include <cstdlib>
#include <cstring>
#include <error.h>
#include <fstream>
#include <getopt.h>
#include <iostream>
#include <queue>
#include <set>
#include <stack>
#include <string>
#include <sys/socket.h>
#include <sys/types.h>
#include <sysexits.h>
#include <vector>
#include <GeoIP.h>
#include <map>

// Format of GeoIP Country database files
// ======================================
//
// 1. Binary trie mapping IP addresses to countries.
// 2. Optional unused data.
// 3. Optional database-info block.
// 4. Optional structure-info block.
//
// Binary trie
// -----------
//
// The trie treats IP addresses as bit sequences and maps them to
// numbers.  In the country database, each such number is 0xFFFF00 +
// the the country ID that GeoIP_id_by_ipnum() would return.  The
// meanings of country IDs are hardcoded in libGeoIP and cannot be
// overridden by the database.
//
// The root node of the trie is at the beginning of the file, and the
// other nodes then follow it.	Each node has the same size and
// consists of two little-endian pointers that correspond to the two
// possible values of a bit.  In the country database, the pointers are
// 24-bit, and each node is thus 6 bytes long.
//
// Each pointer is one of:
// - The number that the whole lookup should return, i.e. 0xFFFF00 + id
//   in the country database.
// - The number of the node that should be examined next, counting from
//   0 at the beginning of the file.  Pointing back to nodes with
//   smaller numbers is allowed, but loops are not allowed.
//
// Optional unused data
// --------------------
//
// The file format seems to permit extra data between the binary trie
// and the optional blocks.
//
// Optional database-info block
// ----------------------------
//
// Near the end of the file, there may be a three-byte tag (0x00 0x00
// 0x00) followed by at most DATABASE_INFO_MAX_SIZE - 1 = 99 bytes of
// text that describes the database.  GeoIP_database_info() returns
// this text and appends a terminating '\0'.
//
// The GeoLite Country IPv4 database downloadable from MaxMind
// includes this database-info block.
//
// Optional structure-info block
// -----------------------------
//
// At the very end of the file, there may be a three-byte tag (0xFF
// 0xFF 0xFF) followed by at most STRUCTURE_INFO_MAX_SIZE - 1 = 19
// bytes.  The first byte is the database type,
// e.g. GEOIP_COUNTRY_EDITION = 1 or GEOIP_COUNTRY_EDITION_V6 = 12,
// possibly with 105 added to it.  Type-specific information then
// follows.  There is no type-specific information for the country
// editions.
//
// The GeoLite Country IPv4 database downloadable from MaxMind does
// not include this structure-info block.

namespace {
	class binary_trie
	{
	public:
		typedef uint_fast32_t edge_type;

		explicit binary_trie(edge_type leaf);
		void set_range(
			const uint8_t range_min[],
			const uint8_t range_max[],
			std::size_t bit_count,
			edge_type leaf);
		void reorder_depth_first();
		void reorder_in_blocks(std::size_t bytes_per_block);
		void write_binary(std::ostream &dat_stream) const;
		void write_segment(std::ostream &dat_stream) const;
		void update_records();

	private:
		struct node
		{
			edge_type edges[2];
		};
		std::vector<node> nodes;

		// This could be std::vector<bool> but that seems slower.
		typedef std::vector<uint8_t> bits_vector;

		void set_range_in_node(
			const bits_vector *min_bits,
			const bits_vector *max_bits,
			std::size_t bit_pos,
			edge_type edit_node,
			edge_type leaf);
		void set_range_in_edge(
			const bits_vector *min_bits,
			const bits_vector *max_bits,
			std::size_t bit_pos,
			edge_type edit_node,
			bool bit,
			edge_type leaf);
		void reorder(
			const std::vector<edge_type> &old_to_new,
			const std::vector<edge_type> &new_to_old);
	};
}

/** Construct a binary trie and its root node.
 *
 * \param leaf
 * Both edges of the root node will initially point to this leaf.
 * The caller should provide a value that means nothing was found.  */
binary_trie::binary_trie(edge_type leaf)
{
	const node node = {{ leaf, leaf }};
	nodes.push_back(node);
}

/** Edit the trie so it maps a range of bit sequences to the same
 * leaf.
 *
 * \param range_min
 * The first bit sequence in the range.  Eight bits are packed in each
 * byte.  The most significant bit of the whole sequence is in the
 * most significant bit of the first byte.
 *
 * \param range_max
 * The last bit sequence in the range.
 *
 * \param bit_count
 * The number of bits in both sequences.
 *
 * \param leaf
 * The leaf to which all the bit sequences in the range should be
 * mapped.  */
void
binary_trie::set_range(
	const uint8_t range_min[],
	const uint8_t range_max[],
	std::size_t bit_count,
	edge_type leaf)
{
	bits_vector min_bits(bit_count);
	bits_vector max_bits(bit_count);
	for (std::size_t i = 0; i < bit_count; ++i) {
		std::size_t byte_pos = i / 8;
		uint8_t mask = 1 << ((~i) % 8);
		min_bits[i] = ((range_min[byte_pos] & mask) != 0);
		max_bits[i] = ((range_max[byte_pos] & mask) != 0);
	}
	set_range_in_node(&min_bits, &max_bits, 0, 0, leaf);
}

/** Edit a node in the trie so it maps a range of bit sequences to the
 * same leaf.
 *
 * \param min_bits
 * The first bit sequence in the range, or NULL if unbounded.
 *
 * \param max_bits
 * The last bit sequence in the range, or NULL if unbounded.
 *
 * \param bit_pos
 * Which bit in the sequences corresponds to \a edit_node.
 *
 * \param edit_node
 * The node to be modified.
 *
 * \param leaf
 * The leaf to which all the bit sequences in the range should be
 * mapped.  */
void
binary_trie::set_range_in_node(
	const bits_vector *min_bits,
	const bits_vector *max_bits,
	std::size_t bit_pos,
	edge_type edit_node,
	edge_type leaf)
{
	if (!min_bits || (*min_bits)[bit_pos] == false) {
		set_range_in_edge(min_bits,
				  (max_bits && (*max_bits)[bit_pos] == false)
				  ? max_bits : NULL,
				  bit_pos + 1, edit_node, false, leaf);
	}
	if (!max_bits || (*max_bits)[bit_pos] == true) {
		set_range_in_edge((min_bits && (*min_bits)[bit_pos] == true)
				  ? min_bits : NULL,
				  max_bits,
				  bit_pos + 1, edit_node, true, leaf);
	}
}

/** Edit an edge in the trie so it maps a range of bit sequences to
 * the same leaf.
 *
 * \param min_bits
 * The first bit sequence in the range, or NULL if unbounded.
 *
 * \param max_bits
 * The last bit sequence in the range, or NULL if unbounded.
 *
 * \param bit_pos
 * Which bit in the sequences corresponds to \a bit.
 *
 * \param edit_node
 * The node in which the edge to be modified is located.
 *
 * \param bit
 * Which edge of \a edit_node should be modified.
 *
 * \param leaf
 * The leaf to which all the bit sequences in the range should be
 * mapped.  */
void
binary_trie::set_range_in_edge(
	const bits_vector *min_bits,
	const bits_vector *max_bits,
	std::size_t bit_pos,
	edge_type edit_node,
	bool bit,
	edge_type leaf)
{
	// Check if the range fills this edge entirely.
	bool entire = true;
	if (min_bits
	    && std::find(min_bits->begin() + bit_pos, min_bits->end(),
			 true) != min_bits->end())
		entire = false;
	if (max_bits
	    && std::find(max_bits->begin() + bit_pos, max_bits->end(),
			 false) != max_bits->end())
		entire = false;

	if (entire) {
		nodes[edit_node].edges[bit] = leaf;
	} else {
		edge_type next = nodes[edit_node].edges[bit];
		if (next >= nodes.size()) {
			const node new_node = {{ next, next }};
			next = nodes.size();
			nodes.push_back(new_node);
			nodes[edit_node].edges[bit] = next;
		}

		set_range_in_node(min_bits, max_bits, bit_pos,
				  next, leaf);
	}
}

/** Renumber the nodes in depth-first order.  */
void
binary_trie::reorder_depth_first()
{
	std::vector<edge_type> old_to_new, new_to_old;
	std::stack<edge_type> depth_first;
	old_to_new.resize(nodes.size(), -1);
	new_to_old.reserve(nodes.size());
	depth_first.push(0);
	while (!depth_first.empty()) {
		const edge_type edge = depth_first.top();
		depth_first.pop();
		if (edge < nodes.size()) {
			old_to_new[edge] = new_to_old.size();
			new_to_old.push_back(edge);
			depth_first.push(nodes[edge].edges[1]);
			depth_first.push(nodes[edge].edges[0]);
		}
	}
	reorder(old_to_new, new_to_old);
}

/** Renumber the nodes to make lookups use CPU and disk caches more
 * effectively.
 *
 * First group the nodes into blocks so that each block contains the
 * root of a subtrie and as many levels of its descendants as will
 * fit.  This way, after the root is paged in, the next few lookup
 * steps need not page in anything else.  Then, sort the nodes of each
 * block in depth-first order.  That should give each lookup almost
 * 1/2 chance to find the next node immediately adjacent.
 *
 * With a block size of 1024 bytes, this renumbering reduces the time
 * required for random lookups by about 1.1%, compared to a plain
 * depth-first order.  However, it's still 2.3% slower than the
 * database optimized by MaxMind.  */
void
binary_trie::reorder_in_blocks(
	std::size_t bytes_per_block)
{
	const edge_type none = -1;
	std::vector<edge_type> old_to_new, new_to_old;
	ssize_t bytes_left = bytes_per_block;
	old_to_new.resize(nodes.size(), none);
	new_to_old.reserve(nodes.size());
	for (edge_type subtrie = 0; subtrie < nodes.size(); ++subtrie) {
		// If subtrie has already been added to the output,
		// ignore it.
		if (old_to_new[subtrie] != none)
			continue;

		// Walk breadth-first from subtrie until we have a
		// block full of nodes or the subtrie runs out.  Don't
		// add these nodes immediately to the output, however.
		// Instead just list them in nodes_in_block.
		std::set<edge_type> nodes_in_block;
		std::queue<edge_type> breadth_first;
		breadth_first.push(subtrie);
		if (bytes_left <= 0)
			bytes_left += bytes_per_block;
		while (bytes_left > 0 && !breadth_first.empty()) {
			edge_type edge = breadth_first.front();
			breadth_first.pop();
			if (edge >= nodes.size())
				continue;

			// Let the last node of the block straddle the
			// block boundary.  That's better than making
			// the hotter first node do so.
			bytes_left -= 6;
			nodes_in_block.insert(edge);

			breadth_first.push(nodes[edge].edges[0]);
			breadth_first.push(nodes[edge].edges[1]);
		}

		// Add the nodes from nodes_in_block to the output in
		// depth-first order.  This assumes they are all
		// reachable from subtrie.
		std::stack<edge_type> depth_first;
		depth_first.push(subtrie);
		while (!depth_first.empty()) {
			edge_type edge = depth_first.top();
			depth_first.pop();
			if (nodes_in_block.find(edge)
			    == nodes_in_block.end())
				continue;

			old_to_new[edge] = new_to_old.size();
			new_to_old.push_back(edge);

			depth_first.push(nodes[edge].edges[1]);
			depth_first.push(nodes[edge].edges[0]);
		}
	}
	reorder(old_to_new, new_to_old);
}

void
binary_trie::reorder(
	const std::vector<edge_type> &old_to_new,
	const std::vector<edge_type> &new_to_old)
{
	std::vector<node> new_nodes;
	new_nodes.reserve(new_to_old.size());
	for (std::vector<edge_type>::const_iterator
		     it = new_to_old.begin();
	     it != new_to_old.end(); ++it) {
		node new_node;
		for (int bit = 0; bit <= 1; ++bit) {
			edge_type old_edge = nodes[*it].edges[bit];
			if (old_edge < nodes.size())
				new_node.edges[bit] = old_to_new[old_edge];
			else
				new_node.edges[bit] = old_edge;
		}
		new_nodes.push_back(new_node);
	}
	swap(new_nodes, nodes);
}

/** Add the size of the trie (number of nodes) to data records*/
void
binary_trie::update_records()
{
	for (std::vector<node>::iterator it = nodes.begin();
	     it != nodes.end(); ++it) {
	// previously, we commandeered the MSB in order to indicate which records
	// were data records, rather than pointers to other nodes in the trie.
	// Here, we remove that bit, and increment the record by the number of nodes,
	// because this is how libGeoIP determines whether a node points to an entry
	// inside the data section or another node.
		for (int i = 0;i<2;++i) {
			if (it->edges[i] == 0x80000000)
				it->edges[i] = nodes.size();
			else if (it->edges[i] & 0x80000000)
			{
				it->edges[i] = (it->edges[i] & 0x7FFFFFFF) + nodes.size();
			}
		}
	}
}

/** Write the 3 byte segment offset. **/
void
binary_trie::write_segment(std::ostream &dat_stream) const
{
	int len = nodes.size();
	dat_stream << (char) (0xFF & len);
	dat_stream << (char) (0xFF & (len >> 8));
	dat_stream << (char) (0xFF & (len >> 16));
}

/** Write the trie to a stream in GeoIP binary format.	*/
void
binary_trie::write_binary(std::ostream &dat_stream) const
{
	for (std::vector<node>::const_iterator it = nodes.begin();
	     it != nodes.end(); ++it) {
		union {
			uint8_t bytes[6];
			char chars[6];
		} binary = {{
			(it->edges[0]	   ) & 0xFF,
			(it->edges[0] >>  8) & 0xFF,
			(it->edges[0] >> 16) & 0xFF,
			(it->edges[1]	   ) & 0xFF,
			(it->edges[1] >>  8) & 0xFF,
			(it->edges[1] >> 16) & 0xFF
		}};
		dat_stream.write(binary.chars, 6);
		if (dat_stream.bad())
			return;
	}
}


namespace {
	void
	v4_csv_line_to_vector(
		const std::string line,
		std::vector<std::string> &fields)
	{
		std::string buf(line);
		std::string delim = ",";
		std::size_t fs;
		for(int i = 0; i<2;++i) {
			fs = buf.find(delim);
			fields.push_back(buf.substr(0,fs));
			buf.erase(0,fs + delim.length());
		}
		if (buf[0] == '"' || buf[0] == '\'')
		fields.push_back(buf.substr(1, buf.length() - 2));
		else
			fields.push_back(buf);
	}

	void
	v6_csv_line_to_vector(
		const std::string line,
		std::vector<std::string> & fields)
	{
		std::string buf(line);
		std::string delim = ",";
		std::size_t fs;
		for(int i = 0; i<3;++i) {
			fs = buf.rfind(delim);
			fields.push_back(buf.substr(fs+delim.length(), buf.length()));
			buf.erase(fs,buf.length());
		}
		if (buf[0] == '"' || buf[0] == '\'')
			fields.push_back(buf.substr(1, buf.length() - 2));
		else
		fields.push_back(buf.substr(0, buf.length()));
	}

	/** Load ranges of IP addresses from a CSV-formatted stream to
	 * a trie.
	 *
	 * \param trie
	 * Load the ranges to this trie, overwriting original values.
	 *
	 * \param csv_file_name
	 * The name of the file that \a csv_stream is reading.
	 * This string is used only for error messages.
	 *
	 * \param csv_stream
	 * Load the ranges from this stream.
	 *
	 * \param address_family
	 * The type of IP addresses in the CSV data: either AF_INET * for IPv4 or AF_INET6 for IPv6.  */
	void
	csv_stream_to_trie_db(
		binary_trie &trie,
		std::string &database_segment,
		const char *csv_file_name,
		std::istream &csv_stream,
		int address_family)
	{
		enum {
			V4_CSV_FIELD_MIN_DECIMAL,
			V4_CSV_FIELD_MAX_DECIMAL,
			V4_CSV_FIELD_ASNUM_DESCRIPTION,
			V4_CSV_FIELDS
		};

		enum {
			V6_CSV_FIELD_NET_BITS,
			V6_CSV_FIELD_MAX_TEXT,
			V6_CSV_FIELD_MIN_TEXT,
			V6_CSV_FIELD_ASNUM_DESCRIPTION,
			V6_CSV_FIELDS
		};

		std::string csv_line;
		std::map<std::string,int> segment_offset;
		std::vector<std::string> csv_fields;
		// create a map to track which as descriptions are added already
		int csv_line_number = 0;
		database_segment += '\0'; // padding so that record 0 is not at the start of the db.
		while (getline(csv_stream, csv_line)) {
			++csv_line_number;
			std::string as;
			switch (address_family) {
			case AF_INET:
				v4_csv_line_to_vector(csv_line, csv_fields);
				if (csv_fields.size() != V4_CSV_FIELDS) {
					error_at_line(EX_DATAERR, 0, csv_file_name, csv_line_number,
						      "Wrong number of fields");
				}
				as = csv_fields[V4_CSV_FIELD_ASNUM_DESCRIPTION];
				if (segment_offset.find(as) == segment_offset.end()) { // no key found
					segment_offset[as] = database_segment.length(); // start of record
					database_segment += csv_fields[V4_CSV_FIELD_ASNUM_DESCRIPTION] + '\x00';
				}
				break;
			case AF_INET6:
				v6_csv_line_to_vector(csv_line, csv_fields);
				if (csv_fields.size() != V6_CSV_FIELDS) {
					error_at_line(EX_DATAERR, 0, csv_file_name, csv_line_number,
						      "Wrong number of fields");
				}
				as = csv_fields[V6_CSV_FIELD_ASNUM_DESCRIPTION];
				if (segment_offset.find(as) == segment_offset.end()) { // no key found
					segment_offset[as] = database_segment.length(); // start of record
					database_segment += csv_fields[V6_CSV_FIELD_ASNUM_DESCRIPTION] + '\x00';
				}
				break;
			default:
				abort();
			}

			// use the MSB to indicate that this is a data record. Later, the field
			// is set to the segment offset + the nubmer of nodes in the trie.
			const binary_trie::edge_type leaf = 0x80000000 | segment_offset[as];
			union {
				struct in_addr inet;
				uint8_t inetbytes[4];
				struct in6_addr inet6;
			} minaddr, maxaddr;

			switch (address_family) {
			case AF_INET:
				inet_aton(csv_fields[V4_CSV_FIELD_MIN_DECIMAL].c_str(), &(minaddr.inet));
				inet_aton(csv_fields[V4_CSV_FIELD_MAX_DECIMAL].c_str(), &(maxaddr.inet));
				trie.set_range(minaddr.inetbytes, maxaddr.inetbytes, 32, leaf);
				break;

			case AF_INET6:
				if (inet_pton(address_family, csv_fields[V6_CSV_FIELD_MIN_TEXT].c_str(), &minaddr) <= 0) {
					error_at_line(EX_DATAERR, 0, csv_file_name, csv_line_number,
						      "Cannot parse minimum address: %s",
						      csv_fields[V6_CSV_FIELD_MIN_TEXT].c_str());
				}
				if (inet_pton(address_family, csv_fields[V6_CSV_FIELD_MAX_TEXT].c_str(), &maxaddr) <= 0) {
					error_at_line(EX_DATAERR, 0, csv_file_name, csv_line_number,
						      "Cannot parse maximum address: %s",
						      csv_fields[V6_CSV_FIELD_MAX_TEXT].c_str());
				}
				trie.set_range(minaddr.inet6.s6_addr, maxaddr.inet6.s6_addr,
					       128, leaf);
				break;

			default:
				abort();
			}
			csv_fields.clear();
		}
		if (csv_stream.bad()) {
			error(EX_IOERR, errno, "%s", csv_file_name);
		}
	}

	/** Load ranges of IP addresses from a CSV-formatted file or
	 * standard input to a trie.
	 *
	 * \param trie
	 * Load the ranges to this trie, overwriting original values.
	 *
	 * \param csv_file_name
	 * The name of the CSV file that should be read, or "-" for
	 * standard input.
	 *
	 * \param address_family
	 * The type of IP addresses in the CSV data: either AF_INET
	 * for IPv4 or AF_INET6 for IPv6.  */
	void
	csv_file_to_trie_db(
		binary_trie &trie,
		std::string &database_segment,
		const char *csv_file_name,
		int address_family)
	{
		if (std::strcmp(csv_file_name, "-") == 0) {
			csv_stream_to_trie_db(trie, database_segment, csv_file_name, std::cin, address_family);
		} else {
			std::ifstream csv_stream(csv_file_name, std::ios::in);
			if (!csv_stream) {
				error(EX_NOINPUT, errno, "%s", csv_file_name);
			}
			csv_stream_to_trie_db(trie, database_segment, csv_file_name, csv_stream, address_family);
		}
	}

	/** Write a GeoIP binary database to a stream.
	 *
	 * \param trie
	 * Mapping from IP addresses to country codes or other values.
	 *
	 * \param dat_file_name
	 * The name of the file that \a dat_stream is writing.
	 * This string is used only for error messages.
	 *
	 * \param dat_stream
	 * Write the database to this stream.
	 *
	 * \param database_info
	 * Copyright or other information about the database, or NULL.
	 * GeoIP_database_info() will return this string.
	 *
	 * \param address_family
	 * The type of IP addresses in the database: either AF_INET
	 * for IPv4 or AF_INET6 for IPv6.  */
	void
	write_dat_stream(
		const binary_trie &trie,
		const char *dat_file_name,
		std::ostream &dat_stream,
		const char *database_info,
		std::string database_segment,
		int address_family)
	{

		trie.write_binary(dat_stream);
		if (dat_stream.bad()) {
			error(EX_IOERR, errno, "%s", dat_file_name);
		}

		// or open the file and read the length
		if (database_segment.length() > 0) {
			dat_stream << database_segment;
			if (dat_stream.bad()) {
				error(EX_IOERR, errno, "%s", dat_file_name);
			}

		}

		// write the metadata section
		if (database_info) {
			const char tag[3] = { 0, 0, 0 };
			dat_stream.write(tag, 3);
			dat_stream.write(database_info, std::strlen(database_info));
			if (dat_stream.bad()) {
				error(EX_IOERR, errno, "%s", dat_file_name);
			}
		}

		switch (address_family) {
		case AF_INET: {
			const unsigned char structure_info[4] = { 0xFF, 0xFF, 0xFF, 9 };
			dat_stream.write((const char *)structure_info, 4);
			break;
		}
		case AF_INET6: {
			const unsigned char structure_info[4] = { 0xFF, 0xFF, 0xFF, 21 };
			dat_stream.write((const char *)structure_info, 4);
			break;
		}
		default:
			abort();
		}

		trie.write_segment(dat_stream);
		if (dat_stream.bad()) {
			error(EX_IOERR, errno, "%s", dat_file_name);
		}
	}

	/** Write a GeoIP binary database to a file or standard output.
	 *
	 * \param trie
	 * Mapping from IP addresses to country codes or other values.
	 *
	 * \param csv_file_name
	 * The name of the file that should be written, or "-" for
	 * standard output.
	 *
	 * \param database_info
	 * Copyright or other information about the database, or NULL.
	 * GeoIP_database_info() will return this string.
	 *
	 * \param address_family
	 * The type of IP addresses in the database: either AF_INET
	 * for IPv4 or AF_INET6 for IPv6.  */
	void
	write_dat_file(
		const binary_trie &trie,
		const char *dat_file_name,
		const char *database_info,
		std::string  database_segment,
		int address_family)
	{
		if (std::strcmp(dat_file_name, "-") == 0) {
			write_dat_stream(trie, dat_file_name, std::cout,
					 database_info, database_segment, address_family);
		} else {
			std::ofstream dat_stream(
				dat_file_name,
				std::ios::out | std::ios::binary);
			if (!dat_stream) {
				error(EX_CANTCREAT, errno, "%s", dat_file_name);
			}
			write_dat_stream(trie, dat_file_name, dat_stream,
					 database_info, database_segment, address_family);
		}
	}

	struct cmdline {
		const char *csv_file_name;
		const char *dat_file_name;
		int address_family;
		const char *database_info;
		bool verbose;

		cmdline(int argc, char **argv);
	};
}

cmdline::cmdline(int argc, char **argv):
	csv_file_name("-"),
	dat_file_name("-"),
	address_family(AF_INET),
	database_info(NULL),
	verbose(false)
{
	enum {
		OPT_HELP = -2
	};

	static const struct option long_options[] = {
		{ "info", required_argument, NULL, 'i' },
		{ "output", required_argument, NULL, 'o' },
		{ "verbose", no_argument, NULL, 'v' },
		{ "help", no_argument, NULL, OPT_HELP },
		{ NULL, 0, NULL, 0 }
	};
	static const char *const usage = "\
Usage: %s [OPTION] [CSV-FILE]...\n\
Convert a country database from CSV to GeoIP binary format.\n\
\n\
  -i, --info=TEXT       add copyright or other info TEXT to output\n\
  -o, --output=FILE     write the binary data to FILE, not stdout\n\
  -v, --verbose         show what is going on\n\
      --help            display this help and exit\n";

	for (;;) {
		int optret = getopt_long(argc, argv, "46i:o:v", long_options, NULL);

		if (optret == -1)
			break;
		switch (optret) {
		case '4':
			address_family = AF_INET;
			break;
		case '6':
			address_family = AF_INET6;
			break;
		case 'i':
			database_info = optarg;
			if (std::strlen(database_info) > 99) {
				error(EX_USAGE, 0,
				      "Database info must not be longer than 99 bytes");
			}
			break;
		case 'o':
			dat_file_name = optarg;
			break;
		case 'v':
			verbose = true;
			break;
		case OPT_HELP:
			std::printf(usage, program_invocation_name);
			std::exit(EX_OK);
		case '?':
			std::fprintf(stderr,
				     "Try `%s --help' for more information.\n",
				     program_invocation_name);
			std::exit(EX_USAGE);
		default:
			std::abort();
		}
	}

	if (optind < argc)
		csv_file_name = argv[optind++];

	if (optind < argc) {
		error(EX_USAGE, 0,
		      "Only one non-option argument is allowed");
	}
}

int
main(int argc, char **argv)
{
	cmdline cmdline(argc, argv);

	std::ostream *verbose_stream;
	if (!cmdline.verbose)
		verbose_stream = NULL;
	else if (strcmp(cmdline.dat_file_name, "-") == 0)
		verbose_stream = &std::cerr;
	else
		verbose_stream = &std::cout;

	if (verbose_stream) {
		*verbose_stream << program_invocation_name
				<< ": Reading CSV and building the trie"
				<< std::endl;
	}
	/* Initialize the trie with a value that will point to the start of the
	 * data section, e.g. an empty record. See binary_trie::update_records()
	 */
	binary_trie trie(0x80000000);
	std::string database_segment;
	csv_file_to_trie_db(trie, database_segment, cmdline.csv_file_name, cmdline.address_family);

	if (verbose_stream) {
		*verbose_stream << program_invocation_name
				<< ": Optimizing" << std::endl;
	}
	trie.reorder_depth_first();
	trie.reorder_in_blocks(1024);
	trie.update_records();

	if (verbose_stream) {
		*verbose_stream << program_invocation_name
				<< ": Writing output" << std::endl;
	}
	write_dat_file(trie, cmdline.dat_file_name, cmdline.database_info,
			database_segment, cmdline.address_family);

	if (verbose_stream) {
		*verbose_stream << program_invocation_name
				<< ": All done" << std::endl; }
}