diff options
author | Antonio Juarez <antonio.maria.juarez@live.com> | 2014-03-03 22:07:58 +0000 |
---|---|---|
committer | Antonio Juarez <antonio.maria.juarez@live.com> | 2014-03-03 22:07:58 +0000 |
commit | 296ae46ed8f8f6e5f986f978febad302e3df231a (patch) | |
tree | 1629164454a239308f33c9e12afb22e7f3cd8eeb /src/common/base58.cpp | |
parent | changed name (diff) | |
download | monero-296ae46ed8f8f6e5f986f978febad302e3df231a.tar.xz |
moved all stuff to github
Diffstat (limited to 'src/common/base58.cpp')
-rw-r--r-- | src/common/base58.cpp | 246 |
1 files changed, 246 insertions, 0 deletions
diff --git a/src/common/base58.cpp b/src/common/base58.cpp new file mode 100644 index 000000000..30042eeba --- /dev/null +++ b/src/common/base58.cpp @@ -0,0 +1,246 @@ +// Copyright (c) 2012-2013 The Cryptonote developers +// Distributed under the MIT/X11 software license, see the accompanying +// file COPYING or http://www.opensource.org/licenses/mit-license.php. + + +#include "base58.h" + +#include <assert.h> +#include <string> +#include <vector> + +#include "crypto/hash.h" +#include "int-util.h" +#include "util.h" +#include "varint.h" + +namespace tools +{ + namespace base58 + { + namespace + { + const char alphabet[] = "123456789ABCDEFGHJKLMNPQRSTUVWXYZabcdefghijkmnopqrstuvwxyz"; + const size_t alphabet_size = sizeof(alphabet) - 1; + const size_t encoded_block_sizes[] = {0, 2, 3, 5, 6, 7, 9, 10, 11}; + const size_t full_block_size = sizeof(encoded_block_sizes) / sizeof(encoded_block_sizes[0]) - 1; + const size_t full_encoded_block_size = encoded_block_sizes[full_block_size]; + const size_t addr_checksum_size = 4; + + struct reverse_alphabet + { + reverse_alphabet() + { + m_data.resize(alphabet[alphabet_size - 1] - alphabet[0] + 1, -1); + + for (size_t i = 0; i < alphabet_size; ++i) + { + size_t idx = static_cast<size_t>(alphabet[i] - alphabet[0]); + m_data[idx] = static_cast<int8_t>(i); + } + } + + int operator()(char letter) const + { + size_t idx = static_cast<size_t>(letter - alphabet[0]); + return idx < m_data.size() ? m_data[idx] : -1; + } + + static reverse_alphabet instance; + + private: + std::vector<int8_t> m_data; + }; + + reverse_alphabet reverse_alphabet::instance; + + struct decoded_block_sizes + { + decoded_block_sizes() + { + m_data.resize(encoded_block_sizes[full_block_size] + 1, -1); + for (size_t i = 0; i <= full_block_size; ++i) + { + m_data[encoded_block_sizes[i]] = static_cast<int>(i); + } + } + + int operator()(size_t encoded_block_size) const + { + assert(encoded_block_size <= full_encoded_block_size); + return m_data[encoded_block_size]; + } + + static decoded_block_sizes instance; + + private: + std::vector<int> m_data; + }; + + decoded_block_sizes decoded_block_sizes::instance; + + uint64_t uint_8be_to_64(const uint8_t* data, size_t size) + { + assert(1 <= size && size <= sizeof(uint64_t)); + + uint64_t res = 0; + switch (9 - size) + { + case 1: res |= *data++; + case 2: res <<= 8; res |= *data++; + case 3: res <<= 8; res |= *data++; + case 4: res <<= 8; res |= *data++; + case 5: res <<= 8; res |= *data++; + case 6: res <<= 8; res |= *data++; + case 7: res <<= 8; res |= *data++; + case 8: res <<= 8; res |= *data; break; + default: assert(false); + } + + return res; + } + + void uint_64_to_8be(uint64_t num, size_t size, uint8_t* data) + { + assert(1 <= size && size <= sizeof(uint64_t)); + + uint64_t num_be = SWAP64BE(num); + memcpy(data, reinterpret_cast<uint8_t*>(&num_be) + sizeof(uint64_t) - size, size); + } + + void encode_block(const char* block, size_t size, char* res) + { + assert(1 <= size && size <= sizeof(full_block_size)); + + uint64_t num = uint_8be_to_64(reinterpret_cast<const uint8_t*>(block), size); + int i = static_cast<int>(encoded_block_sizes[size]) - 1; + while (0 < num) + { + uint64_t remainder = num % alphabet_size; + num /= alphabet_size; + res[i] = alphabet[remainder]; + --i; + } + } + + bool decode_block(const char* block, size_t size, char* res) + { + assert(1 <= size && size <= full_encoded_block_size); + + int res_size = decoded_block_sizes::instance(size); + if (res_size <= 0) + return false; // Invalid block size + + uint64_t res_num = 0; + uint64_t order = 1; + for (size_t i = size - 1; i < size; --i) + { + int digit = reverse_alphabet::instance(block[i]); + if (digit < 0) + return false; // Invalid symbol + + uint64_t product_hi; + uint64_t tmp = res_num + mul128(order, digit, &product_hi); + if (tmp < res_num || 0 != product_hi) + return false; // Overflow + + res_num = tmp; + order *= alphabet_size; // Never overflows, 58^10 < 2^64 + } + + if (static_cast<size_t>(res_size) < full_block_size && (UINT64_C(1) << (8 * res_size)) <= res_num) + return false; // Overflow + + uint_64_to_8be(res_num, res_size, reinterpret_cast<uint8_t*>(res)); + + return true; + } + } + + std::string encode(const std::string& data) + { + if (data.empty()) + return std::string(); + + size_t full_block_count = data.size() / full_block_size; + size_t last_block_size = data.size() % full_block_size; + size_t res_size = full_block_count * full_encoded_block_size + encoded_block_sizes[last_block_size]; + + std::string res(res_size, alphabet[0]); + for (size_t i = 0; i < full_block_count; ++i) + { + encode_block(data.data() + i * full_block_size, full_block_size, &res[i * full_encoded_block_size]); + } + + if (0 < last_block_size) + { + encode_block(data.data() + full_block_count * full_block_size, last_block_size, &res[full_block_count * full_encoded_block_size]); + } + + return res; + } + + bool decode(const std::string& enc, std::string& data) + { + if (enc.empty()) + { + data.clear(); + return true; + } + + size_t full_block_count = enc.size() / full_encoded_block_size; + size_t last_block_size = enc.size() % full_encoded_block_size; + int last_block_decoded_size = decoded_block_sizes::instance(last_block_size); + if (last_block_decoded_size < 0) + return false; // Invalid enc length + size_t data_size = full_block_count * full_block_size + last_block_decoded_size; + + data.resize(data_size, 0); + for (size_t i = 0; i < full_block_count; ++i) + { + if (!decode_block(enc.data() + i * full_encoded_block_size, full_encoded_block_size, &data[i * full_block_size])) + return false; + } + + if (0 < last_block_size) + { + if (!decode_block(enc.data() + full_block_count * full_encoded_block_size, last_block_size, + &data[full_block_count * full_block_size])) + return false; + } + + return true; + } + + std::string encode_addr(uint64_t tag, const std::string& data) + { + std::string buf = get_varint_data(tag); + buf += data; + crypto::hash hash = crypto::cn_fast_hash(buf.data(), buf.size()); + const char* hash_data = reinterpret_cast<const char*>(&hash); + buf.append(hash_data, addr_checksum_size); + return encode(buf); + } + + bool decode_addr(std::string addr, uint64_t& tag, std::string& data) + { + std::string addr_data; + bool r = decode(addr, addr_data); + if (!r) return false; + + std::string checksum(addr_checksum_size, '\0'); + checksum = addr_data.substr(addr_data.size() - addr_checksum_size); + + addr_data.resize(addr_data.size() - addr_checksum_size); + crypto::hash hash = crypto::cn_fast_hash(addr_data.data(), addr_data.size()); + std::string expected_checksum(reinterpret_cast<const char*>(&hash), addr_checksum_size); + if (expected_checksum != checksum) return false; + + int read = tools::read_varint(addr_data.begin(), addr_data.end(), tag); + if (read <= 0) return false; + + data = addr_data.substr(read); + return true; + } + } +} |