base64.cpp

Go to the documentation of this file.

/*
* Base64 Encoding and Decoding
* (C) 2010,2015,2020 Jack Lloyd
*
* Botan is released under the Simplified BSD License (see license.txt)
*/
 
#include <botan/base64.h>
 
#include <botan/exceptn.h>
#include <botan/internal/charset.h>
#include <botan/internal/codec_base.h>
#include <botan/internal/fmt.h>
#include <botan/internal/int_utils.h>
#include <botan/internal/loadstor.h>
#include <botan/internal/rounding.h>
 
namespace Botan {
 
namespace {
 
class Base64 final {
   public:
      static std::string name() { return "base64"; }
 
      static constexpr size_t encoding_bytes_in() noexcept { return m_encoding_bytes_in; }
 
      static constexpr size_t encoding_bytes_out() noexcept { return m_encoding_bytes_out; }
 
      static constexpr size_t decoding_bytes_in() noexcept { return m_encoding_bytes_out; }
 
      static constexpr size_t decoding_bytes_out() noexcept { return m_encoding_bytes_in; }
 
      static constexpr size_t bits_consumed() noexcept { return m_encoding_bits; }
 
      static constexpr size_t remaining_bits_before_padding() noexcept { return m_remaining_bits_before_padding; }
 
      static constexpr size_t encode_max_output(size_t input_length) {
         const size_t encoding_blocks = round_up(input_length, m_encoding_bytes_in) / m_encoding_bytes_in;
         return mul_or_throw(encoding_blocks, m_encoding_bytes_out, "Input too large to base64 encode");
      }
 
      static constexpr size_t decode_max_output(size_t input_length) {
         // Divide before multiply to avoid overflow; round_up makes the division exact.
         return (round_up(input_length, m_encoding_bytes_out) / m_encoding_bytes_out) * m_encoding_bytes_in;
      }
 
      static void encode(char out[4], const uint8_t in[3]) noexcept;
 
      static uint8_t lookup_binary_value(char input) noexcept;
 
      static bool check_bad_char(uint8_t bin, char input, bool ignore_ws);
 
      static void decode(uint8_t* out_ptr, const uint8_t decode_buf[4]) {
         out_ptr[0] = (decode_buf[0] << 2) | (decode_buf[1] >> 4);
         out_ptr[1] = (decode_buf[1] << 4) | (decode_buf[2] >> 2);
         out_ptr[2] = (decode_buf[2] << 6) | decode_buf[3];
      }
 
      static size_t bytes_to_remove(size_t final_truncate) { return final_truncate; }
 
   private:
      static constexpr size_t m_encoding_bits = 6;
      static constexpr size_t m_remaining_bits_before_padding = 8;
 
      static constexpr size_t m_encoding_bytes_in = 3;
      static constexpr size_t m_encoding_bytes_out = 4;
};
 
uint32_t lookup_base64_chars(uint32_t x32) {
   /*
   * The basic insight of this approach is that our goal is computing
   * f(x) = y where x is in [0,63) and y is the correct base64 encoding.
   *
   * Instead of doing this directly, we compute
   * offset(x) such that f(x) = x + offset(x)
   *
   * This is described in
   * http://0x80.pl/notesen/2016-01-12-sse-base64-encoding.html#improved-version
   *
   * Here we do a SWAR (simd within a register) implementation of Wojciech's lookup_version2_swar
   */
 
   uint32_t r = x32 + 0x41414141;
 
   r += (~swar_lt<uint32_t>(x32, 0x1A1A1A1A)) & 0x06060606;
   r -= (~swar_lt<uint32_t>(x32, 0x34343434)) & 0x4B4B4B4B;
   r -= (~swar_lt<uint32_t>(x32, 0x3E3E3E3E)) & 0x0F0F0F0F;
   r += (~swar_lt<uint32_t>(x32, 0x3F3F3F3F)) & 0x03030303;
 
   return r;
}
 
//static
void Base64::encode(char out[4], const uint8_t in[3]) noexcept {
   const uint32_t b0 = (in[0] & 0xFC) >> 2;
   const uint32_t b1 = ((in[0] & 0x03) << 4) | (in[1] >> 4);
   const uint32_t b2 = ((in[1] & 0x0F) << 2) | (in[2] >> 6);
   const uint32_t b3 = in[2] & 0x3F;
 
   const uint32_t z = (b0 << 24) | (b1 << 16) | (b2 << 8) | b3;
 
   const uint32_t b64 = lookup_base64_chars(z);
 
   out[0] = static_cast<char>(get_byte<0>(b64));
   out[1] = static_cast<char>(get_byte<1>(b64));
   out[2] = static_cast<char>(get_byte<2>(b64));
   out[3] = static_cast<char>(get_byte<3>(b64));
}
 
//static
uint8_t Base64::lookup_binary_value(char input) noexcept {
   auto has_zero_byte = [](uint64_t v) { return ((v - 0x0101010101010101) & ~(v) & 0x8080808080808080); };
 
   // Assumes each byte is either 0x00 or 0x80
   auto index_of_first_set_byte = [](uint64_t v) {
      return ((((v - 1) & 0x0101010101010101) * 0x0101010101010101) >> 56) - 1;
   };
 
   constexpr uint64_t lo = 0x0101010101010101;
 
   const uint8_t x = static_cast<uint8_t>(input);
 
   const uint64_t x8 = x * lo;
 
   // Defines the valid ASCII ranges of base64, except the special chars (below)
   constexpr uint64_t val_l = make_uint64(0, 0, 0, 0, 0, 'A', 'a', '0');
   constexpr uint64_t val_u = make_uint64(0, 0, 0, 0, 0, 26, 26, 10);
 
   // If x is in one of the ranges return a mask. Otherwise we xor in at the
   // high word which will be our invalid marker
   auto v_mask = swar_in_range<uint64_t>(x8, val_l, val_u) ^ 0x80000000;
 
   // This is the offset added to x to get the value
   const uint64_t val_v = 0xbfb904 ^ (0xFF000000 - (x << 24));
 
   const uint8_t z = x + static_cast<uint8_t>(val_v >> (8 * index_of_first_set_byte(v_mask)));
 
   // Valid base64 special characters, and some whitespace chars
   constexpr uint64_t specials_i = make_uint64(0, '+', '/', '=', ' ', '\n', '\t', '\r');
 
   const uint64_t specials_v = 0x3e3f8180808080 ^ (static_cast<uint64_t>(z) << 56);
 
   const uint64_t smask = has_zero_byte(x8 ^ specials_i) ^ 0x8000000000000000;
 
   return static_cast<uint8_t>(specials_v >> (8 * index_of_first_set_byte(smask)));
}
 
//static
bool Base64::check_bad_char(uint8_t bin, char input, bool ignore_ws) {
   if(bin <= 0x3F) {
      return true;
   } else if(!(bin == 0x81 || (bin == 0x80 && ignore_ws))) {
      throw Invalid_Argument(fmt("base64_decode: invalid character '{}'", format_char_for_display(input)));
   }
   return false;
}
 
}  // namespace
 
size_t base64_encode(char out[], const uint8_t in[], size_t input_length, size_t& input_consumed, bool final_inputs) {
   return base_encode(Base64(), out, in, input_length, input_consumed, final_inputs);
}
 
std::string base64_encode(const uint8_t input[], size_t input_length) {
   return base_encode_to_string(Base64(), input, input_length);
}
 
size_t base64_decode(
   uint8_t out[], const char in[], size_t input_length, size_t& input_consumed, bool final_inputs, bool ignore_ws) {
   return base_decode(Base64(), out, in, input_length, input_consumed, final_inputs, ignore_ws);
}
 
size_t base64_decode(uint8_t output[], const char input[], size_t input_length, bool ignore_ws) {
   return base_decode_full(Base64(), output, input, input_length, ignore_ws);
}
 
size_t base64_decode(uint8_t output[], std::string_view input, bool ignore_ws) {
   return base64_decode(output, input.data(), input.length(), ignore_ws);
}
 
size_t base64_decode(std::span<uint8_t> output, std::string_view input, bool ignore_ws) {
   if(output.size() < base64_decode_max_output(input.size())) {
      throw Invalid_Argument("base64_decode: output buffer is too short");
   }
   return base64_decode(output.data(), input.data(), input.length(), ignore_ws);
}
 
secure_vector<uint8_t> base64_decode(const char input[], size_t input_length, bool ignore_ws) {
   return base_decode_to_vec<secure_vector<uint8_t>>(Base64(), input, input_length, ignore_ws);
}
 
secure_vector<uint8_t> base64_decode(std::string_view input, bool ignore_ws) {
   return base64_decode(input.data(), input.size(), ignore_ws);
}
 
size_t base64_encode_max_output(size_t input_length) {
   return Base64::encode_max_output(input_length);
}
 
size_t base64_decode_max_output(size_t input_length) {
   return Base64::decode_max_output(input_length);
}
 
}  // namespace Botan