Aes.c 8.1 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295
  1. /* Aes.c -- AES encryption / decryption
  2. 2016-05-21 : Igor Pavlov : Public domain */
  3. #include "Precomp.h"
  4. #include "Aes.h"
  5. #include "CpuArch.h"
  6. static UInt32 T[256 * 4];
  7. static const Byte Sbox[256] = {
  8. 0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5, 0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76,
  9. 0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0, 0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0,
  10. 0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc, 0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15,
  11. 0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a, 0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75,
  12. 0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0, 0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84,
  13. 0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b, 0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf,
  14. 0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85, 0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8,
  15. 0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5, 0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2,
  16. 0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17, 0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73,
  17. 0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88, 0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb,
  18. 0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c, 0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79,
  19. 0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9, 0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08,
  20. 0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6, 0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a,
  21. 0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e, 0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e,
  22. 0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94, 0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf,
  23. 0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68, 0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16};
  24. void MY_FAST_CALL AesCbc_Encode(UInt32 *ivAes, Byte *data, size_t numBlocks);
  25. void MY_FAST_CALL AesCbc_Decode(UInt32 *ivAes, Byte *data, size_t numBlocks);
  26. void MY_FAST_CALL AesCtr_Code(UInt32 *ivAes, Byte *data, size_t numBlocks);
  27. void MY_FAST_CALL AesCbc_Encode_Intel(UInt32 *ivAes, Byte *data, size_t numBlocks);
  28. void MY_FAST_CALL AesCbc_Decode_Intel(UInt32 *ivAes, Byte *data, size_t numBlocks);
  29. void MY_FAST_CALL AesCtr_Code_Intel(UInt32 *ivAes, Byte *data, size_t numBlocks);
  30. AES_CODE_FUNC g_AesCbc_Encode;
  31. AES_CODE_FUNC g_AesCbc_Decode;
  32. AES_CODE_FUNC g_AesCtr_Code;
  33. static UInt32 D[256 * 4];
  34. static Byte InvS[256];
  35. static const Byte Rcon[11] = { 0x00, 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x1b, 0x36 };
  36. #define xtime(x) ((((x) << 1) ^ (((x) & 0x80) != 0 ? 0x1B : 0)) & 0xFF)
  37. #define Ui32(a0, a1, a2, a3) ((UInt32)(a0) | ((UInt32)(a1) << 8) | ((UInt32)(a2) << 16) | ((UInt32)(a3) << 24))
  38. #define gb0(x) ( (x) & 0xFF)
  39. #define gb1(x) (((x) >> ( 8)) & 0xFF)
  40. #define gb2(x) (((x) >> (16)) & 0xFF)
  41. #define gb3(x) (((x) >> (24)) & 0xFF)
  42. void AesGenTables(void)
  43. {
  44. unsigned i;
  45. for (i = 0; i < 256; i++)
  46. InvS[Sbox[i]] = (Byte)i;
  47. for (i = 0; i < 256; i++)
  48. {
  49. {
  50. UInt32 a1 = Sbox[i];
  51. UInt32 a2 = xtime(a1);
  52. UInt32 a3 = a2 ^ a1;
  53. T[ i] = Ui32(a2, a1, a1, a3);
  54. T[0x100 + i] = Ui32(a3, a2, a1, a1);
  55. T[0x200 + i] = Ui32(a1, a3, a2, a1);
  56. T[0x300 + i] = Ui32(a1, a1, a3, a2);
  57. }
  58. {
  59. UInt32 a1 = InvS[i];
  60. UInt32 a2 = xtime(a1);
  61. UInt32 a4 = xtime(a2);
  62. UInt32 a8 = xtime(a4);
  63. UInt32 a9 = a8 ^ a1;
  64. UInt32 aB = a8 ^ a2 ^ a1;
  65. UInt32 aD = a8 ^ a4 ^ a1;
  66. UInt32 aE = a8 ^ a4 ^ a2;
  67. D[ i] = Ui32(aE, a9, aD, aB);
  68. D[0x100 + i] = Ui32(aB, aE, a9, aD);
  69. D[0x200 + i] = Ui32(aD, aB, aE, a9);
  70. D[0x300 + i] = Ui32(a9, aD, aB, aE);
  71. }
  72. }
  73. g_AesCbc_Encode = AesCbc_Encode;
  74. g_AesCbc_Decode = AesCbc_Decode;
  75. g_AesCtr_Code = AesCtr_Code;
  76. #ifdef MY_CPU_X86_OR_AMD64
  77. if (CPU_Is_Aes_Supported())
  78. {
  79. g_AesCbc_Encode = AesCbc_Encode_Intel;
  80. g_AesCbc_Decode = AesCbc_Decode_Intel;
  81. g_AesCtr_Code = AesCtr_Code_Intel;
  82. }
  83. #endif
  84. }
  85. #define HT(i, x, s) (T + (x << 8))[gb ## x(s[(i + x) & 3])]
  86. #define HT4(m, i, s, p) m[i] = \
  87. HT(i, 0, s) ^ \
  88. HT(i, 1, s) ^ \
  89. HT(i, 2, s) ^ \
  90. HT(i, 3, s) ^ w[p + i]
  91. #define HT16(m, s, p) \
  92. HT4(m, 0, s, p); \
  93. HT4(m, 1, s, p); \
  94. HT4(m, 2, s, p); \
  95. HT4(m, 3, s, p); \
  96. #define FT(i, x) Sbox[gb ## x(m[(i + x) & 3])]
  97. #define FT4(i) dest[i] = Ui32(FT(i, 0), FT(i, 1), FT(i, 2), FT(i, 3)) ^ w[i];
  98. #define HD(i, x, s) (D + (x << 8))[gb ## x(s[(i - x) & 3])]
  99. #define HD4(m, i, s, p) m[i] = \
  100. HD(i, 0, s) ^ \
  101. HD(i, 1, s) ^ \
  102. HD(i, 2, s) ^ \
  103. HD(i, 3, s) ^ w[p + i];
  104. #define HD16(m, s, p) \
  105. HD4(m, 0, s, p); \
  106. HD4(m, 1, s, p); \
  107. HD4(m, 2, s, p); \
  108. HD4(m, 3, s, p); \
  109. #define FD(i, x) InvS[gb ## x(m[(i - x) & 3])]
  110. #define FD4(i) dest[i] = Ui32(FD(i, 0), FD(i, 1), FD(i, 2), FD(i, 3)) ^ w[i];
  111. void MY_FAST_CALL Aes_SetKey_Enc(UInt32 *w, const Byte *key, unsigned keySize)
  112. {
  113. unsigned i, wSize;
  114. wSize = keySize + 28;
  115. keySize /= 4;
  116. w[0] = ((UInt32)keySize / 2) + 3;
  117. w += 4;
  118. for (i = 0; i < keySize; i++, key += 4)
  119. w[i] = GetUi32(key);
  120. for (; i < wSize; i++)
  121. {
  122. UInt32 t = w[i - 1];
  123. unsigned rem = i % keySize;
  124. if (rem == 0)
  125. t = Ui32(Sbox[gb1(t)] ^ Rcon[i / keySize], Sbox[gb2(t)], Sbox[gb3(t)], Sbox[gb0(t)]);
  126. else if (keySize > 6 && rem == 4)
  127. t = Ui32(Sbox[gb0(t)], Sbox[gb1(t)], Sbox[gb2(t)], Sbox[gb3(t)]);
  128. w[i] = w[i - keySize] ^ t;
  129. }
  130. }
  131. void MY_FAST_CALL Aes_SetKey_Dec(UInt32 *w, const Byte *key, unsigned keySize)
  132. {
  133. unsigned i, num;
  134. Aes_SetKey_Enc(w, key, keySize);
  135. num = keySize + 20;
  136. w += 8;
  137. for (i = 0; i < num; i++)
  138. {
  139. UInt32 r = w[i];
  140. w[i] =
  141. D[ (unsigned)Sbox[gb0(r)]] ^
  142. D[0x100 + (unsigned)Sbox[gb1(r)]] ^
  143. D[0x200 + (unsigned)Sbox[gb2(r)]] ^
  144. D[0x300 + (unsigned)Sbox[gb3(r)]];
  145. }
  146. }
  147. /* Aes_Encode and Aes_Decode functions work with little-endian words.
  148. src and dest are pointers to 4 UInt32 words.
  149. src and dest can point to same block */
  150. static void Aes_Encode(const UInt32 *w, UInt32 *dest, const UInt32 *src)
  151. {
  152. UInt32 s[4];
  153. UInt32 m[4];
  154. UInt32 numRounds2 = w[0];
  155. w += 4;
  156. s[0] = src[0] ^ w[0];
  157. s[1] = src[1] ^ w[1];
  158. s[2] = src[2] ^ w[2];
  159. s[3] = src[3] ^ w[3];
  160. w += 4;
  161. for (;;)
  162. {
  163. HT16(m, s, 0);
  164. if (--numRounds2 == 0)
  165. break;
  166. HT16(s, m, 4);
  167. w += 8;
  168. }
  169. w += 4;
  170. FT4(0); FT4(1); FT4(2); FT4(3);
  171. }
  172. static void Aes_Decode(const UInt32 *w, UInt32 *dest, const UInt32 *src)
  173. {
  174. UInt32 s[4];
  175. UInt32 m[4];
  176. UInt32 numRounds2 = w[0];
  177. w += 4 + numRounds2 * 8;
  178. s[0] = src[0] ^ w[0];
  179. s[1] = src[1] ^ w[1];
  180. s[2] = src[2] ^ w[2];
  181. s[3] = src[3] ^ w[3];
  182. for (;;)
  183. {
  184. w -= 8;
  185. HD16(m, s, 4);
  186. if (--numRounds2 == 0)
  187. break;
  188. HD16(s, m, 0);
  189. }
  190. FD4(0); FD4(1); FD4(2); FD4(3);
  191. }
  192. void AesCbc_Init(UInt32 *p, const Byte *iv)
  193. {
  194. unsigned i;
  195. for (i = 0; i < 4; i++)
  196. p[i] = GetUi32(iv + i * 4);
  197. }
  198. void MY_FAST_CALL AesCbc_Encode(UInt32 *p, Byte *data, size_t numBlocks)
  199. {
  200. for (; numBlocks != 0; numBlocks--, data += AES_BLOCK_SIZE)
  201. {
  202. p[0] ^= GetUi32(data);
  203. p[1] ^= GetUi32(data + 4);
  204. p[2] ^= GetUi32(data + 8);
  205. p[3] ^= GetUi32(data + 12);
  206. Aes_Encode(p + 4, p, p);
  207. SetUi32(data, p[0]);
  208. SetUi32(data + 4, p[1]);
  209. SetUi32(data + 8, p[2]);
  210. SetUi32(data + 12, p[3]);
  211. }
  212. }
  213. void MY_FAST_CALL AesCbc_Decode(UInt32 *p, Byte *data, size_t numBlocks)
  214. {
  215. UInt32 in[4], out[4];
  216. for (; numBlocks != 0; numBlocks--, data += AES_BLOCK_SIZE)
  217. {
  218. in[0] = GetUi32(data);
  219. in[1] = GetUi32(data + 4);
  220. in[2] = GetUi32(data + 8);
  221. in[3] = GetUi32(data + 12);
  222. Aes_Decode(p + 4, out, in);
  223. SetUi32(data, p[0] ^ out[0]);
  224. SetUi32(data + 4, p[1] ^ out[1]);
  225. SetUi32(data + 8, p[2] ^ out[2]);
  226. SetUi32(data + 12, p[3] ^ out[3]);
  227. p[0] = in[0];
  228. p[1] = in[1];
  229. p[2] = in[2];
  230. p[3] = in[3];
  231. }
  232. }
  233. void MY_FAST_CALL AesCtr_Code(UInt32 *p, Byte *data, size_t numBlocks)
  234. {
  235. for (; numBlocks != 0; numBlocks--)
  236. {
  237. UInt32 temp[4];
  238. Byte buf[16];
  239. int i;
  240. if (++p[0] == 0)
  241. p[1]++;
  242. Aes_Encode(p + 4, temp, p);
  243. SetUi32(buf, temp[0]);
  244. SetUi32(buf + 4, temp[1]);
  245. SetUi32(buf + 8, temp[2]);
  246. SetUi32(buf + 12, temp[3]);
  247. for (i = 0; i < 16; i++)
  248. *data++ ^= buf[i];
  249. }
  250. }