Home | History | Annotate | Download | only in C
      1 /* Aes.c -- AES encryption / decryption
      2 2016-05-21 : Igor Pavlov : Public domain */
      3 
      4 #include "Precomp.h"
      5 
      6 #include "Aes.h"
      7 #include "CpuArch.h"
      8 
      9 static UInt32 T[256 * 4];
     10 static const Byte Sbox[256] = {
     11   0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5, 0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76,
     12   0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0, 0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0,
     13   0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc, 0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15,
     14   0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a, 0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75,
     15   0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0, 0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84,
     16   0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b, 0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf,
     17   0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85, 0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8,
     18   0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5, 0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2,
     19   0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17, 0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73,
     20   0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88, 0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb,
     21   0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c, 0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79,
     22   0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9, 0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08,
     23   0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6, 0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a,
     24   0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e, 0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e,
     25   0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94, 0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf,
     26   0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68, 0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16};
     27 
     28 void MY_FAST_CALL AesCbc_Encode(UInt32 *ivAes, Byte *data, size_t numBlocks);
     29 void MY_FAST_CALL AesCbc_Decode(UInt32 *ivAes, Byte *data, size_t numBlocks);
     30 void MY_FAST_CALL AesCtr_Code(UInt32 *ivAes, Byte *data, size_t numBlocks);
     31 
     32 void MY_FAST_CALL AesCbc_Encode_Intel(UInt32 *ivAes, Byte *data, size_t numBlocks);
     33 void MY_FAST_CALL AesCbc_Decode_Intel(UInt32 *ivAes, Byte *data, size_t numBlocks);
     34 void MY_FAST_CALL AesCtr_Code_Intel(UInt32 *ivAes, Byte *data, size_t numBlocks);
     35 
     36 AES_CODE_FUNC g_AesCbc_Encode;
     37 AES_CODE_FUNC g_AesCbc_Decode;
     38 AES_CODE_FUNC g_AesCtr_Code;
     39 
     40 static UInt32 D[256 * 4];
     41 static Byte InvS[256];
     42 
     43 static const Byte Rcon[11] = { 0x00, 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x1b, 0x36 };
     44 
     45 #define xtime(x) ((((x) << 1) ^ (((x) & 0x80) != 0 ? 0x1B : 0)) & 0xFF)
     46 
     47 #define Ui32(a0, a1, a2, a3) ((UInt32)(a0) | ((UInt32)(a1) << 8) | ((UInt32)(a2) << 16) | ((UInt32)(a3) << 24))
     48 
     49 #define gb0(x) ( (x)          & 0xFF)
     50 #define gb1(x) (((x) >> ( 8)) & 0xFF)
     51 #define gb2(x) (((x) >> (16)) & 0xFF)
     52 #define gb3(x) (((x) >> (24)) & 0xFF)
     53 
     54 void AesGenTables(void)
     55 {
     56   unsigned i;
     57   for (i = 0; i < 256; i++)
     58     InvS[Sbox[i]] = (Byte)i;
     59 
     60   for (i = 0; i < 256; i++)
     61   {
     62     {
     63       UInt32 a1 = Sbox[i];
     64       UInt32 a2 = xtime(a1);
     65       UInt32 a3 = a2 ^ a1;
     66       T[        i] = Ui32(a2, a1, a1, a3);
     67       T[0x100 + i] = Ui32(a3, a2, a1, a1);
     68       T[0x200 + i] = Ui32(a1, a3, a2, a1);
     69       T[0x300 + i] = Ui32(a1, a1, a3, a2);
     70     }
     71     {
     72       UInt32 a1 = InvS[i];
     73       UInt32 a2 = xtime(a1);
     74       UInt32 a4 = xtime(a2);
     75       UInt32 a8 = xtime(a4);
     76       UInt32 a9 = a8 ^ a1;
     77       UInt32 aB = a8 ^ a2 ^ a1;
     78       UInt32 aD = a8 ^ a4 ^ a1;
     79       UInt32 aE = a8 ^ a4 ^ a2;
     80       D[        i] = Ui32(aE, a9, aD, aB);
     81       D[0x100 + i] = Ui32(aB, aE, a9, aD);
     82       D[0x200 + i] = Ui32(aD, aB, aE, a9);
     83       D[0x300 + i] = Ui32(a9, aD, aB, aE);
     84     }
     85   }
     86 
     87   g_AesCbc_Encode = AesCbc_Encode;
     88   g_AesCbc_Decode = AesCbc_Decode;
     89   g_AesCtr_Code = AesCtr_Code;
     90 
     91   #ifdef MY_CPU_X86_OR_AMD64
     92   if (CPU_Is_Aes_Supported())
     93   {
     94     g_AesCbc_Encode = AesCbc_Encode_Intel;
     95     g_AesCbc_Decode = AesCbc_Decode_Intel;
     96     g_AesCtr_Code = AesCtr_Code_Intel;
     97   }
     98   #endif
     99 }
    100 
    101 
    102 #define HT(i, x, s) (T + (x << 8))[gb ## x(s[(i + x) & 3])]
    103 
    104 #define HT4(m, i, s, p) m[i] = \
    105     HT(i, 0, s) ^ \
    106     HT(i, 1, s) ^ \
    107     HT(i, 2, s) ^ \
    108     HT(i, 3, s) ^ w[p + i]
    109 
    110 #define HT16(m, s, p) \
    111     HT4(m, 0, s, p); \
    112     HT4(m, 1, s, p); \
    113     HT4(m, 2, s, p); \
    114     HT4(m, 3, s, p); \
    115 
    116 #define FT(i, x) Sbox[gb ## x(m[(i + x) & 3])]
    117 #define FT4(i) dest[i] = Ui32(FT(i, 0), FT(i, 1), FT(i, 2), FT(i, 3)) ^ w[i];
    118 
    119 
    120 #define HD(i, x, s) (D + (x << 8))[gb ## x(s[(i - x) & 3])]
    121 
    122 #define HD4(m, i, s, p) m[i] = \
    123     HD(i, 0, s) ^ \
    124     HD(i, 1, s) ^ \
    125     HD(i, 2, s) ^ \
    126     HD(i, 3, s) ^ w[p + i];
    127 
    128 #define HD16(m, s, p) \
    129     HD4(m, 0, s, p); \
    130     HD4(m, 1, s, p); \
    131     HD4(m, 2, s, p); \
    132     HD4(m, 3, s, p); \
    133 
    134 #define FD(i, x) InvS[gb ## x(m[(i - x) & 3])]
    135 #define FD4(i) dest[i] = Ui32(FD(i, 0), FD(i, 1), FD(i, 2), FD(i, 3)) ^ w[i];
    136 
    137 void MY_FAST_CALL Aes_SetKey_Enc(UInt32 *w, const Byte *key, unsigned keySize)
    138 {
    139   unsigned i, wSize;
    140   wSize = keySize + 28;
    141   keySize /= 4;
    142   w[0] = ((UInt32)keySize / 2) + 3;
    143   w += 4;
    144 
    145   for (i = 0; i < keySize; i++, key += 4)
    146     w[i] = GetUi32(key);
    147 
    148   for (; i < wSize; i++)
    149   {
    150     UInt32 t = w[i - 1];
    151     unsigned rem = i % keySize;
    152     if (rem == 0)
    153       t = Ui32(Sbox[gb1(t)] ^ Rcon[i / keySize], Sbox[gb2(t)], Sbox[gb3(t)], Sbox[gb0(t)]);
    154     else if (keySize > 6 && rem == 4)
    155       t = Ui32(Sbox[gb0(t)], Sbox[gb1(t)], Sbox[gb2(t)], Sbox[gb3(t)]);
    156     w[i] = w[i - keySize] ^ t;
    157   }
    158 }
    159 
    160 void MY_FAST_CALL Aes_SetKey_Dec(UInt32 *w, const Byte *key, unsigned keySize)
    161 {
    162   unsigned i, num;
    163   Aes_SetKey_Enc(w, key, keySize);
    164   num = keySize + 20;
    165   w += 8;
    166   for (i = 0; i < num; i++)
    167   {
    168     UInt32 r = w[i];
    169     w[i] =
    170       D[        (unsigned)Sbox[gb0(r)]] ^
    171       D[0x100 + (unsigned)Sbox[gb1(r)]] ^
    172       D[0x200 + (unsigned)Sbox[gb2(r)]] ^
    173       D[0x300 + (unsigned)Sbox[gb3(r)]];
    174   }
    175 }
    176 
    177 /* Aes_Encode and Aes_Decode functions work with little-endian words.
    178   src and dest are pointers to 4 UInt32 words.
    179   src and dest can point to same block */
    180 
    181 static void Aes_Encode(const UInt32 *w, UInt32 *dest, const UInt32 *src)
    182 {
    183   UInt32 s[4];
    184   UInt32 m[4];
    185   UInt32 numRounds2 = w[0];
    186   w += 4;
    187   s[0] = src[0] ^ w[0];
    188   s[1] = src[1] ^ w[1];
    189   s[2] = src[2] ^ w[2];
    190   s[3] = src[3] ^ w[3];
    191   w += 4;
    192   for (;;)
    193   {
    194     HT16(m, s, 0);
    195     if (--numRounds2 == 0)
    196       break;
    197     HT16(s, m, 4);
    198     w += 8;
    199   }
    200   w += 4;
    201   FT4(0); FT4(1); FT4(2); FT4(3);
    202 }
    203 
    204 static void Aes_Decode(const UInt32 *w, UInt32 *dest, const UInt32 *src)
    205 {
    206   UInt32 s[4];
    207   UInt32 m[4];
    208   UInt32 numRounds2 = w[0];
    209   w += 4 + numRounds2 * 8;
    210   s[0] = src[0] ^ w[0];
    211   s[1] = src[1] ^ w[1];
    212   s[2] = src[2] ^ w[2];
    213   s[3] = src[3] ^ w[3];
    214   for (;;)
    215   {
    216     w -= 8;
    217     HD16(m, s, 4);
    218     if (--numRounds2 == 0)
    219       break;
    220     HD16(s, m, 0);
    221   }
    222   FD4(0); FD4(1); FD4(2); FD4(3);
    223 }
    224 
    225 void AesCbc_Init(UInt32 *p, const Byte *iv)
    226 {
    227   unsigned i;
    228   for (i = 0; i < 4; i++)
    229     p[i] = GetUi32(iv + i * 4);
    230 }
    231 
    232 void MY_FAST_CALL AesCbc_Encode(UInt32 *p, Byte *data, size_t numBlocks)
    233 {
    234   for (; numBlocks != 0; numBlocks--, data += AES_BLOCK_SIZE)
    235   {
    236     p[0] ^= GetUi32(data);
    237     p[1] ^= GetUi32(data + 4);
    238     p[2] ^= GetUi32(data + 8);
    239     p[3] ^= GetUi32(data + 12);
    240 
    241     Aes_Encode(p + 4, p, p);
    242 
    243     SetUi32(data,      p[0]);
    244     SetUi32(data + 4,  p[1]);
    245     SetUi32(data + 8,  p[2]);
    246     SetUi32(data + 12, p[3]);
    247   }
    248 }
    249 
    250 void MY_FAST_CALL AesCbc_Decode(UInt32 *p, Byte *data, size_t numBlocks)
    251 {
    252   UInt32 in[4], out[4];
    253   for (; numBlocks != 0; numBlocks--, data += AES_BLOCK_SIZE)
    254   {
    255     in[0] = GetUi32(data);
    256     in[1] = GetUi32(data + 4);
    257     in[2] = GetUi32(data + 8);
    258     in[3] = GetUi32(data + 12);
    259 
    260     Aes_Decode(p + 4, out, in);
    261 
    262     SetUi32(data,      p[0] ^ out[0]);
    263     SetUi32(data + 4,  p[1] ^ out[1]);
    264     SetUi32(data + 8,  p[2] ^ out[2]);
    265     SetUi32(data + 12, p[3] ^ out[3]);
    266 
    267     p[0] = in[0];
    268     p[1] = in[1];
    269     p[2] = in[2];
    270     p[3] = in[3];
    271   }
    272 }
    273 
    274 void MY_FAST_CALL AesCtr_Code(UInt32 *p, Byte *data, size_t numBlocks)
    275 {
    276   for (; numBlocks != 0; numBlocks--)
    277   {
    278     UInt32 temp[4];
    279     Byte buf[16];
    280     int i;
    281 
    282     if (++p[0] == 0)
    283       p[1]++;
    284 
    285     Aes_Encode(p + 4, temp, p);
    286 
    287     SetUi32(buf,      temp[0]);
    288     SetUi32(buf + 4,  temp[1]);
    289     SetUi32(buf + 8,  temp[2]);
    290     SetUi32(buf + 12, temp[3]);
    291 
    292     for (i = 0; i < 16; i++)
    293       *data++ ^= buf[i];
    294   }
    295 }
    296