src/liblzma/check/sha256.c

   1 ///////////////////////////////////////////////////////////////////////////////
   2 //
   3 /// \file       sha256.c
   4 /// \brief      SHA256
   5 //
   6 //  Based on the public domain code found from Wei Dai's Crypto++ library
   7 //  version 5.5.1: http://www.cryptopp.com/
   8 //  This code has been put into the public domain.
   9 //
  10 /// \todo       Crypto++ has x86 ASM optimizations. They use SSE so if they
  11 ///             are imported to liblzma, SSE instructions need to be used
  12 ///             conditionally to keep the code working on older boxes.
  13 //
  14 //  This library is distributed in the hope that it will be useful,
  15 //  but WITHOUT ANY WARRANTY; without even the implied warranty of
  16 //  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  17 //
  18 ///////////////////////////////////////////////////////////////////////////////
  19
  20 #include "check.h"
  21
  22 #ifndef WORDS_BIGENDIAN
  23 #       include "../../common/bswap.h"
  24 #endif
  25
  26 // At least on x86, GCC is able to optimize this to a rotate instruction.
  27 #define rotr_32(num, amount) ((num) >> (amount) | (num) << (32 - (amount)))
  28
  29 #define blk0(i) (W[i] = data[i])
  30 #define blk2(i) (W[i & 15] += s1(W[(i - 2) & 15]) + W[(i - 7) & 15] \
  31                 + s0(W[(i - 15) & 15]))
  32
  33 #define Ch(x, y, z) (z ^ (x & (y ^ z)))
  34 #define Maj(x, y, z) ((x & y) | (z & (x | y)))
  35
  36 #define a(i) T[(0 - i) & 7]
  37 #define b(i) T[(1 - i) & 7]
  38 #define c(i) T[(2 - i) & 7]
  39 #define d(i) T[(3 - i) & 7]
  40 #define e(i) T[(4 - i) & 7]
  41 #define f(i) T[(5 - i) & 7]
  42 #define g(i) T[(6 - i) & 7]
  43 #define h(i) T[(7 - i) & 7]
  44
  45 #define R(i) \
  46         h(i) += S1(e(i)) + Ch(e(i), f(i), g(i)) + SHA256_K[i + j] \
  47                 + (j ? blk2(i) : blk0(i)); \
  48         d(i) += h(i); \
  49         h(i) += S0(a(i)) + Maj(a(i), b(i), c(i))
  50
  51 #define S0(x) (rotr_32(x, 2) ^ rotr_32(x, 13) ^ rotr_32(x, 22))
  52 #define S1(x) (rotr_32(x, 6) ^ rotr_32(x, 11) ^ rotr_32(x, 25))
  53 #define s0(x) (rotr_32(x, 7) ^ rotr_32(x, 18) ^ (x >> 3))
  54 #define s1(x) (rotr_32(x, 17) ^ rotr_32(x, 19) ^ (x >> 10))
  55
  56
  57 static const uint32_t SHA256_K[64] = {
  58         0x428A2F98, 0x71374491, 0xB5C0FBCF, 0xE9B5DBA5,
  59         0x3956C25B, 0x59F111F1, 0x923F82A4, 0xAB1C5ED5,
  60         0xD807AA98, 0x12835B01, 0x243185BE, 0x550C7DC3,
  61         0x72BE5D74, 0x80DEB1FE, 0x9BDC06A7, 0xC19BF174,
  62         0xE49B69C1, 0xEFBE4786, 0x0FC19DC6, 0x240CA1CC,
  63         0x2DE92C6F, 0x4A7484AA, 0x5CB0A9DC, 0x76F988DA,
  64         0x983E5152, 0xA831C66D, 0xB00327C8, 0xBF597FC7,
  65         0xC6E00BF3, 0xD5A79147, 0x06CA6351, 0x14292967,
  66         0x27B70A85, 0x2E1B2138, 0x4D2C6DFC, 0x53380D13,
  67         0x650A7354, 0x766A0ABB, 0x81C2C92E, 0x92722C85,
  68         0xA2BFE8A1, 0xA81A664B, 0xC24B8B70, 0xC76C51A3,
  69         0xD192E819, 0xD6990624, 0xF40E3585, 0x106AA070,
  70         0x19A4C116, 0x1E376C08, 0x2748774C, 0x34B0BCB5,
  71         0x391C0CB3, 0x4ED8AA4A, 0x5B9CCA4F, 0x682E6FF3,
  72         0x748F82EE, 0x78A5636F, 0x84C87814, 0x8CC70208,
  73         0x90BEFFFA, 0xA4506CEB, 0xBEF9A3F7, 0xC67178F2,
  74 };
  75
  76
  77 static void
  78 transform(uint32_t state[static 8], const uint32_t data[static 16])
  79 {
  80         uint32_t W[16];
  81         uint32_t T[8];
  82
  83         // Copy state[] to working vars.
  84         memcpy(T, state, sizeof(T));
  85
  86         // 64 operations, partially loop unrolled
  87         for (unsigned int j = 0; j < 64; j += 16) {
  88                 R( 0); R( 1); R( 2); R( 3);
  89                 R( 4); R( 5); R( 6); R( 7);
  90                 R( 8); R( 9); R(10); R(11);
  91                 R(12); R(13); R(14); R(15);
  92         }
  93
  94         // Add the working vars back into state[].
  95         state[0] += a(0);
  96         state[1] += b(0);
  97         state[2] += c(0);
  98         state[3] += d(0);
  99         state[4] += e(0);
 100         state[5] += f(0);
 101         state[6] += g(0);
 102         state[7] += h(0);
 103 }
 104
 105
 106 static void
 107 process(lzma_check *check)
 108 {
 109 #ifdef WORDS_BIGENDIAN
 110         transform(check->state.sha256.state, (uint32_t *)(check->buffer));
 111
 112 #else
 113         uint32_t data[16];
 114
 115         for (size_t i = 0; i < 16; ++i)
 116                 data[i] = bswap_32(*((uint32_t*)(check->buffer) + i));
 117
 118         transform(check->state.sha256.state, data);
 119 #endif
 120
 121         return;
 122 }
 123
 124
 125 extern void
 126 lzma_sha256_init(lzma_check *check)
 127 {
 128         static const uint32_t s[8] = {
 129                 0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A,
 130                 0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19,
 131         };
 132
 133         memcpy(check->state.sha256.state, s, sizeof(s));
 134         check->state.sha256.size = 0;
 135
 136         return;
 137 }
 138
 139
 140 extern void
 141 lzma_sha256_update(const uint8_t *buf, size_t size, lzma_check *check)
 142 {
 143         // Copy the input data into a properly aligned temporary buffer.
 144         // This way we can be called with arbitrarily sized buffers
 145         // (no need to be multiple of 64 bytes), and the code works also
 146         // on architectures that don't allow unaligned memory access.
 147         while (size > 0) {
 148                 const size_t copy_start = check->state.sha256.size & 0x3F;
 149                 size_t copy_size = 64 - copy_start;
 150                 if (copy_size > size)
 151                         copy_size = size;
 152
 153                 memcpy(check->buffer + copy_start, buf, copy_size);
 154
 155                 buf += copy_size;
 156                 size -= copy_size;
 157                 check->state.sha256.size += copy_size;
 158
 159                 if ((check->state.sha256.size & 0x3F) == 0)
 160                         process(check);
 161         }
 162
 163         return;
 164 }
 165
 166
 167 extern void
 168 lzma_sha256_finish(lzma_check *check)
 169 {
 170         // Add padding as described in RFC 3174 (it describes SHA-1 but
 171         // the same padding style is used for SHA-256 too).
 172         size_t pos = check->state.sha256.size & 0x3F;
 173         check->buffer[pos++] = 0x80;
 174
 175         while (pos != 64 - 8) {
 176                 if (pos == 64) {
 177                         process(check);
 178                         pos = 0;
 179                 }
 180
 181                 check->buffer[pos++] = 0x00;
 182         }
 183
 184         // Convert the message size from bytes to bits.
 185         check->state.sha256.size *= 8;
 186
 187 #ifdef WORDS_BIGENDIAN
 188         *(uint64_t *)(check->buffer + 64 - 8) = check->state.sha256.size;
 189 #else
 190         *(uint64_t *)(check->buffer + 64 - 8)
 191                         = bswap_64(check->state.sha256.size);
 192 #endif
 193
 194         process(check);
 195
 196         for (size_t i = 0; i < 8; ++i)
 197 #ifdef WORDS_BIGENDIAN
 198                 ((uint32_t *)(check->buffer))[i]
 199                                 = check->state.sha256.state[i];
 200 #else
 201                 ((uint32_t *)(check->buffer))[i]
 202                                 = bswap_32(check->state.sha256.state[i]);
 203 #endif
 204
 205         return;
 206 }