block-sha1/sha1.c

   1 /*
   2  * SHA1 routine optimized to do word accesses rather than byte accesses,
   3  * and to avoid unnecessary copies into the context array.
   4  *
   5  * This was initially based on the Mozilla SHA1 implementation, although
   6  * none of the original Mozilla code remains.
   7  */
   8
   9 /* this is only to get definitions for memcpy(), ntohl() and htonl() */
  10 #include "../git-compat-util.h"
  11
  12 #include "sha1.h"
  13
  14 #if defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
  15
  16 /*
  17  * Force usage of rol or ror by selecting the one with the smaller constant.
  18  * It _can_ generate slightly smaller code (a constant of 1 is special), but
  19  * perhaps more importantly it's possibly faster on any uarch that does a
  20  * rotate with a loop.
  21  */
  22
  23 #define SHA_ASM(op, x, n) ({ unsigned int __res; __asm__(op " %1,%0":"=r" (__res):"i" (n), "0" (x)); __res; })
  24 #define SHA_ROL(x,n)    SHA_ASM("rol", x, n)
  25 #define SHA_ROR(x,n)    SHA_ASM("ror", x, n)
  26
  27 #else
  28
  29 #define SHA_ROT(X,l,r)  (((X) << (l)) | ((X) >> (r)))
  30 #define SHA_ROL(X,n)    SHA_ROT(X,n,32-(n))
  31 #define SHA_ROR(X,n)    SHA_ROT(X,32-(n),n)
  32
  33 #endif
  34
  35 /*
  36  * If you have 32 registers or more, the compiler can (and should)
  37  * try to change the array[] accesses into registers. However, on
  38  * machines with less than ~25 registers, that won't really work,
  39  * and at least gcc will make an unholy mess of it.
  40  *
  41  * So to avoid that mess which just slows things down, we force
  42  * the stores to memory to actually happen (we might be better off
  43  * with a 'W(t)=(val);asm("":"+m" (W(t))' there instead, as
  44  * suggested by Artur Skawina - that will also make gcc unable to
  45  * try to do the silly "optimize away loads" part because it won't
  46  * see what the value will be).
  47  *
  48  * Ben Herrenschmidt reports that on PPC, the C version comes close
  49  * to the optimized asm with this (ie on PPC you don't want that
  50  * 'volatile', since there are lots of registers).
  51  *
  52  * On ARM we get the best code generation by forcing a full memory barrier
  53  * between each SHA_ROUND, otherwise gcc happily get wild with spilling and
  54  * the stack frame size simply explode and performance goes down the drain.
  55  */
  56
  57 #if defined(__i386__) || defined(__x86_64__)
  58   #define setW(x, val) (*(volatile unsigned int *)&W(x) = (val))
  59 #elif defined(__GNUC__) && defined(__arm__)
  60   #define setW(x, val) do { W(x) = (val); __asm__("":::"memory"); } while (0)
  61 #else
  62   #define setW(x, val) (W(x) = (val))
  63 #endif
  64
  65 /* This "rolls" over the 512-bit array */
  66 #define W(x) (array[(x)&15])
  67
  68 /*
  69  * Where do we get the source from? The first 16 iterations get it from
  70  * the input data, the next mix it from the 512-bit array.
  71  */
  72 #define SHA_SRC(t) get_be32((unsigned char *) block + (t)*4)
  73 #define SHA_MIX(t) SHA_ROL(W((t)+13) ^ W((t)+8) ^ W((t)+2) ^ W(t), 1);
  74
  75 #define SHA_ROUND(t, input, fn, constant, A, B, C, D, E) do { \
  76         unsigned int TEMP = input(t); setW(t, TEMP); \
  77         E += TEMP + SHA_ROL(A,5) + (fn) + (constant); \
  78         B = SHA_ROR(B, 2); } while (0)
  79
  80 #define T_0_15(t, A, B, C, D, E)  SHA_ROUND(t, SHA_SRC, (((C^D)&B)^D) , 0x5a827999, A, B, C, D, E )
  81 #define T_16_19(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, (((C^D)&B)^D) , 0x5a827999, A, B, C, D, E )
  82 #define T_20_39(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, (B^C^D) , 0x6ed9eba1, A, B, C, D, E )
  83 #define T_40_59(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, ((B&C)+(D&(B^C))) , 0x8f1bbcdc, A, B, C, D, E )
  84 #define T_60_79(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, (B^C^D) ,  0xca62c1d6, A, B, C, D, E )
  85
  86 static void blk_SHA1_Block(blk_SHA_CTX *ctx, const void *block)
  87 {
  88         unsigned int A,B,C,D,E;
  89         unsigned int array[16];
  90
  91         A = ctx->H[0];
  92         B = ctx->H[1];
  93         C = ctx->H[2];
  94         D = ctx->H[3];
  95         E = ctx->H[4];
  96
  97         /* Round 1 - iterations 0-16 take their input from 'block' */
  98         T_0_15( 0, A, B, C, D, E);
  99         T_0_15( 1, E, A, B, C, D);
 100         T_0_15( 2, D, E, A, B, C);
 101         T_0_15( 3, C, D, E, A, B);
 102         T_0_15( 4, B, C, D, E, A);
 103         T_0_15( 5, A, B, C, D, E);
 104         T_0_15( 6, E, A, B, C, D);
 105         T_0_15( 7, D, E, A, B, C);
 106         T_0_15( 8, C, D, E, A, B);
 107         T_0_15( 9, B, C, D, E, A);
 108         T_0_15(10, A, B, C, D, E);
 109         T_0_15(11, E, A, B, C, D);
 110         T_0_15(12, D, E, A, B, C);
 111         T_0_15(13, C, D, E, A, B);
 112         T_0_15(14, B, C, D, E, A);
 113         T_0_15(15, A, B, C, D, E);
 114
 115         /* Round 1 - tail. Input from 512-bit mixing array */
 116         T_16_19(16, E, A, B, C, D);
 117         T_16_19(17, D, E, A, B, C);
 118         T_16_19(18, C, D, E, A, B);
 119         T_16_19(19, B, C, D, E, A);
 120
 121         /* Round 2 */
 122         T_20_39(20, A, B, C, D, E);
 123         T_20_39(21, E, A, B, C, D);
 124         T_20_39(22, D, E, A, B, C);
 125         T_20_39(23, C, D, E, A, B);
 126         T_20_39(24, B, C, D, E, A);
 127         T_20_39(25, A, B, C, D, E);
 128         T_20_39(26, E, A, B, C, D);
 129         T_20_39(27, D, E, A, B, C);
 130         T_20_39(28, C, D, E, A, B);
 131         T_20_39(29, B, C, D, E, A);
 132         T_20_39(30, A, B, C, D, E);
 133         T_20_39(31, E, A, B, C, D);
 134         T_20_39(32, D, E, A, B, C);
 135         T_20_39(33, C, D, E, A, B);
 136         T_20_39(34, B, C, D, E, A);
 137         T_20_39(35, A, B, C, D, E);
 138         T_20_39(36, E, A, B, C, D);
 139         T_20_39(37, D, E, A, B, C);
 140         T_20_39(38, C, D, E, A, B);
 141         T_20_39(39, B, C, D, E, A);
 142
 143         /* Round 3 */
 144         T_40_59(40, A, B, C, D, E);
 145         T_40_59(41, E, A, B, C, D);
 146         T_40_59(42, D, E, A, B, C);
 147         T_40_59(43, C, D, E, A, B);
 148         T_40_59(44, B, C, D, E, A);
 149         T_40_59(45, A, B, C, D, E);
 150         T_40_59(46, E, A, B, C, D);
 151         T_40_59(47, D, E, A, B, C);
 152         T_40_59(48, C, D, E, A, B);
 153         T_40_59(49, B, C, D, E, A);
 154         T_40_59(50, A, B, C, D, E);
 155         T_40_59(51, E, A, B, C, D);
 156         T_40_59(52, D, E, A, B, C);
 157         T_40_59(53, C, D, E, A, B);
 158         T_40_59(54, B, C, D, E, A);
 159         T_40_59(55, A, B, C, D, E);
 160         T_40_59(56, E, A, B, C, D);
 161         T_40_59(57, D, E, A, B, C);
 162         T_40_59(58, C, D, E, A, B);
 163         T_40_59(59, B, C, D, E, A);
 164
 165         /* Round 4 */
 166         T_60_79(60, A, B, C, D, E);
 167         T_60_79(61, E, A, B, C, D);
 168         T_60_79(62, D, E, A, B, C);
 169         T_60_79(63, C, D, E, A, B);
 170         T_60_79(64, B, C, D, E, A);
 171         T_60_79(65, A, B, C, D, E);
 172         T_60_79(66, E, A, B, C, D);
 173         T_60_79(67, D, E, A, B, C);
 174         T_60_79(68, C, D, E, A, B);
 175         T_60_79(69, B, C, D, E, A);
 176         T_60_79(70, A, B, C, D, E);
 177         T_60_79(71, E, A, B, C, D);
 178         T_60_79(72, D, E, A, B, C);
 179         T_60_79(73, C, D, E, A, B);
 180         T_60_79(74, B, C, D, E, A);
 181         T_60_79(75, A, B, C, D, E);
 182         T_60_79(76, E, A, B, C, D);
 183         T_60_79(77, D, E, A, B, C);
 184         T_60_79(78, C, D, E, A, B);
 185         T_60_79(79, B, C, D, E, A);
 186
 187         ctx->H[0] += A;
 188         ctx->H[1] += B;
 189         ctx->H[2] += C;
 190         ctx->H[3] += D;
 191         ctx->H[4] += E;
 192 }
 193
 194 void blk_SHA1_Init(blk_SHA_CTX *ctx)
 195 {
 196         ctx->size = 0;
 197
 198         /* Initialize H with the magic constants (see FIPS180 for constants) */
 199         ctx->H[0] = 0x67452301;
 200         ctx->H[1] = 0xefcdab89;
 201         ctx->H[2] = 0x98badcfe;
 202         ctx->H[3] = 0x10325476;
 203         ctx->H[4] = 0xc3d2e1f0;
 204 }
 205
 206 void blk_SHA1_Update(blk_SHA_CTX *ctx, const void *data, unsigned long len)
 207 {
 208         unsigned int lenW = ctx->size & 63;
 209
 210         ctx->size += len;
 211
 212         /* Read the data into W and process blocks as they get full */
 213         if (lenW) {
 214                 unsigned int left = 64 - lenW;
 215                 if (len < left)
 216                         left = len;
 217                 memcpy(lenW + (char *)ctx->W, data, left);
 218                 lenW = (lenW + left) & 63;
 219                 len -= left;
 220                 data = ((const char *)data + left);
 221                 if (lenW)
 222                         return;
 223                 blk_SHA1_Block(ctx, ctx->W);
 224         }
 225         while (len >= 64) {
 226                 blk_SHA1_Block(ctx, data);
 227                 data = ((const char *)data + 64);
 228                 len -= 64;
 229         }
 230         if (len)
 231                 memcpy(ctx->W, data, len);
 232 }
 233
 234 void blk_SHA1_Final(unsigned char hashout[20], blk_SHA_CTX *ctx)
 235 {
 236         static const unsigned char pad[64] = { 0x80 };
 237         unsigned int padlen[2];
 238         int i;
 239
 240         /* Pad with a binary 1 (ie 0x80), then zeroes, then length */
 241         padlen[0] = htonl((uint32_t)(ctx->size >> 29));
 242         padlen[1] = htonl((uint32_t)(ctx->size << 3));
 243
 244         i = ctx->size & 63;
 245         blk_SHA1_Update(ctx, pad, 1 + (63 & (55 - i)));
 246         blk_SHA1_Update(ctx, padlen, 8);
 247
 248         /* Output hash */
 249         for (i = 0; i < 5; i++)
 250                 put_be32(hashout + i * 4, ctx->H[i]);
 251 }