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/* 66 * Performance might be improved if the CPU architecture is OK with 67 * unaligned 32-bit loads and a fast ntohl() is available. 68 * Otherwise fall back to byte loads and shifts which is portable, 69 * and is faster on architectures with memory alignment issues. 70 */ 71 72#if defined(__i386__) || defined(__x86_64__) || \ 73 defined(_M_IX86) || defined(_M_X64) || \ 74 defined(__ppc__) || defined(__ppc64__) || \ 75 defined(__powerpc__) || defined(__powerpc64__) || \ 76 defined(__s390__) || defined(__s390x__) 77 78#define get_be32(p) ntohl(*(unsigned int *)(p)) 79#define put_be32(p, v) do { *(unsigned int *)(p) = htonl(v); } while (0) 80 81#else 82 83#define get_be32(p) ( \ 84 (*((unsigned char *)(p) + 0) << 24) | \ 85 (*((unsigned char *)(p) + 1) << 16) | \ 86 (*((unsigned char *)(p) + 2) << 8) | \ 87 (*((unsigned char *)(p) + 3) << 0) ) 88#define put_be32(p, v) do { \ 89 unsigned int __v = (v); \ 90 *((unsigned char *)(p) + 0) = __v >> 24; \ 91 *((unsigned char *)(p) + 1) = __v >> 16; \ 92 *((unsigned char *)(p) + 2) = __v >> 8; \ 93 *((unsigned char *)(p) + 3) = __v >> 0; } while (0) 94 95#endif 96 97/* This "rolls" over the 512-bit array */ 98#define W(x) (array[(x)&15]) 99 100/* 101 * Where do we get the source from? The first 16 iterations get it from 102 * the input data, the next mix it from the 512-bit array. 103 */ 104#define SHA_SRC(t) get_be32((unsigned char *) block + (t)*4) 105#define SHA_MIX(t) SHA_ROL(W((t)+13) ^ W((t)+8) ^ W((t)+2) ^ W(t), 1); 106 107#define SHA_ROUND(t, input, fn, constant, A, B, C, D, E) do { \ 108 unsigned int TEMP = input(t); setW(t, TEMP); \ 109 E += TEMP + SHA_ROL(A,5) + (fn) + (constant); \ 110 B = SHA_ROR(B, 2); } while (0) 111 112#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 ) 113#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 ) 114#define T_20_39(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, (B^C^D) , 0x6ed9eba1, A, B, C, D, E ) 115#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 ) 116#define T_60_79(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, (B^C^D) , 0xca62c1d6, A, B, C, D, E ) 117 118static void blk_SHA1_Block(blk_SHA_CTX *ctx, const void *block) 119{ 120 unsigned int A,B,C,D,E; 121 unsigned int array[16]; 122 123 A = ctx->H[0]; 124 B = ctx->H[1]; 125 C = ctx->H[2]; 126 D = ctx->H[3]; 127 E = ctx->H[4]; 128 129 /* Round 1 - iterations 0-16 take their input from 'block' */ 130 T_0_15( 0, A, B, C, D, E); 131 T_0_15( 1, E, A, B, C, D); 132 T_0_15( 2, D, E, A, B, C); 133 T_0_15( 3, C, D, E, A, B); 134 T_0_15( 4, B, C, D, E, A); 135 T_0_15( 5, A, B, C, D, E); 136 T_0_15( 6, E, A, B, C, D); 137 T_0_15( 7, D, E, A, B, C); 138 T_0_15( 8, C, D, E, A, B); 139 T_0_15( 9, B, C, D, E, A); 140 T_0_15(10, A, B, C, D, E); 141 T_0_15(11, E, A, B, C, D); 142 T_0_15(12, D, E, A, B, C); 143 T_0_15(13, C, D, E, A, B); 144 T_0_15(14, B, C, D, E, A); 145 T_0_15(15, A, B, C, D, E); 146 147 /* Round 1 - tail. Input from 512-bit mixing array */ 148 T_16_19(16, E, A, B, C, D); 149 T_16_19(17, D, E, A, B, C); 150 T_16_19(18, C, D, E, A, B); 151 T_16_19(19, B, C, D, E, A); 152 153 /* Round 2 */ 154 T_20_39(20, A, B, C, D, E); 155 T_20_39(21, E, A, B, C, D); 156 T_20_39(22, D, E, A, B, C); 157 T_20_39(23, C, D, E, A, B); 158 T_20_39(24, B, C, D, E, A); 159 T_20_39(25, A, B, C, D, E); 160 T_20_39(26, E, A, B, C, D); 161 T_20_39(27, D, E, A, B, C); 162 T_20_39(28, C, D, E, A, B); 163 T_20_39(29, B, C, D, E, A); 164 T_20_39(30, A, B, C, D, E); 165 T_20_39(31, E, A, B, C, D); 166 T_20_39(32, D, E, A, B, C); 167 T_20_39(33, C, D, E, A, B); 168 T_20_39(34, B, C, D, E, A); 169 T_20_39(35, A, B, C, D, E); 170 T_20_39(36, E, A, B, C, D); 171 T_20_39(37, D, E, A, B, C); 172 T_20_39(38, C, D, E, A, B); 173 T_20_39(39, B, C, D, E, A); 174 175 /* Round 3 */ 176 T_40_59(40, A, B, C, D, E); 177 T_40_59(41, E, A, B, C, D); 178 T_40_59(42, D, E, A, B, C); 179 T_40_59(43, C, D, E, A, B); 180 T_40_59(44, B, C, D, E, A); 181 T_40_59(45, A, B, C, D, E); 182 T_40_59(46, E, A, B, C, D); 183 T_40_59(47, D, E, A, B, C); 184 T_40_59(48, C, D, E, A, B); 185 T_40_59(49, B, C, D, E, A); 186 T_40_59(50, A, B, C, D, E); 187 T_40_59(51, E, A, B, C, D); 188 T_40_59(52, D, E, A, B, C); 189 T_40_59(53, C, D, E, A, B); 190 T_40_59(54, B, C, D, E, A); 191 T_40_59(55, A, B, C, D, E); 192 T_40_59(56, E, A, B, C, D); 193 T_40_59(57, D, E, A, B, C); 194 T_40_59(58, C, D, E, A, B); 195 T_40_59(59, B, C, D, E, A); 196 197 /* Round 4 */ 198 T_60_79(60, A, B, C, D, E); 199 T_60_79(61, E, A, B, C, D); 200 T_60_79(62, D, E, A, B, C); 201 T_60_79(63, C, D, E, A, B); 202 T_60_79(64, B, C, D, E, A); 203 T_60_79(65, A, B, C, D, E); 204 T_60_79(66, E, A, B, C, D); 205 T_60_79(67, D, E, A, B, C); 206 T_60_79(68, C, D, E, A, B); 207 T_60_79(69, B, C, D, E, A); 208 T_60_79(70, A, B, C, D, E); 209 T_60_79(71, E, A, B, C, D); 210 T_60_79(72, D, E, A, B, C); 211 T_60_79(73, C, D, E, A, B); 212 T_60_79(74, B, C, D, E, A); 213 T_60_79(75, A, B, C, D, E); 214 T_60_79(76, E, A, B, C, D); 215 T_60_79(77, D, E, A, B, C); 216 T_60_79(78, C, D, E, A, B); 217 T_60_79(79, B, C, D, E, A); 218 219 ctx->H[0] += A; 220 ctx->H[1] += B; 221 ctx->H[2] += C; 222 ctx->H[3] += D; 223 ctx->H[4] += E; 224} 225 226void blk_SHA1_Init(blk_SHA_CTX *ctx) 227{ 228 ctx->size = 0; 229 230 /* Initialize H with the magic constants (see FIPS180 for constants) */ 231 ctx->H[0] = 0x67452301; 232 ctx->H[1] = 0xefcdab89; 233 ctx->H[2] = 0x98badcfe; 234 ctx->H[3] = 0x10325476; 235 ctx->H[4] = 0xc3d2e1f0; 236} 237 238void blk_SHA1_Update(blk_SHA_CTX *ctx, const void *data, unsigned long len) 239{ 240 unsigned int lenW = ctx->size & 63; 241 242 ctx->size += len; 243 244 /* Read the data into W and process blocks as they get full */ 245 if (lenW) { 246 unsigned int left = 64 - lenW; 247 if (len < left) 248 left = len; 249 memcpy(lenW + (char *)ctx->W, data, left); 250 lenW = (lenW + left) & 63; 251 len -= left; 252 data = ((const char *)data + left); 253 if (lenW) 254 return; 255 blk_SHA1_Block(ctx, ctx->W); 256 } 257 while (len >= 64) { 258 blk_SHA1_Block(ctx, data); 259 data = ((const char *)data + 64); 260 len -= 64; 261 } 262 if (len) 263 memcpy(ctx->W, data, len); 264} 265 266void blk_SHA1_Final(unsigned char hashout[20], blk_SHA_CTX *ctx) 267{ 268 static const unsigned char pad[64] = { 0x80 }; 269 unsigned int padlen[2]; 270 int i; 271 272 /* Pad with a binary 1 (ie 0x80), then zeroes, then length */ 273 padlen[0] = htonl((uint32_t)(ctx->size >> 29)); 274 padlen[1] = htonl((uint32_t)(ctx->size << 3)); 275 276 i = ctx->size & 63; 277 blk_SHA1_Update(ctx, pad, 1+ (63 & (55 - i))); 278 blk_SHA1_Update(ctx, padlen, 8); 279 280 /* Output hash */ 281 for (i = 0; i < 5; i++) 282 put_be32(hashout + i*4, ctx->H[i]); 283}