1 /*
2 * Copyright (C) the libgit2 contributors. All rights reserved.
3 *
4 * This file is part of libgit2, distributed under the GNU GPL v2 with
5 * a Linking Exception. For full terms see the included COPYING file.
6 */
7
8 #include "generic.h"
9
10 #if defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
11
12 /*
13 * Force usage of rol or ror by selecting the one with the smaller constant.
14 * It _can_ generate slightly smaller code (a constant of 1 is special), but
15 * perhaps more importantly it's possibly faster on any uarch that does a
16 * rotate with a loop.
17 */
18
19 #define SHA_ASM(op, x, n) (__extension__ ({ unsigned int __res; __asm__(op " %1,%0":"=r" (__res):"i" (n), "0" (x)); __res; }))
20 #define SHA_ROL(x,n) SHA_ASM("rol", x, n)
21 #define SHA_ROR(x,n) SHA_ASM("ror", x, n)
22
23 #else
24
25 #define SHA_ROT(X,l,r) (((X) << (l)) | ((X) >> (r)))
26 #define SHA_ROL(X,n) SHA_ROT(X,n,32-(n))
27 #define SHA_ROR(X,n) SHA_ROT(X,32-(n),n)
28
29 #endif
30
31 /*
32 * If you have 32 registers or more, the compiler can (and should)
33 * try to change the array[] accesses into registers. However, on
34 * machines with less than ~25 registers, that won't really work,
35 * and at least gcc will make an unholy mess of it.
36 *
37 * So to avoid that mess which just slows things down, we force
38 * the stores to memory to actually happen (we might be better off
39 * with a 'W(t)=(val);asm("":"+m" (W(t))' there instead, as
40 * suggested by Artur Skawina - that will also make gcc unable to
41 * try to do the silly "optimize away loads" part because it won't
42 * see what the value will be).
43 *
44 * Ben Herrenschmidt reports that on PPC, the C version comes close
45 * to the optimized asm with this (ie on PPC you don't want that
46 * 'volatile', since there are lots of registers).
47 *
48 * On ARM we get the best code generation by forcing a full memory barrier
49 * between each SHA_ROUND, otherwise gcc happily get wild with spilling and
50 * the stack frame size simply explode and performance goes down the drain.
51 */
52
53 #if defined(__i386__) || defined(__x86_64__)
54 #define setW(x, val) (*(volatile unsigned int *)&W(x) = (val))
55 #elif defined(__GNUC__) && defined(__arm__)
56 #define setW(x, val) do { W(x) = (val); __asm__("":::"memory"); } while (0)
57 #else
58 #define setW(x, val) (W(x) = (val))
59 #endif
60
61 /*
62 * Performance might be improved if the CPU architecture is OK with
63 * unaligned 32-bit loads and a fast ntohl() is available.
64 * Otherwise fall back to byte loads and shifts which is portable,
65 * and is faster on architectures with memory alignment issues.
66 */
67
68 #if defined(__i386__) || defined(__x86_64__) || \
69 defined(_M_IX86) || defined(_M_X64) || \
70 defined(__ppc__) || defined(__ppc64__) || \
71 defined(__powerpc__) || defined(__powerpc64__) || \
72 defined(__s390__) || defined(__s390x__)
73
74 #define get_be32(p) ntohl(*(const unsigned int *)(p))
75 #define put_be32(p, v) do { *(unsigned int *)(p) = htonl(v); } while (0)
76
77 #else
78
79 #define get_be32(p) ( \
80 (*((const unsigned char *)(p) + 0) << 24) | \
81 (*((const unsigned char *)(p) + 1) << 16) | \
82 (*((const unsigned char *)(p) + 2) << 8) | \
83 (*((const unsigned char *)(p) + 3) << 0) )
84 #define put_be32(p, v) do { \
85 unsigned int __v = (v); \
86 *((unsigned char *)(p) + 0) = __v >> 24; \
87 *((unsigned char *)(p) + 1) = __v >> 16; \
88 *((unsigned char *)(p) + 2) = __v >> 8; \
89 *((unsigned char *)(p) + 3) = __v >> 0; } while (0)
90
91 #endif
92
93 /* This "rolls" over the 512-bit array */
94 #define W(x) (array[(x)&15])
95
96 /*
97 * Where do we get the source from? The first 16 iterations get it from
98 * the input data, the next mix it from the 512-bit array.
99 */
100 #define SHA_SRC(t) get_be32(data + t)
101 #define SHA_MIX(t) SHA_ROL(W(t+13) ^ W(t+8) ^ W(t+2) ^ W(t), 1)
102
103 #define SHA_ROUND(t, input, fn, constant, A, B, C, D, E) do { \
104 unsigned int TEMP = input(t); setW(t, TEMP); \
105 E += TEMP + SHA_ROL(A,5) + (fn) + (constant); \
106 B = SHA_ROR(B, 2); } while (0)
107
108 #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 )
109 #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 )
110 #define T_20_39(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, (B^C^D) , 0x6ed9eba1, A, B, C, D, E )
111 #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 )
112 #define T_60_79(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, (B^C^D) , 0xca62c1d6, A, B, C, D, E )
113
hash__block(git_hash_sha1_ctx * ctx,const unsigned int * data)114 static void hash__block(git_hash_sha1_ctx *ctx, const unsigned int *data)
115 {
116 unsigned int A,B,C,D,E;
117 unsigned int array[16];
118
119 A = ctx->H[0];
120 B = ctx->H[1];
121 C = ctx->H[2];
122 D = ctx->H[3];
123 E = ctx->H[4];
124
125 /* Round 1 - iterations 0-16 take their input from 'data' */
126 T_0_15( 0, A, B, C, D, E);
127 T_0_15( 1, E, A, B, C, D);
128 T_0_15( 2, D, E, A, B, C);
129 T_0_15( 3, C, D, E, A, B);
130 T_0_15( 4, B, C, D, E, A);
131 T_0_15( 5, A, B, C, D, E);
132 T_0_15( 6, E, A, B, C, D);
133 T_0_15( 7, D, E, A, B, C);
134 T_0_15( 8, C, D, E, A, B);
135 T_0_15( 9, B, C, D, E, A);
136 T_0_15(10, A, B, C, D, E);
137 T_0_15(11, E, A, B, C, D);
138 T_0_15(12, D, E, A, B, C);
139 T_0_15(13, C, D, E, A, B);
140 T_0_15(14, B, C, D, E, A);
141 T_0_15(15, A, B, C, D, E);
142
143 /* Round 1 - tail. Input from 512-bit mixing array */
144 T_16_19(16, E, A, B, C, D);
145 T_16_19(17, D, E, A, B, C);
146 T_16_19(18, C, D, E, A, B);
147 T_16_19(19, B, C, D, E, A);
148
149 /* Round 2 */
150 T_20_39(20, A, B, C, D, E);
151 T_20_39(21, E, A, B, C, D);
152 T_20_39(22, D, E, A, B, C);
153 T_20_39(23, C, D, E, A, B);
154 T_20_39(24, B, C, D, E, A);
155 T_20_39(25, A, B, C, D, E);
156 T_20_39(26, E, A, B, C, D);
157 T_20_39(27, D, E, A, B, C);
158 T_20_39(28, C, D, E, A, B);
159 T_20_39(29, B, C, D, E, A);
160 T_20_39(30, A, B, C, D, E);
161 T_20_39(31, E, A, B, C, D);
162 T_20_39(32, D, E, A, B, C);
163 T_20_39(33, C, D, E, A, B);
164 T_20_39(34, B, C, D, E, A);
165 T_20_39(35, A, B, C, D, E);
166 T_20_39(36, E, A, B, C, D);
167 T_20_39(37, D, E, A, B, C);
168 T_20_39(38, C, D, E, A, B);
169 T_20_39(39, B, C, D, E, A);
170
171 /* Round 3 */
172 T_40_59(40, A, B, C, D, E);
173 T_40_59(41, E, A, B, C, D);
174 T_40_59(42, D, E, A, B, C);
175 T_40_59(43, C, D, E, A, B);
176 T_40_59(44, B, C, D, E, A);
177 T_40_59(45, A, B, C, D, E);
178 T_40_59(46, E, A, B, C, D);
179 T_40_59(47, D, E, A, B, C);
180 T_40_59(48, C, D, E, A, B);
181 T_40_59(49, B, C, D, E, A);
182 T_40_59(50, A, B, C, D, E);
183 T_40_59(51, E, A, B, C, D);
184 T_40_59(52, D, E, A, B, C);
185 T_40_59(53, C, D, E, A, B);
186 T_40_59(54, B, C, D, E, A);
187 T_40_59(55, A, B, C, D, E);
188 T_40_59(56, E, A, B, C, D);
189 T_40_59(57, D, E, A, B, C);
190 T_40_59(58, C, D, E, A, B);
191 T_40_59(59, B, C, D, E, A);
192
193 /* Round 4 */
194 T_60_79(60, A, B, C, D, E);
195 T_60_79(61, E, A, B, C, D);
196 T_60_79(62, D, E, A, B, C);
197 T_60_79(63, C, D, E, A, B);
198 T_60_79(64, B, C, D, E, A);
199 T_60_79(65, A, B, C, D, E);
200 T_60_79(66, E, A, B, C, D);
201 T_60_79(67, D, E, A, B, C);
202 T_60_79(68, C, D, E, A, B);
203 T_60_79(69, B, C, D, E, A);
204 T_60_79(70, A, B, C, D, E);
205 T_60_79(71, E, A, B, C, D);
206 T_60_79(72, D, E, A, B, C);
207 T_60_79(73, C, D, E, A, B);
208 T_60_79(74, B, C, D, E, A);
209 T_60_79(75, A, B, C, D, E);
210 T_60_79(76, E, A, B, C, D);
211 T_60_79(77, D, E, A, B, C);
212 T_60_79(78, C, D, E, A, B);
213 T_60_79(79, B, C, D, E, A);
214
215 ctx->H[0] += A;
216 ctx->H[1] += B;
217 ctx->H[2] += C;
218 ctx->H[3] += D;
219 ctx->H[4] += E;
220 }
221
git_hash_sha1_global_init(void)222 int git_hash_sha1_global_init(void)
223 {
224 return 0;
225 }
226
git_hash_sha1_ctx_init(git_hash_sha1_ctx * ctx)227 int git_hash_sha1_ctx_init(git_hash_sha1_ctx *ctx)
228 {
229 return git_hash_sha1_init(ctx);
230 }
231
git_hash_sha1_ctx_cleanup(git_hash_sha1_ctx * ctx)232 void git_hash_sha1_ctx_cleanup(git_hash_sha1_ctx *ctx)
233 {
234 GIT_UNUSED(ctx);
235 }
236
git_hash_sha1_init(git_hash_sha1_ctx * ctx)237 int git_hash_sha1_init(git_hash_sha1_ctx *ctx)
238 {
239 ctx->size = 0;
240
241 /* Initialize H with the magic constants (see FIPS180 for constants) */
242 ctx->H[0] = 0x67452301;
243 ctx->H[1] = 0xefcdab89;
244 ctx->H[2] = 0x98badcfe;
245 ctx->H[3] = 0x10325476;
246 ctx->H[4] = 0xc3d2e1f0;
247
248 return 0;
249 }
250
git_hash_sha1_update(git_hash_sha1_ctx * ctx,const void * data,size_t len)251 int git_hash_sha1_update(git_hash_sha1_ctx *ctx, const void *data, size_t len)
252 {
253 unsigned int lenW = ctx->size & 63;
254
255 ctx->size += len;
256
257 /* Read the data into W and process blocks as they get full */
258 if (lenW) {
259 unsigned int left = 64 - lenW;
260 if (len < left)
261 left = (unsigned int)len;
262 memcpy(lenW + (char *)ctx->W, data, left);
263 lenW = (lenW + left) & 63;
264 len -= left;
265 data = ((const char *)data + left);
266 if (lenW)
267 return 0;
268 hash__block(ctx, ctx->W);
269 }
270 while (len >= 64) {
271 hash__block(ctx, data);
272 data = ((const char *)data + 64);
273 len -= 64;
274 }
275 if (len)
276 memcpy(ctx->W, data, len);
277
278 return 0;
279 }
280
git_hash_sha1_final(git_oid * out,git_hash_sha1_ctx * ctx)281 int git_hash_sha1_final(git_oid *out, git_hash_sha1_ctx *ctx)
282 {
283 static const unsigned char pad[64] = { 0x80 };
284 unsigned int padlen[2];
285 int i;
286
287 /* Pad with a binary 1 (ie 0x80), then zeroes, then length */
288 padlen[0] = htonl((uint32_t)(ctx->size >> 29));
289 padlen[1] = htonl((uint32_t)(ctx->size << 3));
290
291 i = ctx->size & 63;
292 git_hash_sha1_update(ctx, pad, 1+ (63 & (55 - i)));
293 git_hash_sha1_update(ctx, padlen, 8);
294
295 /* Output hash */
296 for (i = 0; i < 5; i++)
297 put_be32(out->id + i*4, ctx->H[i]);
298
299 return 0;
300 }
301