day 23 optimize again
[aoc_eblake.git] / 2016 / md5.c
blob47d9953c955bdf76b4ad2e8612330e6314921ffd
1 /* Functions to compute MD5 message digest of files or memory blocks.
2 according to the definition of MD5 in RFC 1321 from April 1992.
3 Copyright (C) 1995-1997, 1999-2001, 2005-2006, 2008-2017 Free Software
4 Foundation, Inc.
5 This file is part of the GNU C Library.
7 This program is free software; you can redistribute it and/or modify it
8 under the terms of the GNU General Public License as published by the
9 Free Software Foundation; either version 2, or (at your option) any
10 later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, see <http://www.gnu.org/licenses/>. */
20 /* Written by Ulrich Drepper <drepper@gnu.ai.mit.edu>, 1995. */
22 // hacked for reuse in advent code
23 //#include <config.h>
25 #if HAVE_OPENSSL_MD5
26 # define GL_OPENSSL_INLINE _GL_EXTERN_INLINE
27 #endif
28 #include "md5.h"
30 #include <stdalign.h>
31 #include <stdint.h>
32 #include <stdlib.h>
33 #include <string.h>
34 #include <sys/types.h>
36 #if USE_UNLOCKED_IO
37 # include "unlocked-io.h"
38 #endif
40 #ifdef _LIBC
41 # include <endian.h>
42 # if __BYTE_ORDER == __BIG_ENDIAN
43 # define WORDS_BIGENDIAN 1
44 # endif
45 /* We need to keep the namespace clean so define the MD5 function
46 protected using leading __ . */
47 # define md5_init_ctx __md5_init_ctx
48 # define md5_process_block __md5_process_block
49 # define md5_process_bytes __md5_process_bytes
50 # define md5_finish_ctx __md5_finish_ctx
51 # define md5_read_ctx __md5_read_ctx
52 # define md5_stream __md5_stream
53 # define md5_buffer __md5_buffer
54 #endif
56 #ifdef WORDS_BIGENDIAN
57 # define SWAP(n) \
58 (((n) << 24) | (((n) & 0xff00) << 8) | (((n) >> 8) & 0xff00) | ((n) >> 24))
59 #else
60 # define SWAP(n) (n)
61 #endif
63 #define BLOCKSIZE 32768
64 #if BLOCKSIZE % 64 != 0
65 # error "invalid BLOCKSIZE"
66 #endif
68 #if ! HAVE_OPENSSL_MD5
69 /* This array contains the bytes used to pad the buffer to the next
70 64-byte boundary. (RFC 1321, 3.1: Step 1) */
71 static const unsigned char fillbuf[64] = { 0x80, 0 /* , 0, 0, ... */ };
74 /* Initialize structure containing state of computation.
75 (RFC 1321, 3.3: Step 3) */
76 void
77 md5_init_ctx (struct md5_ctx *ctx)
79 ctx->A = 0x67452301;
80 ctx->B = 0xefcdab89;
81 ctx->C = 0x98badcfe;
82 ctx->D = 0x10325476;
84 ctx->total[0] = ctx->total[1] = 0;
85 ctx->buflen = 0;
88 /* Copy the 4 byte value from v into the memory location pointed to by *cp,
89 If your architecture allows unaligned access this is equivalent to
90 * (uint32_t *) cp = v */
91 static void
92 set_uint32 (char *cp, uint32_t v)
94 memcpy (cp, &v, sizeof v);
97 /* Put result from CTX in first 16 bytes following RESBUF. The result
98 must be in little endian byte order. */
99 void *
100 md5_read_ctx (const struct md5_ctx *ctx, void *resbuf)
102 char *r = resbuf;
103 set_uint32 (r + 0 * sizeof ctx->A, SWAP (ctx->A));
104 set_uint32 (r + 1 * sizeof ctx->B, SWAP (ctx->B));
105 set_uint32 (r + 2 * sizeof ctx->C, SWAP (ctx->C));
106 set_uint32 (r + 3 * sizeof ctx->D, SWAP (ctx->D));
108 return resbuf;
111 /* Process the remaining bytes in the internal buffer and the usual
112 prolog according to the standard and write the result to RESBUF. */
113 void *
114 md5_finish_ctx (struct md5_ctx *ctx, void *resbuf)
116 /* Take yet unprocessed bytes into account. */
117 uint32_t bytes = ctx->buflen;
118 size_t size = (bytes < 56) ? 64 / 4 : 64 * 2 / 4;
120 /* Now count remaining bytes. */
121 ctx->total[0] += bytes;
122 if (ctx->total[0] < bytes)
123 ++ctx->total[1];
125 /* Put the 64-bit file length in *bits* at the end of the buffer. */
126 ctx->buffer[size - 2] = SWAP (ctx->total[0] << 3);
127 ctx->buffer[size - 1] = SWAP ((ctx->total[1] << 3) | (ctx->total[0] >> 29));
129 memcpy (&((char *) ctx->buffer)[bytes], fillbuf, (size - 2) * 4 - bytes);
131 /* Process last bytes. */
132 md5_process_block (ctx->buffer, size * 4, ctx);
134 return md5_read_ctx (ctx, resbuf);
136 #endif
138 /* Compute MD5 message digest for bytes read from STREAM. The
139 resulting message digest number will be written into the 16 bytes
140 beginning at RESBLOCK. */
142 md5_stream (FILE *stream, void *resblock)
144 struct md5_ctx ctx;
145 size_t sum;
147 char *buffer = malloc (BLOCKSIZE + 72);
148 if (!buffer)
149 return 1;
151 /* Initialize the computation context. */
152 md5_init_ctx (&ctx);
154 /* Iterate over full file contents. */
155 while (1)
157 /* We read the file in blocks of BLOCKSIZE bytes. One call of the
158 computation function processes the whole buffer so that with the
159 next round of the loop another block can be read. */
160 size_t n;
161 sum = 0;
163 /* Read block. Take care for partial reads. */
164 while (1)
166 n = fread (buffer + sum, 1, BLOCKSIZE - sum, stream);
168 sum += n;
170 if (sum == BLOCKSIZE)
171 break;
173 if (n == 0)
175 /* Check for the error flag IFF N == 0, so that we don't
176 exit the loop after a partial read due to e.g., EAGAIN
177 or EWOULDBLOCK. */
178 if (ferror (stream))
180 free (buffer);
181 return 1;
183 goto process_partial_block;
186 /* We've read at least one byte, so ignore errors. But always
187 check for EOF, since feof may be true even though N > 0.
188 Otherwise, we could end up calling fread after EOF. */
189 if (feof (stream))
190 goto process_partial_block;
193 /* Process buffer with BLOCKSIZE bytes. Note that
194 BLOCKSIZE % 64 == 0
196 md5_process_block (buffer, BLOCKSIZE, &ctx);
199 process_partial_block:
201 /* Process any remaining bytes. */
202 if (sum > 0)
203 md5_process_bytes (buffer, sum, &ctx);
205 /* Construct result in desired memory. */
206 md5_finish_ctx (&ctx, resblock);
207 free (buffer);
208 return 0;
211 #if ! HAVE_OPENSSL_MD5
212 /* Compute MD5 message digest for LEN bytes beginning at BUFFER. The
213 result is always in little endian byte order, so that a byte-wise
214 output yields to the wanted ASCII representation of the message
215 digest. */
216 void *
217 md5_buffer (const char *buffer, size_t len, void *resblock)
219 struct md5_ctx ctx;
221 /* Initialize the computation context. */
222 md5_init_ctx (&ctx);
224 /* Process whole buffer but last len % 64 bytes. */
225 md5_process_bytes (buffer, len, &ctx);
227 /* Put result in desired memory area. */
228 return md5_finish_ctx (&ctx, resblock);
232 void
233 md5_process_bytes (const void *buffer, size_t len, struct md5_ctx *ctx)
235 /* When we already have some bits in our internal buffer concatenate
236 both inputs first. */
237 if (ctx->buflen != 0)
239 size_t left_over = ctx->buflen;
240 size_t add = 128 - left_over > len ? len : 128 - left_over;
242 memcpy (&((char *) ctx->buffer)[left_over], buffer, add);
243 ctx->buflen += add;
245 if (ctx->buflen > 64)
247 md5_process_block (ctx->buffer, ctx->buflen & ~63, ctx);
249 ctx->buflen &= 63;
250 /* The regions in the following copy operation cannot overlap,
251 because ctx->buflen < 64 ≤ (left_over + add) & ~63. */
252 memcpy (ctx->buffer,
253 &((char *) ctx->buffer)[(left_over + add) & ~63],
254 ctx->buflen);
257 buffer = (const char *) buffer + add;
258 len -= add;
261 /* Process available complete blocks. */
262 if (len >= 64)
264 #if !(_STRING_ARCH_unaligned || _STRING_INLINE_unaligned)
265 # define UNALIGNED_P(p) ((uintptr_t) (p) % alignof (uint32_t) != 0)
266 if (UNALIGNED_P (buffer))
267 while (len > 64)
269 md5_process_block (memcpy (ctx->buffer, buffer, 64), 64, ctx);
270 buffer = (const char *) buffer + 64;
271 len -= 64;
273 else
274 #endif
276 md5_process_block (buffer, len & ~63, ctx);
277 buffer = (const char *) buffer + (len & ~63);
278 len &= 63;
282 /* Move remaining bytes in internal buffer. */
283 if (len > 0)
285 size_t left_over = ctx->buflen;
287 memcpy (&((char *) ctx->buffer)[left_over], buffer, len);
288 left_over += len;
289 if (left_over >= 64)
291 md5_process_block (ctx->buffer, 64, ctx);
292 left_over -= 64;
293 /* The regions in the following copy operation cannot overlap,
294 because left_over ≤ 64. */
295 memcpy (ctx->buffer, &ctx->buffer[16], left_over);
297 ctx->buflen = left_over;
302 /* These are the four functions used in the four steps of the MD5 algorithm
303 and defined in the RFC 1321. The first function is a little bit optimized
304 (as found in Colin Plumbs public domain implementation). */
305 /* #define FF(b, c, d) ((b & c) | (~b & d)) */
306 #define FF(b, c, d) (d ^ (b & (c ^ d)))
307 #define FG(b, c, d) FF (d, b, c)
308 #define FH(b, c, d) (b ^ c ^ d)
309 #define FI(b, c, d) (c ^ (b | ~d))
311 /* Process LEN bytes of BUFFER, accumulating context into CTX.
312 It is assumed that LEN % 64 == 0. */
314 void
315 md5_process_block (const void *buffer, size_t len, struct md5_ctx *ctx)
317 uint32_t correct_words[16];
318 const uint32_t *words = buffer;
319 size_t nwords = len / sizeof (uint32_t);
320 const uint32_t *endp = words + nwords;
321 uint32_t A = ctx->A;
322 uint32_t B = ctx->B;
323 uint32_t C = ctx->C;
324 uint32_t D = ctx->D;
325 uint32_t lolen = len;
327 /* First increment the byte count. RFC 1321 specifies the possible
328 length of the file up to 2^64 bits. Here we only compute the
329 number of bytes. Do a double word increment. */
330 ctx->total[0] += lolen;
331 ctx->total[1] += (len >> 31 >> 1) + (ctx->total[0] < lolen);
333 /* Process all bytes in the buffer with 64 bytes in each round of
334 the loop. */
335 while (words < endp)
337 uint32_t *cwp = correct_words;
338 uint32_t A_save = A;
339 uint32_t B_save = B;
340 uint32_t C_save = C;
341 uint32_t D_save = D;
343 /* First round: using the given function, the context and a constant
344 the next context is computed. Because the algorithms processing
345 unit is a 32-bit word and it is determined to work on words in
346 little endian byte order we perhaps have to change the byte order
347 before the computation. To reduce the work for the next steps
348 we store the swapped words in the array CORRECT_WORDS. */
350 #define OP(a, b, c, d, s, T) \
351 do \
353 a += FF (b, c, d) + (*cwp++ = SWAP (*words)) + T; \
354 ++words; \
355 CYCLIC (a, s); \
356 a += b; \
358 while (0)
360 /* It is unfortunate that C does not provide an operator for
361 cyclic rotation. Hope the C compiler is smart enough. */
362 #define CYCLIC(w, s) (w = (w << s) | (w >> (32 - s)))
364 /* Before we start, one word to the strange constants.
365 They are defined in RFC 1321 as
367 T[i] = (int) (4294967296.0 * fabs (sin (i))), i=1..64
369 Here is an equivalent invocation using Perl:
371 perl -e 'foreach(1..64){printf "0x%08x\n", int (4294967296 * abs (sin $_))}'
374 /* Round 1. */
375 OP (A, B, C, D, 7, 0xd76aa478);
376 OP (D, A, B, C, 12, 0xe8c7b756);
377 OP (C, D, A, B, 17, 0x242070db);
378 OP (B, C, D, A, 22, 0xc1bdceee);
379 OP (A, B, C, D, 7, 0xf57c0faf);
380 OP (D, A, B, C, 12, 0x4787c62a);
381 OP (C, D, A, B, 17, 0xa8304613);
382 OP (B, C, D, A, 22, 0xfd469501);
383 OP (A, B, C, D, 7, 0x698098d8);
384 OP (D, A, B, C, 12, 0x8b44f7af);
385 OP (C, D, A, B, 17, 0xffff5bb1);
386 OP (B, C, D, A, 22, 0x895cd7be);
387 OP (A, B, C, D, 7, 0x6b901122);
388 OP (D, A, B, C, 12, 0xfd987193);
389 OP (C, D, A, B, 17, 0xa679438e);
390 OP (B, C, D, A, 22, 0x49b40821);
392 /* For the second to fourth round we have the possibly swapped words
393 in CORRECT_WORDS. Redefine the macro to take an additional first
394 argument specifying the function to use. */
395 #undef OP
396 #define OP(f, a, b, c, d, k, s, T) \
397 do \
399 a += f (b, c, d) + correct_words[k] + T; \
400 CYCLIC (a, s); \
401 a += b; \
403 while (0)
405 /* Round 2. */
406 OP (FG, A, B, C, D, 1, 5, 0xf61e2562);
407 OP (FG, D, A, B, C, 6, 9, 0xc040b340);
408 OP (FG, C, D, A, B, 11, 14, 0x265e5a51);
409 OP (FG, B, C, D, A, 0, 20, 0xe9b6c7aa);
410 OP (FG, A, B, C, D, 5, 5, 0xd62f105d);
411 OP (FG, D, A, B, C, 10, 9, 0x02441453);
412 OP (FG, C, D, A, B, 15, 14, 0xd8a1e681);
413 OP (FG, B, C, D, A, 4, 20, 0xe7d3fbc8);
414 OP (FG, A, B, C, D, 9, 5, 0x21e1cde6);
415 OP (FG, D, A, B, C, 14, 9, 0xc33707d6);
416 OP (FG, C, D, A, B, 3, 14, 0xf4d50d87);
417 OP (FG, B, C, D, A, 8, 20, 0x455a14ed);
418 OP (FG, A, B, C, D, 13, 5, 0xa9e3e905);
419 OP (FG, D, A, B, C, 2, 9, 0xfcefa3f8);
420 OP (FG, C, D, A, B, 7, 14, 0x676f02d9);
421 OP (FG, B, C, D, A, 12, 20, 0x8d2a4c8a);
423 /* Round 3. */
424 OP (FH, A, B, C, D, 5, 4, 0xfffa3942);
425 OP (FH, D, A, B, C, 8, 11, 0x8771f681);
426 OP (FH, C, D, A, B, 11, 16, 0x6d9d6122);
427 OP (FH, B, C, D, A, 14, 23, 0xfde5380c);
428 OP (FH, A, B, C, D, 1, 4, 0xa4beea44);
429 OP (FH, D, A, B, C, 4, 11, 0x4bdecfa9);
430 OP (FH, C, D, A, B, 7, 16, 0xf6bb4b60);
431 OP (FH, B, C, D, A, 10, 23, 0xbebfbc70);
432 OP (FH, A, B, C, D, 13, 4, 0x289b7ec6);
433 OP (FH, D, A, B, C, 0, 11, 0xeaa127fa);
434 OP (FH, C, D, A, B, 3, 16, 0xd4ef3085);
435 OP (FH, B, C, D, A, 6, 23, 0x04881d05);
436 OP (FH, A, B, C, D, 9, 4, 0xd9d4d039);
437 OP (FH, D, A, B, C, 12, 11, 0xe6db99e5);
438 OP (FH, C, D, A, B, 15, 16, 0x1fa27cf8);
439 OP (FH, B, C, D, A, 2, 23, 0xc4ac5665);
441 /* Round 4. */
442 OP (FI, A, B, C, D, 0, 6, 0xf4292244);
443 OP (FI, D, A, B, C, 7, 10, 0x432aff97);
444 OP (FI, C, D, A, B, 14, 15, 0xab9423a7);
445 OP (FI, B, C, D, A, 5, 21, 0xfc93a039);
446 OP (FI, A, B, C, D, 12, 6, 0x655b59c3);
447 OP (FI, D, A, B, C, 3, 10, 0x8f0ccc92);
448 OP (FI, C, D, A, B, 10, 15, 0xffeff47d);
449 OP (FI, B, C, D, A, 1, 21, 0x85845dd1);
450 OP (FI, A, B, C, D, 8, 6, 0x6fa87e4f);
451 OP (FI, D, A, B, C, 15, 10, 0xfe2ce6e0);
452 OP (FI, C, D, A, B, 6, 15, 0xa3014314);
453 OP (FI, B, C, D, A, 13, 21, 0x4e0811a1);
454 OP (FI, A, B, C, D, 4, 6, 0xf7537e82);
455 OP (FI, D, A, B, C, 11, 10, 0xbd3af235);
456 OP (FI, C, D, A, B, 2, 15, 0x2ad7d2bb);
457 OP (FI, B, C, D, A, 9, 21, 0xeb86d391);
459 /* Add the starting values of the context. */
460 A += A_save;
461 B += B_save;
462 C += C_save;
463 D += D_save;
466 /* Put checksum in context given as argument. */
467 ctx->A = A;
468 ctx->B = B;
469 ctx->C = C;
470 ctx->D = D;
472 #endif