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) 1995-2011, 2015 Red Hat, Inc.
4   This file is part of elfutils.
5   Written by Ulrich Drepper <drepper@redhat.com>, 1995.
6
7   This file is free software; you can redistribute it and/or modify
8   it under the terms of either
9
10     * the GNU Lesser General Public License as published by the Free
11       Software Foundation; either version 3 of the License, or (at
12       your option) any later version
13
14   or
15
16     * the GNU General Public License as published by the Free
17       Software Foundation; either version 2 of the License, or (at
18       your option) any later version
19
20   or both in parallel, as here.
21
22   elfutils is distributed in the hope that it will be useful, but
23   WITHOUT ANY WARRANTY; without even the implied warranty of
24   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
25   General Public License for more details.
26
27   You should have received copies of the GNU General Public License and
28   the GNU Lesser General Public License along with this program.  If
29   not, see <http://www.gnu.org/licenses/>.  */
30
31#ifdef HAVE_CONFIG_H
32# include <config.h>
33#endif
34
35#include <stdlib.h>
36#include <string.h>
37#include <sys/types.h>
38
39#include "md5.h"
40#include "system.h"
41
42#define SWAP(n) LE32 (n)
43
44/* This array contains the bytes used to pad the buffer to the next
45   64-byte boundary.  (RFC 1321, 3.1: Step 1)  */
46static const unsigned char fillbuf[64] = { 0x80, 0 /* , 0, 0, ...  */ };
47
48
49/* Initialize structure containing state of computation.
50   (RFC 1321, 3.3: Step 3)  */
51void
52md5_init_ctx (struct md5_ctx *ctx)
53{
54  ctx->A = 0x67452301;
55  ctx->B = 0xefcdab89;
56  ctx->C = 0x98badcfe;
57  ctx->D = 0x10325476;
58
59  ctx->total[0] = ctx->total[1] = 0;
60  ctx->buflen = 0;
61}
62
63/* Put result from CTX in first 16 bytes following RESBUF.  The result
64   must be in little endian byte order.
65
66   IMPORTANT: On some systems it is required that RESBUF is correctly
67   aligned for a 32 bits value.  */
68void *
69md5_read_ctx (const struct md5_ctx *ctx, void *resbuf)
70{
71  ((md5_uint32 *) resbuf)[0] = SWAP (ctx->A);
72  ((md5_uint32 *) resbuf)[1] = SWAP (ctx->B);
73  ((md5_uint32 *) resbuf)[2] = SWAP (ctx->C);
74  ((md5_uint32 *) resbuf)[3] = SWAP (ctx->D);
75
76  return resbuf;
77}
78
79static void
80le64_copy (char *dest, uint64_t x)
81{
82  for (size_t i = 0; i < 8; ++i)
83    {
84      dest[i] = (uint8_t) x;
85      x >>= 8;
86    }
87}
88
89/* Process the remaining bytes in the internal buffer and the usual
90   prolog according to the standard and write the result to RESBUF.
91
92   IMPORTANT: On some systems it is required that RESBUF is correctly
93   aligned for a 32 bits value.  */
94void *
95md5_finish_ctx (struct md5_ctx *ctx, void *resbuf)
96{
97  /* Take yet unprocessed bytes into account.  */
98  md5_uint32 bytes = ctx->buflen;
99  size_t pad;
100
101  /* Now count remaining bytes.  */
102  ctx->total[0] += bytes;
103  if (ctx->total[0] < bytes)
104    ++ctx->total[1];
105
106  pad = bytes >= 56 ? 64 + 56 - bytes : 56 - bytes;
107  memcpy (&ctx->buffer[bytes], fillbuf, pad);
108
109  /* Put the 64-bit file length in *bits* at the end of the buffer.  */
110  const uint64_t bit_length = ((ctx->total[0] << 3)
111			       + ((uint64_t) ((ctx->total[1] << 3) |
112					      (ctx->total[0] >> 29)) << 32));
113  le64_copy (&ctx->buffer[bytes + pad], bit_length);
114
115  /* Process last bytes.  */
116  md5_process_block (ctx->buffer, bytes + pad + 8, ctx);
117
118  return md5_read_ctx (ctx, resbuf);
119}
120
121
122#ifdef NEED_MD5_STREAM
123/* Compute MD5 message digest for bytes read from STREAM.  The
124   resulting message digest number will be written into the 16 bytes
125   beginning at RESBLOCK.  */
126int
127md5_stream (FILE *stream, void *resblock)
128{
129  /* Important: BLOCKSIZE must be a multiple of 64.  */
130#define BLOCKSIZE 4096
131  struct md5_ctx ctx;
132  char buffer[BLOCKSIZE + 72];
133  size_t sum;
134
135  /* Initialize the computation context.  */
136  md5_init_ctx (&ctx);
137
138  /* Iterate over full file contents.  */
139  while (1)
140    {
141      /* We read the file in blocks of BLOCKSIZE bytes.  One call of the
142	 computation function processes the whole buffer so that with the
143	 next round of the loop another block can be read.  */
144      size_t n;
145      sum = 0;
146
147      /* Read block.  Take care for partial reads.  */
148      do
149	{
150	  n = fread (buffer + sum, 1, BLOCKSIZE - sum, stream);
151
152	  sum += n;
153	}
154      while (sum < BLOCKSIZE && n != 0);
155      if (n == 0 && ferror (stream))
156        return 1;
157
158      /* If end of file is reached, end the loop.  */
159      if (n == 0)
160	break;
161
162      /* Process buffer with BLOCKSIZE bytes.  Note that
163			BLOCKSIZE % 64 == 0
164       */
165      md5_process_block (buffer, BLOCKSIZE, &ctx);
166    }
167
168  /* Add the last bytes if necessary.  */
169  if (sum > 0)
170    md5_process_bytes (buffer, sum, &ctx);
171
172  /* Construct result in desired memory.  */
173  md5_finish_ctx (&ctx, resblock);
174  return 0;
175}
176#endif
177
178
179#ifdef NEED_MD5_BUFFER
180/* Compute MD5 message digest for LEN bytes beginning at BUFFER.  The
181   result is always in little endian byte order, so that a byte-wise
182   output yields to the wanted ASCII representation of the message
183   digest.  */
184void *
185md5_buffer (const char *buffer, size_t len, void *resblock)
186{
187  struct md5_ctx ctx;
188
189  /* Initialize the computation context.  */
190  md5_init_ctx (&ctx);
191
192  /* Process whole buffer but last len % 64 bytes.  */
193  md5_process_bytes (buffer, len, &ctx);
194
195  /* Put result in desired memory area.  */
196  return md5_finish_ctx (&ctx, resblock);
197}
198#endif
199
200
201void
202md5_process_bytes (const void *buffer, size_t len, struct md5_ctx *ctx)
203{
204  /* When we already have some bits in our internal buffer concatenate
205     both inputs first.  */
206  if (ctx->buflen != 0)
207    {
208      size_t left_over = ctx->buflen;
209      size_t add = 128 - left_over > len ? len : 128 - left_over;
210
211      memcpy (&ctx->buffer[left_over], buffer, add);
212      ctx->buflen += add;
213
214      if (ctx->buflen > 64)
215	{
216	  md5_process_block (ctx->buffer, ctx->buflen & ~63, ctx);
217
218	  ctx->buflen &= 63;
219	  /* The regions in the following copy operation cannot overlap.  */
220	  memcpy (ctx->buffer, &ctx->buffer[(left_over + add) & ~63],
221		  ctx->buflen);
222	}
223
224      buffer = (const char *) buffer + add;
225      len -= add;
226    }
227
228  /* Process available complete blocks.  */
229  if (len >= 64)
230    {
231#if !_STRING_ARCH_unaligned
232/* To check alignment gcc has an appropriate operator.  Other
233   compilers don't.  */
234# if __GNUC__ >= 2
235#  define UNALIGNED_P(p) (((md5_uintptr) p) % __alignof__ (md5_uint32) != 0)
236# else
237#  define UNALIGNED_P(p) (((md5_uintptr) p) % sizeof (md5_uint32) != 0)
238# endif
239      if (UNALIGNED_P (buffer))
240	while (len > 64)
241	  {
242	    md5_process_block (memcpy (ctx->buffer, buffer, 64), 64, ctx);
243	    buffer = (const char *) buffer + 64;
244	    len -= 64;
245	  }
246      else
247#endif
248	{
249	  md5_process_block (buffer, len & ~63, ctx);
250	  buffer = (const char *) buffer + (len & ~63);
251	  len &= 63;
252	}
253    }
254
255  /* Move remaining bytes in internal buffer.  */
256  if (len > 0)
257    {
258      size_t left_over = ctx->buflen;
259
260      memcpy (&ctx->buffer[left_over], buffer, len);
261      left_over += len;
262      if (left_over >= 64)
263	{
264	  md5_process_block (ctx->buffer, 64, ctx);
265	  left_over -= 64;
266	  memcpy (ctx->buffer, &ctx->buffer[64], left_over);
267	}
268      ctx->buflen = left_over;
269    }
270}
271
272
273/* These are the four functions used in the four steps of the MD5 algorithm
274   and defined in the RFC 1321.  The first function is a little bit optimized
275   (as found in Colin Plumbs public domain implementation).  */
276/* #define FF(b, c, d) ((b & c) | (~b & d)) */
277#define FF(b, c, d) (d ^ (b & (c ^ d)))
278#define FG(b, c, d) FF (d, b, c)
279#define FH(b, c, d) (b ^ c ^ d)
280#define FI(b, c, d) (c ^ (b | ~d))
281
282/* Process LEN bytes of BUFFER, accumulating context into CTX.
283   It is assumed that LEN % 64 == 0.  */
284
285void
286md5_process_block (const void *buffer, size_t len, struct md5_ctx *ctx)
287{
288  md5_uint32 correct_words[16];
289  const md5_uint32 *words = buffer;
290  size_t nwords = len / sizeof (md5_uint32);
291  const md5_uint32 *endp = words + nwords;
292  md5_uint32 A = ctx->A;
293  md5_uint32 B = ctx->B;
294  md5_uint32 C = ctx->C;
295  md5_uint32 D = ctx->D;
296
297  /* First increment the byte count.  RFC 1321 specifies the possible
298     length of the file up to 2^64 bits.  Here we only compute the
299     number of bytes.  Do a double word increment.  */
300  ctx->total[0] += len;
301  if (ctx->total[0] < len)
302    ++ctx->total[1];
303
304  /* Process all bytes in the buffer with 64 bytes in each round of
305     the loop.  */
306  while (words < endp)
307    {
308      md5_uint32 *cwp = correct_words;
309      md5_uint32 A_save = A;
310      md5_uint32 B_save = B;
311      md5_uint32 C_save = C;
312      md5_uint32 D_save = D;
313
314      /* First round: using the given function, the context and a constant
315	 the next context is computed.  Because the algorithms processing
316	 unit is a 32-bit word and it is determined to work on words in
317	 little endian byte order we perhaps have to change the byte order
318	 before the computation.  To reduce the work for the next steps
319	 we store the swapped words in the array CORRECT_WORDS.  */
320
321#define OP(a, b, c, d, s, T)						\
322      do								\
323        {								\
324	  a += FF (b, c, d) + (*cwp++ = SWAP (*words)) + T;		\
325	  ++words;							\
326	  CYCLIC (a, s);						\
327	  a += b;							\
328        }								\
329      while (0)
330
331      /* It is unfortunate that C does not provide an operator for
332	 cyclic rotation.  Hope the C compiler is smart enough.  */
333#define CYCLIC(w, s) (w = (w << s) | (w >> (32 - s)))
334
335      /* Before we start, one word to the strange constants.
336	 They are defined in RFC 1321 as
337
338	 T[i] = (int) (4294967296.0 * fabs (sin (i))), i=1..64
339       */
340
341      /* Round 1.  */
342      OP (A, B, C, D,  7, 0xd76aa478);
343      OP (D, A, B, C, 12, 0xe8c7b756);
344      OP (C, D, A, B, 17, 0x242070db);
345      OP (B, C, D, A, 22, 0xc1bdceee);
346      OP (A, B, C, D,  7, 0xf57c0faf);
347      OP (D, A, B, C, 12, 0x4787c62a);
348      OP (C, D, A, B, 17, 0xa8304613);
349      OP (B, C, D, A, 22, 0xfd469501);
350      OP (A, B, C, D,  7, 0x698098d8);
351      OP (D, A, B, C, 12, 0x8b44f7af);
352      OP (C, D, A, B, 17, 0xffff5bb1);
353      OP (B, C, D, A, 22, 0x895cd7be);
354      OP (A, B, C, D,  7, 0x6b901122);
355      OP (D, A, B, C, 12, 0xfd987193);
356      OP (C, D, A, B, 17, 0xa679438e);
357      OP (B, C, D, A, 22, 0x49b40821);
358
359      /* For the second to fourth round we have the possibly swapped words
360	 in CORRECT_WORDS.  Redefine the macro to take an additional first
361	 argument specifying the function to use.  */
362#undef OP
363#define OP(f, a, b, c, d, k, s, T)					\
364      do 								\
365	{								\
366	  a += f (b, c, d) + correct_words[k] + T;			\
367	  CYCLIC (a, s);						\
368	  a += b;							\
369	}								\
370      while (0)
371
372      /* Round 2.  */
373      OP (FG, A, B, C, D,  1,  5, 0xf61e2562);
374      OP (FG, D, A, B, C,  6,  9, 0xc040b340);
375      OP (FG, C, D, A, B, 11, 14, 0x265e5a51);
376      OP (FG, B, C, D, A,  0, 20, 0xe9b6c7aa);
377      OP (FG, A, B, C, D,  5,  5, 0xd62f105d);
378      OP (FG, D, A, B, C, 10,  9, 0x02441453);
379      OP (FG, C, D, A, B, 15, 14, 0xd8a1e681);
380      OP (FG, B, C, D, A,  4, 20, 0xe7d3fbc8);
381      OP (FG, A, B, C, D,  9,  5, 0x21e1cde6);
382      OP (FG, D, A, B, C, 14,  9, 0xc33707d6);
383      OP (FG, C, D, A, B,  3, 14, 0xf4d50d87);
384      OP (FG, B, C, D, A,  8, 20, 0x455a14ed);
385      OP (FG, A, B, C, D, 13,  5, 0xa9e3e905);
386      OP (FG, D, A, B, C,  2,  9, 0xfcefa3f8);
387      OP (FG, C, D, A, B,  7, 14, 0x676f02d9);
388      OP (FG, B, C, D, A, 12, 20, 0x8d2a4c8a);
389
390      /* Round 3.  */
391      OP (FH, A, B, C, D,  5,  4, 0xfffa3942);
392      OP (FH, D, A, B, C,  8, 11, 0x8771f681);
393      OP (FH, C, D, A, B, 11, 16, 0x6d9d6122);
394      OP (FH, B, C, D, A, 14, 23, 0xfde5380c);
395      OP (FH, A, B, C, D,  1,  4, 0xa4beea44);
396      OP (FH, D, A, B, C,  4, 11, 0x4bdecfa9);
397      OP (FH, C, D, A, B,  7, 16, 0xf6bb4b60);
398      OP (FH, B, C, D, A, 10, 23, 0xbebfbc70);
399      OP (FH, A, B, C, D, 13,  4, 0x289b7ec6);
400      OP (FH, D, A, B, C,  0, 11, 0xeaa127fa);
401      OP (FH, C, D, A, B,  3, 16, 0xd4ef3085);
402      OP (FH, B, C, D, A,  6, 23, 0x04881d05);
403      OP (FH, A, B, C, D,  9,  4, 0xd9d4d039);
404      OP (FH, D, A, B, C, 12, 11, 0xe6db99e5);
405      OP (FH, C, D, A, B, 15, 16, 0x1fa27cf8);
406      OP (FH, B, C, D, A,  2, 23, 0xc4ac5665);
407
408      /* Round 4.  */
409      OP (FI, A, B, C, D,  0,  6, 0xf4292244);
410      OP (FI, D, A, B, C,  7, 10, 0x432aff97);
411      OP (FI, C, D, A, B, 14, 15, 0xab9423a7);
412      OP (FI, B, C, D, A,  5, 21, 0xfc93a039);
413      OP (FI, A, B, C, D, 12,  6, 0x655b59c3);
414      OP (FI, D, A, B, C,  3, 10, 0x8f0ccc92);
415      OP (FI, C, D, A, B, 10, 15, 0xffeff47d);
416      OP (FI, B, C, D, A,  1, 21, 0x85845dd1);
417      OP (FI, A, B, C, D,  8,  6, 0x6fa87e4f);
418      OP (FI, D, A, B, C, 15, 10, 0xfe2ce6e0);
419      OP (FI, C, D, A, B,  6, 15, 0xa3014314);
420      OP (FI, B, C, D, A, 13, 21, 0x4e0811a1);
421      OP (FI, A, B, C, D,  4,  6, 0xf7537e82);
422      OP (FI, D, A, B, C, 11, 10, 0xbd3af235);
423      OP (FI, C, D, A, B,  2, 15, 0x2ad7d2bb);
424      OP (FI, B, C, D, A,  9, 21, 0xeb86d391);
425
426      /* Add the starting values of the context.  */
427      A += A_save;
428      B += B_save;
429      C += C_save;
430      D += D_save;
431    }
432
433  /* Put checksum in context given as argument.  */
434  ctx->A = A;
435  ctx->B = B;
436  ctx->C = C;
437  ctx->D = D;
438}
439