memxor.c 7.81 KB
Newer Older
Niels Möller's avatar
Niels Möller committed
1
2
3
4
/* memxor.c
 *
 */

5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
/* nettle, low-level cryptographics library
 *
 * Copyright (C) 1991, 1993, 1995 Free Software Foundation, Inc.
 * Copyright (C) 2010 Niels Mller
 *  
 * The nettle library is free software; you can redistribute it and/or modify
 * it under the terms of the GNU Lesser General Public License as published by
 * the Free Software Foundation; either version 2.1 of the License, or (at your
 * option) any later version.
 * 
 * The nettle library is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
 * or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
 * License for more details.
 * 
 * You should have received a copy of the GNU Lesser General Public License
 * along with the nettle library; see the file COPYING.LIB.  If not, write to
 * the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
 * MA 02111-1307, USA.
 */

Niels Möller's avatar
Niels Möller committed
26
27
/* Implementation inspired by memcmp in glibc, contributed to the FSF
   by Torbjorn Granlund.
28
 */
Niels Möller's avatar
Niels Möller committed
29

30
#if HAVE_CONFIG_H
31
# include "config.h"
32
33
#endif

34
35
#include <limits.h>

Niels Möller's avatar
Niels Möller committed
36
37
#include "memxor.h"

38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
typedef unsigned long int word_t;

#if SIZEOF_LONG & (SIZEOF_LONG - 1)
#error Word size must be a power of two
#endif

#define ALIGN_OFFSET(p) ((uintptr_t) (p) % sizeof(word_t))

#ifndef WORDS_BIGENDIAN
#define MERGE(w0, sh_1, w1, sh_2) (((w0) >> (sh_1)) | ((w1) << (sh_2)))
#else
#define MERGE(w0, sh_1, w1, sh_2) (((w0) << (sh_1)) | ((w1) >> (sh_2)))
#endif

#define WORD_T_THRESH 16

/* XOR word-aligned areas. n is the number of words, not bytes. */
static void
memxor_common_alignment (word_t *dst, const word_t *src, size_t n)
{
Niels Möller's avatar
Niels Möller committed
58
59
60
  /* FIXME: Require n > 0? */
  /* FIXME: Unroll four times, like memcmp? Probably not worth the
     effort. */
61

Niels Möller's avatar
Niels Möller committed
62
  if (n & 1)
63
    {
Niels Möller's avatar
Niels Möller committed
64
65
66
67
68
69
70
      *dst++ ^= *src++;
      n--;
    }
  for (; n >= 2; dst += 2, src += 2, n -= 2)
    {
      dst[0] ^= src[0];
      dst[1] ^= src[1];
71
72
73
    }
}

Niels Möller's avatar
Niels Möller committed
74
/* XOR *un-aligned* src-area onto aligned dst area. n is number of
75
76
77
78
79
80
81
82
83
84
85
86
87
88
   words, not bytes. Assumes we can read complete words at the start
   and end of the src operand. */
static void
memxor_different_alignment (word_t *dst, const uint8_t *src, size_t n)
{
  size_t i;
  int shl, shr;
  const word_t *src_word;
  unsigned offset = ALIGN_OFFSET (src);
  word_t s0, s1;

  shl = CHAR_BIT * offset;
  shr = CHAR_BIT * (sizeof(word_t) - offset);

Niels Möller's avatar
Niels Möller committed
89
  src_word = (const word_t *) ((uintptr_t) src & -SIZEOF_LONG);
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108

  /* FIXME: Unroll four times, like memcmp? */
  i = n & 1;
  s0 = src_word[i];
  if (i)
    {
      s1 = src_word[0];
      dst[0] ^= MERGE (s1, shl, s0, shr);
    }

  for (; i < n; i += 2)
    {
      s1 = src_word[i+1];
      dst[i] ^= MERGE(s0, shl, s1, shr);
      s0 = src_word[i+2];
      dst[i+1] ^= MERGE(s1, shl, s0, shr);
    }
}

Niels Möller's avatar
Niels Möller committed
109
110
111
/* Performance, Intel SU1400 (x86_64): 0.25 cycles/byte aligned, 0.45
   cycles/byte unaligned. */

112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
/* XOR LEN bytes starting at SRCADDR onto DESTADDR. Result undefined
   if the source overlaps with the destination. Return DESTADDR. */
uint8_t *
memxor(uint8_t *dst, const uint8_t *src, size_t n)
{
  uint8_t *orig_dst = dst;

  if (n >= WORD_T_THRESH)
    {
      /* There are at least some bytes to compare.  No need to test
	 for N == 0 in this alignment loop.  */
      while (ALIGN_OFFSET (dst))
	{
	  *dst++ ^= *src++;
	  n--;
	}
      if (ALIGN_OFFSET (src))
	memxor_different_alignment ((word_t *) dst, src, n / sizeof(word_t));
      else
	memxor_common_alignment ((word_t *) dst, (const word_t *) src, n / sizeof(word_t));

Niels Möller's avatar
Niels Möller committed
133
134
135
      dst += n & -SIZEOF_LONG;
      src += n & -SIZEOF_LONG;
      n = n & (SIZEOF_LONG - 1);
136
137
138
139
140
141
    }
  for (; n > 0; n--)
    *dst++ ^= *src++;

  return orig_dst;
}
Niels Möller's avatar
Niels Möller committed
142
143
144
145
146
147


/* XOR word-aligned areas. n is the number of words, not bytes. */
static void
memxor3_common_alignment (word_t *dst,
			  const word_t *a, const word_t *b, size_t n)
Niels Möller's avatar
Niels Möller committed
148
{
Niels Möller's avatar
Niels Möller committed
149
150
151
152
  /* FIXME: Require n > 0? */
  while (n-- > 0)
    dst[n] = a[n] ^ b[n];
}
Niels Möller's avatar
Niels Möller committed
153

Niels Möller's avatar
Niels Möller committed
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
static void
memxor3_different_alignment_b (word_t *dst,
			       const word_t *a, const uint8_t *b, unsigned offset, size_t n)
{
  int shl, shr;
  const word_t *b_word;

  word_t s0, s1;

  shl = CHAR_BIT * offset;
  shr = CHAR_BIT * (sizeof(word_t) - offset);

  b_word = (const word_t *) ((uintptr_t) b & -SIZEOF_LONG);

  if (n & 1)
    {
      n--;
      s1 = b_word[n];
      s0 = b_word[n+1];
      dst[n] = a[n] ^ MERGE (s1, shl, s0, shr);
    }
  else
    s1 = b_word[n];
  
  while (n > 0)
    {
      n -= 2;
      s0 = b_word[n+1]; 
      dst[n+1] = a[n+1] ^ MERGE(s0, shl, s1, shr);
      s1 = b_word[n];
      dst[n] = a[n] ^ MERGE(s1, shl, s0, shr);
    }
Niels Möller's avatar
Niels Möller committed
186
}
187

Niels Möller's avatar
Niels Möller committed
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
static void
memxor3_different_alignment_ab (word_t *dst,
				const uint8_t *a, const uint8_t *b,
				unsigned offset, size_t n)
{
  int shl, shr;
  const word_t *a_word;
  const word_t *b_word;
  
  word_t s0, s1;

  shl = CHAR_BIT * offset;
  shr = CHAR_BIT * (sizeof(word_t) - offset);

  a_word = (const word_t *) ((uintptr_t) a & -SIZEOF_LONG);
  b_word = (const word_t *) ((uintptr_t) b & -SIZEOF_LONG);

  if (n & 1)
    {
      n--;
      s1 = a_word[n] ^ b_word[n];
      s0 = a_word[n+1] ^ b_word[n+1];
      dst[n] = MERGE (s1, shl, s0, shr);
    }
  else    
    s1 = a_word[n] ^ b_word[n];
  
  while (n > 0)
    {
      n -= 2;
      s0 = a_word[n+1] ^ b_word[n+1]; 
      dst[n+1] = MERGE(s0, shl, s1, shr);
      s1 = a_word[n] ^ b_word[n];
      dst[n] = MERGE(s1, shl, s0, shr);
    }
}

static void
memxor3_different_alignment_all (word_t *dst,
				 const uint8_t *a, const uint8_t *b,
				 unsigned a_offset, unsigned b_offset,
				 size_t n)
{
  int al, ar, bl, br;
  const word_t *a_word;
  const word_t *b_word;
  
  word_t a0, a1, b0, b1;

  al = CHAR_BIT * a_offset;
  ar = CHAR_BIT * (sizeof(word_t) - a_offset);
  bl = CHAR_BIT * b_offset;
  br = CHAR_BIT * (sizeof(word_t) - b_offset);

  a_word = (const word_t *) ((uintptr_t) a & -SIZEOF_LONG);
  b_word = (const word_t *) ((uintptr_t) b & -SIZEOF_LONG);

  if (n & 1)
    {
      n--;
      a1 = a_word[n]; a0 = a_word[n+1];
      b1 = b_word[n]; b0 = b_word[n+1];
      
      dst[n] = MERGE (a1, al, a0, ar) ^ MERGE (b1, bl, b0, br);
    }
  else    
    {
      a1 = a_word[n];
      b1 = b_word[n];
    }
  
  while (n > 0)
    {
      n -= 2;
      a0 = a_word[n+1]; b0 = b_word[n+1]; 
      dst[n+1] = MERGE(a0, al, a1, ar) ^ MERGE(b0, bl, b1, br);
      a1 = a_word[n]; b1 = b_word[n];
      dst[n] = MERGE(a1, al, a0, ar) ^ MERGE(b1, bl, b0, br);
    }
}

/* Current implementation processes data in descending order, to
   support overlapping operation with one of the sources overlapping
   the start of the destination area. This feature is used only
   internally by cbc decrypt, and it is not advertised or documented
   to nettle users. */
274
275
276
uint8_t *
memxor3(uint8_t *dst, const uint8_t *a, const uint8_t *b, size_t n)
{
Niels Möller's avatar
Niels Möller committed
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
  if (n >= WORD_T_THRESH)
    {
      unsigned i;
      unsigned a_offset;
      unsigned b_offset;
      size_t nwords;

      for (i = ALIGN_OFFSET(dst + n); i > 0; i--)
	{
	  n--;
	  dst[n] = a[n] ^ b[n];
	}

      a_offset = ALIGN_OFFSET(a + n);
      b_offset = ALIGN_OFFSET(b + n);

      nwords = n / sizeof (word_t);
      n %= sizeof (word_t);

      if (a_offset == b_offset)
	{
	  if (!a_offset)
	    memxor3_common_alignment((word_t *) (dst + n),
				     (const word_t *) (a + n),
				     (const word_t *) (b + n), nwords);
	  else
	    memxor3_different_alignment_ab((word_t *) (dst + n),
					   a + n, b + n, a_offset,
					   nwords);
	}
      else if (!a_offset)
	memxor3_different_alignment_b((word_t *) (dst + n),
				      (const word_t *) (a + n), b + n,
				      b_offset, nwords);
      else if (!b_offset)
	memxor3_different_alignment_b((word_t *) (dst + n),
				      (const word_t *) (b + n), a + n,
				      a_offset, nwords);
      else
	memxor3_different_alignment_all((word_t *) (dst + n), a + n, b + n,
					a_offset, b_offset, nwords);
					
    }
  while (n-- > 0)
    dst[n] = a[n] ^ b[n];
322
323
324

  return dst;
}