File: csprng.c

package info (click to toggle)
libmath-prime-util-perl 0.73-1
  • links: PTS, VCS
  • area: main
  • in suites: bullseye, buster, sid
  • size: 2,800 kB
  • sloc: perl: 24,676; ansic: 11,471; python: 24; makefile: 18
file content (303 lines) | stat: -rw-r--r-- 8,658 bytes parent folder | download
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
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
186
187
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
274
275
276
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

/* Our API for random numbers.
 *
 * We can use ISAAC, ChaCha20, or something else.
 *
 * 3700    ns/word  ChaCha20 in Perl
 * 3100    ns/word  Salsa20 in Perl
 * 1600    ns/word  ChaCha8 in Perl
 *  760    ns/word  ISAAC in Perl
 *
 *   11.20 ns/word  ChaCha20 (openbsd)
 *   10.31 ns/word  ChaCha20 (dj)
 *    8.66 ns/word  ChaCha20 (sse2 Peters)
 *    6.85 ns/word  ChaCha12 (dj)
 *    5.99 ns/word  Tyche
 *    5.11 ns/word  ChaCha8 (dj)
 *    4.37 ns/word  MT19937 (Cokus)
 *    4.14 ns/word  Tyche-i
 *    3.26 ns/word  ISAAC
 *    3.18 ns/word  PCG64 (64-bit state, 64-bit types)
 *    1.95 ns/word  PCG64 (64-bit state, 128-bit types)
 *    1.84 ns/word  ChaCha20 (AVX2 chacha-opt)
 *    1.48 ns/word  Xoroshiro128+
 *    1.16 ns/word  SplitMix64
 *
 * These functions do locking, the underlying library does not.
 */

#include <stdio.h>
#include <stddef.h>
#include <string.h>
#include "ptypes.h"
#include "csprng.h"

#include "chacha.h"
#define SEED_BYTES (32+8)
#define CSEED(ctx,bytes,data,good)  chacha_seed(ctx,bytes,data,good)
#define CRBYTES(ctx,bytes,data)     chacha_rand_bytes(ctx,bytes,data)
#define CIRAND32(ctx)               chacha_irand32(ctx)
#define CIRAND64(ctx)               chacha_irand64(ctx)
#define CSELFTEST()                 chacha_selftest()

/* Helper macros, similar to ChaCha, so we're consistent. */
#ifndef U8TO32_LE
#define U8TO32_LE(p) \
  (((uint32_t)((p)[0])      ) | \
   ((uint32_t)((p)[1]) <<  8) | \
   ((uint32_t)((p)[2]) << 16) | \
   ((uint32_t)((p)[3]) << 24))
#endif
#define U32TO8_LE(p, v) \
  do { \
    uint32_t _v = v; \
    (p)[0] = (((_v)      ) & 0xFFU); \
    (p)[1] = (((_v) >>  8) & 0xFFU); \
    (p)[2] = (((_v) >> 16) & 0xFFU); \
    (p)[3] = (((_v) >> 24) & 0xFFU); \
  } while (0)

/*****************************************************************************/

/* We put a simple 32-bit non-CS PRNG here to help fill small seeds. */
#if 0
/* XOSHIRO128**  32-bit output, 32-bit types, 128-bit state */
static INLINE uint32_t rotl(const uint32_t x, int k) {
  return (x << k) | (x >> (32 - k));
}
uint32_t prng_next(char* ctx) {
  uint32_t *s = (uint32_t*) ctx;
  const uint32_t result_starstar = rotl(s[0] * 5, 7) * 9;
  const uint32_t t = s[1] << 9;
  s[2] ^= s[0];  s[3] ^= s[1];  s[1] ^= s[2];  s[0] ^= s[3];
  s[2] ^= t;
  s[3] = rotl(s[3], 11);
  return result_starstar;
}
char* prng_new(uint32_t a, uint32_t b, uint32_t c, uint32_t d) {
  uint32_t *state;
  New(0, state, 4, uint32_t);
  state[0] = 1;  state[1] = b;  state[2] = c;  state[3] = d;
  (void) prng_next((char*)state);
  state[0] += a;
  (void) prng_next((char*)state);
  return (char*) state;
}
#else
/* PCG RXS M XS 32.  32-bit output, 32-bit state and types. */
uint32_t prng_next(char* ctx) {
  uint32_t *rng = (uint32_t*) ctx;
  uint32_t word, oldstate = rng[0];
  rng[0] = rng[0] * 747796405U + rng[1];
  word = ((oldstate >> ((oldstate >> 28u) + 4u)) ^ oldstate) * 277803737u;
  return (word >> 22u) ^ word;
}
char* prng_new(uint32_t a, uint32_t b, uint32_t c, uint32_t d) {
  uint32_t *state;
  New(0, state, 2, uint32_t);
  state[0] = 0U;
  state[1] = (b << 1u) | 1u;
  (void) prng_next((char*)state);
  state[0] += a;
  (void) prng_next((char*)state);
  state[0] ^= c;
  (void) prng_next((char*)state);
  state[0] ^= d;
  (void) prng_next((char*)state);
  return (char*) state;
}
#endif

/*****************************************************************************/

uint32_t csprng_context_size(void)
{
  return sizeof(chacha_context_t);
}
static char _has_selftest_run = 0;

void csprng_seed(void *ctx, uint32_t bytes, const unsigned char* data)
{
  unsigned char seed[SEED_BYTES + 4];

  /* If given a short seed, minimize zeros in state */
  if (bytes >= SEED_BYTES) {
    memcpy(seed, data, SEED_BYTES);
  } else {
    char* rng;
    uint32_t a, b, c, d, i;
    memcpy(seed, data, bytes);
    memset(seed+bytes, 0, sizeof(seed)-bytes);
    a = U8TO32_LE((seed +  0));
    b = U8TO32_LE((seed +  4));
    c = U8TO32_LE((seed +  8));
    d = U8TO32_LE((seed + 12));
    rng = prng_new(a,b,c,d);
    for (i = 4*((bytes+3)/4); i < SEED_BYTES; i += 4)
      U32TO8_LE(seed + i, prng_next(rng));
    Safefree(rng);
#if 0
    printf("got %u bytes in expanded to %u\n", bytes, SEED_BYTES);
    printf("from: ");for(i=0;i<bytes;i++)printf("%02x",data[i]);printf("\n");
    printf("to:   ");for(i=0;i<SEED_BYTES;i++)printf("%02x",seed[i]);printf("\n");
#endif
  }

  if (!_has_selftest_run) {
    _has_selftest_run = 1;
    CSELFTEST();
  }
  CSEED(ctx, SEED_BYTES, seed, (bytes >= 16));
}

extern void csprng_srand(void* ctx, UV insecure_seed)
{
#if BITS_PER_WORD == 32
  unsigned char seed[4] = {0};
  U32TO8_LE(seed, insecure_seed);
  csprng_seed(ctx, 4, seed);
#else
  unsigned char seed[8] = {0};
  if (insecure_seed <= UVCONST(4294967295)) {
    U32TO8_LE(seed, insecure_seed);
    csprng_seed(ctx, 4, seed);
  } else {
    U32TO8_LE(seed, insecure_seed);
    U32TO8_LE(seed + 4, (insecure_seed >> 32));
    csprng_seed(ctx, 8, seed);
  }
#endif
}

void csprng_rand_bytes(void* ctx, uint32_t bytes, unsigned char* data)
{
  CRBYTES(ctx, bytes, data);
}

uint32_t irand32(void* ctx)
{
  return CIRAND32(ctx);
}
UV irand64(void* ctx)
{
#if BITS_PER_WORD < 64
  croak("irand64 too many bits for UV");
#else
  return CIRAND64(ctx);
#endif
}


/*****************************************************************************/

int is_csprng_well_seeded(void *ctx)
{
  chacha_context_t *cs = ctx;
  return cs->goodseed;
}

/* There are many ways to get floats from integers.  A few good, many bad.
 *
 * Vigna recommends (x64 >> 11) * (1.0 / (1ULL<<53)).
 * http://xoroshiro.di.unimi.it
 * Also see alternatives discussed:
 * http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/VERSIONS/C-LANG/speed-up-real.html
 *
 * Melissa O'Neill notes the problem is harder than it looks, doesn't address.
 * http://www.pcg-random.org/pdf/toms-oneill-pcg-family-v1.02.pdf
 *
 * randomstate for numpy uses separate code for each generator.
 * With the exception of dSFMT, they each one one of:
 *     (x64 >> 11) * (1 / 9007199254740992.0)
 *     ((x32 >> 5) * 67108864.0 + (y32 >> 6)) / 9007199254740992.0
 * where the first one is identical to Vigna.
 *
 * David Jones recommends the minor 32-bit variant:
 *     ((x32 >> 6) * 134217728.0 + (y32 >> 5)) / 9007199254740992.0
 * http://www0.cs.ucl.ac.uk/staff/d.jones/GoodPracticeRNG.pdf
 *
 * Taylor Campbell discusses this in:
 * http://mumble.net/~campbell/tmp/random_real.c
 * He points out that there are two common non-broken choices,
 * div by 2^-53  or   div by 2^-64, and each are slightly flawed in
 * different ways.  He shows a theoretically better method.
 */

/*
 * We prefer the x64 / 2^-64 method.  It seems to produce the best results
 * and is easiest for ensuring we fill up all the bits.
 * It is similar to what Geoff Kuenning does in MTwist, though he computes
 * the constants at runtime to ensure a dodgy compiler won't munge them.
 */
#define TO_NV_32    2.3283064365386962890625000000000000000E-10L
#define TO_NV_64    5.4210108624275221700372640043497085571E-20L
#define TO_NV_96    1.2621774483536188886587657044524579675E-29L
#define TO_NV_128   2.9387358770557187699218413430556141945E-39L

#define DRAND_32_32  (CIRAND32(ctx) * TO_NV_32)
#define DRAND_64_32  (((CIRAND32(ctx)>>5) * 67108864.0 + (CIRAND32(ctx)>>6)) / 9007199254740992.0)
#define DRAND_64_64  (CIRAND64(ctx) * TO_NV_64)
#define DRAND_128_32 (CIRAND32(ctx) * TO_NV_32 + CIRAND32(ctx) * TO_NV_64 + CIRAND32(ctx) * TO_NV_96 + CIRAND32(ctx) * TO_NV_128)
#define DRAND_128_64 (CIRAND64(ctx) * TO_NV_64 + CIRAND64(ctx) * TO_NV_128)

NV drand64(void* ctx)
{
  NV r;
#if NVMANTBITS <= 32
  r = DRAND_32_32;
#elif NVMANTBITS <= 64
  r = (BITS_PER_WORD <= 32)  ?  DRAND_64_32  :  DRAND_64_64;
#else
  r = (BITS_PER_WORD <= 32)  ?  DRAND_128_32 :  DRAND_128_64;
#endif
  return r;
}

/* Return rand 32-bit integer between 0 to n-1 inclusive */
uint32_t urandomm32(void *ctx, uint32_t n)
{
  uint32_t r, rmin;

  if (n <= 1)
    return 0;

  rmin = -n % n;
  while (1) {
    r = CIRAND32(ctx);
    if (r >= rmin)
      break;
  }
  return r % n;
}

UV urandomm64(void* ctx, UV n)
{
  UV r, rmin;

  if (n <= 4294967295UL)
    return urandomm32(ctx,n);
  if (n-1 == 4294967295UL)
    return irand32(ctx);

  rmin = -n % n;
  while (1) {
    r = CIRAND64(ctx);
    if (r >= rmin)
      break;
  }
  return r % n;
}

UV urandomb(void* ctx, int nbits)
{
  if (nbits == 0) {
    return 0;
  } else if (nbits <= 32) {
    return irand32(ctx) >> (32-nbits);
#if BITS_PER_WORD == 64
  } else if (nbits <= 64) {
    return irand64(ctx) >> (64-nbits);
#endif
  }
  croak("irand64 too many bits for UV");
}