aboutsummaryrefslogtreecommitdiffstats
path: root/contrib/restricted/aws/s2n/utils/s2n_random.c
blob: 70b02eff7f370ff7047f9fa0a2d0ff10fcd661e5 (plain) (blame)
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
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
/*
 * Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
 *
 * Licensed under the Apache License, Version 2.0 (the "License").
 * You may not use this file except in compliance with the License.
 * A copy of the License is located at
 *
 *  http://aws.amazon.com/apache2.0
 *
 * or in the "license" file accompanying this file. This file is distributed
 * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either
 * express or implied. See the License for the specific language governing
 * permissions and limitations under the License.
 */

#include <errno.h>
#include <fcntl.h>
#include <limits.h>
#include <openssl/engine.h>
#include <openssl/rand.h>
#include <pthread.h>
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
#include <sys/param.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <time.h>
#include <unistd.h>

#if defined(S2N_CPUID_AVAILABLE)
    #include <cpuid.h>
#endif

#include "api/s2n.h"
#include "crypto/s2n_drbg.h"
#include "error/s2n_errno.h"
#include "stuffer/s2n_stuffer.h"
#include "utils/s2n_fork_detection.h"
#include "utils/s2n_mem.h"
#include "utils/s2n_random.h"
#include "utils/s2n_result.h"
#include "utils/s2n_safety.h"

#define ENTROPY_SOURCE "/dev/urandom"

/* See https://en.wikipedia.org/wiki/CPUID */
#define RDRAND_ECX_FLAG 0x40000000

/* One second in nanoseconds */
#define ONE_S INT64_C(1000000000)

/* Placeholder value for an uninitialized entropy file descriptor */
#define UNINITIALIZED_ENTROPY_FD -1

static int entropy_fd = UNINITIALIZED_ENTROPY_FD;

struct s2n_rand_state {
    uint64_t cached_fork_generation_number;
    struct s2n_drbg public_drbg;
    struct s2n_drbg private_drbg;
    bool drbgs_initialized;
};

/* Key which will control per-thread freeing of drbg memory */
static pthread_key_t s2n_per_thread_rand_state_key;
/* Needed to ensure key is initialized only once */
static pthread_once_t s2n_per_thread_rand_state_key_once = PTHREAD_ONCE_INIT;

static __thread struct s2n_rand_state s2n_per_thread_rand_state = {
    .cached_fork_generation_number = 0,
    .public_drbg = { 0 },
    .private_drbg = { 0 },
    .drbgs_initialized = false
};

static int s2n_rand_init_impl(void);
static int s2n_rand_cleanup_impl(void);
static int s2n_rand_urandom_impl(void *ptr, uint32_t size);
static int s2n_rand_rdrand_impl(void *ptr, uint32_t size);

static s2n_rand_init_callback s2n_rand_init_cb = s2n_rand_init_impl;
static s2n_rand_cleanup_callback s2n_rand_cleanup_cb = s2n_rand_cleanup_impl;
static s2n_rand_seed_callback s2n_rand_seed_cb = s2n_rand_urandom_impl;
static s2n_rand_mix_callback s2n_rand_mix_cb = s2n_rand_urandom_impl;

/* non-static for SAW proof */
bool s2n_cpu_supports_rdrand()
{
#if defined(S2N_CPUID_AVAILABLE)
    uint32_t eax, ebx, ecx, edx;
    if (!__get_cpuid(1, &eax, &ebx, &ecx, &edx)) {
        return false;
    }

    if (ecx & RDRAND_ECX_FLAG) {
        return true;
    }
#endif
    return false;
}

int s2n_rand_set_callbacks(s2n_rand_init_callback rand_init_callback,
        s2n_rand_cleanup_callback rand_cleanup_callback,
        s2n_rand_seed_callback rand_seed_callback,
        s2n_rand_mix_callback rand_mix_callback)
{
    POSIX_ENSURE_REF(rand_init_callback);
    POSIX_ENSURE_REF(rand_cleanup_callback);
    POSIX_ENSURE_REF(rand_seed_callback);
    POSIX_ENSURE_REF(rand_mix_callback);
    s2n_rand_init_cb = rand_init_callback;
    s2n_rand_cleanup_cb = rand_cleanup_callback;
    s2n_rand_seed_cb = rand_seed_callback;
    s2n_rand_mix_cb = rand_mix_callback;

    return S2N_SUCCESS;
}

S2N_RESULT s2n_get_seed_entropy(struct s2n_blob *blob)
{
    RESULT_ENSURE_REF(blob);

    RESULT_ENSURE(s2n_rand_seed_cb(blob->data, blob->size) >= S2N_SUCCESS, S2N_ERR_CANCELLED);

    return S2N_RESULT_OK;
}

S2N_RESULT s2n_get_mix_entropy(struct s2n_blob *blob)
{
    RESULT_ENSURE_REF(blob);

    RESULT_GUARD_POSIX(s2n_rand_mix_cb(blob->data, blob->size));

    return S2N_RESULT_OK;
}

static void s2n_drbg_destructor(void *_unused_argument)
{
    (void) _unused_argument;

    s2n_result_ignore(s2n_rand_cleanup_thread());
}

static void s2n_drbg_make_rand_state_key(void)
{
    (void) pthread_key_create(&s2n_per_thread_rand_state_key, s2n_drbg_destructor);
}

static S2N_RESULT s2n_init_drbgs(void)
{
    uint8_t s2n_public_drbg[] = "s2n public drbg";
    uint8_t s2n_private_drbg[] = "s2n private drbg";
    struct s2n_blob public = { 0 };
    RESULT_GUARD_POSIX(s2n_blob_init(&public, s2n_public_drbg, sizeof(s2n_public_drbg)));
    struct s2n_blob private = { 0 };
    RESULT_GUARD_POSIX(s2n_blob_init(&private, s2n_private_drbg, sizeof(s2n_private_drbg)));

    RESULT_ENSURE(pthread_once(&s2n_per_thread_rand_state_key_once, s2n_drbg_make_rand_state_key) == 0, S2N_ERR_DRBG);

    RESULT_GUARD(s2n_drbg_instantiate(&s2n_per_thread_rand_state.public_drbg, &public, S2N_AES_128_CTR_NO_DF_PR));
    RESULT_GUARD(s2n_drbg_instantiate(&s2n_per_thread_rand_state.private_drbg, &private, S2N_AES_256_CTR_NO_DF_PR));

    RESULT_ENSURE(pthread_setspecific(s2n_per_thread_rand_state_key, &s2n_per_thread_rand_state) == 0, S2N_ERR_DRBG);

    s2n_per_thread_rand_state.drbgs_initialized = true;

    return S2N_RESULT_OK;
}

static S2N_RESULT s2n_ensure_initialized_drbgs(void)
{
    if (s2n_per_thread_rand_state.drbgs_initialized == false) {
        RESULT_GUARD(s2n_init_drbgs());

        /* Then cache the fork generation number. We just initialized the drbg
         * states with new entropy and forking is not an external event.
         */
        uint64_t returned_fork_generation_number = 0;
        RESULT_GUARD(s2n_get_fork_generation_number(&returned_fork_generation_number));
        s2n_per_thread_rand_state.cached_fork_generation_number = returned_fork_generation_number;
    }

    return S2N_RESULT_OK;
}

/* s2n_ensure_uniqueness() implements defenses against uniqueness
 * breaking events that might cause duplicated drbg states. Currently, only
 * implements fork detection.
 */
static S2N_RESULT s2n_ensure_uniqueness(void)
{
    uint64_t returned_fork_generation_number = 0;
    RESULT_GUARD(s2n_get_fork_generation_number(&returned_fork_generation_number));

    if (returned_fork_generation_number != s2n_per_thread_rand_state.cached_fork_generation_number) {
        /* This assumes that s2n_rand_cleanup_thread() doesn't mutate any other
         * state than the drbg states and it resets the drbg initialization
         * boolean to false. s2n_ensure_initialized_drbgs() will cache the new
         * fork generation number in the per thread state.
         */
        RESULT_GUARD(s2n_rand_cleanup_thread());
        RESULT_GUARD(s2n_ensure_initialized_drbgs());
    }

    return S2N_RESULT_OK;
}

static S2N_RESULT s2n_get_random_data(struct s2n_blob *out_blob,
        struct s2n_drbg *drbg_state)
{
    RESULT_GUARD(s2n_ensure_initialized_drbgs());
    RESULT_GUARD(s2n_ensure_uniqueness());

    uint32_t offset = 0;
    uint32_t remaining = out_blob->size;

    while (remaining) {
        struct s2n_blob slice = { 0 };

        RESULT_GUARD_POSIX(s2n_blob_slice(out_blob, &slice, offset, MIN(remaining, S2N_DRBG_GENERATE_LIMIT)));
        RESULT_GUARD(s2n_drbg_generate(drbg_state, &slice));

        remaining -= slice.size;
        offset += slice.size;
    }

    return S2N_RESULT_OK;
}

S2N_RESULT s2n_get_public_random_data(struct s2n_blob *blob)
{
    RESULT_GUARD(s2n_get_random_data(blob, &s2n_per_thread_rand_state.public_drbg));

    return S2N_RESULT_OK;
}

S2N_RESULT s2n_get_private_random_data(struct s2n_blob *blob)
{
    RESULT_GUARD(s2n_get_random_data(blob, &s2n_per_thread_rand_state.private_drbg));

    return S2N_RESULT_OK;
}

S2N_RESULT s2n_get_public_random_bytes_used(uint64_t *bytes_used)
{
    RESULT_GUARD(s2n_drbg_bytes_used(&s2n_per_thread_rand_state.public_drbg, bytes_used));
    return S2N_RESULT_OK;
}

S2N_RESULT s2n_get_private_random_bytes_used(uint64_t *bytes_used)
{
    RESULT_GUARD(s2n_drbg_bytes_used(&s2n_per_thread_rand_state.private_drbg, bytes_used));
    return S2N_RESULT_OK;
}

static int s2n_rand_urandom_impl(void *ptr, uint32_t size)
{
    POSIX_ENSURE(entropy_fd != UNINITIALIZED_ENTROPY_FD, S2N_ERR_NOT_INITIALIZED);

    uint8_t *data = ptr;
    uint32_t n = size;
    struct timespec sleep_time = { .tv_sec = 0, .tv_nsec = 0 };
    long backoff = 1;

    while (n) {
        errno = 0;
        int r = read(entropy_fd, data, n);
        if (r <= 0) {
            /*
             * A non-blocking read() on /dev/urandom should "never" fail,
             * except for EINTR. If it does, briefly pause and use
             * exponential backoff to avoid creating a tight spinning loop.
             *
             * iteration          delay
             * ---------    -----------------
             *    1         10          nsec
             *    2         100         nsec
             *    3         1,000       nsec
             *    4         10,000      nsec
             *    5         100,000     nsec
             *    6         1,000,000   nsec
             *    7         10,000,000  nsec
             *    8         99,999,999  nsec
             *    9         99,999,999  nsec
             *    ...
             */
            if (errno != EINTR) {
                backoff = MIN(backoff * 10, ONE_S - 1);
                sleep_time.tv_nsec = backoff;
                do {
                    r = nanosleep(&sleep_time, &sleep_time);
                } while (r != 0);
            }

            continue;
        }

        data += r;
        n -= r;
    }

    return S2N_SUCCESS;
}

/*
 * Return a random number in the range [0, bound)
 */
S2N_RESULT s2n_public_random(int64_t bound, uint64_t *output)
{
    uint64_t r;

    RESULT_ENSURE_GT(bound, 0);

    while (1) {
        struct s2n_blob blob = { 0 };
        RESULT_GUARD_POSIX(s2n_blob_init(&blob, (void *) &r, sizeof(r)));
        RESULT_GUARD(s2n_get_public_random_data(&blob));

        /* Imagine an int was one byte and UINT_MAX was 256. If the
         * caller asked for s2n_random(129, ...) we'd end up in
         * trouble. Each number in the range 0...127 would be twice
         * as likely as 128. That's because r == 0 % 129 -> 0, and
         * r == 129 % 129 -> 0, but only r == 128 returns 128,
         * r == 257 is out of range.
         *
         * To de-bias the dice, we discard values of r that are higher
         * that the highest multiple of 'bound' an int can support. If
         * bound is a uint, then in the worst case we discard 50% - 1 r's.
         * But since 'bound' is an int and INT_MAX is <= UINT_MAX / 2,
         * in the worst case we discard 25% - 1 r's.
         */
        if (r < (UINT64_MAX - (UINT64_MAX % bound))) {
            *output = r % bound;
            return S2N_RESULT_OK;
        }
    }
}

#if S2N_LIBCRYPTO_SUPPORTS_CUSTOM_RAND

    #define S2N_RAND_ENGINE_ID "s2n_rand"

int s2n_openssl_compat_rand(unsigned char *buf, int num)
{
    struct s2n_blob out = { 0 };
    POSIX_GUARD(s2n_blob_init(&out, buf, num));

    if (s2n_result_is_error(s2n_get_private_random_data(&out))) {
        return 0;
    }
    return 1;
}

int s2n_openssl_compat_status(void)
{
    return 1;
}

int s2n_openssl_compat_init(ENGINE *unused)
{
    return 1;
}

RAND_METHOD s2n_openssl_rand_method = {
    .seed = NULL,
    .bytes = s2n_openssl_compat_rand,
    .cleanup = NULL,
    .add = NULL,
    .pseudorand = s2n_openssl_compat_rand,
    .status = s2n_openssl_compat_status
};
#endif

static int s2n_rand_init_impl(void)
{
OPEN:
    entropy_fd = open(ENTROPY_SOURCE, O_RDONLY);
    if (entropy_fd == -1) {
        if (errno == EINTR) {
            goto OPEN;
        }
        POSIX_BAIL(S2N_ERR_OPEN_RANDOM);
    }

    if (s2n_cpu_supports_rdrand()) {
        s2n_rand_mix_cb = s2n_rand_rdrand_impl;
    }

    return S2N_SUCCESS;
}

S2N_RESULT s2n_rand_init(void)
{
    RESULT_ENSURE(s2n_rand_init_cb() >= S2N_SUCCESS, S2N_ERR_CANCELLED);

    RESULT_GUARD(s2n_ensure_initialized_drbgs());

#if S2N_LIBCRYPTO_SUPPORTS_CUSTOM_RAND
    /* Create an engine */
    ENGINE *e = ENGINE_new();

    RESULT_ENSURE(e != NULL, S2N_ERR_OPEN_RANDOM);
    RESULT_GUARD_OSSL(ENGINE_set_id(e, S2N_RAND_ENGINE_ID), S2N_ERR_OPEN_RANDOM);
    RESULT_GUARD_OSSL(ENGINE_set_name(e, "s2n entropy generator"), S2N_ERR_OPEN_RANDOM);
    RESULT_GUARD_OSSL(ENGINE_set_flags(e, ENGINE_FLAGS_NO_REGISTER_ALL), S2N_ERR_OPEN_RANDOM);
    RESULT_GUARD_OSSL(ENGINE_set_init_function(e, s2n_openssl_compat_init), S2N_ERR_OPEN_RANDOM);
    RESULT_GUARD_OSSL(ENGINE_set_RAND(e, &s2n_openssl_rand_method), S2N_ERR_OPEN_RANDOM);
    RESULT_GUARD_OSSL(ENGINE_add(e), S2N_ERR_OPEN_RANDOM);
    RESULT_GUARD_OSSL(ENGINE_free(e), S2N_ERR_OPEN_RANDOM);

    /* Use that engine for rand() */
    e = ENGINE_by_id(S2N_RAND_ENGINE_ID);
    RESULT_ENSURE(e != NULL, S2N_ERR_OPEN_RANDOM);
    RESULT_GUARD_OSSL(ENGINE_init(e), S2N_ERR_OPEN_RANDOM);
    RESULT_GUARD_OSSL(ENGINE_set_default(e, ENGINE_METHOD_RAND), S2N_ERR_OPEN_RANDOM);
    RESULT_GUARD_OSSL(ENGINE_free(e), S2N_ERR_OPEN_RANDOM);
#endif

    return S2N_RESULT_OK;
}

static int s2n_rand_cleanup_impl(void)
{
    POSIX_ENSURE(entropy_fd != UNINITIALIZED_ENTROPY_FD, S2N_ERR_NOT_INITIALIZED);

    POSIX_GUARD(close(entropy_fd));
    entropy_fd = UNINITIALIZED_ENTROPY_FD;

    return S2N_SUCCESS;
}

S2N_RESULT s2n_rand_cleanup(void)
{
    RESULT_ENSURE(s2n_rand_cleanup_cb() >= S2N_SUCCESS, S2N_ERR_CANCELLED);

#if S2N_LIBCRYPTO_SUPPORTS_CUSTOM_RAND
    /* Cleanup our rand ENGINE in libcrypto */
    ENGINE *rand_engine = ENGINE_by_id(S2N_RAND_ENGINE_ID);
    if (rand_engine) {
        ENGINE_remove(rand_engine);
        ENGINE_finish(rand_engine);
        ENGINE_unregister_RAND(rand_engine);
        ENGINE_free(rand_engine);
        ENGINE_cleanup();
        RAND_set_rand_engine(NULL);
        RAND_set_rand_method(NULL);
    }
#endif

    s2n_rand_init_cb = s2n_rand_init_impl;
    s2n_rand_cleanup_cb = s2n_rand_cleanup_impl;
    s2n_rand_seed_cb = s2n_rand_urandom_impl;
    s2n_rand_mix_cb = s2n_rand_urandom_impl;

    return S2N_RESULT_OK;
}

S2N_RESULT s2n_rand_cleanup_thread(void)
{
    /* Currently, it is only safe for this function to mutate the drbg states
     * in the per thread rand state. See s2n_ensure_uniqueness().
     */
    RESULT_GUARD(s2n_drbg_wipe(&s2n_per_thread_rand_state.private_drbg));
    RESULT_GUARD(s2n_drbg_wipe(&s2n_per_thread_rand_state.public_drbg));

    s2n_per_thread_rand_state.drbgs_initialized = false;

    return S2N_RESULT_OK;
}

/* This must only be used for unit tests. Any real use is dangerous and will be
 * overwritten in s2n_ensure_uniqueness() if it is forked. This was added to
 * support known answer tests that use OpenSSL and s2n_get_private_random_data
 * directly.
 */
S2N_RESULT s2n_set_private_drbg_for_test(struct s2n_drbg drbg)
{
    RESULT_ENSURE(s2n_in_unit_test(), S2N_ERR_NOT_IN_UNIT_TEST);
    RESULT_GUARD(s2n_drbg_wipe(&s2n_per_thread_rand_state.private_drbg));

    s2n_per_thread_rand_state.private_drbg = drbg;

    return S2N_RESULT_OK;
}

/*
 * volatile is important to prevent the compiler from
 * re-ordering or optimizing the use of RDRAND.
 */
static int s2n_rand_rdrand_impl(void *data, uint32_t size)
{
#if defined(__x86_64__) || defined(__i386__)
    struct s2n_blob out = { 0 };
    POSIX_GUARD(s2n_blob_init(&out, data, size));
    size_t space_remaining = 0;
    struct s2n_stuffer stuffer = { 0 };
    union {
        uint64_t u64;
    #if defined(__i386__)
        struct {
            /* since we check first that we're on intel, we can safely assume little endian. */
            uint32_t u_low;
            uint32_t u_high;
        } i386_fields;
    #endif /* defined(__i386__) */
        uint8_t u8[8];
    } output;

    POSIX_GUARD(s2n_stuffer_init(&stuffer, &out));
    while ((space_remaining = s2n_stuffer_space_remaining(&stuffer))) {
        unsigned char success = 0;
        output.u64 = 0;

        for (int tries = 0; tries < 10; tries++) {
    #if defined(__i386__)
            /* execute the rdrand instruction, store the result in a general purpose register (it's assigned to
            * output.i386_fields.u_low). Check the carry bit, which will be set on success. Then clober the register and reset
            * the carry bit. Due to needing to support an ancient assembler we use the opcode syntax.
            * the %b1 is to force compilers to use c1 instead of ecx.
            * Here's a description of how the opcode is encoded:
            * 0x0fc7 (rdrand)
            * 0xf0 (store the result in eax).
            */
            unsigned char success_high = 0, success_low = 0;
            __asm__ __volatile__(
                    ".byte 0x0f, 0xc7, 0xf0;\n"
                    "setc %b1;\n"
                    : "=a"(output.i386_fields.u_low), "=qm"(success_low)
                    :
                    : "cc");

            __asm__ __volatile__(
                    ".byte 0x0f, 0xc7, 0xf0;\n"
                    "setc %b1;\n"
                    : "=a"(output.i386_fields.u_high), "=qm"(success_high)
                    :
                    : "cc");
            /* cppcheck-suppress knownConditionTrueFalse */
            success = success_high & success_low;

            /* Treat either all 1 or all 0 bits in either the high or low order
             * bits as failure */
            if (output.i386_fields.u_low == 0 || output.i386_fields.u_low == UINT32_MAX
                    || output.i386_fields.u_high == 0 || output.i386_fields.u_high == UINT32_MAX) {
                success = 0;
            }
    #else
            /* execute the rdrand instruction, store the result in a general purpose register (it's assigned to
            * output.u64). Check the carry bit, which will be set on success. Then clober the carry bit.
            * Due to needing to support an ancient assembler we use the opcode syntax.
            * the %b1 is to force compilers to use c1 instead of ecx.
            * Here's a description of how the opcode is encoded:
            * 0x48 (pick a 64-bit register it does more too, but that's all that matters there)
            * 0x0fc7 (rdrand)
            * 0xf0 (store the result in rax). */
            __asm__ __volatile__(
                    ".byte 0x48, 0x0f, 0xc7, 0xf0;\n"
                    "setc %b1;\n"
                    : "=a"(output.u64), "=qm"(success)
                    :
                    : "cc");
    #endif /* defined(__i386__) */

            /* Some AMD CPUs will find that RDRAND "sticks" on all 1s but still reports success.
             * Some other very old CPUs use all 0s as an error condition while still reporting success.
             * If we encounter either of these suspicious values (a 1/2^63 chance) we'll treat them as
             * a failure and generate a new value.
             *
             * In the future we could add CPUID checks to detect processors with these known bugs,
             * however it does not appear worth it. The entropy loss is negligible and the
             * corresponding likelihood that a healthy CPU generates either of these values is also
             * negligible (1/2^63). Finally, adding processor specific logic would greatly
             * increase the complexity and would cause us to "miss" any unknown processors with
             * similar bugs. */
            if (output.u64 == UINT64_MAX || output.u64 == 0) {
                success = 0;
            }

            if (success) {
                break;
            }
        }

        POSIX_ENSURE(success, S2N_ERR_RDRAND_FAILED);

        size_t data_to_fill = MIN(sizeof(output), space_remaining);

        POSIX_GUARD(s2n_stuffer_write_bytes(&stuffer, output.u8, data_to_fill));
    }

    return S2N_SUCCESS;
#else
    POSIX_BAIL(S2N_ERR_UNSUPPORTED_CPU);
#endif
}