Loading...
1// SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause)
2/*
3 * Copyright (C) 2017-2024 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
4 * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005
5 * Copyright Theodore Ts'o, 1994, 1995, 1996, 1997, 1998, 1999. All rights reserved.
6 *
7 * This driver produces cryptographically secure pseudorandom data. It is divided
8 * into roughly six sections, each with a section header:
9 *
10 * - Initialization and readiness waiting.
11 * - Fast key erasure RNG, the "crng".
12 * - Entropy accumulation and extraction routines.
13 * - Entropy collection routines.
14 * - Userspace reader/writer interfaces.
15 * - Sysctl interface.
16 *
17 * The high level overview is that there is one input pool, into which
18 * various pieces of data are hashed. Prior to initialization, some of that
19 * data is then "credited" as having a certain number of bits of entropy.
20 * When enough bits of entropy are available, the hash is finalized and
21 * handed as a key to a stream cipher that expands it indefinitely for
22 * various consumers. This key is periodically refreshed as the various
23 * entropy collectors, described below, add data to the input pool.
24 */
25
26#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
27
28#include <linux/utsname.h>
29#include <linux/module.h>
30#include <linux/kernel.h>
31#include <linux/major.h>
32#include <linux/string.h>
33#include <linux/fcntl.h>
34#include <linux/slab.h>
35#include <linux/random.h>
36#include <linux/poll.h>
37#include <linux/init.h>
38#include <linux/fs.h>
39#include <linux/blkdev.h>
40#include <linux/interrupt.h>
41#include <linux/mm.h>
42#include <linux/nodemask.h>
43#include <linux/spinlock.h>
44#include <linux/kthread.h>
45#include <linux/percpu.h>
46#include <linux/ptrace.h>
47#include <linux/workqueue.h>
48#include <linux/irq.h>
49#include <linux/ratelimit.h>
50#include <linux/syscalls.h>
51#include <linux/completion.h>
52#include <linux/uuid.h>
53#include <linux/uaccess.h>
54#include <linux/suspend.h>
55#include <linux/siphash.h>
56#include <linux/sched/isolation.h>
57#include <crypto/chacha.h>
58#include <crypto/blake2s.h>
59#ifdef CONFIG_VDSO_GETRANDOM
60#include <vdso/getrandom.h>
61#include <vdso/datapage.h>
62#include <vdso/vsyscall.h>
63#endif
64#include <asm/archrandom.h>
65#include <asm/processor.h>
66#include <asm/irq.h>
67#include <asm/irq_regs.h>
68#include <asm/io.h>
69
70/*********************************************************************
71 *
72 * Initialization and readiness waiting.
73 *
74 * Much of the RNG infrastructure is devoted to various dependencies
75 * being able to wait until the RNG has collected enough entropy and
76 * is ready for safe consumption.
77 *
78 *********************************************************************/
79
80/*
81 * crng_init is protected by base_crng->lock, and only increases
82 * its value (from empty->early->ready).
83 */
84static enum {
85 CRNG_EMPTY = 0, /* Little to no entropy collected */
86 CRNG_EARLY = 1, /* At least POOL_EARLY_BITS collected */
87 CRNG_READY = 2 /* Fully initialized with POOL_READY_BITS collected */
88} crng_init __read_mostly = CRNG_EMPTY;
89static DEFINE_STATIC_KEY_FALSE(crng_is_ready);
90#define crng_ready() (static_branch_likely(&crng_is_ready) || crng_init >= CRNG_READY)
91/* Various types of waiters for crng_init->CRNG_READY transition. */
92static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
93static struct fasync_struct *fasync;
94static ATOMIC_NOTIFIER_HEAD(random_ready_notifier);
95
96/* Control how we warn userspace. */
97static struct ratelimit_state urandom_warning =
98 RATELIMIT_STATE_INIT_FLAGS("urandom_warning", HZ, 3, RATELIMIT_MSG_ON_RELEASE);
99static int ratelimit_disable __read_mostly =
100 IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM);
101module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
102MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
103
104/*
105 * Returns whether or not the input pool has been seeded and thus guaranteed
106 * to supply cryptographically secure random numbers. This applies to: the
107 * /dev/urandom device, the get_random_bytes function, and the get_random_{u8,
108 * u16,u32,u64,long} family of functions.
109 *
110 * Returns: true if the input pool has been seeded.
111 * false if the input pool has not been seeded.
112 */
113bool rng_is_initialized(void)
114{
115 return crng_ready();
116}
117EXPORT_SYMBOL(rng_is_initialized);
118
119static void __cold crng_set_ready(struct work_struct *work)
120{
121 static_branch_enable(&crng_is_ready);
122}
123
124/* Used by wait_for_random_bytes(), and considered an entropy collector, below. */
125static void try_to_generate_entropy(void);
126
127/*
128 * Wait for the input pool to be seeded and thus guaranteed to supply
129 * cryptographically secure random numbers. This applies to: the /dev/urandom
130 * device, the get_random_bytes function, and the get_random_{u8,u16,u32,u64,
131 * long} family of functions. Using any of these functions without first
132 * calling this function forfeits the guarantee of security.
133 *
134 * Returns: 0 if the input pool has been seeded.
135 * -ERESTARTSYS if the function was interrupted by a signal.
136 */
137int wait_for_random_bytes(void)
138{
139 while (!crng_ready()) {
140 int ret;
141
142 try_to_generate_entropy();
143 ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ);
144 if (ret)
145 return ret > 0 ? 0 : ret;
146 }
147 return 0;
148}
149EXPORT_SYMBOL(wait_for_random_bytes);
150
151/*
152 * Add a callback function that will be invoked when the crng is initialised,
153 * or immediately if it already has been. Only use this is you are absolutely
154 * sure it is required. Most users should instead be able to test
155 * `rng_is_initialized()` on demand, or make use of `get_random_bytes_wait()`.
156 */
157int __cold execute_with_initialized_rng(struct notifier_block *nb)
158{
159 unsigned long flags;
160 int ret = 0;
161
162 spin_lock_irqsave(&random_ready_notifier.lock, flags);
163 if (crng_ready())
164 nb->notifier_call(nb, 0, NULL);
165 else
166 ret = raw_notifier_chain_register((struct raw_notifier_head *)&random_ready_notifier.head, nb);
167 spin_unlock_irqrestore(&random_ready_notifier.lock, flags);
168 return ret;
169}
170
171#define warn_unseeded_randomness() \
172 if (IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM) && !crng_ready()) \
173 printk_deferred(KERN_NOTICE "random: %s called from %pS with crng_init=%d\n", \
174 __func__, (void *)_RET_IP_, crng_init)
175
176
177/*********************************************************************
178 *
179 * Fast key erasure RNG, the "crng".
180 *
181 * These functions expand entropy from the entropy extractor into
182 * long streams for external consumption using the "fast key erasure"
183 * RNG described at <https://blog.cr.yp.to/20170723-random.html>.
184 *
185 * There are a few exported interfaces for use by other drivers:
186 *
187 * void get_random_bytes(void *buf, size_t len)
188 * u8 get_random_u8()
189 * u16 get_random_u16()
190 * u32 get_random_u32()
191 * u32 get_random_u32_below(u32 ceil)
192 * u32 get_random_u32_above(u32 floor)
193 * u32 get_random_u32_inclusive(u32 floor, u32 ceil)
194 * u64 get_random_u64()
195 * unsigned long get_random_long()
196 *
197 * These interfaces will return the requested number of random bytes
198 * into the given buffer or as a return value. This is equivalent to
199 * a read from /dev/urandom. The u8, u16, u32, u64, long family of
200 * functions may be higher performance for one-off random integers,
201 * because they do a bit of buffering and do not invoke reseeding
202 * until the buffer is emptied.
203 *
204 *********************************************************************/
205
206enum {
207 CRNG_RESEED_START_INTERVAL = HZ,
208 CRNG_RESEED_INTERVAL = 60 * HZ
209};
210
211static struct {
212 u8 key[CHACHA_KEY_SIZE] __aligned(__alignof__(long));
213 unsigned long generation;
214 spinlock_t lock;
215} base_crng = {
216 .lock = __SPIN_LOCK_UNLOCKED(base_crng.lock)
217};
218
219struct crng {
220 u8 key[CHACHA_KEY_SIZE];
221 unsigned long generation;
222 local_lock_t lock;
223};
224
225static DEFINE_PER_CPU(struct crng, crngs) = {
226 .generation = ULONG_MAX,
227 .lock = INIT_LOCAL_LOCK(crngs.lock),
228};
229
230/*
231 * Return the interval until the next reseeding, which is normally
232 * CRNG_RESEED_INTERVAL, but during early boot, it is at an interval
233 * proportional to the uptime.
234 */
235static unsigned int crng_reseed_interval(void)
236{
237 static bool early_boot = true;
238
239 if (unlikely(READ_ONCE(early_boot))) {
240 time64_t uptime = ktime_get_seconds();
241 if (uptime >= CRNG_RESEED_INTERVAL / HZ * 2)
242 WRITE_ONCE(early_boot, false);
243 else
244 return max_t(unsigned int, CRNG_RESEED_START_INTERVAL,
245 (unsigned int)uptime / 2 * HZ);
246 }
247 return CRNG_RESEED_INTERVAL;
248}
249
250/* Used by crng_reseed() and crng_make_state() to extract a new seed from the input pool. */
251static void extract_entropy(void *buf, size_t len);
252
253/* This extracts a new crng key from the input pool. */
254static void crng_reseed(struct work_struct *work)
255{
256 static DECLARE_DELAYED_WORK(next_reseed, crng_reseed);
257 unsigned long flags;
258 unsigned long next_gen;
259 u8 key[CHACHA_KEY_SIZE];
260
261 /* Immediately schedule the next reseeding, so that it fires sooner rather than later. */
262 if (likely(system_unbound_wq))
263 queue_delayed_work(system_unbound_wq, &next_reseed, crng_reseed_interval());
264
265 extract_entropy(key, sizeof(key));
266
267 /*
268 * We copy the new key into the base_crng, overwriting the old one,
269 * and update the generation counter. We avoid hitting ULONG_MAX,
270 * because the per-cpu crngs are initialized to ULONG_MAX, so this
271 * forces new CPUs that come online to always initialize.
272 */
273 spin_lock_irqsave(&base_crng.lock, flags);
274 memcpy(base_crng.key, key, sizeof(base_crng.key));
275 next_gen = base_crng.generation + 1;
276 if (next_gen == ULONG_MAX)
277 ++next_gen;
278 WRITE_ONCE(base_crng.generation, next_gen);
279#ifdef CONFIG_VDSO_GETRANDOM
280 /* base_crng.generation's invalid value is ULONG_MAX, while
281 * _vdso_rng_data.generation's invalid value is 0, so add one to the
282 * former to arrive at the latter. Use smp_store_release so that this
283 * is ordered with the write above to base_crng.generation. Pairs with
284 * the smp_rmb() before the syscall in the vDSO code.
285 *
286 * Cast to unsigned long for 32-bit architectures, since atomic 64-bit
287 * operations are not supported on those architectures. This is safe
288 * because base_crng.generation is a 32-bit value. On big-endian
289 * architectures it will be stored in the upper 32 bits, but that's okay
290 * because the vDSO side only checks whether the value changed, without
291 * actually using or interpreting the value.
292 */
293 smp_store_release((unsigned long *)&__arch_get_k_vdso_rng_data()->generation, next_gen + 1);
294#endif
295 if (!static_branch_likely(&crng_is_ready))
296 crng_init = CRNG_READY;
297 spin_unlock_irqrestore(&base_crng.lock, flags);
298 memzero_explicit(key, sizeof(key));
299}
300
301/*
302 * This generates a ChaCha block using the provided key, and then
303 * immediately overwrites that key with half the block. It returns
304 * the resultant ChaCha state to the user, along with the second
305 * half of the block containing 32 bytes of random data that may
306 * be used; random_data_len may not be greater than 32.
307 *
308 * The returned ChaCha state contains within it a copy of the old
309 * key value, at index 4, so the state should always be zeroed out
310 * immediately after using in order to maintain forward secrecy.
311 * If the state cannot be erased in a timely manner, then it is
312 * safer to set the random_data parameter to &chacha_state[4] so
313 * that this function overwrites it before returning.
314 */
315static void crng_fast_key_erasure(u8 key[CHACHA_KEY_SIZE],
316 u32 chacha_state[CHACHA_STATE_WORDS],
317 u8 *random_data, size_t random_data_len)
318{
319 u8 first_block[CHACHA_BLOCK_SIZE];
320
321 BUG_ON(random_data_len > 32);
322
323 chacha_init_consts(chacha_state);
324 memcpy(&chacha_state[4], key, CHACHA_KEY_SIZE);
325 memset(&chacha_state[12], 0, sizeof(u32) * 4);
326 chacha20_block(chacha_state, first_block);
327
328 memcpy(key, first_block, CHACHA_KEY_SIZE);
329 memcpy(random_data, first_block + CHACHA_KEY_SIZE, random_data_len);
330 memzero_explicit(first_block, sizeof(first_block));
331}
332
333/*
334 * This function returns a ChaCha state that you may use for generating
335 * random data. It also returns up to 32 bytes on its own of random data
336 * that may be used; random_data_len may not be greater than 32.
337 */
338static void crng_make_state(u32 chacha_state[CHACHA_STATE_WORDS],
339 u8 *random_data, size_t random_data_len)
340{
341 unsigned long flags;
342 struct crng *crng;
343
344 BUG_ON(random_data_len > 32);
345
346 /*
347 * For the fast path, we check whether we're ready, unlocked first, and
348 * then re-check once locked later. In the case where we're really not
349 * ready, we do fast key erasure with the base_crng directly, extracting
350 * when crng_init is CRNG_EMPTY.
351 */
352 if (!crng_ready()) {
353 bool ready;
354
355 spin_lock_irqsave(&base_crng.lock, flags);
356 ready = crng_ready();
357 if (!ready) {
358 if (crng_init == CRNG_EMPTY)
359 extract_entropy(base_crng.key, sizeof(base_crng.key));
360 crng_fast_key_erasure(base_crng.key, chacha_state,
361 random_data, random_data_len);
362 }
363 spin_unlock_irqrestore(&base_crng.lock, flags);
364 if (!ready)
365 return;
366 }
367
368 local_lock_irqsave(&crngs.lock, flags);
369 crng = raw_cpu_ptr(&crngs);
370
371 /*
372 * If our per-cpu crng is older than the base_crng, then it means
373 * somebody reseeded the base_crng. In that case, we do fast key
374 * erasure on the base_crng, and use its output as the new key
375 * for our per-cpu crng. This brings us up to date with base_crng.
376 */
377 if (unlikely(crng->generation != READ_ONCE(base_crng.generation))) {
378 spin_lock(&base_crng.lock);
379 crng_fast_key_erasure(base_crng.key, chacha_state,
380 crng->key, sizeof(crng->key));
381 crng->generation = base_crng.generation;
382 spin_unlock(&base_crng.lock);
383 }
384
385 /*
386 * Finally, when we've made it this far, our per-cpu crng has an up
387 * to date key, and we can do fast key erasure with it to produce
388 * some random data and a ChaCha state for the caller. All other
389 * branches of this function are "unlikely", so most of the time we
390 * should wind up here immediately.
391 */
392 crng_fast_key_erasure(crng->key, chacha_state, random_data, random_data_len);
393 local_unlock_irqrestore(&crngs.lock, flags);
394}
395
396static void _get_random_bytes(void *buf, size_t len)
397{
398 u32 chacha_state[CHACHA_STATE_WORDS];
399 u8 tmp[CHACHA_BLOCK_SIZE];
400 size_t first_block_len;
401
402 if (!len)
403 return;
404
405 first_block_len = min_t(size_t, 32, len);
406 crng_make_state(chacha_state, buf, first_block_len);
407 len -= first_block_len;
408 buf += first_block_len;
409
410 while (len) {
411 if (len < CHACHA_BLOCK_SIZE) {
412 chacha20_block(chacha_state, tmp);
413 memcpy(buf, tmp, len);
414 memzero_explicit(tmp, sizeof(tmp));
415 break;
416 }
417
418 chacha20_block(chacha_state, buf);
419 if (unlikely(chacha_state[12] == 0))
420 ++chacha_state[13];
421 len -= CHACHA_BLOCK_SIZE;
422 buf += CHACHA_BLOCK_SIZE;
423 }
424
425 memzero_explicit(chacha_state, sizeof(chacha_state));
426}
427
428/*
429 * This returns random bytes in arbitrary quantities. The quality of the
430 * random bytes is good as /dev/urandom. In order to ensure that the
431 * randomness provided by this function is okay, the function
432 * wait_for_random_bytes() should be called and return 0 at least once
433 * at any point prior.
434 */
435void get_random_bytes(void *buf, size_t len)
436{
437 warn_unseeded_randomness();
438 _get_random_bytes(buf, len);
439}
440EXPORT_SYMBOL(get_random_bytes);
441
442static ssize_t get_random_bytes_user(struct iov_iter *iter)
443{
444 u32 chacha_state[CHACHA_STATE_WORDS];
445 u8 block[CHACHA_BLOCK_SIZE];
446 size_t ret = 0, copied;
447
448 if (unlikely(!iov_iter_count(iter)))
449 return 0;
450
451 /*
452 * Immediately overwrite the ChaCha key at index 4 with random
453 * bytes, in case userspace causes copy_to_iter() below to sleep
454 * forever, so that we still retain forward secrecy in that case.
455 */
456 crng_make_state(chacha_state, (u8 *)&chacha_state[4], CHACHA_KEY_SIZE);
457 /*
458 * However, if we're doing a read of len <= 32, we don't need to
459 * use chacha_state after, so we can simply return those bytes to
460 * the user directly.
461 */
462 if (iov_iter_count(iter) <= CHACHA_KEY_SIZE) {
463 ret = copy_to_iter(&chacha_state[4], CHACHA_KEY_SIZE, iter);
464 goto out_zero_chacha;
465 }
466
467 for (;;) {
468 chacha20_block(chacha_state, block);
469 if (unlikely(chacha_state[12] == 0))
470 ++chacha_state[13];
471
472 copied = copy_to_iter(block, sizeof(block), iter);
473 ret += copied;
474 if (!iov_iter_count(iter) || copied != sizeof(block))
475 break;
476
477 BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0);
478 if (ret % PAGE_SIZE == 0) {
479 if (signal_pending(current))
480 break;
481 cond_resched();
482 }
483 }
484
485 memzero_explicit(block, sizeof(block));
486out_zero_chacha:
487 memzero_explicit(chacha_state, sizeof(chacha_state));
488 return ret ? ret : -EFAULT;
489}
490
491/*
492 * Batched entropy returns random integers. The quality of the random
493 * number is good as /dev/urandom. In order to ensure that the randomness
494 * provided by this function is okay, the function wait_for_random_bytes()
495 * should be called and return 0 at least once at any point prior.
496 */
497
498#define DEFINE_BATCHED_ENTROPY(type) \
499struct batch_ ##type { \
500 /* \
501 * We make this 1.5x a ChaCha block, so that we get the \
502 * remaining 32 bytes from fast key erasure, plus one full \
503 * block from the detached ChaCha state. We can increase \
504 * the size of this later if needed so long as we keep the \
505 * formula of (integer_blocks + 0.5) * CHACHA_BLOCK_SIZE. \
506 */ \
507 type entropy[CHACHA_BLOCK_SIZE * 3 / (2 * sizeof(type))]; \
508 local_lock_t lock; \
509 unsigned long generation; \
510 unsigned int position; \
511}; \
512 \
513static DEFINE_PER_CPU(struct batch_ ##type, batched_entropy_ ##type) = { \
514 .lock = INIT_LOCAL_LOCK(batched_entropy_ ##type.lock), \
515 .position = UINT_MAX \
516}; \
517 \
518type get_random_ ##type(void) \
519{ \
520 type ret; \
521 unsigned long flags; \
522 struct batch_ ##type *batch; \
523 unsigned long next_gen; \
524 \
525 warn_unseeded_randomness(); \
526 \
527 if (!crng_ready()) { \
528 _get_random_bytes(&ret, sizeof(ret)); \
529 return ret; \
530 } \
531 \
532 local_lock_irqsave(&batched_entropy_ ##type.lock, flags); \
533 batch = raw_cpu_ptr(&batched_entropy_##type); \
534 \
535 next_gen = READ_ONCE(base_crng.generation); \
536 if (batch->position >= ARRAY_SIZE(batch->entropy) || \
537 next_gen != batch->generation) { \
538 _get_random_bytes(batch->entropy, sizeof(batch->entropy)); \
539 batch->position = 0; \
540 batch->generation = next_gen; \
541 } \
542 \
543 ret = batch->entropy[batch->position]; \
544 batch->entropy[batch->position] = 0; \
545 ++batch->position; \
546 local_unlock_irqrestore(&batched_entropy_ ##type.lock, flags); \
547 return ret; \
548} \
549EXPORT_SYMBOL(get_random_ ##type);
550
551DEFINE_BATCHED_ENTROPY(u8)
552DEFINE_BATCHED_ENTROPY(u16)
553DEFINE_BATCHED_ENTROPY(u32)
554DEFINE_BATCHED_ENTROPY(u64)
555
556u32 __get_random_u32_below(u32 ceil)
557{
558 /*
559 * This is the slow path for variable ceil. It is still fast, most of
560 * the time, by doing traditional reciprocal multiplication and
561 * opportunistically comparing the lower half to ceil itself, before
562 * falling back to computing a larger bound, and then rejecting samples
563 * whose lower half would indicate a range indivisible by ceil. The use
564 * of `-ceil % ceil` is analogous to `2^32 % ceil`, but is computable
565 * in 32-bits.
566 */
567 u32 rand = get_random_u32();
568 u64 mult;
569
570 /*
571 * This function is technically undefined for ceil == 0, and in fact
572 * for the non-underscored constant version in the header, we build bug
573 * on that. But for the non-constant case, it's convenient to have that
574 * evaluate to being a straight call to get_random_u32(), so that
575 * get_random_u32_inclusive() can work over its whole range without
576 * undefined behavior.
577 */
578 if (unlikely(!ceil))
579 return rand;
580
581 mult = (u64)ceil * rand;
582 if (unlikely((u32)mult < ceil)) {
583 u32 bound = -ceil % ceil;
584 while (unlikely((u32)mult < bound))
585 mult = (u64)ceil * get_random_u32();
586 }
587 return mult >> 32;
588}
589EXPORT_SYMBOL(__get_random_u32_below);
590
591#ifdef CONFIG_SMP
592/*
593 * This function is called when the CPU is coming up, with entry
594 * CPUHP_RANDOM_PREPARE, which comes before CPUHP_WORKQUEUE_PREP.
595 */
596int __cold random_prepare_cpu(unsigned int cpu)
597{
598 /*
599 * When the cpu comes back online, immediately invalidate both
600 * the per-cpu crng and all batches, so that we serve fresh
601 * randomness.
602 */
603 per_cpu_ptr(&crngs, cpu)->generation = ULONG_MAX;
604 per_cpu_ptr(&batched_entropy_u8, cpu)->position = UINT_MAX;
605 per_cpu_ptr(&batched_entropy_u16, cpu)->position = UINT_MAX;
606 per_cpu_ptr(&batched_entropy_u32, cpu)->position = UINT_MAX;
607 per_cpu_ptr(&batched_entropy_u64, cpu)->position = UINT_MAX;
608 return 0;
609}
610#endif
611
612
613/**********************************************************************
614 *
615 * Entropy accumulation and extraction routines.
616 *
617 * Callers may add entropy via:
618 *
619 * static void mix_pool_bytes(const void *buf, size_t len)
620 *
621 * After which, if added entropy should be credited:
622 *
623 * static void credit_init_bits(size_t bits)
624 *
625 * Finally, extract entropy via:
626 *
627 * static void extract_entropy(void *buf, size_t len)
628 *
629 **********************************************************************/
630
631enum {
632 POOL_BITS = BLAKE2S_HASH_SIZE * 8,
633 POOL_READY_BITS = POOL_BITS, /* When crng_init->CRNG_READY */
634 POOL_EARLY_BITS = POOL_READY_BITS / 2 /* When crng_init->CRNG_EARLY */
635};
636
637static struct {
638 struct blake2s_state hash;
639 spinlock_t lock;
640 unsigned int init_bits;
641} input_pool = {
642 .hash.h = { BLAKE2S_IV0 ^ (0x01010000 | BLAKE2S_HASH_SIZE),
643 BLAKE2S_IV1, BLAKE2S_IV2, BLAKE2S_IV3, BLAKE2S_IV4,
644 BLAKE2S_IV5, BLAKE2S_IV6, BLAKE2S_IV7 },
645 .hash.outlen = BLAKE2S_HASH_SIZE,
646 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
647};
648
649static void _mix_pool_bytes(const void *buf, size_t len)
650{
651 blake2s_update(&input_pool.hash, buf, len);
652}
653
654/*
655 * This function adds bytes into the input pool. It does not
656 * update the initialization bit counter; the caller should call
657 * credit_init_bits if this is appropriate.
658 */
659static void mix_pool_bytes(const void *buf, size_t len)
660{
661 unsigned long flags;
662
663 spin_lock_irqsave(&input_pool.lock, flags);
664 _mix_pool_bytes(buf, len);
665 spin_unlock_irqrestore(&input_pool.lock, flags);
666}
667
668/*
669 * This is an HKDF-like construction for using the hashed collected entropy
670 * as a PRF key, that's then expanded block-by-block.
671 */
672static void extract_entropy(void *buf, size_t len)
673{
674 unsigned long flags;
675 u8 seed[BLAKE2S_HASH_SIZE], next_key[BLAKE2S_HASH_SIZE];
676 struct {
677 unsigned long rdseed[32 / sizeof(long)];
678 size_t counter;
679 } block;
680 size_t i, longs;
681
682 for (i = 0; i < ARRAY_SIZE(block.rdseed);) {
683 longs = arch_get_random_seed_longs(&block.rdseed[i], ARRAY_SIZE(block.rdseed) - i);
684 if (longs) {
685 i += longs;
686 continue;
687 }
688 longs = arch_get_random_longs(&block.rdseed[i], ARRAY_SIZE(block.rdseed) - i);
689 if (longs) {
690 i += longs;
691 continue;
692 }
693 block.rdseed[i++] = random_get_entropy();
694 }
695
696 spin_lock_irqsave(&input_pool.lock, flags);
697
698 /* seed = HASHPRF(last_key, entropy_input) */
699 blake2s_final(&input_pool.hash, seed);
700
701 /* next_key = HASHPRF(seed, RDSEED || 0) */
702 block.counter = 0;
703 blake2s(next_key, (u8 *)&block, seed, sizeof(next_key), sizeof(block), sizeof(seed));
704 blake2s_init_key(&input_pool.hash, BLAKE2S_HASH_SIZE, next_key, sizeof(next_key));
705
706 spin_unlock_irqrestore(&input_pool.lock, flags);
707 memzero_explicit(next_key, sizeof(next_key));
708
709 while (len) {
710 i = min_t(size_t, len, BLAKE2S_HASH_SIZE);
711 /* output = HASHPRF(seed, RDSEED || ++counter) */
712 ++block.counter;
713 blake2s(buf, (u8 *)&block, seed, i, sizeof(block), sizeof(seed));
714 len -= i;
715 buf += i;
716 }
717
718 memzero_explicit(seed, sizeof(seed));
719 memzero_explicit(&block, sizeof(block));
720}
721
722#define credit_init_bits(bits) if (!crng_ready()) _credit_init_bits(bits)
723
724static void __cold _credit_init_bits(size_t bits)
725{
726 static DECLARE_WORK(set_ready, crng_set_ready);
727 unsigned int new, orig, add;
728 unsigned long flags;
729
730 if (!bits)
731 return;
732
733 add = min_t(size_t, bits, POOL_BITS);
734
735 orig = READ_ONCE(input_pool.init_bits);
736 do {
737 new = min_t(unsigned int, POOL_BITS, orig + add);
738 } while (!try_cmpxchg(&input_pool.init_bits, &orig, new));
739
740 if (orig < POOL_READY_BITS && new >= POOL_READY_BITS) {
741 crng_reseed(NULL); /* Sets crng_init to CRNG_READY under base_crng.lock. */
742 if (static_key_initialized && system_unbound_wq)
743 queue_work(system_unbound_wq, &set_ready);
744 atomic_notifier_call_chain(&random_ready_notifier, 0, NULL);
745#ifdef CONFIG_VDSO_GETRANDOM
746 WRITE_ONCE(__arch_get_k_vdso_rng_data()->is_ready, true);
747#endif
748 wake_up_interruptible(&crng_init_wait);
749 kill_fasync(&fasync, SIGIO, POLL_IN);
750 pr_notice("crng init done\n");
751 if (urandom_warning.missed)
752 pr_notice("%d urandom warning(s) missed due to ratelimiting\n",
753 urandom_warning.missed);
754 } else if (orig < POOL_EARLY_BITS && new >= POOL_EARLY_BITS) {
755 spin_lock_irqsave(&base_crng.lock, flags);
756 /* Check if crng_init is CRNG_EMPTY, to avoid race with crng_reseed(). */
757 if (crng_init == CRNG_EMPTY) {
758 extract_entropy(base_crng.key, sizeof(base_crng.key));
759 crng_init = CRNG_EARLY;
760 }
761 spin_unlock_irqrestore(&base_crng.lock, flags);
762 }
763}
764
765
766/**********************************************************************
767 *
768 * Entropy collection routines.
769 *
770 * The following exported functions are used for pushing entropy into
771 * the above entropy accumulation routines:
772 *
773 * void add_device_randomness(const void *buf, size_t len);
774 * void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy, bool sleep_after);
775 * void add_bootloader_randomness(const void *buf, size_t len);
776 * void add_vmfork_randomness(const void *unique_vm_id, size_t len);
777 * void add_interrupt_randomness(int irq);
778 * void add_input_randomness(unsigned int type, unsigned int code, unsigned int value);
779 * void add_disk_randomness(struct gendisk *disk);
780 *
781 * add_device_randomness() adds data to the input pool that
782 * is likely to differ between two devices (or possibly even per boot).
783 * This would be things like MAC addresses or serial numbers, or the
784 * read-out of the RTC. This does *not* credit any actual entropy to
785 * the pool, but it initializes the pool to different values for devices
786 * that might otherwise be identical and have very little entropy
787 * available to them (particularly common in the embedded world).
788 *
789 * add_hwgenerator_randomness() is for true hardware RNGs, and will credit
790 * entropy as specified by the caller. If the entropy pool is full it will
791 * block until more entropy is needed.
792 *
793 * add_bootloader_randomness() is called by bootloader drivers, such as EFI
794 * and device tree, and credits its input depending on whether or not the
795 * command line option 'random.trust_bootloader'.
796 *
797 * add_vmfork_randomness() adds a unique (but not necessarily secret) ID
798 * representing the current instance of a VM to the pool, without crediting,
799 * and then force-reseeds the crng so that it takes effect immediately.
800 *
801 * add_interrupt_randomness() uses the interrupt timing as random
802 * inputs to the entropy pool. Using the cycle counters and the irq source
803 * as inputs, it feeds the input pool roughly once a second or after 64
804 * interrupts, crediting 1 bit of entropy for whichever comes first.
805 *
806 * add_input_randomness() uses the input layer interrupt timing, as well
807 * as the event type information from the hardware.
808 *
809 * add_disk_randomness() uses what amounts to the seek time of block
810 * layer request events, on a per-disk_devt basis, as input to the
811 * entropy pool. Note that high-speed solid state drives with very low
812 * seek times do not make for good sources of entropy, as their seek
813 * times are usually fairly consistent.
814 *
815 * The last two routines try to estimate how many bits of entropy
816 * to credit. They do this by keeping track of the first and second
817 * order deltas of the event timings.
818 *
819 **********************************************************************/
820
821static bool trust_cpu __initdata = true;
822static bool trust_bootloader __initdata = true;
823static int __init parse_trust_cpu(char *arg)
824{
825 return kstrtobool(arg, &trust_cpu);
826}
827static int __init parse_trust_bootloader(char *arg)
828{
829 return kstrtobool(arg, &trust_bootloader);
830}
831early_param("random.trust_cpu", parse_trust_cpu);
832early_param("random.trust_bootloader", parse_trust_bootloader);
833
834static int random_pm_notification(struct notifier_block *nb, unsigned long action, void *data)
835{
836 unsigned long flags, entropy = random_get_entropy();
837
838 /*
839 * Encode a representation of how long the system has been suspended,
840 * in a way that is distinct from prior system suspends.
841 */
842 ktime_t stamps[] = { ktime_get(), ktime_get_boottime(), ktime_get_real() };
843
844 spin_lock_irqsave(&input_pool.lock, flags);
845 _mix_pool_bytes(&action, sizeof(action));
846 _mix_pool_bytes(stamps, sizeof(stamps));
847 _mix_pool_bytes(&entropy, sizeof(entropy));
848 spin_unlock_irqrestore(&input_pool.lock, flags);
849
850 if (crng_ready() && (action == PM_RESTORE_PREPARE ||
851 (action == PM_POST_SUSPEND && !IS_ENABLED(CONFIG_PM_AUTOSLEEP) &&
852 !IS_ENABLED(CONFIG_PM_USERSPACE_AUTOSLEEP)))) {
853 crng_reseed(NULL);
854 pr_notice("crng reseeded on system resumption\n");
855 }
856 return 0;
857}
858
859static struct notifier_block pm_notifier = { .notifier_call = random_pm_notification };
860
861/*
862 * This is called extremely early, before time keeping functionality is
863 * available, but arch randomness is. Interrupts are not yet enabled.
864 */
865void __init random_init_early(const char *command_line)
866{
867 unsigned long entropy[BLAKE2S_BLOCK_SIZE / sizeof(long)];
868 size_t i, longs, arch_bits;
869
870#if defined(LATENT_ENTROPY_PLUGIN)
871 static const u8 compiletime_seed[BLAKE2S_BLOCK_SIZE] __initconst __latent_entropy;
872 _mix_pool_bytes(compiletime_seed, sizeof(compiletime_seed));
873#endif
874
875 for (i = 0, arch_bits = sizeof(entropy) * 8; i < ARRAY_SIZE(entropy);) {
876 longs = arch_get_random_seed_longs(entropy, ARRAY_SIZE(entropy) - i);
877 if (longs) {
878 _mix_pool_bytes(entropy, sizeof(*entropy) * longs);
879 i += longs;
880 continue;
881 }
882 longs = arch_get_random_longs(entropy, ARRAY_SIZE(entropy) - i);
883 if (longs) {
884 _mix_pool_bytes(entropy, sizeof(*entropy) * longs);
885 i += longs;
886 continue;
887 }
888 arch_bits -= sizeof(*entropy) * 8;
889 ++i;
890 }
891
892 _mix_pool_bytes(init_utsname(), sizeof(*(init_utsname())));
893 _mix_pool_bytes(command_line, strlen(command_line));
894
895 /* Reseed if already seeded by earlier phases. */
896 if (crng_ready())
897 crng_reseed(NULL);
898 else if (trust_cpu)
899 _credit_init_bits(arch_bits);
900}
901
902/*
903 * This is called a little bit after the prior function, and now there is
904 * access to timestamps counters. Interrupts are not yet enabled.
905 */
906void __init random_init(void)
907{
908 unsigned long entropy = random_get_entropy();
909 ktime_t now = ktime_get_real();
910
911 _mix_pool_bytes(&now, sizeof(now));
912 _mix_pool_bytes(&entropy, sizeof(entropy));
913 add_latent_entropy();
914
915 /*
916 * If we were initialized by the cpu or bootloader before jump labels
917 * or workqueues are initialized, then we should enable the static
918 * branch here, where it's guaranteed that these have been initialized.
919 */
920 if (!static_branch_likely(&crng_is_ready) && crng_init >= CRNG_READY)
921 crng_set_ready(NULL);
922
923 /* Reseed if already seeded by earlier phases. */
924 if (crng_ready())
925 crng_reseed(NULL);
926
927 WARN_ON(register_pm_notifier(&pm_notifier));
928
929 WARN(!entropy, "Missing cycle counter and fallback timer; RNG "
930 "entropy collection will consequently suffer.");
931}
932
933/*
934 * Add device- or boot-specific data to the input pool to help
935 * initialize it.
936 *
937 * None of this adds any entropy; it is meant to avoid the problem of
938 * the entropy pool having similar initial state across largely
939 * identical devices.
940 */
941void add_device_randomness(const void *buf, size_t len)
942{
943 unsigned long entropy = random_get_entropy();
944 unsigned long flags;
945
946 spin_lock_irqsave(&input_pool.lock, flags);
947 _mix_pool_bytes(&entropy, sizeof(entropy));
948 _mix_pool_bytes(buf, len);
949 spin_unlock_irqrestore(&input_pool.lock, flags);
950}
951EXPORT_SYMBOL(add_device_randomness);
952
953/*
954 * Interface for in-kernel drivers of true hardware RNGs. Those devices
955 * may produce endless random bits, so this function will sleep for
956 * some amount of time after, if the sleep_after parameter is true.
957 */
958void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy, bool sleep_after)
959{
960 mix_pool_bytes(buf, len);
961 credit_init_bits(entropy);
962
963 /*
964 * Throttle writing to once every reseed interval, unless we're not yet
965 * initialized or no entropy is credited.
966 */
967 if (sleep_after && !kthread_should_stop() && (crng_ready() || !entropy))
968 schedule_timeout_interruptible(crng_reseed_interval());
969}
970EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
971
972/*
973 * Handle random seed passed by bootloader, and credit it depending
974 * on the command line option 'random.trust_bootloader'.
975 */
976void __init add_bootloader_randomness(const void *buf, size_t len)
977{
978 mix_pool_bytes(buf, len);
979 if (trust_bootloader)
980 credit_init_bits(len * 8);
981}
982
983#if IS_ENABLED(CONFIG_VMGENID)
984static BLOCKING_NOTIFIER_HEAD(vmfork_chain);
985
986/*
987 * Handle a new unique VM ID, which is unique, not secret, so we
988 * don't credit it, but we do immediately force a reseed after so
989 * that it's used by the crng posthaste.
990 */
991void __cold add_vmfork_randomness(const void *unique_vm_id, size_t len)
992{
993 add_device_randomness(unique_vm_id, len);
994 if (crng_ready()) {
995 crng_reseed(NULL);
996 pr_notice("crng reseeded due to virtual machine fork\n");
997 }
998 blocking_notifier_call_chain(&vmfork_chain, 0, NULL);
999}
1000#if IS_MODULE(CONFIG_VMGENID)
1001EXPORT_SYMBOL_GPL(add_vmfork_randomness);
1002#endif
1003
1004int __cold register_random_vmfork_notifier(struct notifier_block *nb)
1005{
1006 return blocking_notifier_chain_register(&vmfork_chain, nb);
1007}
1008EXPORT_SYMBOL_GPL(register_random_vmfork_notifier);
1009
1010int __cold unregister_random_vmfork_notifier(struct notifier_block *nb)
1011{
1012 return blocking_notifier_chain_unregister(&vmfork_chain, nb);
1013}
1014EXPORT_SYMBOL_GPL(unregister_random_vmfork_notifier);
1015#endif
1016
1017struct fast_pool {
1018 unsigned long pool[4];
1019 unsigned long last;
1020 unsigned int count;
1021 struct timer_list mix;
1022};
1023
1024static void mix_interrupt_randomness(struct timer_list *work);
1025
1026static DEFINE_PER_CPU(struct fast_pool, irq_randomness) = {
1027#ifdef CONFIG_64BIT
1028#define FASTMIX_PERM SIPHASH_PERMUTATION
1029 .pool = { SIPHASH_CONST_0, SIPHASH_CONST_1, SIPHASH_CONST_2, SIPHASH_CONST_3 },
1030#else
1031#define FASTMIX_PERM HSIPHASH_PERMUTATION
1032 .pool = { HSIPHASH_CONST_0, HSIPHASH_CONST_1, HSIPHASH_CONST_2, HSIPHASH_CONST_3 },
1033#endif
1034 .mix = __TIMER_INITIALIZER(mix_interrupt_randomness, 0)
1035};
1036
1037/*
1038 * This is [Half]SipHash-1-x, starting from an empty key. Because
1039 * the key is fixed, it assumes that its inputs are non-malicious,
1040 * and therefore this has no security on its own. s represents the
1041 * four-word SipHash state, while v represents a two-word input.
1042 */
1043static void fast_mix(unsigned long s[4], unsigned long v1, unsigned long v2)
1044{
1045 s[3] ^= v1;
1046 FASTMIX_PERM(s[0], s[1], s[2], s[3]);
1047 s[0] ^= v1;
1048 s[3] ^= v2;
1049 FASTMIX_PERM(s[0], s[1], s[2], s[3]);
1050 s[0] ^= v2;
1051}
1052
1053#ifdef CONFIG_SMP
1054/*
1055 * This function is called when the CPU has just come online, with
1056 * entry CPUHP_AP_RANDOM_ONLINE, just after CPUHP_AP_WORKQUEUE_ONLINE.
1057 */
1058int __cold random_online_cpu(unsigned int cpu)
1059{
1060 /*
1061 * During CPU shutdown and before CPU onlining, add_interrupt_
1062 * randomness() may schedule mix_interrupt_randomness(), and
1063 * set the MIX_INFLIGHT flag. However, because the worker can
1064 * be scheduled on a different CPU during this period, that
1065 * flag will never be cleared. For that reason, we zero out
1066 * the flag here, which runs just after workqueues are onlined
1067 * for the CPU again. This also has the effect of setting the
1068 * irq randomness count to zero so that new accumulated irqs
1069 * are fresh.
1070 */
1071 per_cpu_ptr(&irq_randomness, cpu)->count = 0;
1072 return 0;
1073}
1074#endif
1075
1076static void mix_interrupt_randomness(struct timer_list *work)
1077{
1078 struct fast_pool *fast_pool = container_of(work, struct fast_pool, mix);
1079 /*
1080 * The size of the copied stack pool is explicitly 2 longs so that we
1081 * only ever ingest half of the siphash output each time, retaining
1082 * the other half as the next "key" that carries over. The entropy is
1083 * supposed to be sufficiently dispersed between bits so on average
1084 * we don't wind up "losing" some.
1085 */
1086 unsigned long pool[2];
1087 unsigned int count;
1088
1089 /* Check to see if we're running on the wrong CPU due to hotplug. */
1090 local_irq_disable();
1091 if (fast_pool != this_cpu_ptr(&irq_randomness)) {
1092 local_irq_enable();
1093 return;
1094 }
1095
1096 /*
1097 * Copy the pool to the stack so that the mixer always has a
1098 * consistent view, before we reenable irqs again.
1099 */
1100 memcpy(pool, fast_pool->pool, sizeof(pool));
1101 count = fast_pool->count;
1102 fast_pool->count = 0;
1103 fast_pool->last = jiffies;
1104 local_irq_enable();
1105
1106 mix_pool_bytes(pool, sizeof(pool));
1107 credit_init_bits(clamp_t(unsigned int, (count & U16_MAX) / 64, 1, sizeof(pool) * 8));
1108
1109 memzero_explicit(pool, sizeof(pool));
1110}
1111
1112void add_interrupt_randomness(int irq)
1113{
1114 enum { MIX_INFLIGHT = 1U << 31 };
1115 unsigned long entropy = random_get_entropy();
1116 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1117 struct pt_regs *regs = get_irq_regs();
1118 unsigned int new_count;
1119
1120 fast_mix(fast_pool->pool, entropy,
1121 (regs ? instruction_pointer(regs) : _RET_IP_) ^ swab(irq));
1122 new_count = ++fast_pool->count;
1123
1124 if (new_count & MIX_INFLIGHT)
1125 return;
1126
1127 if (new_count < 1024 && !time_is_before_jiffies(fast_pool->last + HZ))
1128 return;
1129
1130 fast_pool->count |= MIX_INFLIGHT;
1131 if (!timer_pending(&fast_pool->mix)) {
1132 fast_pool->mix.expires = jiffies;
1133 add_timer_on(&fast_pool->mix, raw_smp_processor_id());
1134 }
1135}
1136EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1137
1138/* There is one of these per entropy source */
1139struct timer_rand_state {
1140 unsigned long last_time;
1141 long last_delta, last_delta2;
1142};
1143
1144/*
1145 * This function adds entropy to the entropy "pool" by using timing
1146 * delays. It uses the timer_rand_state structure to make an estimate
1147 * of how many bits of entropy this call has added to the pool. The
1148 * value "num" is also added to the pool; it should somehow describe
1149 * the type of event that just happened.
1150 */
1151static void add_timer_randomness(struct timer_rand_state *state, unsigned int num)
1152{
1153 unsigned long entropy = random_get_entropy(), now = jiffies, flags;
1154 long delta, delta2, delta3;
1155 unsigned int bits;
1156
1157 /*
1158 * If we're in a hard IRQ, add_interrupt_randomness() will be called
1159 * sometime after, so mix into the fast pool.
1160 */
1161 if (in_hardirq()) {
1162 fast_mix(this_cpu_ptr(&irq_randomness)->pool, entropy, num);
1163 } else {
1164 spin_lock_irqsave(&input_pool.lock, flags);
1165 _mix_pool_bytes(&entropy, sizeof(entropy));
1166 _mix_pool_bytes(&num, sizeof(num));
1167 spin_unlock_irqrestore(&input_pool.lock, flags);
1168 }
1169
1170 if (crng_ready())
1171 return;
1172
1173 /*
1174 * Calculate number of bits of randomness we probably added.
1175 * We take into account the first, second and third-order deltas
1176 * in order to make our estimate.
1177 */
1178 delta = now - READ_ONCE(state->last_time);
1179 WRITE_ONCE(state->last_time, now);
1180
1181 delta2 = delta - READ_ONCE(state->last_delta);
1182 WRITE_ONCE(state->last_delta, delta);
1183
1184 delta3 = delta2 - READ_ONCE(state->last_delta2);
1185 WRITE_ONCE(state->last_delta2, delta2);
1186
1187 if (delta < 0)
1188 delta = -delta;
1189 if (delta2 < 0)
1190 delta2 = -delta2;
1191 if (delta3 < 0)
1192 delta3 = -delta3;
1193 if (delta > delta2)
1194 delta = delta2;
1195 if (delta > delta3)
1196 delta = delta3;
1197
1198 /*
1199 * delta is now minimum absolute delta. Round down by 1 bit
1200 * on general principles, and limit entropy estimate to 11 bits.
1201 */
1202 bits = min(fls(delta >> 1), 11);
1203
1204 /*
1205 * As mentioned above, if we're in a hard IRQ, add_interrupt_randomness()
1206 * will run after this, which uses a different crediting scheme of 1 bit
1207 * per every 64 interrupts. In order to let that function do accounting
1208 * close to the one in this function, we credit a full 64/64 bit per bit,
1209 * and then subtract one to account for the extra one added.
1210 */
1211 if (in_hardirq())
1212 this_cpu_ptr(&irq_randomness)->count += max(1u, bits * 64) - 1;
1213 else
1214 _credit_init_bits(bits);
1215}
1216
1217void add_input_randomness(unsigned int type, unsigned int code, unsigned int value)
1218{
1219 static unsigned char last_value;
1220 static struct timer_rand_state input_timer_state = { INITIAL_JIFFIES };
1221
1222 /* Ignore autorepeat and the like. */
1223 if (value == last_value)
1224 return;
1225
1226 last_value = value;
1227 add_timer_randomness(&input_timer_state,
1228 (type << 4) ^ code ^ (code >> 4) ^ value);
1229}
1230EXPORT_SYMBOL_GPL(add_input_randomness);
1231
1232#ifdef CONFIG_BLOCK
1233void add_disk_randomness(struct gendisk *disk)
1234{
1235 if (!disk || !disk->random)
1236 return;
1237 /* First major is 1, so we get >= 0x200 here. */
1238 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1239}
1240EXPORT_SYMBOL_GPL(add_disk_randomness);
1241
1242void __cold rand_initialize_disk(struct gendisk *disk)
1243{
1244 struct timer_rand_state *state;
1245
1246 /*
1247 * If kzalloc returns null, we just won't use that entropy
1248 * source.
1249 */
1250 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1251 if (state) {
1252 state->last_time = INITIAL_JIFFIES;
1253 disk->random = state;
1254 }
1255}
1256#endif
1257
1258struct entropy_timer_state {
1259 unsigned long entropy;
1260 struct timer_list timer;
1261 atomic_t samples;
1262 unsigned int samples_per_bit;
1263};
1264
1265/*
1266 * Each time the timer fires, we expect that we got an unpredictable jump in
1267 * the cycle counter. Even if the timer is running on another CPU, the timer
1268 * activity will be touching the stack of the CPU that is generating entropy.
1269 *
1270 * Note that we don't re-arm the timer in the timer itself - we are happy to be
1271 * scheduled away, since that just makes the load more complex, but we do not
1272 * want the timer to keep ticking unless the entropy loop is running.
1273 *
1274 * So the re-arming always happens in the entropy loop itself.
1275 */
1276static void __cold entropy_timer(struct timer_list *timer)
1277{
1278 struct entropy_timer_state *state = container_of(timer, struct entropy_timer_state, timer);
1279 unsigned long entropy = random_get_entropy();
1280
1281 mix_pool_bytes(&entropy, sizeof(entropy));
1282 if (atomic_inc_return(&state->samples) % state->samples_per_bit == 0)
1283 credit_init_bits(1);
1284}
1285
1286/*
1287 * If we have an actual cycle counter, see if we can generate enough entropy
1288 * with timing noise.
1289 */
1290static void __cold try_to_generate_entropy(void)
1291{
1292 enum { NUM_TRIAL_SAMPLES = 8192, MAX_SAMPLES_PER_BIT = HZ / 15 };
1293 u8 stack_bytes[sizeof(struct entropy_timer_state) + SMP_CACHE_BYTES - 1];
1294 struct entropy_timer_state *stack = PTR_ALIGN((void *)stack_bytes, SMP_CACHE_BYTES);
1295 unsigned int i, num_different = 0;
1296 unsigned long last = random_get_entropy();
1297 int cpu = -1;
1298
1299 for (i = 0; i < NUM_TRIAL_SAMPLES - 1; ++i) {
1300 stack->entropy = random_get_entropy();
1301 if (stack->entropy != last)
1302 ++num_different;
1303 last = stack->entropy;
1304 }
1305 stack->samples_per_bit = DIV_ROUND_UP(NUM_TRIAL_SAMPLES, num_different + 1);
1306 if (stack->samples_per_bit > MAX_SAMPLES_PER_BIT)
1307 return;
1308
1309 atomic_set(&stack->samples, 0);
1310 timer_setup_on_stack(&stack->timer, entropy_timer, 0);
1311 while (!crng_ready() && !signal_pending(current)) {
1312 /*
1313 * Check !timer_pending() and then ensure that any previous callback has finished
1314 * executing by checking try_to_del_timer_sync(), before queueing the next one.
1315 */
1316 if (!timer_pending(&stack->timer) && try_to_del_timer_sync(&stack->timer) >= 0) {
1317 struct cpumask timer_cpus;
1318 unsigned int num_cpus;
1319
1320 /*
1321 * Preemption must be disabled here, both to read the current CPU number
1322 * and to avoid scheduling a timer on a dead CPU.
1323 */
1324 preempt_disable();
1325
1326 /* Only schedule callbacks on timer CPUs that are online. */
1327 cpumask_and(&timer_cpus, housekeeping_cpumask(HK_TYPE_TIMER), cpu_online_mask);
1328 num_cpus = cpumask_weight(&timer_cpus);
1329 /* In very bizarre case of misconfiguration, fallback to all online. */
1330 if (unlikely(num_cpus == 0)) {
1331 timer_cpus = *cpu_online_mask;
1332 num_cpus = cpumask_weight(&timer_cpus);
1333 }
1334
1335 /* Basic CPU round-robin, which avoids the current CPU. */
1336 do {
1337 cpu = cpumask_next(cpu, &timer_cpus);
1338 if (cpu >= nr_cpu_ids)
1339 cpu = cpumask_first(&timer_cpus);
1340 } while (cpu == smp_processor_id() && num_cpus > 1);
1341
1342 /* Expiring the timer at `jiffies` means it's the next tick. */
1343 stack->timer.expires = jiffies;
1344
1345 add_timer_on(&stack->timer, cpu);
1346
1347 preempt_enable();
1348 }
1349 mix_pool_bytes(&stack->entropy, sizeof(stack->entropy));
1350 schedule();
1351 stack->entropy = random_get_entropy();
1352 }
1353 mix_pool_bytes(&stack->entropy, sizeof(stack->entropy));
1354
1355 del_timer_sync(&stack->timer);
1356 destroy_timer_on_stack(&stack->timer);
1357}
1358
1359
1360/**********************************************************************
1361 *
1362 * Userspace reader/writer interfaces.
1363 *
1364 * getrandom(2) is the primary modern interface into the RNG and should
1365 * be used in preference to anything else.
1366 *
1367 * Reading from /dev/random has the same functionality as calling
1368 * getrandom(2) with flags=0. In earlier versions, however, it had
1369 * vastly different semantics and should therefore be avoided, to
1370 * prevent backwards compatibility issues.
1371 *
1372 * Reading from /dev/urandom has the same functionality as calling
1373 * getrandom(2) with flags=GRND_INSECURE. Because it does not block
1374 * waiting for the RNG to be ready, it should not be used.
1375 *
1376 * Writing to either /dev/random or /dev/urandom adds entropy to
1377 * the input pool but does not credit it.
1378 *
1379 * Polling on /dev/random indicates when the RNG is initialized, on
1380 * the read side, and when it wants new entropy, on the write side.
1381 *
1382 * Both /dev/random and /dev/urandom have the same set of ioctls for
1383 * adding entropy, getting the entropy count, zeroing the count, and
1384 * reseeding the crng.
1385 *
1386 **********************************************************************/
1387
1388SYSCALL_DEFINE3(getrandom, char __user *, ubuf, size_t, len, unsigned int, flags)
1389{
1390 struct iov_iter iter;
1391 int ret;
1392
1393 if (flags & ~(GRND_NONBLOCK | GRND_RANDOM | GRND_INSECURE))
1394 return -EINVAL;
1395
1396 /*
1397 * Requesting insecure and blocking randomness at the same time makes
1398 * no sense.
1399 */
1400 if ((flags & (GRND_INSECURE | GRND_RANDOM)) == (GRND_INSECURE | GRND_RANDOM))
1401 return -EINVAL;
1402
1403 if (!crng_ready() && !(flags & GRND_INSECURE)) {
1404 if (flags & GRND_NONBLOCK)
1405 return -EAGAIN;
1406 ret = wait_for_random_bytes();
1407 if (unlikely(ret))
1408 return ret;
1409 }
1410
1411 ret = import_ubuf(ITER_DEST, ubuf, len, &iter);
1412 if (unlikely(ret))
1413 return ret;
1414 return get_random_bytes_user(&iter);
1415}
1416
1417static __poll_t random_poll(struct file *file, poll_table *wait)
1418{
1419 poll_wait(file, &crng_init_wait, wait);
1420 return crng_ready() ? EPOLLIN | EPOLLRDNORM : EPOLLOUT | EPOLLWRNORM;
1421}
1422
1423static ssize_t write_pool_user(struct iov_iter *iter)
1424{
1425 u8 block[BLAKE2S_BLOCK_SIZE];
1426 ssize_t ret = 0;
1427 size_t copied;
1428
1429 if (unlikely(!iov_iter_count(iter)))
1430 return 0;
1431
1432 for (;;) {
1433 copied = copy_from_iter(block, sizeof(block), iter);
1434 ret += copied;
1435 mix_pool_bytes(block, copied);
1436 if (!iov_iter_count(iter) || copied != sizeof(block))
1437 break;
1438
1439 BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0);
1440 if (ret % PAGE_SIZE == 0) {
1441 if (signal_pending(current))
1442 break;
1443 cond_resched();
1444 }
1445 }
1446
1447 memzero_explicit(block, sizeof(block));
1448 return ret ? ret : -EFAULT;
1449}
1450
1451static ssize_t random_write_iter(struct kiocb *kiocb, struct iov_iter *iter)
1452{
1453 return write_pool_user(iter);
1454}
1455
1456static ssize_t urandom_read_iter(struct kiocb *kiocb, struct iov_iter *iter)
1457{
1458 static int maxwarn = 10;
1459
1460 /*
1461 * Opportunistically attempt to initialize the RNG on platforms that
1462 * have fast cycle counters, but don't (for now) require it to succeed.
1463 */
1464 if (!crng_ready())
1465 try_to_generate_entropy();
1466
1467 if (!crng_ready()) {
1468 if (!ratelimit_disable && maxwarn <= 0)
1469 ++urandom_warning.missed;
1470 else if (ratelimit_disable || __ratelimit(&urandom_warning)) {
1471 --maxwarn;
1472 pr_notice("%s: uninitialized urandom read (%zu bytes read)\n",
1473 current->comm, iov_iter_count(iter));
1474 }
1475 }
1476
1477 return get_random_bytes_user(iter);
1478}
1479
1480static ssize_t random_read_iter(struct kiocb *kiocb, struct iov_iter *iter)
1481{
1482 int ret;
1483
1484 if (!crng_ready() &&
1485 ((kiocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO)) ||
1486 (kiocb->ki_filp->f_flags & O_NONBLOCK)))
1487 return -EAGAIN;
1488
1489 ret = wait_for_random_bytes();
1490 if (ret != 0)
1491 return ret;
1492 return get_random_bytes_user(iter);
1493}
1494
1495static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1496{
1497 int __user *p = (int __user *)arg;
1498 int ent_count;
1499
1500 switch (cmd) {
1501 case RNDGETENTCNT:
1502 /* Inherently racy, no point locking. */
1503 if (put_user(input_pool.init_bits, p))
1504 return -EFAULT;
1505 return 0;
1506 case RNDADDTOENTCNT:
1507 if (!capable(CAP_SYS_ADMIN))
1508 return -EPERM;
1509 if (get_user(ent_count, p))
1510 return -EFAULT;
1511 if (ent_count < 0)
1512 return -EINVAL;
1513 credit_init_bits(ent_count);
1514 return 0;
1515 case RNDADDENTROPY: {
1516 struct iov_iter iter;
1517 ssize_t ret;
1518 int len;
1519
1520 if (!capable(CAP_SYS_ADMIN))
1521 return -EPERM;
1522 if (get_user(ent_count, p++))
1523 return -EFAULT;
1524 if (ent_count < 0)
1525 return -EINVAL;
1526 if (get_user(len, p++))
1527 return -EFAULT;
1528 ret = import_ubuf(ITER_SOURCE, p, len, &iter);
1529 if (unlikely(ret))
1530 return ret;
1531 ret = write_pool_user(&iter);
1532 if (unlikely(ret < 0))
1533 return ret;
1534 /* Since we're crediting, enforce that it was all written into the pool. */
1535 if (unlikely(ret != len))
1536 return -EFAULT;
1537 credit_init_bits(ent_count);
1538 return 0;
1539 }
1540 case RNDZAPENTCNT:
1541 case RNDCLEARPOOL:
1542 /* No longer has any effect. */
1543 if (!capable(CAP_SYS_ADMIN))
1544 return -EPERM;
1545 return 0;
1546 case RNDRESEEDCRNG:
1547 if (!capable(CAP_SYS_ADMIN))
1548 return -EPERM;
1549 if (!crng_ready())
1550 return -ENODATA;
1551 crng_reseed(NULL);
1552 return 0;
1553 default:
1554 return -EINVAL;
1555 }
1556}
1557
1558static int random_fasync(int fd, struct file *filp, int on)
1559{
1560 return fasync_helper(fd, filp, on, &fasync);
1561}
1562
1563const struct file_operations random_fops = {
1564 .read_iter = random_read_iter,
1565 .write_iter = random_write_iter,
1566 .poll = random_poll,
1567 .unlocked_ioctl = random_ioctl,
1568 .compat_ioctl = compat_ptr_ioctl,
1569 .fasync = random_fasync,
1570 .llseek = noop_llseek,
1571 .splice_read = copy_splice_read,
1572 .splice_write = iter_file_splice_write,
1573};
1574
1575const struct file_operations urandom_fops = {
1576 .read_iter = urandom_read_iter,
1577 .write_iter = random_write_iter,
1578 .unlocked_ioctl = random_ioctl,
1579 .compat_ioctl = compat_ptr_ioctl,
1580 .fasync = random_fasync,
1581 .llseek = noop_llseek,
1582 .splice_read = copy_splice_read,
1583 .splice_write = iter_file_splice_write,
1584};
1585
1586
1587/********************************************************************
1588 *
1589 * Sysctl interface.
1590 *
1591 * These are partly unused legacy knobs with dummy values to not break
1592 * userspace and partly still useful things. They are usually accessible
1593 * in /proc/sys/kernel/random/ and are as follows:
1594 *
1595 * - boot_id - a UUID representing the current boot.
1596 *
1597 * - uuid - a random UUID, different each time the file is read.
1598 *
1599 * - poolsize - the number of bits of entropy that the input pool can
1600 * hold, tied to the POOL_BITS constant.
1601 *
1602 * - entropy_avail - the number of bits of entropy currently in the
1603 * input pool. Always <= poolsize.
1604 *
1605 * - write_wakeup_threshold - the amount of entropy in the input pool
1606 * below which write polls to /dev/random will unblock, requesting
1607 * more entropy, tied to the POOL_READY_BITS constant. It is writable
1608 * to avoid breaking old userspaces, but writing to it does not
1609 * change any behavior of the RNG.
1610 *
1611 * - urandom_min_reseed_secs - fixed to the value CRNG_RESEED_INTERVAL.
1612 * It is writable to avoid breaking old userspaces, but writing
1613 * to it does not change any behavior of the RNG.
1614 *
1615 ********************************************************************/
1616
1617#ifdef CONFIG_SYSCTL
1618
1619#include <linux/sysctl.h>
1620
1621static int sysctl_random_min_urandom_seed = CRNG_RESEED_INTERVAL / HZ;
1622static int sysctl_random_write_wakeup_bits = POOL_READY_BITS;
1623static int sysctl_poolsize = POOL_BITS;
1624static u8 sysctl_bootid[UUID_SIZE];
1625
1626/*
1627 * This function is used to return both the bootid UUID, and random
1628 * UUID. The difference is in whether table->data is NULL; if it is,
1629 * then a new UUID is generated and returned to the user.
1630 */
1631static int proc_do_uuid(const struct ctl_table *table, int write, void *buf,
1632 size_t *lenp, loff_t *ppos)
1633{
1634 u8 tmp_uuid[UUID_SIZE], *uuid;
1635 char uuid_string[UUID_STRING_LEN + 1];
1636 struct ctl_table fake_table = {
1637 .data = uuid_string,
1638 .maxlen = UUID_STRING_LEN
1639 };
1640
1641 if (write)
1642 return -EPERM;
1643
1644 uuid = table->data;
1645 if (!uuid) {
1646 uuid = tmp_uuid;
1647 generate_random_uuid(uuid);
1648 } else {
1649 static DEFINE_SPINLOCK(bootid_spinlock);
1650
1651 spin_lock(&bootid_spinlock);
1652 if (!uuid[8])
1653 generate_random_uuid(uuid);
1654 spin_unlock(&bootid_spinlock);
1655 }
1656
1657 snprintf(uuid_string, sizeof(uuid_string), "%pU", uuid);
1658 return proc_dostring(&fake_table, 0, buf, lenp, ppos);
1659}
1660
1661/* The same as proc_dointvec, but writes don't change anything. */
1662static int proc_do_rointvec(const struct ctl_table *table, int write, void *buf,
1663 size_t *lenp, loff_t *ppos)
1664{
1665 return write ? 0 : proc_dointvec(table, 0, buf, lenp, ppos);
1666}
1667
1668static struct ctl_table random_table[] = {
1669 {
1670 .procname = "poolsize",
1671 .data = &sysctl_poolsize,
1672 .maxlen = sizeof(int),
1673 .mode = 0444,
1674 .proc_handler = proc_dointvec,
1675 },
1676 {
1677 .procname = "entropy_avail",
1678 .data = &input_pool.init_bits,
1679 .maxlen = sizeof(int),
1680 .mode = 0444,
1681 .proc_handler = proc_dointvec,
1682 },
1683 {
1684 .procname = "write_wakeup_threshold",
1685 .data = &sysctl_random_write_wakeup_bits,
1686 .maxlen = sizeof(int),
1687 .mode = 0644,
1688 .proc_handler = proc_do_rointvec,
1689 },
1690 {
1691 .procname = "urandom_min_reseed_secs",
1692 .data = &sysctl_random_min_urandom_seed,
1693 .maxlen = sizeof(int),
1694 .mode = 0644,
1695 .proc_handler = proc_do_rointvec,
1696 },
1697 {
1698 .procname = "boot_id",
1699 .data = &sysctl_bootid,
1700 .mode = 0444,
1701 .proc_handler = proc_do_uuid,
1702 },
1703 {
1704 .procname = "uuid",
1705 .mode = 0444,
1706 .proc_handler = proc_do_uuid,
1707 },
1708};
1709
1710/*
1711 * random_init() is called before sysctl_init(),
1712 * so we cannot call register_sysctl_init() in random_init()
1713 */
1714static int __init random_sysctls_init(void)
1715{
1716 register_sysctl_init("kernel/random", random_table);
1717 return 0;
1718}
1719device_initcall(random_sysctls_init);
1720#endif
1/*
2 * random.c -- A strong random number generator
3 *
4 * Copyright (C) 2017 Jason A. Donenfeld <Jason@zx2c4.com>. All
5 * Rights Reserved.
6 *
7 * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005
8 *
9 * Copyright Theodore Ts'o, 1994, 1995, 1996, 1997, 1998, 1999. All
10 * rights reserved.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, and the entire permission notice in its entirety,
17 * including the disclaimer of warranties.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. The name of the author may not be used to endorse or promote
22 * products derived from this software without specific prior
23 * written permission.
24 *
25 * ALTERNATIVELY, this product may be distributed under the terms of
26 * the GNU General Public License, in which case the provisions of the GPL are
27 * required INSTEAD OF the above restrictions. (This clause is
28 * necessary due to a potential bad interaction between the GPL and
29 * the restrictions contained in a BSD-style copyright.)
30 *
31 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
32 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
33 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ALL OF
34 * WHICH ARE HEREBY DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE
35 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
36 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT
37 * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
38 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
39 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
40 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
41 * USE OF THIS SOFTWARE, EVEN IF NOT ADVISED OF THE POSSIBILITY OF SUCH
42 * DAMAGE.
43 */
44
45/*
46 * (now, with legal B.S. out of the way.....)
47 *
48 * This routine gathers environmental noise from device drivers, etc.,
49 * and returns good random numbers, suitable for cryptographic use.
50 * Besides the obvious cryptographic uses, these numbers are also good
51 * for seeding TCP sequence numbers, and other places where it is
52 * desirable to have numbers which are not only random, but hard to
53 * predict by an attacker.
54 *
55 * Theory of operation
56 * ===================
57 *
58 * Computers are very predictable devices. Hence it is extremely hard
59 * to produce truly random numbers on a computer --- as opposed to
60 * pseudo-random numbers, which can easily generated by using a
61 * algorithm. Unfortunately, it is very easy for attackers to guess
62 * the sequence of pseudo-random number generators, and for some
63 * applications this is not acceptable. So instead, we must try to
64 * gather "environmental noise" from the computer's environment, which
65 * must be hard for outside attackers to observe, and use that to
66 * generate random numbers. In a Unix environment, this is best done
67 * from inside the kernel.
68 *
69 * Sources of randomness from the environment include inter-keyboard
70 * timings, inter-interrupt timings from some interrupts, and other
71 * events which are both (a) non-deterministic and (b) hard for an
72 * outside observer to measure. Randomness from these sources are
73 * added to an "entropy pool", which is mixed using a CRC-like function.
74 * This is not cryptographically strong, but it is adequate assuming
75 * the randomness is not chosen maliciously, and it is fast enough that
76 * the overhead of doing it on every interrupt is very reasonable.
77 * As random bytes are mixed into the entropy pool, the routines keep
78 * an *estimate* of how many bits of randomness have been stored into
79 * the random number generator's internal state.
80 *
81 * When random bytes are desired, they are obtained by taking the SHA
82 * hash of the contents of the "entropy pool". The SHA hash avoids
83 * exposing the internal state of the entropy pool. It is believed to
84 * be computationally infeasible to derive any useful information
85 * about the input of SHA from its output. Even if it is possible to
86 * analyze SHA in some clever way, as long as the amount of data
87 * returned from the generator is less than the inherent entropy in
88 * the pool, the output data is totally unpredictable. For this
89 * reason, the routine decreases its internal estimate of how many
90 * bits of "true randomness" are contained in the entropy pool as it
91 * outputs random numbers.
92 *
93 * If this estimate goes to zero, the routine can still generate
94 * random numbers; however, an attacker may (at least in theory) be
95 * able to infer the future output of the generator from prior
96 * outputs. This requires successful cryptanalysis of SHA, which is
97 * not believed to be feasible, but there is a remote possibility.
98 * Nonetheless, these numbers should be useful for the vast majority
99 * of purposes.
100 *
101 * Exported interfaces ---- output
102 * ===============================
103 *
104 * There are three exported interfaces; the first is one designed to
105 * be used from within the kernel:
106 *
107 * void get_random_bytes(void *buf, int nbytes);
108 *
109 * This interface will return the requested number of random bytes,
110 * and place it in the requested buffer.
111 *
112 * The two other interfaces are two character devices /dev/random and
113 * /dev/urandom. /dev/random is suitable for use when very high
114 * quality randomness is desired (for example, for key generation or
115 * one-time pads), as it will only return a maximum of the number of
116 * bits of randomness (as estimated by the random number generator)
117 * contained in the entropy pool.
118 *
119 * The /dev/urandom device does not have this limit, and will return
120 * as many bytes as are requested. As more and more random bytes are
121 * requested without giving time for the entropy pool to recharge,
122 * this will result in random numbers that are merely cryptographically
123 * strong. For many applications, however, this is acceptable.
124 *
125 * Exported interfaces ---- input
126 * ==============================
127 *
128 * The current exported interfaces for gathering environmental noise
129 * from the devices are:
130 *
131 * void add_device_randomness(const void *buf, unsigned int size);
132 * void add_input_randomness(unsigned int type, unsigned int code,
133 * unsigned int value);
134 * void add_interrupt_randomness(int irq, int irq_flags);
135 * void add_disk_randomness(struct gendisk *disk);
136 *
137 * add_device_randomness() is for adding data to the random pool that
138 * is likely to differ between two devices (or possibly even per boot).
139 * This would be things like MAC addresses or serial numbers, or the
140 * read-out of the RTC. This does *not* add any actual entropy to the
141 * pool, but it initializes the pool to different values for devices
142 * that might otherwise be identical and have very little entropy
143 * available to them (particularly common in the embedded world).
144 *
145 * add_input_randomness() uses the input layer interrupt timing, as well as
146 * the event type information from the hardware.
147 *
148 * add_interrupt_randomness() uses the interrupt timing as random
149 * inputs to the entropy pool. Using the cycle counters and the irq source
150 * as inputs, it feeds the randomness roughly once a second.
151 *
152 * add_disk_randomness() uses what amounts to the seek time of block
153 * layer request events, on a per-disk_devt basis, as input to the
154 * entropy pool. Note that high-speed solid state drives with very low
155 * seek times do not make for good sources of entropy, as their seek
156 * times are usually fairly consistent.
157 *
158 * All of these routines try to estimate how many bits of randomness a
159 * particular randomness source. They do this by keeping track of the
160 * first and second order deltas of the event timings.
161 *
162 * Ensuring unpredictability at system startup
163 * ============================================
164 *
165 * When any operating system starts up, it will go through a sequence
166 * of actions that are fairly predictable by an adversary, especially
167 * if the start-up does not involve interaction with a human operator.
168 * This reduces the actual number of bits of unpredictability in the
169 * entropy pool below the value in entropy_count. In order to
170 * counteract this effect, it helps to carry information in the
171 * entropy pool across shut-downs and start-ups. To do this, put the
172 * following lines an appropriate script which is run during the boot
173 * sequence:
174 *
175 * echo "Initializing random number generator..."
176 * random_seed=/var/run/random-seed
177 * # Carry a random seed from start-up to start-up
178 * # Load and then save the whole entropy pool
179 * if [ -f $random_seed ]; then
180 * cat $random_seed >/dev/urandom
181 * else
182 * touch $random_seed
183 * fi
184 * chmod 600 $random_seed
185 * dd if=/dev/urandom of=$random_seed count=1 bs=512
186 *
187 * and the following lines in an appropriate script which is run as
188 * the system is shutdown:
189 *
190 * # Carry a random seed from shut-down to start-up
191 * # Save the whole entropy pool
192 * echo "Saving random seed..."
193 * random_seed=/var/run/random-seed
194 * touch $random_seed
195 * chmod 600 $random_seed
196 * dd if=/dev/urandom of=$random_seed count=1 bs=512
197 *
198 * For example, on most modern systems using the System V init
199 * scripts, such code fragments would be found in
200 * /etc/rc.d/init.d/random. On older Linux systems, the correct script
201 * location might be in /etc/rcb.d/rc.local or /etc/rc.d/rc.0.
202 *
203 * Effectively, these commands cause the contents of the entropy pool
204 * to be saved at shut-down time and reloaded into the entropy pool at
205 * start-up. (The 'dd' in the addition to the bootup script is to
206 * make sure that /etc/random-seed is different for every start-up,
207 * even if the system crashes without executing rc.0.) Even with
208 * complete knowledge of the start-up activities, predicting the state
209 * of the entropy pool requires knowledge of the previous history of
210 * the system.
211 *
212 * Configuring the /dev/random driver under Linux
213 * ==============================================
214 *
215 * The /dev/random driver under Linux uses minor numbers 8 and 9 of
216 * the /dev/mem major number (#1). So if your system does not have
217 * /dev/random and /dev/urandom created already, they can be created
218 * by using the commands:
219 *
220 * mknod /dev/random c 1 8
221 * mknod /dev/urandom c 1 9
222 *
223 * Acknowledgements:
224 * =================
225 *
226 * Ideas for constructing this random number generator were derived
227 * from Pretty Good Privacy's random number generator, and from private
228 * discussions with Phil Karn. Colin Plumb provided a faster random
229 * number generator, which speed up the mixing function of the entropy
230 * pool, taken from PGPfone. Dale Worley has also contributed many
231 * useful ideas and suggestions to improve this driver.
232 *
233 * Any flaws in the design are solely my responsibility, and should
234 * not be attributed to the Phil, Colin, or any of authors of PGP.
235 *
236 * Further background information on this topic may be obtained from
237 * RFC 1750, "Randomness Recommendations for Security", by Donald
238 * Eastlake, Steve Crocker, and Jeff Schiller.
239 */
240
241#include <linux/utsname.h>
242#include <linux/module.h>
243#include <linux/kernel.h>
244#include <linux/major.h>
245#include <linux/string.h>
246#include <linux/fcntl.h>
247#include <linux/slab.h>
248#include <linux/random.h>
249#include <linux/poll.h>
250#include <linux/init.h>
251#include <linux/fs.h>
252#include <linux/genhd.h>
253#include <linux/interrupt.h>
254#include <linux/mm.h>
255#include <linux/nodemask.h>
256#include <linux/spinlock.h>
257#include <linux/kthread.h>
258#include <linux/percpu.h>
259#include <linux/cryptohash.h>
260#include <linux/fips.h>
261#include <linux/ptrace.h>
262#include <linux/workqueue.h>
263#include <linux/irq.h>
264#include <linux/ratelimit.h>
265#include <linux/syscalls.h>
266#include <linux/completion.h>
267#include <linux/uuid.h>
268#include <crypto/chacha20.h>
269
270#include <asm/processor.h>
271#include <linux/uaccess.h>
272#include <asm/irq.h>
273#include <asm/irq_regs.h>
274#include <asm/io.h>
275
276#define CREATE_TRACE_POINTS
277#include <trace/events/random.h>
278
279/* #define ADD_INTERRUPT_BENCH */
280
281/*
282 * Configuration information
283 */
284#define INPUT_POOL_SHIFT 12
285#define INPUT_POOL_WORDS (1 << (INPUT_POOL_SHIFT-5))
286#define OUTPUT_POOL_SHIFT 10
287#define OUTPUT_POOL_WORDS (1 << (OUTPUT_POOL_SHIFT-5))
288#define SEC_XFER_SIZE 512
289#define EXTRACT_SIZE 10
290
291
292#define LONGS(x) (((x) + sizeof(unsigned long) - 1)/sizeof(unsigned long))
293
294/*
295 * To allow fractional bits to be tracked, the entropy_count field is
296 * denominated in units of 1/8th bits.
297 *
298 * 2*(ENTROPY_SHIFT + log2(poolbits)) must <= 31, or the multiply in
299 * credit_entropy_bits() needs to be 64 bits wide.
300 */
301#define ENTROPY_SHIFT 3
302#define ENTROPY_BITS(r) ((r)->entropy_count >> ENTROPY_SHIFT)
303
304/*
305 * The minimum number of bits of entropy before we wake up a read on
306 * /dev/random. Should be enough to do a significant reseed.
307 */
308static int random_read_wakeup_bits = 64;
309
310/*
311 * If the entropy count falls under this number of bits, then we
312 * should wake up processes which are selecting or polling on write
313 * access to /dev/random.
314 */
315static int random_write_wakeup_bits = 28 * OUTPUT_POOL_WORDS;
316
317/*
318 * Originally, we used a primitive polynomial of degree .poolwords
319 * over GF(2). The taps for various sizes are defined below. They
320 * were chosen to be evenly spaced except for the last tap, which is 1
321 * to get the twisting happening as fast as possible.
322 *
323 * For the purposes of better mixing, we use the CRC-32 polynomial as
324 * well to make a (modified) twisted Generalized Feedback Shift
325 * Register. (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR
326 * generators. ACM Transactions on Modeling and Computer Simulation
327 * 2(3):179-194. Also see M. Matsumoto & Y. Kurita, 1994. Twisted
328 * GFSR generators II. ACM Transactions on Modeling and Computer
329 * Simulation 4:254-266)
330 *
331 * Thanks to Colin Plumb for suggesting this.
332 *
333 * The mixing operation is much less sensitive than the output hash,
334 * where we use SHA-1. All that we want of mixing operation is that
335 * it be a good non-cryptographic hash; i.e. it not produce collisions
336 * when fed "random" data of the sort we expect to see. As long as
337 * the pool state differs for different inputs, we have preserved the
338 * input entropy and done a good job. The fact that an intelligent
339 * attacker can construct inputs that will produce controlled
340 * alterations to the pool's state is not important because we don't
341 * consider such inputs to contribute any randomness. The only
342 * property we need with respect to them is that the attacker can't
343 * increase his/her knowledge of the pool's state. Since all
344 * additions are reversible (knowing the final state and the input,
345 * you can reconstruct the initial state), if an attacker has any
346 * uncertainty about the initial state, he/she can only shuffle that
347 * uncertainty about, but never cause any collisions (which would
348 * decrease the uncertainty).
349 *
350 * Our mixing functions were analyzed by Lacharme, Roeck, Strubel, and
351 * Videau in their paper, "The Linux Pseudorandom Number Generator
352 * Revisited" (see: http://eprint.iacr.org/2012/251.pdf). In their
353 * paper, they point out that we are not using a true Twisted GFSR,
354 * since Matsumoto & Kurita used a trinomial feedback polynomial (that
355 * is, with only three taps, instead of the six that we are using).
356 * As a result, the resulting polynomial is neither primitive nor
357 * irreducible, and hence does not have a maximal period over
358 * GF(2**32). They suggest a slight change to the generator
359 * polynomial which improves the resulting TGFSR polynomial to be
360 * irreducible, which we have made here.
361 */
362static struct poolinfo {
363 int poolbitshift, poolwords, poolbytes, poolbits, poolfracbits;
364#define S(x) ilog2(x)+5, (x), (x)*4, (x)*32, (x) << (ENTROPY_SHIFT+5)
365 int tap1, tap2, tap3, tap4, tap5;
366} poolinfo_table[] = {
367 /* was: x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 */
368 /* x^128 + x^104 + x^76 + x^51 +x^25 + x + 1 */
369 { S(128), 104, 76, 51, 25, 1 },
370 /* was: x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 */
371 /* x^32 + x^26 + x^19 + x^14 + x^7 + x + 1 */
372 { S(32), 26, 19, 14, 7, 1 },
373#if 0
374 /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1 -- 115 */
375 { S(2048), 1638, 1231, 819, 411, 1 },
376
377 /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
378 { S(1024), 817, 615, 412, 204, 1 },
379
380 /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
381 { S(1024), 819, 616, 410, 207, 2 },
382
383 /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
384 { S(512), 411, 308, 208, 104, 1 },
385
386 /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
387 { S(512), 409, 307, 206, 102, 2 },
388 /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
389 { S(512), 409, 309, 205, 103, 2 },
390
391 /* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
392 { S(256), 205, 155, 101, 52, 1 },
393
394 /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
395 { S(128), 103, 78, 51, 27, 2 },
396
397 /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
398 { S(64), 52, 39, 26, 14, 1 },
399#endif
400};
401
402/*
403 * Static global variables
404 */
405static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
406static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
407static struct fasync_struct *fasync;
408
409static DEFINE_SPINLOCK(random_ready_list_lock);
410static LIST_HEAD(random_ready_list);
411
412struct crng_state {
413 __u32 state[16];
414 unsigned long init_time;
415 spinlock_t lock;
416};
417
418struct crng_state primary_crng = {
419 .lock = __SPIN_LOCK_UNLOCKED(primary_crng.lock),
420};
421
422/*
423 * crng_init = 0 --> Uninitialized
424 * 1 --> Initialized
425 * 2 --> Initialized from input_pool
426 *
427 * crng_init is protected by primary_crng->lock, and only increases
428 * its value (from 0->1->2).
429 */
430static int crng_init = 0;
431#define crng_ready() (likely(crng_init > 1))
432static int crng_init_cnt = 0;
433static unsigned long crng_global_init_time = 0;
434#define CRNG_INIT_CNT_THRESH (2*CHACHA20_KEY_SIZE)
435static void _extract_crng(struct crng_state *crng,
436 __u32 out[CHACHA20_BLOCK_WORDS]);
437static void _crng_backtrack_protect(struct crng_state *crng,
438 __u32 tmp[CHACHA20_BLOCK_WORDS], int used);
439static void process_random_ready_list(void);
440static void _get_random_bytes(void *buf, int nbytes);
441
442static struct ratelimit_state unseeded_warning =
443 RATELIMIT_STATE_INIT("warn_unseeded_randomness", HZ, 3);
444static struct ratelimit_state urandom_warning =
445 RATELIMIT_STATE_INIT("warn_urandom_randomness", HZ, 3);
446
447static int ratelimit_disable __read_mostly;
448
449module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
450MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
451
452/**********************************************************************
453 *
454 * OS independent entropy store. Here are the functions which handle
455 * storing entropy in an entropy pool.
456 *
457 **********************************************************************/
458
459struct entropy_store;
460struct entropy_store {
461 /* read-only data: */
462 const struct poolinfo *poolinfo;
463 __u32 *pool;
464 const char *name;
465 struct entropy_store *pull;
466 struct work_struct push_work;
467
468 /* read-write data: */
469 unsigned long last_pulled;
470 spinlock_t lock;
471 unsigned short add_ptr;
472 unsigned short input_rotate;
473 int entropy_count;
474 int entropy_total;
475 unsigned int initialized:1;
476 unsigned int last_data_init:1;
477 __u8 last_data[EXTRACT_SIZE];
478};
479
480static ssize_t extract_entropy(struct entropy_store *r, void *buf,
481 size_t nbytes, int min, int rsvd);
482static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
483 size_t nbytes, int fips);
484
485static void crng_reseed(struct crng_state *crng, struct entropy_store *r);
486static void push_to_pool(struct work_struct *work);
487static __u32 input_pool_data[INPUT_POOL_WORDS] __latent_entropy;
488static __u32 blocking_pool_data[OUTPUT_POOL_WORDS] __latent_entropy;
489
490static struct entropy_store input_pool = {
491 .poolinfo = &poolinfo_table[0],
492 .name = "input",
493 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
494 .pool = input_pool_data
495};
496
497static struct entropy_store blocking_pool = {
498 .poolinfo = &poolinfo_table[1],
499 .name = "blocking",
500 .pull = &input_pool,
501 .lock = __SPIN_LOCK_UNLOCKED(blocking_pool.lock),
502 .pool = blocking_pool_data,
503 .push_work = __WORK_INITIALIZER(blocking_pool.push_work,
504 push_to_pool),
505};
506
507static __u32 const twist_table[8] = {
508 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
509 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
510
511/*
512 * This function adds bytes into the entropy "pool". It does not
513 * update the entropy estimate. The caller should call
514 * credit_entropy_bits if this is appropriate.
515 *
516 * The pool is stirred with a primitive polynomial of the appropriate
517 * degree, and then twisted. We twist by three bits at a time because
518 * it's cheap to do so and helps slightly in the expected case where
519 * the entropy is concentrated in the low-order bits.
520 */
521static void _mix_pool_bytes(struct entropy_store *r, const void *in,
522 int nbytes)
523{
524 unsigned long i, tap1, tap2, tap3, tap4, tap5;
525 int input_rotate;
526 int wordmask = r->poolinfo->poolwords - 1;
527 const char *bytes = in;
528 __u32 w;
529
530 tap1 = r->poolinfo->tap1;
531 tap2 = r->poolinfo->tap2;
532 tap3 = r->poolinfo->tap3;
533 tap4 = r->poolinfo->tap4;
534 tap5 = r->poolinfo->tap5;
535
536 input_rotate = r->input_rotate;
537 i = r->add_ptr;
538
539 /* mix one byte at a time to simplify size handling and churn faster */
540 while (nbytes--) {
541 w = rol32(*bytes++, input_rotate);
542 i = (i - 1) & wordmask;
543
544 /* XOR in the various taps */
545 w ^= r->pool[i];
546 w ^= r->pool[(i + tap1) & wordmask];
547 w ^= r->pool[(i + tap2) & wordmask];
548 w ^= r->pool[(i + tap3) & wordmask];
549 w ^= r->pool[(i + tap4) & wordmask];
550 w ^= r->pool[(i + tap5) & wordmask];
551
552 /* Mix the result back in with a twist */
553 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
554
555 /*
556 * Normally, we add 7 bits of rotation to the pool.
557 * At the beginning of the pool, add an extra 7 bits
558 * rotation, so that successive passes spread the
559 * input bits across the pool evenly.
560 */
561 input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
562 }
563
564 r->input_rotate = input_rotate;
565 r->add_ptr = i;
566}
567
568static void __mix_pool_bytes(struct entropy_store *r, const void *in,
569 int nbytes)
570{
571 trace_mix_pool_bytes_nolock(r->name, nbytes, _RET_IP_);
572 _mix_pool_bytes(r, in, nbytes);
573}
574
575static void mix_pool_bytes(struct entropy_store *r, const void *in,
576 int nbytes)
577{
578 unsigned long flags;
579
580 trace_mix_pool_bytes(r->name, nbytes, _RET_IP_);
581 spin_lock_irqsave(&r->lock, flags);
582 _mix_pool_bytes(r, in, nbytes);
583 spin_unlock_irqrestore(&r->lock, flags);
584}
585
586struct fast_pool {
587 __u32 pool[4];
588 unsigned long last;
589 unsigned short reg_idx;
590 unsigned char count;
591};
592
593/*
594 * This is a fast mixing routine used by the interrupt randomness
595 * collector. It's hardcoded for an 128 bit pool and assumes that any
596 * locks that might be needed are taken by the caller.
597 */
598static void fast_mix(struct fast_pool *f)
599{
600 __u32 a = f->pool[0], b = f->pool[1];
601 __u32 c = f->pool[2], d = f->pool[3];
602
603 a += b; c += d;
604 b = rol32(b, 6); d = rol32(d, 27);
605 d ^= a; b ^= c;
606
607 a += b; c += d;
608 b = rol32(b, 16); d = rol32(d, 14);
609 d ^= a; b ^= c;
610
611 a += b; c += d;
612 b = rol32(b, 6); d = rol32(d, 27);
613 d ^= a; b ^= c;
614
615 a += b; c += d;
616 b = rol32(b, 16); d = rol32(d, 14);
617 d ^= a; b ^= c;
618
619 f->pool[0] = a; f->pool[1] = b;
620 f->pool[2] = c; f->pool[3] = d;
621 f->count++;
622}
623
624static void process_random_ready_list(void)
625{
626 unsigned long flags;
627 struct random_ready_callback *rdy, *tmp;
628
629 spin_lock_irqsave(&random_ready_list_lock, flags);
630 list_for_each_entry_safe(rdy, tmp, &random_ready_list, list) {
631 struct module *owner = rdy->owner;
632
633 list_del_init(&rdy->list);
634 rdy->func(rdy);
635 module_put(owner);
636 }
637 spin_unlock_irqrestore(&random_ready_list_lock, flags);
638}
639
640/*
641 * Credit (or debit) the entropy store with n bits of entropy.
642 * Use credit_entropy_bits_safe() if the value comes from userspace
643 * or otherwise should be checked for extreme values.
644 */
645static void credit_entropy_bits(struct entropy_store *r, int nbits)
646{
647 int entropy_count, orig;
648 const int pool_size = r->poolinfo->poolfracbits;
649 int nfrac = nbits << ENTROPY_SHIFT;
650
651 if (!nbits)
652 return;
653
654retry:
655 entropy_count = orig = READ_ONCE(r->entropy_count);
656 if (nfrac < 0) {
657 /* Debit */
658 entropy_count += nfrac;
659 } else {
660 /*
661 * Credit: we have to account for the possibility of
662 * overwriting already present entropy. Even in the
663 * ideal case of pure Shannon entropy, new contributions
664 * approach the full value asymptotically:
665 *
666 * entropy <- entropy + (pool_size - entropy) *
667 * (1 - exp(-add_entropy/pool_size))
668 *
669 * For add_entropy <= pool_size/2 then
670 * (1 - exp(-add_entropy/pool_size)) >=
671 * (add_entropy/pool_size)*0.7869...
672 * so we can approximate the exponential with
673 * 3/4*add_entropy/pool_size and still be on the
674 * safe side by adding at most pool_size/2 at a time.
675 *
676 * The use of pool_size-2 in the while statement is to
677 * prevent rounding artifacts from making the loop
678 * arbitrarily long; this limits the loop to log2(pool_size)*2
679 * turns no matter how large nbits is.
680 */
681 int pnfrac = nfrac;
682 const int s = r->poolinfo->poolbitshift + ENTROPY_SHIFT + 2;
683 /* The +2 corresponds to the /4 in the denominator */
684
685 do {
686 unsigned int anfrac = min(pnfrac, pool_size/2);
687 unsigned int add =
688 ((pool_size - entropy_count)*anfrac*3) >> s;
689
690 entropy_count += add;
691 pnfrac -= anfrac;
692 } while (unlikely(entropy_count < pool_size-2 && pnfrac));
693 }
694
695 if (unlikely(entropy_count < 0)) {
696 pr_warn("random: negative entropy/overflow: pool %s count %d\n",
697 r->name, entropy_count);
698 WARN_ON(1);
699 entropy_count = 0;
700 } else if (entropy_count > pool_size)
701 entropy_count = pool_size;
702 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
703 goto retry;
704
705 r->entropy_total += nbits;
706 if (!r->initialized && r->entropy_total > 128) {
707 r->initialized = 1;
708 r->entropy_total = 0;
709 }
710
711 trace_credit_entropy_bits(r->name, nbits,
712 entropy_count >> ENTROPY_SHIFT,
713 r->entropy_total, _RET_IP_);
714
715 if (r == &input_pool) {
716 int entropy_bits = entropy_count >> ENTROPY_SHIFT;
717
718 if (crng_init < 2 && entropy_bits >= 128) {
719 crng_reseed(&primary_crng, r);
720 entropy_bits = r->entropy_count >> ENTROPY_SHIFT;
721 }
722
723 /* should we wake readers? */
724 if (entropy_bits >= random_read_wakeup_bits &&
725 wq_has_sleeper(&random_read_wait)) {
726 wake_up_interruptible(&random_read_wait);
727 kill_fasync(&fasync, SIGIO, POLL_IN);
728 }
729 /* If the input pool is getting full, send some
730 * entropy to the blocking pool until it is 75% full.
731 */
732 if (entropy_bits > random_write_wakeup_bits &&
733 r->initialized &&
734 r->entropy_total >= 2*random_read_wakeup_bits) {
735 struct entropy_store *other = &blocking_pool;
736
737 if (other->entropy_count <=
738 3 * other->poolinfo->poolfracbits / 4) {
739 schedule_work(&other->push_work);
740 r->entropy_total = 0;
741 }
742 }
743 }
744}
745
746static int credit_entropy_bits_safe(struct entropy_store *r, int nbits)
747{
748 const int nbits_max = r->poolinfo->poolwords * 32;
749
750 if (nbits < 0)
751 return -EINVAL;
752
753 /* Cap the value to avoid overflows */
754 nbits = min(nbits, nbits_max);
755
756 credit_entropy_bits(r, nbits);
757 return 0;
758}
759
760/*********************************************************************
761 *
762 * CRNG using CHACHA20
763 *
764 *********************************************************************/
765
766#define CRNG_RESEED_INTERVAL (300*HZ)
767
768static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
769
770#ifdef CONFIG_NUMA
771/*
772 * Hack to deal with crazy userspace progams when they are all trying
773 * to access /dev/urandom in parallel. The programs are almost
774 * certainly doing something terribly wrong, but we'll work around
775 * their brain damage.
776 */
777static struct crng_state **crng_node_pool __read_mostly;
778#endif
779
780static void invalidate_batched_entropy(void);
781
782static void crng_initialize(struct crng_state *crng)
783{
784 int i;
785 unsigned long rv;
786
787 memcpy(&crng->state[0], "expand 32-byte k", 16);
788 if (crng == &primary_crng)
789 _extract_entropy(&input_pool, &crng->state[4],
790 sizeof(__u32) * 12, 0);
791 else
792 _get_random_bytes(&crng->state[4], sizeof(__u32) * 12);
793 for (i = 4; i < 16; i++) {
794 if (!arch_get_random_seed_long(&rv) &&
795 !arch_get_random_long(&rv))
796 rv = random_get_entropy();
797 crng->state[i] ^= rv;
798 }
799 crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
800}
801
802#ifdef CONFIG_NUMA
803static void do_numa_crng_init(struct work_struct *work)
804{
805 int i;
806 struct crng_state *crng;
807 struct crng_state **pool;
808
809 pool = kcalloc(nr_node_ids, sizeof(*pool), GFP_KERNEL|__GFP_NOFAIL);
810 for_each_online_node(i) {
811 crng = kmalloc_node(sizeof(struct crng_state),
812 GFP_KERNEL | __GFP_NOFAIL, i);
813 spin_lock_init(&crng->lock);
814 crng_initialize(crng);
815 pool[i] = crng;
816 }
817 mb();
818 if (cmpxchg(&crng_node_pool, NULL, pool)) {
819 for_each_node(i)
820 kfree(pool[i]);
821 kfree(pool);
822 }
823}
824
825static DECLARE_WORK(numa_crng_init_work, do_numa_crng_init);
826
827static void numa_crng_init(void)
828{
829 schedule_work(&numa_crng_init_work);
830}
831#else
832static void numa_crng_init(void) {}
833#endif
834
835/*
836 * crng_fast_load() can be called by code in the interrupt service
837 * path. So we can't afford to dilly-dally.
838 */
839static int crng_fast_load(const char *cp, size_t len)
840{
841 unsigned long flags;
842 char *p;
843
844 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
845 return 0;
846 if (crng_init != 0) {
847 spin_unlock_irqrestore(&primary_crng.lock, flags);
848 return 0;
849 }
850 p = (unsigned char *) &primary_crng.state[4];
851 while (len > 0 && crng_init_cnt < CRNG_INIT_CNT_THRESH) {
852 p[crng_init_cnt % CHACHA20_KEY_SIZE] ^= *cp;
853 cp++; crng_init_cnt++; len--;
854 }
855 spin_unlock_irqrestore(&primary_crng.lock, flags);
856 if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
857 invalidate_batched_entropy();
858 crng_init = 1;
859 wake_up_interruptible(&crng_init_wait);
860 pr_notice("random: fast init done\n");
861 }
862 return 1;
863}
864
865/*
866 * crng_slow_load() is called by add_device_randomness, which has two
867 * attributes. (1) We can't trust the buffer passed to it is
868 * guaranteed to be unpredictable (so it might not have any entropy at
869 * all), and (2) it doesn't have the performance constraints of
870 * crng_fast_load().
871 *
872 * So we do something more comprehensive which is guaranteed to touch
873 * all of the primary_crng's state, and which uses a LFSR with a
874 * period of 255 as part of the mixing algorithm. Finally, we do
875 * *not* advance crng_init_cnt since buffer we may get may be something
876 * like a fixed DMI table (for example), which might very well be
877 * unique to the machine, but is otherwise unvarying.
878 */
879static int crng_slow_load(const char *cp, size_t len)
880{
881 unsigned long flags;
882 static unsigned char lfsr = 1;
883 unsigned char tmp;
884 unsigned i, max = CHACHA20_KEY_SIZE;
885 const char * src_buf = cp;
886 char * dest_buf = (char *) &primary_crng.state[4];
887
888 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
889 return 0;
890 if (crng_init != 0) {
891 spin_unlock_irqrestore(&primary_crng.lock, flags);
892 return 0;
893 }
894 if (len > max)
895 max = len;
896
897 for (i = 0; i < max ; i++) {
898 tmp = lfsr;
899 lfsr >>= 1;
900 if (tmp & 1)
901 lfsr ^= 0xE1;
902 tmp = dest_buf[i % CHACHA20_KEY_SIZE];
903 dest_buf[i % CHACHA20_KEY_SIZE] ^= src_buf[i % len] ^ lfsr;
904 lfsr += (tmp << 3) | (tmp >> 5);
905 }
906 spin_unlock_irqrestore(&primary_crng.lock, flags);
907 return 1;
908}
909
910static void crng_reseed(struct crng_state *crng, struct entropy_store *r)
911{
912 unsigned long flags;
913 int i, num;
914 union {
915 __u32 block[CHACHA20_BLOCK_WORDS];
916 __u32 key[8];
917 } buf;
918
919 if (r) {
920 num = extract_entropy(r, &buf, 32, 16, 0);
921 if (num == 0)
922 return;
923 } else {
924 _extract_crng(&primary_crng, buf.block);
925 _crng_backtrack_protect(&primary_crng, buf.block,
926 CHACHA20_KEY_SIZE);
927 }
928 spin_lock_irqsave(&crng->lock, flags);
929 for (i = 0; i < 8; i++) {
930 unsigned long rv;
931 if (!arch_get_random_seed_long(&rv) &&
932 !arch_get_random_long(&rv))
933 rv = random_get_entropy();
934 crng->state[i+4] ^= buf.key[i] ^ rv;
935 }
936 memzero_explicit(&buf, sizeof(buf));
937 crng->init_time = jiffies;
938 spin_unlock_irqrestore(&crng->lock, flags);
939 if (crng == &primary_crng && crng_init < 2) {
940 invalidate_batched_entropy();
941 numa_crng_init();
942 crng_init = 2;
943 process_random_ready_list();
944 wake_up_interruptible(&crng_init_wait);
945 pr_notice("random: crng init done\n");
946 if (unseeded_warning.missed) {
947 pr_notice("random: %d get_random_xx warning(s) missed "
948 "due to ratelimiting\n",
949 unseeded_warning.missed);
950 unseeded_warning.missed = 0;
951 }
952 if (urandom_warning.missed) {
953 pr_notice("random: %d urandom warning(s) missed "
954 "due to ratelimiting\n",
955 urandom_warning.missed);
956 urandom_warning.missed = 0;
957 }
958 }
959}
960
961static void _extract_crng(struct crng_state *crng,
962 __u32 out[CHACHA20_BLOCK_WORDS])
963{
964 unsigned long v, flags;
965
966 if (crng_ready() &&
967 (time_after(crng_global_init_time, crng->init_time) ||
968 time_after(jiffies, crng->init_time + CRNG_RESEED_INTERVAL)))
969 crng_reseed(crng, crng == &primary_crng ? &input_pool : NULL);
970 spin_lock_irqsave(&crng->lock, flags);
971 if (arch_get_random_long(&v))
972 crng->state[14] ^= v;
973 chacha20_block(&crng->state[0], out);
974 if (crng->state[12] == 0)
975 crng->state[13]++;
976 spin_unlock_irqrestore(&crng->lock, flags);
977}
978
979static void extract_crng(__u32 out[CHACHA20_BLOCK_WORDS])
980{
981 struct crng_state *crng = NULL;
982
983#ifdef CONFIG_NUMA
984 if (crng_node_pool)
985 crng = crng_node_pool[numa_node_id()];
986 if (crng == NULL)
987#endif
988 crng = &primary_crng;
989 _extract_crng(crng, out);
990}
991
992/*
993 * Use the leftover bytes from the CRNG block output (if there is
994 * enough) to mutate the CRNG key to provide backtracking protection.
995 */
996static void _crng_backtrack_protect(struct crng_state *crng,
997 __u32 tmp[CHACHA20_BLOCK_WORDS], int used)
998{
999 unsigned long flags;
1000 __u32 *s, *d;
1001 int i;
1002
1003 used = round_up(used, sizeof(__u32));
1004 if (used + CHACHA20_KEY_SIZE > CHACHA20_BLOCK_SIZE) {
1005 extract_crng(tmp);
1006 used = 0;
1007 }
1008 spin_lock_irqsave(&crng->lock, flags);
1009 s = &tmp[used / sizeof(__u32)];
1010 d = &crng->state[4];
1011 for (i=0; i < 8; i++)
1012 *d++ ^= *s++;
1013 spin_unlock_irqrestore(&crng->lock, flags);
1014}
1015
1016static void crng_backtrack_protect(__u32 tmp[CHACHA20_BLOCK_WORDS], int used)
1017{
1018 struct crng_state *crng = NULL;
1019
1020#ifdef CONFIG_NUMA
1021 if (crng_node_pool)
1022 crng = crng_node_pool[numa_node_id()];
1023 if (crng == NULL)
1024#endif
1025 crng = &primary_crng;
1026 _crng_backtrack_protect(crng, tmp, used);
1027}
1028
1029static ssize_t extract_crng_user(void __user *buf, size_t nbytes)
1030{
1031 ssize_t ret = 0, i = CHACHA20_BLOCK_SIZE;
1032 __u32 tmp[CHACHA20_BLOCK_WORDS];
1033 int large_request = (nbytes > 256);
1034
1035 while (nbytes) {
1036 if (large_request && need_resched()) {
1037 if (signal_pending(current)) {
1038 if (ret == 0)
1039 ret = -ERESTARTSYS;
1040 break;
1041 }
1042 schedule();
1043 }
1044
1045 extract_crng(tmp);
1046 i = min_t(int, nbytes, CHACHA20_BLOCK_SIZE);
1047 if (copy_to_user(buf, tmp, i)) {
1048 ret = -EFAULT;
1049 break;
1050 }
1051
1052 nbytes -= i;
1053 buf += i;
1054 ret += i;
1055 }
1056 crng_backtrack_protect(tmp, i);
1057
1058 /* Wipe data just written to memory */
1059 memzero_explicit(tmp, sizeof(tmp));
1060
1061 return ret;
1062}
1063
1064
1065/*********************************************************************
1066 *
1067 * Entropy input management
1068 *
1069 *********************************************************************/
1070
1071/* There is one of these per entropy source */
1072struct timer_rand_state {
1073 cycles_t last_time;
1074 long last_delta, last_delta2;
1075};
1076
1077#define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
1078
1079/*
1080 * Add device- or boot-specific data to the input pool to help
1081 * initialize it.
1082 *
1083 * None of this adds any entropy; it is meant to avoid the problem of
1084 * the entropy pool having similar initial state across largely
1085 * identical devices.
1086 */
1087void add_device_randomness(const void *buf, unsigned int size)
1088{
1089 unsigned long time = random_get_entropy() ^ jiffies;
1090 unsigned long flags;
1091
1092 if (!crng_ready() && size)
1093 crng_slow_load(buf, size);
1094
1095 trace_add_device_randomness(size, _RET_IP_);
1096 spin_lock_irqsave(&input_pool.lock, flags);
1097 _mix_pool_bytes(&input_pool, buf, size);
1098 _mix_pool_bytes(&input_pool, &time, sizeof(time));
1099 spin_unlock_irqrestore(&input_pool.lock, flags);
1100}
1101EXPORT_SYMBOL(add_device_randomness);
1102
1103static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
1104
1105/*
1106 * This function adds entropy to the entropy "pool" by using timing
1107 * delays. It uses the timer_rand_state structure to make an estimate
1108 * of how many bits of entropy this call has added to the pool.
1109 *
1110 * The number "num" is also added to the pool - it should somehow describe
1111 * the type of event which just happened. This is currently 0-255 for
1112 * keyboard scan codes, and 256 upwards for interrupts.
1113 *
1114 */
1115static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
1116{
1117 struct entropy_store *r;
1118 struct {
1119 long jiffies;
1120 unsigned cycles;
1121 unsigned num;
1122 } sample;
1123 long delta, delta2, delta3;
1124
1125 preempt_disable();
1126
1127 sample.jiffies = jiffies;
1128 sample.cycles = random_get_entropy();
1129 sample.num = num;
1130 r = &input_pool;
1131 mix_pool_bytes(r, &sample, sizeof(sample));
1132
1133 /*
1134 * Calculate number of bits of randomness we probably added.
1135 * We take into account the first, second and third-order deltas
1136 * in order to make our estimate.
1137 */
1138 delta = sample.jiffies - state->last_time;
1139 state->last_time = sample.jiffies;
1140
1141 delta2 = delta - state->last_delta;
1142 state->last_delta = delta;
1143
1144 delta3 = delta2 - state->last_delta2;
1145 state->last_delta2 = delta2;
1146
1147 if (delta < 0)
1148 delta = -delta;
1149 if (delta2 < 0)
1150 delta2 = -delta2;
1151 if (delta3 < 0)
1152 delta3 = -delta3;
1153 if (delta > delta2)
1154 delta = delta2;
1155 if (delta > delta3)
1156 delta = delta3;
1157
1158 /*
1159 * delta is now minimum absolute delta.
1160 * Round down by 1 bit on general principles,
1161 * and limit entropy entimate to 12 bits.
1162 */
1163 credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
1164
1165 preempt_enable();
1166}
1167
1168void add_input_randomness(unsigned int type, unsigned int code,
1169 unsigned int value)
1170{
1171 static unsigned char last_value;
1172
1173 /* ignore autorepeat and the like */
1174 if (value == last_value)
1175 return;
1176
1177 last_value = value;
1178 add_timer_randomness(&input_timer_state,
1179 (type << 4) ^ code ^ (code >> 4) ^ value);
1180 trace_add_input_randomness(ENTROPY_BITS(&input_pool));
1181}
1182EXPORT_SYMBOL_GPL(add_input_randomness);
1183
1184static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
1185
1186#ifdef ADD_INTERRUPT_BENCH
1187static unsigned long avg_cycles, avg_deviation;
1188
1189#define AVG_SHIFT 8 /* Exponential average factor k=1/256 */
1190#define FIXED_1_2 (1 << (AVG_SHIFT-1))
1191
1192static void add_interrupt_bench(cycles_t start)
1193{
1194 long delta = random_get_entropy() - start;
1195
1196 /* Use a weighted moving average */
1197 delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
1198 avg_cycles += delta;
1199 /* And average deviation */
1200 delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
1201 avg_deviation += delta;
1202}
1203#else
1204#define add_interrupt_bench(x)
1205#endif
1206
1207static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
1208{
1209 __u32 *ptr = (__u32 *) regs;
1210 unsigned int idx;
1211
1212 if (regs == NULL)
1213 return 0;
1214 idx = READ_ONCE(f->reg_idx);
1215 if (idx >= sizeof(struct pt_regs) / sizeof(__u32))
1216 idx = 0;
1217 ptr += idx++;
1218 WRITE_ONCE(f->reg_idx, idx);
1219 return *ptr;
1220}
1221
1222void add_interrupt_randomness(int irq, int irq_flags)
1223{
1224 struct entropy_store *r;
1225 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1226 struct pt_regs *regs = get_irq_regs();
1227 unsigned long now = jiffies;
1228 cycles_t cycles = random_get_entropy();
1229 __u32 c_high, j_high;
1230 __u64 ip;
1231 unsigned long seed;
1232 int credit = 0;
1233
1234 if (cycles == 0)
1235 cycles = get_reg(fast_pool, regs);
1236 c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
1237 j_high = (sizeof(now) > 4) ? now >> 32 : 0;
1238 fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
1239 fast_pool->pool[1] ^= now ^ c_high;
1240 ip = regs ? instruction_pointer(regs) : _RET_IP_;
1241 fast_pool->pool[2] ^= ip;
1242 fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
1243 get_reg(fast_pool, regs);
1244
1245 fast_mix(fast_pool);
1246 add_interrupt_bench(cycles);
1247
1248 if (unlikely(crng_init == 0)) {
1249 if ((fast_pool->count >= 64) &&
1250 crng_fast_load((char *) fast_pool->pool,
1251 sizeof(fast_pool->pool))) {
1252 fast_pool->count = 0;
1253 fast_pool->last = now;
1254 }
1255 return;
1256 }
1257
1258 if ((fast_pool->count < 64) &&
1259 !time_after(now, fast_pool->last + HZ))
1260 return;
1261
1262 r = &input_pool;
1263 if (!spin_trylock(&r->lock))
1264 return;
1265
1266 fast_pool->last = now;
1267 __mix_pool_bytes(r, &fast_pool->pool, sizeof(fast_pool->pool));
1268
1269 /*
1270 * If we have architectural seed generator, produce a seed and
1271 * add it to the pool. For the sake of paranoia don't let the
1272 * architectural seed generator dominate the input from the
1273 * interrupt noise.
1274 */
1275 if (arch_get_random_seed_long(&seed)) {
1276 __mix_pool_bytes(r, &seed, sizeof(seed));
1277 credit = 1;
1278 }
1279 spin_unlock(&r->lock);
1280
1281 fast_pool->count = 0;
1282
1283 /* award one bit for the contents of the fast pool */
1284 credit_entropy_bits(r, credit + 1);
1285}
1286EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1287
1288#ifdef CONFIG_BLOCK
1289void add_disk_randomness(struct gendisk *disk)
1290{
1291 if (!disk || !disk->random)
1292 return;
1293 /* first major is 1, so we get >= 0x200 here */
1294 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1295 trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
1296}
1297EXPORT_SYMBOL_GPL(add_disk_randomness);
1298#endif
1299
1300/*********************************************************************
1301 *
1302 * Entropy extraction routines
1303 *
1304 *********************************************************************/
1305
1306/*
1307 * This utility inline function is responsible for transferring entropy
1308 * from the primary pool to the secondary extraction pool. We make
1309 * sure we pull enough for a 'catastrophic reseed'.
1310 */
1311static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes);
1312static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1313{
1314 if (!r->pull ||
1315 r->entropy_count >= (nbytes << (ENTROPY_SHIFT + 3)) ||
1316 r->entropy_count > r->poolinfo->poolfracbits)
1317 return;
1318
1319 _xfer_secondary_pool(r, nbytes);
1320}
1321
1322static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1323{
1324 __u32 tmp[OUTPUT_POOL_WORDS];
1325
1326 int bytes = nbytes;
1327
1328 /* pull at least as much as a wakeup */
1329 bytes = max_t(int, bytes, random_read_wakeup_bits / 8);
1330 /* but never more than the buffer size */
1331 bytes = min_t(int, bytes, sizeof(tmp));
1332
1333 trace_xfer_secondary_pool(r->name, bytes * 8, nbytes * 8,
1334 ENTROPY_BITS(r), ENTROPY_BITS(r->pull));
1335 bytes = extract_entropy(r->pull, tmp, bytes,
1336 random_read_wakeup_bits / 8, 0);
1337 mix_pool_bytes(r, tmp, bytes);
1338 credit_entropy_bits(r, bytes*8);
1339}
1340
1341/*
1342 * Used as a workqueue function so that when the input pool is getting
1343 * full, we can "spill over" some entropy to the output pools. That
1344 * way the output pools can store some of the excess entropy instead
1345 * of letting it go to waste.
1346 */
1347static void push_to_pool(struct work_struct *work)
1348{
1349 struct entropy_store *r = container_of(work, struct entropy_store,
1350 push_work);
1351 BUG_ON(!r);
1352 _xfer_secondary_pool(r, random_read_wakeup_bits/8);
1353 trace_push_to_pool(r->name, r->entropy_count >> ENTROPY_SHIFT,
1354 r->pull->entropy_count >> ENTROPY_SHIFT);
1355}
1356
1357/*
1358 * This function decides how many bytes to actually take from the
1359 * given pool, and also debits the entropy count accordingly.
1360 */
1361static size_t account(struct entropy_store *r, size_t nbytes, int min,
1362 int reserved)
1363{
1364 int entropy_count, orig, have_bytes;
1365 size_t ibytes, nfrac;
1366
1367 BUG_ON(r->entropy_count > r->poolinfo->poolfracbits);
1368
1369 /* Can we pull enough? */
1370retry:
1371 entropy_count = orig = READ_ONCE(r->entropy_count);
1372 ibytes = nbytes;
1373 /* never pull more than available */
1374 have_bytes = entropy_count >> (ENTROPY_SHIFT + 3);
1375
1376 if ((have_bytes -= reserved) < 0)
1377 have_bytes = 0;
1378 ibytes = min_t(size_t, ibytes, have_bytes);
1379 if (ibytes < min)
1380 ibytes = 0;
1381
1382 if (unlikely(entropy_count < 0)) {
1383 pr_warn("random: negative entropy count: pool %s count %d\n",
1384 r->name, entropy_count);
1385 WARN_ON(1);
1386 entropy_count = 0;
1387 }
1388 nfrac = ibytes << (ENTROPY_SHIFT + 3);
1389 if ((size_t) entropy_count > nfrac)
1390 entropy_count -= nfrac;
1391 else
1392 entropy_count = 0;
1393
1394 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
1395 goto retry;
1396
1397 trace_debit_entropy(r->name, 8 * ibytes);
1398 if (ibytes &&
1399 (r->entropy_count >> ENTROPY_SHIFT) < random_write_wakeup_bits) {
1400 wake_up_interruptible(&random_write_wait);
1401 kill_fasync(&fasync, SIGIO, POLL_OUT);
1402 }
1403
1404 return ibytes;
1405}
1406
1407/*
1408 * This function does the actual extraction for extract_entropy and
1409 * extract_entropy_user.
1410 *
1411 * Note: we assume that .poolwords is a multiple of 16 words.
1412 */
1413static void extract_buf(struct entropy_store *r, __u8 *out)
1414{
1415 int i;
1416 union {
1417 __u32 w[5];
1418 unsigned long l[LONGS(20)];
1419 } hash;
1420 __u32 workspace[SHA_WORKSPACE_WORDS];
1421 unsigned long flags;
1422
1423 /*
1424 * If we have an architectural hardware random number
1425 * generator, use it for SHA's initial vector
1426 */
1427 sha_init(hash.w);
1428 for (i = 0; i < LONGS(20); i++) {
1429 unsigned long v;
1430 if (!arch_get_random_long(&v))
1431 break;
1432 hash.l[i] = v;
1433 }
1434
1435 /* Generate a hash across the pool, 16 words (512 bits) at a time */
1436 spin_lock_irqsave(&r->lock, flags);
1437 for (i = 0; i < r->poolinfo->poolwords; i += 16)
1438 sha_transform(hash.w, (__u8 *)(r->pool + i), workspace);
1439
1440 /*
1441 * We mix the hash back into the pool to prevent backtracking
1442 * attacks (where the attacker knows the state of the pool
1443 * plus the current outputs, and attempts to find previous
1444 * ouputs), unless the hash function can be inverted. By
1445 * mixing at least a SHA1 worth of hash data back, we make
1446 * brute-forcing the feedback as hard as brute-forcing the
1447 * hash.
1448 */
1449 __mix_pool_bytes(r, hash.w, sizeof(hash.w));
1450 spin_unlock_irqrestore(&r->lock, flags);
1451
1452 memzero_explicit(workspace, sizeof(workspace));
1453
1454 /*
1455 * In case the hash function has some recognizable output
1456 * pattern, we fold it in half. Thus, we always feed back
1457 * twice as much data as we output.
1458 */
1459 hash.w[0] ^= hash.w[3];
1460 hash.w[1] ^= hash.w[4];
1461 hash.w[2] ^= rol32(hash.w[2], 16);
1462
1463 memcpy(out, &hash, EXTRACT_SIZE);
1464 memzero_explicit(&hash, sizeof(hash));
1465}
1466
1467static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
1468 size_t nbytes, int fips)
1469{
1470 ssize_t ret = 0, i;
1471 __u8 tmp[EXTRACT_SIZE];
1472 unsigned long flags;
1473
1474 while (nbytes) {
1475 extract_buf(r, tmp);
1476
1477 if (fips) {
1478 spin_lock_irqsave(&r->lock, flags);
1479 if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
1480 panic("Hardware RNG duplicated output!\n");
1481 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1482 spin_unlock_irqrestore(&r->lock, flags);
1483 }
1484 i = min_t(int, nbytes, EXTRACT_SIZE);
1485 memcpy(buf, tmp, i);
1486 nbytes -= i;
1487 buf += i;
1488 ret += i;
1489 }
1490
1491 /* Wipe data just returned from memory */
1492 memzero_explicit(tmp, sizeof(tmp));
1493
1494 return ret;
1495}
1496
1497/*
1498 * This function extracts randomness from the "entropy pool", and
1499 * returns it in a buffer.
1500 *
1501 * The min parameter specifies the minimum amount we can pull before
1502 * failing to avoid races that defeat catastrophic reseeding while the
1503 * reserved parameter indicates how much entropy we must leave in the
1504 * pool after each pull to avoid starving other readers.
1505 */
1506static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1507 size_t nbytes, int min, int reserved)
1508{
1509 __u8 tmp[EXTRACT_SIZE];
1510 unsigned long flags;
1511
1512 /* if last_data isn't primed, we need EXTRACT_SIZE extra bytes */
1513 if (fips_enabled) {
1514 spin_lock_irqsave(&r->lock, flags);
1515 if (!r->last_data_init) {
1516 r->last_data_init = 1;
1517 spin_unlock_irqrestore(&r->lock, flags);
1518 trace_extract_entropy(r->name, EXTRACT_SIZE,
1519 ENTROPY_BITS(r), _RET_IP_);
1520 xfer_secondary_pool(r, EXTRACT_SIZE);
1521 extract_buf(r, tmp);
1522 spin_lock_irqsave(&r->lock, flags);
1523 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1524 }
1525 spin_unlock_irqrestore(&r->lock, flags);
1526 }
1527
1528 trace_extract_entropy(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1529 xfer_secondary_pool(r, nbytes);
1530 nbytes = account(r, nbytes, min, reserved);
1531
1532 return _extract_entropy(r, buf, nbytes, fips_enabled);
1533}
1534
1535/*
1536 * This function extracts randomness from the "entropy pool", and
1537 * returns it in a userspace buffer.
1538 */
1539static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
1540 size_t nbytes)
1541{
1542 ssize_t ret = 0, i;
1543 __u8 tmp[EXTRACT_SIZE];
1544 int large_request = (nbytes > 256);
1545
1546 trace_extract_entropy_user(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1547 xfer_secondary_pool(r, nbytes);
1548 nbytes = account(r, nbytes, 0, 0);
1549
1550 while (nbytes) {
1551 if (large_request && need_resched()) {
1552 if (signal_pending(current)) {
1553 if (ret == 0)
1554 ret = -ERESTARTSYS;
1555 break;
1556 }
1557 schedule();
1558 }
1559
1560 extract_buf(r, tmp);
1561 i = min_t(int, nbytes, EXTRACT_SIZE);
1562 if (copy_to_user(buf, tmp, i)) {
1563 ret = -EFAULT;
1564 break;
1565 }
1566
1567 nbytes -= i;
1568 buf += i;
1569 ret += i;
1570 }
1571
1572 /* Wipe data just returned from memory */
1573 memzero_explicit(tmp, sizeof(tmp));
1574
1575 return ret;
1576}
1577
1578#define warn_unseeded_randomness(previous) \
1579 _warn_unseeded_randomness(__func__, (void *) _RET_IP_, (previous))
1580
1581static void _warn_unseeded_randomness(const char *func_name, void *caller,
1582 void **previous)
1583{
1584#ifdef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1585 const bool print_once = false;
1586#else
1587 static bool print_once __read_mostly;
1588#endif
1589
1590 if (print_once ||
1591 crng_ready() ||
1592 (previous && (caller == READ_ONCE(*previous))))
1593 return;
1594 WRITE_ONCE(*previous, caller);
1595#ifndef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1596 print_once = true;
1597#endif
1598 if (__ratelimit(&unseeded_warning))
1599 pr_notice("random: %s called from %pS with crng_init=%d\n",
1600 func_name, caller, crng_init);
1601}
1602
1603/*
1604 * This function is the exported kernel interface. It returns some
1605 * number of good random numbers, suitable for key generation, seeding
1606 * TCP sequence numbers, etc. It does not rely on the hardware random
1607 * number generator. For random bytes direct from the hardware RNG
1608 * (when available), use get_random_bytes_arch(). In order to ensure
1609 * that the randomness provided by this function is okay, the function
1610 * wait_for_random_bytes() should be called and return 0 at least once
1611 * at any point prior.
1612 */
1613static void _get_random_bytes(void *buf, int nbytes)
1614{
1615 __u32 tmp[CHACHA20_BLOCK_WORDS];
1616
1617 trace_get_random_bytes(nbytes, _RET_IP_);
1618
1619 while (nbytes >= CHACHA20_BLOCK_SIZE) {
1620 extract_crng(buf);
1621 buf += CHACHA20_BLOCK_SIZE;
1622 nbytes -= CHACHA20_BLOCK_SIZE;
1623 }
1624
1625 if (nbytes > 0) {
1626 extract_crng(tmp);
1627 memcpy(buf, tmp, nbytes);
1628 crng_backtrack_protect(tmp, nbytes);
1629 } else
1630 crng_backtrack_protect(tmp, CHACHA20_BLOCK_SIZE);
1631 memzero_explicit(tmp, sizeof(tmp));
1632}
1633
1634void get_random_bytes(void *buf, int nbytes)
1635{
1636 static void *previous;
1637
1638 warn_unseeded_randomness(&previous);
1639 _get_random_bytes(buf, nbytes);
1640}
1641EXPORT_SYMBOL(get_random_bytes);
1642
1643/*
1644 * Wait for the urandom pool to be seeded and thus guaranteed to supply
1645 * cryptographically secure random numbers. This applies to: the /dev/urandom
1646 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
1647 * family of functions. Using any of these functions without first calling
1648 * this function forfeits the guarantee of security.
1649 *
1650 * Returns: 0 if the urandom pool has been seeded.
1651 * -ERESTARTSYS if the function was interrupted by a signal.
1652 */
1653int wait_for_random_bytes(void)
1654{
1655 if (likely(crng_ready()))
1656 return 0;
1657 return wait_event_interruptible(crng_init_wait, crng_ready());
1658}
1659EXPORT_SYMBOL(wait_for_random_bytes);
1660
1661/*
1662 * Add a callback function that will be invoked when the nonblocking
1663 * pool is initialised.
1664 *
1665 * returns: 0 if callback is successfully added
1666 * -EALREADY if pool is already initialised (callback not called)
1667 * -ENOENT if module for callback is not alive
1668 */
1669int add_random_ready_callback(struct random_ready_callback *rdy)
1670{
1671 struct module *owner;
1672 unsigned long flags;
1673 int err = -EALREADY;
1674
1675 if (crng_ready())
1676 return err;
1677
1678 owner = rdy->owner;
1679 if (!try_module_get(owner))
1680 return -ENOENT;
1681
1682 spin_lock_irqsave(&random_ready_list_lock, flags);
1683 if (crng_ready())
1684 goto out;
1685
1686 owner = NULL;
1687
1688 list_add(&rdy->list, &random_ready_list);
1689 err = 0;
1690
1691out:
1692 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1693
1694 module_put(owner);
1695
1696 return err;
1697}
1698EXPORT_SYMBOL(add_random_ready_callback);
1699
1700/*
1701 * Delete a previously registered readiness callback function.
1702 */
1703void del_random_ready_callback(struct random_ready_callback *rdy)
1704{
1705 unsigned long flags;
1706 struct module *owner = NULL;
1707
1708 spin_lock_irqsave(&random_ready_list_lock, flags);
1709 if (!list_empty(&rdy->list)) {
1710 list_del_init(&rdy->list);
1711 owner = rdy->owner;
1712 }
1713 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1714
1715 module_put(owner);
1716}
1717EXPORT_SYMBOL(del_random_ready_callback);
1718
1719/*
1720 * This function will use the architecture-specific hardware random
1721 * number generator if it is available. The arch-specific hw RNG will
1722 * almost certainly be faster than what we can do in software, but it
1723 * is impossible to verify that it is implemented securely (as
1724 * opposed, to, say, the AES encryption of a sequence number using a
1725 * key known by the NSA). So it's useful if we need the speed, but
1726 * only if we're willing to trust the hardware manufacturer not to
1727 * have put in a back door.
1728 */
1729void get_random_bytes_arch(void *buf, int nbytes)
1730{
1731 char *p = buf;
1732
1733 trace_get_random_bytes_arch(nbytes, _RET_IP_);
1734 while (nbytes) {
1735 unsigned long v;
1736 int chunk = min(nbytes, (int)sizeof(unsigned long));
1737
1738 if (!arch_get_random_long(&v))
1739 break;
1740
1741 memcpy(p, &v, chunk);
1742 p += chunk;
1743 nbytes -= chunk;
1744 }
1745
1746 if (nbytes)
1747 get_random_bytes(p, nbytes);
1748}
1749EXPORT_SYMBOL(get_random_bytes_arch);
1750
1751
1752/*
1753 * init_std_data - initialize pool with system data
1754 *
1755 * @r: pool to initialize
1756 *
1757 * This function clears the pool's entropy count and mixes some system
1758 * data into the pool to prepare it for use. The pool is not cleared
1759 * as that can only decrease the entropy in the pool.
1760 */
1761static void init_std_data(struct entropy_store *r)
1762{
1763 int i;
1764 ktime_t now = ktime_get_real();
1765 unsigned long rv;
1766
1767 r->last_pulled = jiffies;
1768 mix_pool_bytes(r, &now, sizeof(now));
1769 for (i = r->poolinfo->poolbytes; i > 0; i -= sizeof(rv)) {
1770 if (!arch_get_random_seed_long(&rv) &&
1771 !arch_get_random_long(&rv))
1772 rv = random_get_entropy();
1773 mix_pool_bytes(r, &rv, sizeof(rv));
1774 }
1775 mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1776}
1777
1778/*
1779 * Note that setup_arch() may call add_device_randomness()
1780 * long before we get here. This allows seeding of the pools
1781 * with some platform dependent data very early in the boot
1782 * process. But it limits our options here. We must use
1783 * statically allocated structures that already have all
1784 * initializations complete at compile time. We should also
1785 * take care not to overwrite the precious per platform data
1786 * we were given.
1787 */
1788static int rand_initialize(void)
1789{
1790 init_std_data(&input_pool);
1791 init_std_data(&blocking_pool);
1792 crng_initialize(&primary_crng);
1793 crng_global_init_time = jiffies;
1794 if (ratelimit_disable) {
1795 urandom_warning.interval = 0;
1796 unseeded_warning.interval = 0;
1797 }
1798 return 0;
1799}
1800early_initcall(rand_initialize);
1801
1802#ifdef CONFIG_BLOCK
1803void rand_initialize_disk(struct gendisk *disk)
1804{
1805 struct timer_rand_state *state;
1806
1807 /*
1808 * If kzalloc returns null, we just won't use that entropy
1809 * source.
1810 */
1811 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1812 if (state) {
1813 state->last_time = INITIAL_JIFFIES;
1814 disk->random = state;
1815 }
1816}
1817#endif
1818
1819static ssize_t
1820_random_read(int nonblock, char __user *buf, size_t nbytes)
1821{
1822 ssize_t n;
1823
1824 if (nbytes == 0)
1825 return 0;
1826
1827 nbytes = min_t(size_t, nbytes, SEC_XFER_SIZE);
1828 while (1) {
1829 n = extract_entropy_user(&blocking_pool, buf, nbytes);
1830 if (n < 0)
1831 return n;
1832 trace_random_read(n*8, (nbytes-n)*8,
1833 ENTROPY_BITS(&blocking_pool),
1834 ENTROPY_BITS(&input_pool));
1835 if (n > 0)
1836 return n;
1837
1838 /* Pool is (near) empty. Maybe wait and retry. */
1839 if (nonblock)
1840 return -EAGAIN;
1841
1842 wait_event_interruptible(random_read_wait,
1843 ENTROPY_BITS(&input_pool) >=
1844 random_read_wakeup_bits);
1845 if (signal_pending(current))
1846 return -ERESTARTSYS;
1847 }
1848}
1849
1850static ssize_t
1851random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1852{
1853 return _random_read(file->f_flags & O_NONBLOCK, buf, nbytes);
1854}
1855
1856static ssize_t
1857urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1858{
1859 unsigned long flags;
1860 static int maxwarn = 10;
1861 int ret;
1862
1863 if (!crng_ready() && maxwarn > 0) {
1864 maxwarn--;
1865 if (__ratelimit(&urandom_warning))
1866 printk(KERN_NOTICE "random: %s: uninitialized "
1867 "urandom read (%zd bytes read)\n",
1868 current->comm, nbytes);
1869 spin_lock_irqsave(&primary_crng.lock, flags);
1870 crng_init_cnt = 0;
1871 spin_unlock_irqrestore(&primary_crng.lock, flags);
1872 }
1873 nbytes = min_t(size_t, nbytes, INT_MAX >> (ENTROPY_SHIFT + 3));
1874 ret = extract_crng_user(buf, nbytes);
1875 trace_urandom_read(8 * nbytes, 0, ENTROPY_BITS(&input_pool));
1876 return ret;
1877}
1878
1879static __poll_t
1880random_poll(struct file *file, poll_table * wait)
1881{
1882 __poll_t mask;
1883
1884 poll_wait(file, &random_read_wait, wait);
1885 poll_wait(file, &random_write_wait, wait);
1886 mask = 0;
1887 if (ENTROPY_BITS(&input_pool) >= random_read_wakeup_bits)
1888 mask |= EPOLLIN | EPOLLRDNORM;
1889 if (ENTROPY_BITS(&input_pool) < random_write_wakeup_bits)
1890 mask |= EPOLLOUT | EPOLLWRNORM;
1891 return mask;
1892}
1893
1894static int
1895write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1896{
1897 size_t bytes;
1898 __u32 buf[16];
1899 const char __user *p = buffer;
1900
1901 while (count > 0) {
1902 bytes = min(count, sizeof(buf));
1903 if (copy_from_user(&buf, p, bytes))
1904 return -EFAULT;
1905
1906 count -= bytes;
1907 p += bytes;
1908
1909 mix_pool_bytes(r, buf, bytes);
1910 cond_resched();
1911 }
1912
1913 return 0;
1914}
1915
1916static ssize_t random_write(struct file *file, const char __user *buffer,
1917 size_t count, loff_t *ppos)
1918{
1919 size_t ret;
1920
1921 ret = write_pool(&input_pool, buffer, count);
1922 if (ret)
1923 return ret;
1924
1925 return (ssize_t)count;
1926}
1927
1928static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1929{
1930 int size, ent_count;
1931 int __user *p = (int __user *)arg;
1932 int retval;
1933
1934 switch (cmd) {
1935 case RNDGETENTCNT:
1936 /* inherently racy, no point locking */
1937 ent_count = ENTROPY_BITS(&input_pool);
1938 if (put_user(ent_count, p))
1939 return -EFAULT;
1940 return 0;
1941 case RNDADDTOENTCNT:
1942 if (!capable(CAP_SYS_ADMIN))
1943 return -EPERM;
1944 if (get_user(ent_count, p))
1945 return -EFAULT;
1946 return credit_entropy_bits_safe(&input_pool, ent_count);
1947 case RNDADDENTROPY:
1948 if (!capable(CAP_SYS_ADMIN))
1949 return -EPERM;
1950 if (get_user(ent_count, p++))
1951 return -EFAULT;
1952 if (ent_count < 0)
1953 return -EINVAL;
1954 if (get_user(size, p++))
1955 return -EFAULT;
1956 retval = write_pool(&input_pool, (const char __user *)p,
1957 size);
1958 if (retval < 0)
1959 return retval;
1960 return credit_entropy_bits_safe(&input_pool, ent_count);
1961 case RNDZAPENTCNT:
1962 case RNDCLEARPOOL:
1963 /*
1964 * Clear the entropy pool counters. We no longer clear
1965 * the entropy pool, as that's silly.
1966 */
1967 if (!capable(CAP_SYS_ADMIN))
1968 return -EPERM;
1969 input_pool.entropy_count = 0;
1970 blocking_pool.entropy_count = 0;
1971 return 0;
1972 case RNDRESEEDCRNG:
1973 if (!capable(CAP_SYS_ADMIN))
1974 return -EPERM;
1975 if (crng_init < 2)
1976 return -ENODATA;
1977 crng_reseed(&primary_crng, NULL);
1978 crng_global_init_time = jiffies - 1;
1979 return 0;
1980 default:
1981 return -EINVAL;
1982 }
1983}
1984
1985static int random_fasync(int fd, struct file *filp, int on)
1986{
1987 return fasync_helper(fd, filp, on, &fasync);
1988}
1989
1990const struct file_operations random_fops = {
1991 .read = random_read,
1992 .write = random_write,
1993 .poll = random_poll,
1994 .unlocked_ioctl = random_ioctl,
1995 .fasync = random_fasync,
1996 .llseek = noop_llseek,
1997};
1998
1999const struct file_operations urandom_fops = {
2000 .read = urandom_read,
2001 .write = random_write,
2002 .unlocked_ioctl = random_ioctl,
2003 .fasync = random_fasync,
2004 .llseek = noop_llseek,
2005};
2006
2007SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count,
2008 unsigned int, flags)
2009{
2010 int ret;
2011
2012 if (flags & ~(GRND_NONBLOCK|GRND_RANDOM))
2013 return -EINVAL;
2014
2015 if (count > INT_MAX)
2016 count = INT_MAX;
2017
2018 if (flags & GRND_RANDOM)
2019 return _random_read(flags & GRND_NONBLOCK, buf, count);
2020
2021 if (!crng_ready()) {
2022 if (flags & GRND_NONBLOCK)
2023 return -EAGAIN;
2024 ret = wait_for_random_bytes();
2025 if (unlikely(ret))
2026 return ret;
2027 }
2028 return urandom_read(NULL, buf, count, NULL);
2029}
2030
2031/********************************************************************
2032 *
2033 * Sysctl interface
2034 *
2035 ********************************************************************/
2036
2037#ifdef CONFIG_SYSCTL
2038
2039#include <linux/sysctl.h>
2040
2041static int min_read_thresh = 8, min_write_thresh;
2042static int max_read_thresh = OUTPUT_POOL_WORDS * 32;
2043static int max_write_thresh = INPUT_POOL_WORDS * 32;
2044static int random_min_urandom_seed = 60;
2045static char sysctl_bootid[16];
2046
2047/*
2048 * This function is used to return both the bootid UUID, and random
2049 * UUID. The difference is in whether table->data is NULL; if it is,
2050 * then a new UUID is generated and returned to the user.
2051 *
2052 * If the user accesses this via the proc interface, the UUID will be
2053 * returned as an ASCII string in the standard UUID format; if via the
2054 * sysctl system call, as 16 bytes of binary data.
2055 */
2056static int proc_do_uuid(struct ctl_table *table, int write,
2057 void __user *buffer, size_t *lenp, loff_t *ppos)
2058{
2059 struct ctl_table fake_table;
2060 unsigned char buf[64], tmp_uuid[16], *uuid;
2061
2062 uuid = table->data;
2063 if (!uuid) {
2064 uuid = tmp_uuid;
2065 generate_random_uuid(uuid);
2066 } else {
2067 static DEFINE_SPINLOCK(bootid_spinlock);
2068
2069 spin_lock(&bootid_spinlock);
2070 if (!uuid[8])
2071 generate_random_uuid(uuid);
2072 spin_unlock(&bootid_spinlock);
2073 }
2074
2075 sprintf(buf, "%pU", uuid);
2076
2077 fake_table.data = buf;
2078 fake_table.maxlen = sizeof(buf);
2079
2080 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
2081}
2082
2083/*
2084 * Return entropy available scaled to integral bits
2085 */
2086static int proc_do_entropy(struct ctl_table *table, int write,
2087 void __user *buffer, size_t *lenp, loff_t *ppos)
2088{
2089 struct ctl_table fake_table;
2090 int entropy_count;
2091
2092 entropy_count = *(int *)table->data >> ENTROPY_SHIFT;
2093
2094 fake_table.data = &entropy_count;
2095 fake_table.maxlen = sizeof(entropy_count);
2096
2097 return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
2098}
2099
2100static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
2101extern struct ctl_table random_table[];
2102struct ctl_table random_table[] = {
2103 {
2104 .procname = "poolsize",
2105 .data = &sysctl_poolsize,
2106 .maxlen = sizeof(int),
2107 .mode = 0444,
2108 .proc_handler = proc_dointvec,
2109 },
2110 {
2111 .procname = "entropy_avail",
2112 .maxlen = sizeof(int),
2113 .mode = 0444,
2114 .proc_handler = proc_do_entropy,
2115 .data = &input_pool.entropy_count,
2116 },
2117 {
2118 .procname = "read_wakeup_threshold",
2119 .data = &random_read_wakeup_bits,
2120 .maxlen = sizeof(int),
2121 .mode = 0644,
2122 .proc_handler = proc_dointvec_minmax,
2123 .extra1 = &min_read_thresh,
2124 .extra2 = &max_read_thresh,
2125 },
2126 {
2127 .procname = "write_wakeup_threshold",
2128 .data = &random_write_wakeup_bits,
2129 .maxlen = sizeof(int),
2130 .mode = 0644,
2131 .proc_handler = proc_dointvec_minmax,
2132 .extra1 = &min_write_thresh,
2133 .extra2 = &max_write_thresh,
2134 },
2135 {
2136 .procname = "urandom_min_reseed_secs",
2137 .data = &random_min_urandom_seed,
2138 .maxlen = sizeof(int),
2139 .mode = 0644,
2140 .proc_handler = proc_dointvec,
2141 },
2142 {
2143 .procname = "boot_id",
2144 .data = &sysctl_bootid,
2145 .maxlen = 16,
2146 .mode = 0444,
2147 .proc_handler = proc_do_uuid,
2148 },
2149 {
2150 .procname = "uuid",
2151 .maxlen = 16,
2152 .mode = 0444,
2153 .proc_handler = proc_do_uuid,
2154 },
2155#ifdef ADD_INTERRUPT_BENCH
2156 {
2157 .procname = "add_interrupt_avg_cycles",
2158 .data = &avg_cycles,
2159 .maxlen = sizeof(avg_cycles),
2160 .mode = 0444,
2161 .proc_handler = proc_doulongvec_minmax,
2162 },
2163 {
2164 .procname = "add_interrupt_avg_deviation",
2165 .data = &avg_deviation,
2166 .maxlen = sizeof(avg_deviation),
2167 .mode = 0444,
2168 .proc_handler = proc_doulongvec_minmax,
2169 },
2170#endif
2171 { }
2172};
2173#endif /* CONFIG_SYSCTL */
2174
2175struct batched_entropy {
2176 union {
2177 u64 entropy_u64[CHACHA20_BLOCK_SIZE / sizeof(u64)];
2178 u32 entropy_u32[CHACHA20_BLOCK_SIZE / sizeof(u32)];
2179 };
2180 unsigned int position;
2181};
2182static rwlock_t batched_entropy_reset_lock = __RW_LOCK_UNLOCKED(batched_entropy_reset_lock);
2183
2184/*
2185 * Get a random word for internal kernel use only. The quality of the random
2186 * number is either as good as RDRAND or as good as /dev/urandom, with the
2187 * goal of being quite fast and not depleting entropy. In order to ensure
2188 * that the randomness provided by this function is okay, the function
2189 * wait_for_random_bytes() should be called and return 0 at least once
2190 * at any point prior.
2191 */
2192static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64);
2193u64 get_random_u64(void)
2194{
2195 u64 ret;
2196 bool use_lock;
2197 unsigned long flags = 0;
2198 struct batched_entropy *batch;
2199 static void *previous;
2200
2201#if BITS_PER_LONG == 64
2202 if (arch_get_random_long((unsigned long *)&ret))
2203 return ret;
2204#else
2205 if (arch_get_random_long((unsigned long *)&ret) &&
2206 arch_get_random_long((unsigned long *)&ret + 1))
2207 return ret;
2208#endif
2209
2210 warn_unseeded_randomness(&previous);
2211
2212 use_lock = READ_ONCE(crng_init) < 2;
2213 batch = &get_cpu_var(batched_entropy_u64);
2214 if (use_lock)
2215 read_lock_irqsave(&batched_entropy_reset_lock, flags);
2216 if (batch->position % ARRAY_SIZE(batch->entropy_u64) == 0) {
2217 extract_crng((__u32 *)batch->entropy_u64);
2218 batch->position = 0;
2219 }
2220 ret = batch->entropy_u64[batch->position++];
2221 if (use_lock)
2222 read_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2223 put_cpu_var(batched_entropy_u64);
2224 return ret;
2225}
2226EXPORT_SYMBOL(get_random_u64);
2227
2228static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32);
2229u32 get_random_u32(void)
2230{
2231 u32 ret;
2232 bool use_lock;
2233 unsigned long flags = 0;
2234 struct batched_entropy *batch;
2235 static void *previous;
2236
2237 if (arch_get_random_int(&ret))
2238 return ret;
2239
2240 warn_unseeded_randomness(&previous);
2241
2242 use_lock = READ_ONCE(crng_init) < 2;
2243 batch = &get_cpu_var(batched_entropy_u32);
2244 if (use_lock)
2245 read_lock_irqsave(&batched_entropy_reset_lock, flags);
2246 if (batch->position % ARRAY_SIZE(batch->entropy_u32) == 0) {
2247 extract_crng(batch->entropy_u32);
2248 batch->position = 0;
2249 }
2250 ret = batch->entropy_u32[batch->position++];
2251 if (use_lock)
2252 read_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2253 put_cpu_var(batched_entropy_u32);
2254 return ret;
2255}
2256EXPORT_SYMBOL(get_random_u32);
2257
2258/* It's important to invalidate all potential batched entropy that might
2259 * be stored before the crng is initialized, which we can do lazily by
2260 * simply resetting the counter to zero so that it's re-extracted on the
2261 * next usage. */
2262static void invalidate_batched_entropy(void)
2263{
2264 int cpu;
2265 unsigned long flags;
2266
2267 write_lock_irqsave(&batched_entropy_reset_lock, flags);
2268 for_each_possible_cpu (cpu) {
2269 per_cpu_ptr(&batched_entropy_u32, cpu)->position = 0;
2270 per_cpu_ptr(&batched_entropy_u64, cpu)->position = 0;
2271 }
2272 write_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2273}
2274
2275/**
2276 * randomize_page - Generate a random, page aligned address
2277 * @start: The smallest acceptable address the caller will take.
2278 * @range: The size of the area, starting at @start, within which the
2279 * random address must fall.
2280 *
2281 * If @start + @range would overflow, @range is capped.
2282 *
2283 * NOTE: Historical use of randomize_range, which this replaces, presumed that
2284 * @start was already page aligned. We now align it regardless.
2285 *
2286 * Return: A page aligned address within [start, start + range). On error,
2287 * @start is returned.
2288 */
2289unsigned long
2290randomize_page(unsigned long start, unsigned long range)
2291{
2292 if (!PAGE_ALIGNED(start)) {
2293 range -= PAGE_ALIGN(start) - start;
2294 start = PAGE_ALIGN(start);
2295 }
2296
2297 if (start > ULONG_MAX - range)
2298 range = ULONG_MAX - start;
2299
2300 range >>= PAGE_SHIFT;
2301
2302 if (range == 0)
2303 return start;
2304
2305 return start + (get_random_long() % range << PAGE_SHIFT);
2306}
2307
2308/* Interface for in-kernel drivers of true hardware RNGs.
2309 * Those devices may produce endless random bits and will be throttled
2310 * when our pool is full.
2311 */
2312void add_hwgenerator_randomness(const char *buffer, size_t count,
2313 size_t entropy)
2314{
2315 struct entropy_store *poolp = &input_pool;
2316
2317 if (unlikely(crng_init == 0)) {
2318 crng_fast_load(buffer, count);
2319 return;
2320 }
2321
2322 /* Suspend writing if we're above the trickle threshold.
2323 * We'll be woken up again once below random_write_wakeup_thresh,
2324 * or when the calling thread is about to terminate.
2325 */
2326 wait_event_interruptible(random_write_wait, kthread_should_stop() ||
2327 ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
2328 mix_pool_bytes(poolp, buffer, count);
2329 credit_entropy_bits(poolp, entropy);
2330}
2331EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);