1 // SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause)
3 * Copyright (C) 2017-2022 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.
7 * This driver produces cryptographically secure pseudorandom data. It is divided
8 * into roughly six sections, each with a section header:
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.
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.
26 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
39 #include <linux/genhd.h>
40 #include <linux/interrupt.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/siphash.h>
55 #include <crypto/chacha.h>
56 #include <crypto/blake2s.h>
57 #include <asm/processor.h>
59 #include <asm/irq_regs.h>
62 /*********************************************************************
64 * Initialization and readiness waiting.
66 * Much of the RNG infrastructure is devoted to various dependencies
67 * being able to wait until the RNG has collected enough entropy and
68 * is ready for safe consumption.
70 *********************************************************************/
73 * crng_init = 0 --> Uninitialized
75 * 2 --> Initialized from input_pool
77 * crng_init is protected by base_crng->lock, and only increases
78 * its value (from 0->1->2).
80 static int crng_init = 0;
81 #define crng_ready() (likely(crng_init > 1))
82 /* Various types of waiters for crng_init->2 transition. */
83 static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
84 static struct fasync_struct *fasync;
85 static DEFINE_SPINLOCK(random_ready_chain_lock);
86 static RAW_NOTIFIER_HEAD(random_ready_chain);
88 /* Control how we warn userspace. */
89 static struct ratelimit_state unseeded_warning =
90 RATELIMIT_STATE_INIT("warn_unseeded_randomness", HZ, 3);
91 static struct ratelimit_state urandom_warning =
92 RATELIMIT_STATE_INIT("warn_urandom_randomness", HZ, 3);
93 static int ratelimit_disable __read_mostly;
94 module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
95 MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
98 * Returns whether or not the input pool has been seeded and thus guaranteed
99 * to supply cryptographically secure random numbers. This applies to: the
100 * /dev/urandom device, the get_random_bytes function, and the get_random_{u32,
101 * ,u64,int,long} family of functions.
103 * Returns: true if the input pool has been seeded.
104 * false if the input pool has not been seeded.
106 bool rng_is_initialized(void)
110 EXPORT_SYMBOL(rng_is_initialized);
112 /* Used by wait_for_random_bytes(), and considered an entropy collector, below. */
113 static void try_to_generate_entropy(void);
116 * Wait for the input pool to be seeded and thus guaranteed to supply
117 * cryptographically secure random numbers. This applies to: the /dev/urandom
118 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
119 * family of functions. Using any of these functions without first calling
120 * this function forfeits the guarantee of security.
122 * Returns: 0 if the input pool has been seeded.
123 * -ERESTARTSYS if the function was interrupted by a signal.
125 int wait_for_random_bytes(void)
127 while (!crng_ready()) {
130 try_to_generate_entropy();
131 ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ);
133 return ret > 0 ? 0 : ret;
137 EXPORT_SYMBOL(wait_for_random_bytes);
140 * Add a callback function that will be invoked when the input
141 * pool is initialised.
143 * returns: 0 if callback is successfully added
144 * -EALREADY if pool is already initialised (callback not called)
146 int register_random_ready_notifier(struct notifier_block *nb)
154 spin_lock_irqsave(&random_ready_chain_lock, flags);
156 ret = raw_notifier_chain_register(&random_ready_chain, nb);
157 spin_unlock_irqrestore(&random_ready_chain_lock, flags);
160 EXPORT_SYMBOL(register_random_ready_notifier);
163 * Delete a previously registered readiness callback function.
165 int unregister_random_ready_notifier(struct notifier_block *nb)
170 spin_lock_irqsave(&random_ready_chain_lock, flags);
171 ret = raw_notifier_chain_unregister(&random_ready_chain, nb);
172 spin_unlock_irqrestore(&random_ready_chain_lock, flags);
175 EXPORT_SYMBOL(unregister_random_ready_notifier);
177 static void process_random_ready_list(void)
181 spin_lock_irqsave(&random_ready_chain_lock, flags);
182 raw_notifier_call_chain(&random_ready_chain, 0, NULL);
183 spin_unlock_irqrestore(&random_ready_chain_lock, flags);
186 #define warn_unseeded_randomness(previous) \
187 _warn_unseeded_randomness(__func__, (void *)_RET_IP_, (previous))
189 static void _warn_unseeded_randomness(const char *func_name, void *caller, void **previous)
191 #ifdef CONFIG_WARN_ALL_UNSEEDED_RANDOM
192 const bool print_once = false;
194 static bool print_once __read_mostly;
197 if (print_once || crng_ready() ||
198 (previous && (caller == READ_ONCE(*previous))))
200 WRITE_ONCE(*previous, caller);
201 #ifndef CONFIG_WARN_ALL_UNSEEDED_RANDOM
204 if (__ratelimit(&unseeded_warning))
205 printk_deferred(KERN_NOTICE "random: %s called from %pS with crng_init=%d\n",
206 func_name, caller, crng_init);
210 /*********************************************************************
212 * Fast key erasure RNG, the "crng".
214 * These functions expand entropy from the entropy extractor into
215 * long streams for external consumption using the "fast key erasure"
216 * RNG described at <https://blog.cr.yp.to/20170723-random.html>.
218 * There are a few exported interfaces for use by other drivers:
220 * void get_random_bytes(void *buf, size_t nbytes)
221 * u32 get_random_u32()
222 * u64 get_random_u64()
223 * unsigned int get_random_int()
224 * unsigned long get_random_long()
226 * These interfaces will return the requested number of random bytes
227 * into the given buffer or as a return value. This is equivalent to
228 * a read from /dev/urandom. The u32, u64, int, and long family of
229 * functions may be higher performance for one-off random integers,
230 * because they do a bit of buffering and do not invoke reseeding
231 * until the buffer is emptied.
233 *********************************************************************/
236 CRNG_RESEED_START_INTERVAL = HZ,
237 CRNG_RESEED_INTERVAL = 60 * HZ
241 u8 key[CHACHA_KEY_SIZE] __aligned(__alignof__(long));
243 unsigned long generation;
246 .lock = __SPIN_LOCK_UNLOCKED(base_crng.lock)
250 u8 key[CHACHA_KEY_SIZE];
251 unsigned long generation;
255 static DEFINE_PER_CPU(struct crng, crngs) = {
256 .generation = ULONG_MAX,
257 .lock = INIT_LOCAL_LOCK(crngs.lock),
260 /* Used by crng_reseed() and crng_make_state() to extract a new seed from the input pool. */
261 static void extract_entropy(void *buf, size_t nbytes);
263 /* This extracts a new crng key from the input pool. */
264 static void crng_reseed(void)
267 unsigned long next_gen;
268 u8 key[CHACHA_KEY_SIZE];
269 bool finalize_init = false;
271 extract_entropy(key, sizeof(key));
274 * We copy the new key into the base_crng, overwriting the old one,
275 * and update the generation counter. We avoid hitting ULONG_MAX,
276 * because the per-cpu crngs are initialized to ULONG_MAX, so this
277 * forces new CPUs that come online to always initialize.
279 spin_lock_irqsave(&base_crng.lock, flags);
280 memcpy(base_crng.key, key, sizeof(base_crng.key));
281 next_gen = base_crng.generation + 1;
282 if (next_gen == ULONG_MAX)
284 WRITE_ONCE(base_crng.generation, next_gen);
285 WRITE_ONCE(base_crng.birth, jiffies);
288 finalize_init = true;
290 spin_unlock_irqrestore(&base_crng.lock, flags);
291 memzero_explicit(key, sizeof(key));
293 process_random_ready_list();
294 wake_up_interruptible(&crng_init_wait);
295 kill_fasync(&fasync, SIGIO, POLL_IN);
296 pr_notice("crng init done\n");
297 if (unseeded_warning.missed) {
298 pr_notice("%d get_random_xx warning(s) missed due to ratelimiting\n",
299 unseeded_warning.missed);
300 unseeded_warning.missed = 0;
302 if (urandom_warning.missed) {
303 pr_notice("%d urandom warning(s) missed due to ratelimiting\n",
304 urandom_warning.missed);
305 urandom_warning.missed = 0;
311 * This generates a ChaCha block using the provided key, and then
312 * immediately overwites that key with half the block. It returns
313 * the resultant ChaCha state to the user, along with the second
314 * half of the block containing 32 bytes of random data that may
315 * be used; random_data_len may not be greater than 32.
317 * The returned ChaCha state contains within it a copy of the old
318 * key value, at index 4, so the state should always be zeroed out
319 * immediately after using in order to maintain forward secrecy.
320 * If the state cannot be erased in a timely manner, then it is
321 * safer to set the random_data parameter to &chacha_state[4] so
322 * that this function overwrites it before returning.
324 static void crng_fast_key_erasure(u8 key[CHACHA_KEY_SIZE],
325 u32 chacha_state[CHACHA_STATE_WORDS],
326 u8 *random_data, size_t random_data_len)
328 u8 first_block[CHACHA_BLOCK_SIZE];
330 BUG_ON(random_data_len > 32);
332 chacha_init_consts(chacha_state);
333 memcpy(&chacha_state[4], key, CHACHA_KEY_SIZE);
334 memset(&chacha_state[12], 0, sizeof(u32) * 4);
335 chacha20_block(chacha_state, first_block);
337 memcpy(key, first_block, CHACHA_KEY_SIZE);
338 memcpy(random_data, first_block + CHACHA_KEY_SIZE, random_data_len);
339 memzero_explicit(first_block, sizeof(first_block));
343 * Return whether the crng seed is considered to be sufficiently old
344 * that a reseeding is needed. This happens if the last reseeding
345 * was CRNG_RESEED_INTERVAL ago, or during early boot, at an interval
346 * proportional to the uptime.
348 static bool crng_has_old_seed(void)
350 static bool early_boot = true;
351 unsigned long interval = CRNG_RESEED_INTERVAL;
353 if (unlikely(READ_ONCE(early_boot))) {
354 time64_t uptime = ktime_get_seconds();
355 if (uptime >= CRNG_RESEED_INTERVAL / HZ * 2)
356 WRITE_ONCE(early_boot, false);
358 interval = max_t(unsigned int, CRNG_RESEED_START_INTERVAL,
359 (unsigned int)uptime / 2 * HZ);
361 return time_after(jiffies, READ_ONCE(base_crng.birth) + interval);
365 * This function returns a ChaCha state that you may use for generating
366 * random data. It also returns up to 32 bytes on its own of random data
367 * that may be used; random_data_len may not be greater than 32.
369 static void crng_make_state(u32 chacha_state[CHACHA_STATE_WORDS],
370 u8 *random_data, size_t random_data_len)
375 BUG_ON(random_data_len > 32);
378 * For the fast path, we check whether we're ready, unlocked first, and
379 * then re-check once locked later. In the case where we're really not
380 * ready, we do fast key erasure with the base_crng directly, extracting
386 spin_lock_irqsave(&base_crng.lock, flags);
387 ready = crng_ready();
390 extract_entropy(base_crng.key, sizeof(base_crng.key));
391 crng_fast_key_erasure(base_crng.key, chacha_state,
392 random_data, random_data_len);
394 spin_unlock_irqrestore(&base_crng.lock, flags);
400 * If the base_crng is old enough, we reseed, which in turn bumps the
401 * generation counter that we check below.
403 if (unlikely(crng_has_old_seed()))
406 local_lock_irqsave(&crngs.lock, flags);
407 crng = raw_cpu_ptr(&crngs);
410 * If our per-cpu crng is older than the base_crng, then it means
411 * somebody reseeded the base_crng. In that case, we do fast key
412 * erasure on the base_crng, and use its output as the new key
413 * for our per-cpu crng. This brings us up to date with base_crng.
415 if (unlikely(crng->generation != READ_ONCE(base_crng.generation))) {
416 spin_lock(&base_crng.lock);
417 crng_fast_key_erasure(base_crng.key, chacha_state,
418 crng->key, sizeof(crng->key));
419 crng->generation = base_crng.generation;
420 spin_unlock(&base_crng.lock);
424 * Finally, when we've made it this far, our per-cpu crng has an up
425 * to date key, and we can do fast key erasure with it to produce
426 * some random data and a ChaCha state for the caller. All other
427 * branches of this function are "unlikely", so most of the time we
428 * should wind up here immediately.
430 crng_fast_key_erasure(crng->key, chacha_state, random_data, random_data_len);
431 local_unlock_irqrestore(&crngs.lock, flags);
434 static void _get_random_bytes(void *buf, size_t nbytes)
436 u32 chacha_state[CHACHA_STATE_WORDS];
437 u8 tmp[CHACHA_BLOCK_SIZE];
443 len = min_t(size_t, 32, nbytes);
444 crng_make_state(chacha_state, buf, len);
449 if (nbytes < CHACHA_BLOCK_SIZE) {
450 chacha20_block(chacha_state, tmp);
451 memcpy(buf, tmp, nbytes);
452 memzero_explicit(tmp, sizeof(tmp));
456 chacha20_block(chacha_state, buf);
457 if (unlikely(chacha_state[12] == 0))
459 nbytes -= CHACHA_BLOCK_SIZE;
460 buf += CHACHA_BLOCK_SIZE;
463 memzero_explicit(chacha_state, sizeof(chacha_state));
467 * This function is the exported kernel interface. It returns some
468 * number of good random numbers, suitable for key generation, seeding
469 * TCP sequence numbers, etc. It does not rely on the hardware random
470 * number generator. For random bytes direct from the hardware RNG
471 * (when available), use get_random_bytes_arch(). In order to ensure
472 * that the randomness provided by this function is okay, the function
473 * wait_for_random_bytes() should be called and return 0 at least once
474 * at any point prior.
476 void get_random_bytes(void *buf, size_t nbytes)
478 static void *previous;
480 warn_unseeded_randomness(&previous);
481 _get_random_bytes(buf, nbytes);
483 EXPORT_SYMBOL(get_random_bytes);
485 static ssize_t get_random_bytes_user(void __user *buf, size_t nbytes)
487 size_t len, left, ret = 0;
488 u32 chacha_state[CHACHA_STATE_WORDS];
489 u8 output[CHACHA_BLOCK_SIZE];
495 * Immediately overwrite the ChaCha key at index 4 with random
496 * bytes, in case userspace causes copy_to_user() below to sleep
497 * forever, so that we still retain forward secrecy in that case.
499 crng_make_state(chacha_state, (u8 *)&chacha_state[4], CHACHA_KEY_SIZE);
501 * However, if we're doing a read of len <= 32, we don't need to
502 * use chacha_state after, so we can simply return those bytes to
505 if (nbytes <= CHACHA_KEY_SIZE) {
506 ret = nbytes - copy_to_user(buf, &chacha_state[4], nbytes);
507 goto out_zero_chacha;
511 chacha20_block(chacha_state, output);
512 if (unlikely(chacha_state[12] == 0))
515 len = min_t(size_t, nbytes, CHACHA_BLOCK_SIZE);
516 left = copy_to_user(buf, output, len);
528 BUILD_BUG_ON(PAGE_SIZE % CHACHA_BLOCK_SIZE != 0);
529 if (ret % PAGE_SIZE == 0) {
530 if (signal_pending(current))
536 memzero_explicit(output, sizeof(output));
538 memzero_explicit(chacha_state, sizeof(chacha_state));
539 return ret ? ret : -EFAULT;
543 * Batched entropy returns random integers. The quality of the random
544 * number is good as /dev/urandom. In order to ensure that the randomness
545 * provided by this function is okay, the function wait_for_random_bytes()
546 * should be called and return 0 at least once at any point prior.
548 struct batched_entropy {
551 * We make this 1.5x a ChaCha block, so that we get the
552 * remaining 32 bytes from fast key erasure, plus one full
553 * block from the detached ChaCha state. We can increase
554 * the size of this later if needed so long as we keep the
555 * formula of (integer_blocks + 0.5) * CHACHA_BLOCK_SIZE.
557 u64 entropy_u64[CHACHA_BLOCK_SIZE * 3 / (2 * sizeof(u64))];
558 u32 entropy_u32[CHACHA_BLOCK_SIZE * 3 / (2 * sizeof(u32))];
561 unsigned long generation;
562 unsigned int position;
566 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64) = {
567 .lock = INIT_LOCAL_LOCK(batched_entropy_u64.lock),
571 u64 get_random_u64(void)
575 struct batched_entropy *batch;
576 static void *previous;
577 unsigned long next_gen;
579 warn_unseeded_randomness(&previous);
582 _get_random_bytes(&ret, sizeof(ret));
586 local_lock_irqsave(&batched_entropy_u64.lock, flags);
587 batch = raw_cpu_ptr(&batched_entropy_u64);
589 next_gen = READ_ONCE(base_crng.generation);
590 if (batch->position >= ARRAY_SIZE(batch->entropy_u64) ||
591 next_gen != batch->generation) {
592 _get_random_bytes(batch->entropy_u64, sizeof(batch->entropy_u64));
594 batch->generation = next_gen;
597 ret = batch->entropy_u64[batch->position];
598 batch->entropy_u64[batch->position] = 0;
600 local_unlock_irqrestore(&batched_entropy_u64.lock, flags);
603 EXPORT_SYMBOL(get_random_u64);
605 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32) = {
606 .lock = INIT_LOCAL_LOCK(batched_entropy_u32.lock),
610 u32 get_random_u32(void)
614 struct batched_entropy *batch;
615 static void *previous;
616 unsigned long next_gen;
618 warn_unseeded_randomness(&previous);
621 _get_random_bytes(&ret, sizeof(ret));
625 local_lock_irqsave(&batched_entropy_u32.lock, flags);
626 batch = raw_cpu_ptr(&batched_entropy_u32);
628 next_gen = READ_ONCE(base_crng.generation);
629 if (batch->position >= ARRAY_SIZE(batch->entropy_u32) ||
630 next_gen != batch->generation) {
631 _get_random_bytes(batch->entropy_u32, sizeof(batch->entropy_u32));
633 batch->generation = next_gen;
636 ret = batch->entropy_u32[batch->position];
637 batch->entropy_u32[batch->position] = 0;
639 local_unlock_irqrestore(&batched_entropy_u32.lock, flags);
642 EXPORT_SYMBOL(get_random_u32);
646 * This function is called when the CPU is coming up, with entry
647 * CPUHP_RANDOM_PREPARE, which comes before CPUHP_WORKQUEUE_PREP.
649 int random_prepare_cpu(unsigned int cpu)
652 * When the cpu comes back online, immediately invalidate both
653 * the per-cpu crng and all batches, so that we serve fresh
656 per_cpu_ptr(&crngs, cpu)->generation = ULONG_MAX;
657 per_cpu_ptr(&batched_entropy_u32, cpu)->position = UINT_MAX;
658 per_cpu_ptr(&batched_entropy_u64, cpu)->position = UINT_MAX;
664 * randomize_page - Generate a random, page aligned address
665 * @start: The smallest acceptable address the caller will take.
666 * @range: The size of the area, starting at @start, within which the
667 * random address must fall.
669 * If @start + @range would overflow, @range is capped.
671 * NOTE: Historical use of randomize_range, which this replaces, presumed that
672 * @start was already page aligned. We now align it regardless.
674 * Return: A page aligned address within [start, start + range). On error,
675 * @start is returned.
677 unsigned long randomize_page(unsigned long start, unsigned long range)
679 if (!PAGE_ALIGNED(start)) {
680 range -= PAGE_ALIGN(start) - start;
681 start = PAGE_ALIGN(start);
684 if (start > ULONG_MAX - range)
685 range = ULONG_MAX - start;
687 range >>= PAGE_SHIFT;
692 return start + (get_random_long() % range << PAGE_SHIFT);
696 * This function will use the architecture-specific hardware random
697 * number generator if it is available. It is not recommended for
698 * use. Use get_random_bytes() instead. It returns the number of
701 size_t __must_check get_random_bytes_arch(void *buf, size_t nbytes)
703 size_t left = nbytes;
708 size_t chunk = min_t(size_t, left, sizeof(unsigned long));
710 if (!arch_get_random_long(&v))
713 memcpy(p, &v, chunk);
718 return nbytes - left;
720 EXPORT_SYMBOL(get_random_bytes_arch);
723 /**********************************************************************
725 * Entropy accumulation and extraction routines.
727 * Callers may add entropy via:
729 * static void mix_pool_bytes(const void *in, size_t nbytes)
731 * After which, if added entropy should be credited:
733 * static void credit_init_bits(size_t nbits)
735 * Finally, extract entropy via:
737 * static void extract_entropy(void *buf, size_t nbytes)
739 **********************************************************************/
742 POOL_BITS = BLAKE2S_HASH_SIZE * 8,
743 POOL_INIT_BITS = POOL_BITS, /* No point in settling for less. */
744 POOL_FAST_INIT_BITS = POOL_INIT_BITS / 2
748 struct blake2s_state hash;
750 unsigned int init_bits;
752 .hash.h = { BLAKE2S_IV0 ^ (0x01010000 | BLAKE2S_HASH_SIZE),
753 BLAKE2S_IV1, BLAKE2S_IV2, BLAKE2S_IV3, BLAKE2S_IV4,
754 BLAKE2S_IV5, BLAKE2S_IV6, BLAKE2S_IV7 },
755 .hash.outlen = BLAKE2S_HASH_SIZE,
756 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
759 static void _mix_pool_bytes(const void *in, size_t nbytes)
761 blake2s_update(&input_pool.hash, in, nbytes);
765 * This function adds bytes into the input pool. It does not
766 * update the initialization bit counter; the caller should call
767 * credit_init_bits if this is appropriate.
769 static void mix_pool_bytes(const void *in, size_t nbytes)
773 spin_lock_irqsave(&input_pool.lock, flags);
774 _mix_pool_bytes(in, nbytes);
775 spin_unlock_irqrestore(&input_pool.lock, flags);
779 * This is an HKDF-like construction for using the hashed collected entropy
780 * as a PRF key, that's then expanded block-by-block.
782 static void extract_entropy(void *buf, size_t nbytes)
785 u8 seed[BLAKE2S_HASH_SIZE], next_key[BLAKE2S_HASH_SIZE];
787 unsigned long rdseed[32 / sizeof(long)];
792 for (i = 0; i < ARRAY_SIZE(block.rdseed); ++i) {
793 if (!arch_get_random_seed_long(&block.rdseed[i]) &&
794 !arch_get_random_long(&block.rdseed[i]))
795 block.rdseed[i] = random_get_entropy();
798 spin_lock_irqsave(&input_pool.lock, flags);
800 /* seed = HASHPRF(last_key, entropy_input) */
801 blake2s_final(&input_pool.hash, seed);
803 /* next_key = HASHPRF(seed, RDSEED || 0) */
805 blake2s(next_key, (u8 *)&block, seed, sizeof(next_key), sizeof(block), sizeof(seed));
806 blake2s_init_key(&input_pool.hash, BLAKE2S_HASH_SIZE, next_key, sizeof(next_key));
808 spin_unlock_irqrestore(&input_pool.lock, flags);
809 memzero_explicit(next_key, sizeof(next_key));
812 i = min_t(size_t, nbytes, BLAKE2S_HASH_SIZE);
813 /* output = HASHPRF(seed, RDSEED || ++counter) */
815 blake2s(buf, (u8 *)&block, seed, i, sizeof(block), sizeof(seed));
820 memzero_explicit(seed, sizeof(seed));
821 memzero_explicit(&block, sizeof(block));
824 static void credit_init_bits(size_t nbits)
826 unsigned int init_bits, orig, add;
829 if (crng_ready() || !nbits)
832 add = min_t(size_t, nbits, POOL_BITS);
835 orig = READ_ONCE(input_pool.init_bits);
836 init_bits = min_t(unsigned int, POOL_BITS, orig + add);
837 } while (cmpxchg(&input_pool.init_bits, orig, init_bits) != orig);
839 if (!crng_ready() && init_bits >= POOL_INIT_BITS)
841 else if (unlikely(crng_init == 0 && init_bits >= POOL_FAST_INIT_BITS)) {
842 spin_lock_irqsave(&base_crng.lock, flags);
843 if (crng_init == 0) {
844 extract_entropy(base_crng.key, sizeof(base_crng.key));
847 spin_unlock_irqrestore(&base_crng.lock, flags);
852 /**********************************************************************
854 * Entropy collection routines.
856 * The following exported functions are used for pushing entropy into
857 * the above entropy accumulation routines:
859 * void add_device_randomness(const void *buf, size_t size);
860 * void add_hwgenerator_randomness(const void *buffer, size_t count,
862 * void add_bootloader_randomness(const void *buf, size_t size);
863 * void add_interrupt_randomness(int irq);
864 * void add_input_randomness(unsigned int type, unsigned int code,
865 * unsigned int value);
866 * void add_disk_randomness(struct gendisk *disk);
868 * add_device_randomness() adds data to the input pool that
869 * is likely to differ between two devices (or possibly even per boot).
870 * This would be things like MAC addresses or serial numbers, or the
871 * read-out of the RTC. This does *not* credit any actual entropy to
872 * the pool, but it initializes the pool to different values for devices
873 * that might otherwise be identical and have very little entropy
874 * available to them (particularly common in the embedded world).
876 * add_hwgenerator_randomness() is for true hardware RNGs, and will credit
877 * entropy as specified by the caller. If the entropy pool is full it will
878 * block until more entropy is needed.
880 * add_bootloader_randomness() is called by bootloader drivers, such as EFI
881 * and device tree, and credits its input depending on whether or not the
882 * configuration option CONFIG_RANDOM_TRUST_BOOTLOADER is set.
884 * add_interrupt_randomness() uses the interrupt timing as random
885 * inputs to the entropy pool. Using the cycle counters and the irq source
886 * as inputs, it feeds the input pool roughly once a second or after 64
887 * interrupts, crediting 1 bit of entropy for whichever comes first.
889 * add_input_randomness() uses the input layer interrupt timing, as well
890 * as the event type information from the hardware.
892 * add_disk_randomness() uses what amounts to the seek time of block
893 * layer request events, on a per-disk_devt basis, as input to the
894 * entropy pool. Note that high-speed solid state drives with very low
895 * seek times do not make for good sources of entropy, as their seek
896 * times are usually fairly consistent.
898 * The last two routines try to estimate how many bits of entropy
899 * to credit. They do this by keeping track of the first and second
900 * order deltas of the event timings.
902 **********************************************************************/
904 static bool trust_cpu __ro_after_init = IS_ENABLED(CONFIG_RANDOM_TRUST_CPU);
905 static bool trust_bootloader __ro_after_init = IS_ENABLED(CONFIG_RANDOM_TRUST_BOOTLOADER);
906 static int __init parse_trust_cpu(char *arg)
908 return kstrtobool(arg, &trust_cpu);
910 static int __init parse_trust_bootloader(char *arg)
912 return kstrtobool(arg, &trust_bootloader);
914 early_param("random.trust_cpu", parse_trust_cpu);
915 early_param("random.trust_bootloader", parse_trust_bootloader);
918 * The first collection of entropy occurs at system boot while interrupts
919 * are still turned off. Here we push in RDSEED, a timestamp, and utsname().
920 * Depending on the above configuration knob, RDSEED may be considered
921 * sufficient for initialization. Note that much earlier setup may already
922 * have pushed entropy into the input pool by the time we get here.
924 int __init rand_initialize(void)
927 ktime_t now = ktime_get_real();
928 bool arch_init = true;
931 #if defined(LATENT_ENTROPY_PLUGIN)
932 static const u8 compiletime_seed[BLAKE2S_BLOCK_SIZE] __initconst __latent_entropy;
933 _mix_pool_bytes(compiletime_seed, sizeof(compiletime_seed));
936 for (i = 0; i < BLAKE2S_BLOCK_SIZE; i += sizeof(rv)) {
937 if (!arch_get_random_seed_long_early(&rv) &&
938 !arch_get_random_long_early(&rv)) {
939 rv = random_get_entropy();
942 _mix_pool_bytes(&rv, sizeof(rv));
944 _mix_pool_bytes(&now, sizeof(now));
945 _mix_pool_bytes(utsname(), sizeof(*(utsname())));
949 else if (arch_init && trust_cpu)
950 credit_init_bits(BLAKE2S_BLOCK_SIZE * 8);
952 if (ratelimit_disable) {
953 urandom_warning.interval = 0;
954 unseeded_warning.interval = 0;
960 * Add device- or boot-specific data to the input pool to help
963 * None of this adds any entropy; it is meant to avoid the problem of
964 * the entropy pool having similar initial state across largely
967 void add_device_randomness(const void *buf, size_t size)
969 unsigned long entropy = random_get_entropy();
972 spin_lock_irqsave(&input_pool.lock, flags);
973 _mix_pool_bytes(&entropy, sizeof(entropy));
974 _mix_pool_bytes(buf, size);
975 spin_unlock_irqrestore(&input_pool.lock, flags);
977 EXPORT_SYMBOL(add_device_randomness);
980 * Interface for in-kernel drivers of true hardware RNGs.
981 * Those devices may produce endless random bits and will be throttled
982 * when our pool is full.
984 void add_hwgenerator_randomness(const void *buffer, size_t count,
987 mix_pool_bytes(buffer, count);
988 credit_init_bits(entropy);
991 * Throttle writing to once every CRNG_RESEED_INTERVAL, unless
992 * we're not yet initialized.
994 if (!kthread_should_stop() && crng_ready())
995 schedule_timeout_interruptible(CRNG_RESEED_INTERVAL);
997 EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
1000 * Handle random seed passed by bootloader, and credit it if
1001 * CONFIG_RANDOM_TRUST_BOOTLOADER is set.
1003 void add_bootloader_randomness(const void *buf, size_t size)
1005 mix_pool_bytes(buf, size);
1006 if (trust_bootloader)
1007 credit_init_bits(size * 8);
1009 EXPORT_SYMBOL_GPL(add_bootloader_randomness);
1012 struct work_struct mix;
1013 unsigned long pool[4];
1018 static DEFINE_PER_CPU(struct fast_pool, irq_randomness) = {
1020 #define FASTMIX_PERM SIPHASH_PERMUTATION
1021 .pool = { SIPHASH_CONST_0, SIPHASH_CONST_1, SIPHASH_CONST_2, SIPHASH_CONST_3 }
1023 #define FASTMIX_PERM HSIPHASH_PERMUTATION
1024 .pool = { HSIPHASH_CONST_0, HSIPHASH_CONST_1, HSIPHASH_CONST_2, HSIPHASH_CONST_3 }
1029 * This is [Half]SipHash-1-x, starting from an empty key. Because
1030 * the key is fixed, it assumes that its inputs are non-malicious,
1031 * and therefore this has no security on its own. s represents the
1032 * four-word SipHash state, while v represents a two-word input.
1034 static void fast_mix(unsigned long s[4], unsigned long v1, unsigned long v2)
1037 FASTMIX_PERM(s[0], s[1], s[2], s[3]);
1040 FASTMIX_PERM(s[0], s[1], s[2], s[3]);
1046 * This function is called when the CPU has just come online, with
1047 * entry CPUHP_AP_RANDOM_ONLINE, just after CPUHP_AP_WORKQUEUE_ONLINE.
1049 int random_online_cpu(unsigned int cpu)
1052 * During CPU shutdown and before CPU onlining, add_interrupt_
1053 * randomness() may schedule mix_interrupt_randomness(), and
1054 * set the MIX_INFLIGHT flag. However, because the worker can
1055 * be scheduled on a different CPU during this period, that
1056 * flag will never be cleared. For that reason, we zero out
1057 * the flag here, which runs just after workqueues are onlined
1058 * for the CPU again. This also has the effect of setting the
1059 * irq randomness count to zero so that new accumulated irqs
1062 per_cpu_ptr(&irq_randomness, cpu)->count = 0;
1067 static void mix_interrupt_randomness(struct work_struct *work)
1069 struct fast_pool *fast_pool = container_of(work, struct fast_pool, mix);
1071 * The size of the copied stack pool is explicitly 2 longs so that we
1072 * only ever ingest half of the siphash output each time, retaining
1073 * the other half as the next "key" that carries over. The entropy is
1074 * supposed to be sufficiently dispersed between bits so on average
1075 * we don't wind up "losing" some.
1077 unsigned long pool[2];
1080 /* Check to see if we're running on the wrong CPU due to hotplug. */
1081 local_irq_disable();
1082 if (fast_pool != this_cpu_ptr(&irq_randomness)) {
1088 * Copy the pool to the stack so that the mixer always has a
1089 * consistent view, before we reenable irqs again.
1091 memcpy(pool, fast_pool->pool, sizeof(pool));
1092 count = fast_pool->count;
1093 fast_pool->count = 0;
1094 fast_pool->last = jiffies;
1097 mix_pool_bytes(pool, sizeof(pool));
1098 credit_init_bits(max(1u, (count & U16_MAX) / 64));
1100 memzero_explicit(pool, sizeof(pool));
1103 void add_interrupt_randomness(int irq)
1105 enum { MIX_INFLIGHT = 1U << 31 };
1106 unsigned long entropy = random_get_entropy();
1107 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1108 struct pt_regs *regs = get_irq_regs();
1109 unsigned int new_count;
1111 fast_mix(fast_pool->pool, entropy,
1112 (regs ? instruction_pointer(regs) : _RET_IP_) ^ swab(irq));
1113 new_count = ++fast_pool->count;
1115 if (new_count & MIX_INFLIGHT)
1118 if (new_count < 64 && !time_is_before_jiffies(fast_pool->last + HZ))
1121 if (unlikely(!fast_pool->mix.func))
1122 INIT_WORK(&fast_pool->mix, mix_interrupt_randomness);
1123 fast_pool->count |= MIX_INFLIGHT;
1124 queue_work_on(raw_smp_processor_id(), system_highpri_wq, &fast_pool->mix);
1126 EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1128 /* There is one of these per entropy source */
1129 struct timer_rand_state {
1130 unsigned long last_time;
1131 long last_delta, last_delta2;
1135 * This function adds entropy to the entropy "pool" by using timing
1136 * delays. It uses the timer_rand_state structure to make an estimate
1137 * of how many bits of entropy this call has added to the pool. The
1138 * value "num" is also added to the pool; it should somehow describe
1139 * the type of event that just happened.
1141 static void add_timer_randomness(struct timer_rand_state *state, unsigned int num)
1143 unsigned long entropy = random_get_entropy(), now = jiffies, flags;
1144 long delta, delta2, delta3;
1148 * If we're in a hard IRQ, add_interrupt_randomness() will be called
1149 * sometime after, so mix into the fast pool.
1152 fast_mix(this_cpu_ptr(&irq_randomness)->pool, entropy, num);
1154 spin_lock_irqsave(&input_pool.lock, flags);
1155 _mix_pool_bytes(&entropy, sizeof(entropy));
1156 _mix_pool_bytes(&num, sizeof(num));
1157 spin_unlock_irqrestore(&input_pool.lock, flags);
1164 * Calculate number of bits of randomness we probably added.
1165 * We take into account the first, second and third-order deltas
1166 * in order to make our estimate.
1168 delta = now - READ_ONCE(state->last_time);
1169 WRITE_ONCE(state->last_time, now);
1171 delta2 = delta - READ_ONCE(state->last_delta);
1172 WRITE_ONCE(state->last_delta, delta);
1174 delta3 = delta2 - READ_ONCE(state->last_delta2);
1175 WRITE_ONCE(state->last_delta2, delta2);
1189 * delta is now minimum absolute delta. Round down by 1 bit
1190 * on general principles, and limit entropy estimate to 11 bits.
1192 bits = min(fls(delta >> 1), 11);
1195 * As mentioned above, if we're in a hard IRQ, add_interrupt_randomness()
1196 * will run after this, which uses a different crediting scheme of 1 bit
1197 * per every 64 interrupts. In order to let that function do accounting
1198 * close to the one in this function, we credit a full 64/64 bit per bit,
1199 * and then subtract one to account for the extra one added.
1202 this_cpu_ptr(&irq_randomness)->count += max(1u, bits * 64) - 1;
1204 credit_init_bits(bits);
1207 void add_input_randomness(unsigned int type, unsigned int code,
1210 static unsigned char last_value;
1211 static struct timer_rand_state input_timer_state = { INITIAL_JIFFIES };
1213 /* Ignore autorepeat and the like. */
1214 if (value == last_value)
1218 add_timer_randomness(&input_timer_state,
1219 (type << 4) ^ code ^ (code >> 4) ^ value);
1221 EXPORT_SYMBOL_GPL(add_input_randomness);
1224 void add_disk_randomness(struct gendisk *disk)
1226 if (!disk || !disk->random)
1228 /* First major is 1, so we get >= 0x200 here. */
1229 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1231 EXPORT_SYMBOL_GPL(add_disk_randomness);
1233 void rand_initialize_disk(struct gendisk *disk)
1235 struct timer_rand_state *state;
1238 * If kzalloc returns null, we just won't use that entropy
1241 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1243 state->last_time = INITIAL_JIFFIES;
1244 disk->random = state;
1250 * Each time the timer fires, we expect that we got an unpredictable
1251 * jump in the cycle counter. Even if the timer is running on another
1252 * CPU, the timer activity will be touching the stack of the CPU that is
1253 * generating entropy..
1255 * Note that we don't re-arm the timer in the timer itself - we are
1256 * happy to be scheduled away, since that just makes the load more
1257 * complex, but we do not want the timer to keep ticking unless the
1258 * entropy loop is running.
1260 * So the re-arming always happens in the entropy loop itself.
1262 static void entropy_timer(struct timer_list *t)
1264 credit_init_bits(1);
1268 * If we have an actual cycle counter, see if we can
1269 * generate enough entropy with timing noise
1271 static void try_to_generate_entropy(void)
1274 unsigned long entropy;
1275 struct timer_list timer;
1278 stack.entropy = random_get_entropy();
1280 /* Slow counter - or none. Don't even bother */
1281 if (stack.entropy == random_get_entropy())
1284 timer_setup_on_stack(&stack.timer, entropy_timer, 0);
1285 while (!crng_ready() && !signal_pending(current)) {
1286 if (!timer_pending(&stack.timer))
1287 mod_timer(&stack.timer, jiffies + 1);
1288 mix_pool_bytes(&stack.entropy, sizeof(stack.entropy));
1290 stack.entropy = random_get_entropy();
1293 del_timer_sync(&stack.timer);
1294 destroy_timer_on_stack(&stack.timer);
1295 mix_pool_bytes(&stack.entropy, sizeof(stack.entropy));
1299 /**********************************************************************
1301 * Userspace reader/writer interfaces.
1303 * getrandom(2) is the primary modern interface into the RNG and should
1304 * be used in preference to anything else.
1306 * Reading from /dev/random has the same functionality as calling
1307 * getrandom(2) with flags=0. In earlier versions, however, it had
1308 * vastly different semantics and should therefore be avoided, to
1309 * prevent backwards compatibility issues.
1311 * Reading from /dev/urandom has the same functionality as calling
1312 * getrandom(2) with flags=GRND_INSECURE. Because it does not block
1313 * waiting for the RNG to be ready, it should not be used.
1315 * Writing to either /dev/random or /dev/urandom adds entropy to
1316 * the input pool but does not credit it.
1318 * Polling on /dev/random indicates when the RNG is initialized, on
1319 * the read side, and when it wants new entropy, on the write side.
1321 * Both /dev/random and /dev/urandom have the same set of ioctls for
1322 * adding entropy, getting the entropy count, zeroing the count, and
1323 * reseeding the crng.
1325 **********************************************************************/
1327 SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count, unsigned int,
1330 if (flags & ~(GRND_NONBLOCK | GRND_RANDOM | GRND_INSECURE))
1334 * Requesting insecure and blocking randomness at the same time makes
1337 if ((flags & (GRND_INSECURE | GRND_RANDOM)) == (GRND_INSECURE | GRND_RANDOM))
1340 if (count > INT_MAX)
1343 if (!(flags & GRND_INSECURE) && !crng_ready()) {
1346 if (flags & GRND_NONBLOCK)
1348 ret = wait_for_random_bytes();
1352 return get_random_bytes_user(buf, count);
1355 static __poll_t random_poll(struct file *file, poll_table *wait)
1357 poll_wait(file, &crng_init_wait, wait);
1358 return crng_ready() ? EPOLLIN | EPOLLRDNORM : EPOLLOUT | EPOLLWRNORM;
1361 static int write_pool(const char __user *ubuf, size_t count)
1365 u8 block[BLAKE2S_BLOCK_SIZE];
1368 len = min(count, sizeof(block));
1369 if (copy_from_user(block, ubuf, len)) {
1375 mix_pool_bytes(block, len);
1380 memzero_explicit(block, sizeof(block));
1384 static ssize_t random_write(struct file *file, const char __user *buffer,
1385 size_t count, loff_t *ppos)
1389 ret = write_pool(buffer, count);
1393 return (ssize_t)count;
1396 static ssize_t urandom_read(struct file *file, char __user *buf, size_t nbytes,
1399 static int maxwarn = 10;
1401 if (!crng_ready() && maxwarn > 0) {
1403 if (__ratelimit(&urandom_warning))
1404 pr_notice("%s: uninitialized urandom read (%zd bytes read)\n",
1405 current->comm, nbytes);
1408 return get_random_bytes_user(buf, nbytes);
1411 static ssize_t random_read(struct file *file, char __user *buf, size_t nbytes,
1416 ret = wait_for_random_bytes();
1419 return get_random_bytes_user(buf, nbytes);
1422 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1424 int size, ent_count;
1425 int __user *p = (int __user *)arg;
1430 /* Inherently racy, no point locking. */
1431 if (put_user(input_pool.init_bits, p))
1434 case RNDADDTOENTCNT:
1435 if (!capable(CAP_SYS_ADMIN))
1437 if (get_user(ent_count, p))
1441 credit_init_bits(ent_count);
1444 if (!capable(CAP_SYS_ADMIN))
1446 if (get_user(ent_count, p++))
1450 if (get_user(size, p++))
1452 retval = write_pool((const char __user *)p, size);
1455 credit_init_bits(ent_count);
1459 /* No longer has any effect. */
1460 if (!capable(CAP_SYS_ADMIN))
1464 if (!capable(CAP_SYS_ADMIN))
1475 static int random_fasync(int fd, struct file *filp, int on)
1477 return fasync_helper(fd, filp, on, &fasync);
1480 const struct file_operations random_fops = {
1481 .read = random_read,
1482 .write = random_write,
1483 .poll = random_poll,
1484 .unlocked_ioctl = random_ioctl,
1485 .compat_ioctl = compat_ptr_ioctl,
1486 .fasync = random_fasync,
1487 .llseek = noop_llseek,
1490 const struct file_operations urandom_fops = {
1491 .read = urandom_read,
1492 .write = random_write,
1493 .unlocked_ioctl = random_ioctl,
1494 .compat_ioctl = compat_ptr_ioctl,
1495 .fasync = random_fasync,
1496 .llseek = noop_llseek,
1500 /********************************************************************
1504 * These are partly unused legacy knobs with dummy values to not break
1505 * userspace and partly still useful things. They are usually accessible
1506 * in /proc/sys/kernel/random/ and are as follows:
1508 * - boot_id - a UUID representing the current boot.
1510 * - uuid - a random UUID, different each time the file is read.
1512 * - poolsize - the number of bits of entropy that the input pool can
1513 * hold, tied to the POOL_BITS constant.
1515 * - entropy_avail - the number of bits of entropy currently in the
1516 * input pool. Always <= poolsize.
1518 * - write_wakeup_threshold - the amount of entropy in the input pool
1519 * below which write polls to /dev/random will unblock, requesting
1520 * more entropy, tied to the POOL_INIT_BITS constant. It is writable
1521 * to avoid breaking old userspaces, but writing to it does not
1522 * change any behavior of the RNG.
1524 * - urandom_min_reseed_secs - fixed to the value CRNG_RESEED_INTERVAL.
1525 * It is writable to avoid breaking old userspaces, but writing
1526 * to it does not change any behavior of the RNG.
1528 ********************************************************************/
1530 #ifdef CONFIG_SYSCTL
1532 #include <linux/sysctl.h>
1534 static int sysctl_random_min_urandom_seed = CRNG_RESEED_INTERVAL / HZ;
1535 static int sysctl_random_write_wakeup_bits = POOL_INIT_BITS;
1536 static int sysctl_poolsize = POOL_BITS;
1537 static u8 sysctl_bootid[UUID_SIZE];
1540 * This function is used to return both the bootid UUID, and random
1541 * UUID. The difference is in whether table->data is NULL; if it is,
1542 * then a new UUID is generated and returned to the user.
1544 static int proc_do_uuid(struct ctl_table *table, int write, void *buffer,
1545 size_t *lenp, loff_t *ppos)
1547 u8 tmp_uuid[UUID_SIZE], *uuid;
1548 char uuid_string[UUID_STRING_LEN + 1];
1549 struct ctl_table fake_table = {
1550 .data = uuid_string,
1551 .maxlen = UUID_STRING_LEN
1560 generate_random_uuid(uuid);
1562 static DEFINE_SPINLOCK(bootid_spinlock);
1564 spin_lock(&bootid_spinlock);
1566 generate_random_uuid(uuid);
1567 spin_unlock(&bootid_spinlock);
1570 snprintf(uuid_string, sizeof(uuid_string), "%pU", uuid);
1571 return proc_dostring(&fake_table, 0, buffer, lenp, ppos);
1574 /* The same as proc_dointvec, but writes don't change anything. */
1575 static int proc_do_rointvec(struct ctl_table *table, int write, void *buffer,
1576 size_t *lenp, loff_t *ppos)
1578 return write ? 0 : proc_dointvec(table, 0, buffer, lenp, ppos);
1581 extern struct ctl_table random_table[];
1582 struct ctl_table random_table[] = {
1584 .procname = "poolsize",
1585 .data = &sysctl_poolsize,
1586 .maxlen = sizeof(int),
1588 .proc_handler = proc_dointvec,
1591 .procname = "entropy_avail",
1592 .data = &input_pool.init_bits,
1593 .maxlen = sizeof(int),
1595 .proc_handler = proc_dointvec,
1598 .procname = "write_wakeup_threshold",
1599 .data = &sysctl_random_write_wakeup_bits,
1600 .maxlen = sizeof(int),
1602 .proc_handler = proc_do_rointvec,
1605 .procname = "urandom_min_reseed_secs",
1606 .data = &sysctl_random_min_urandom_seed,
1607 .maxlen = sizeof(int),
1609 .proc_handler = proc_do_rointvec,
1612 .procname = "boot_id",
1613 .data = &sysctl_bootid,
1615 .proc_handler = proc_do_uuid,
1620 .proc_handler = proc_do_uuid,
1624 #endif /* CONFIG_SYSCTL */