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 <linux/uio.h>
56 #include <crypto/chacha.h>
57 #include <crypto/blake2s.h>
58 #include <asm/processor.h>
60 #include <asm/irq_regs.h>
63 /*********************************************************************
65 * Initialization and readiness waiting.
67 * Much of the RNG infrastructure is devoted to various dependencies
68 * being able to wait until the RNG has collected enough entropy and
69 * is ready for safe consumption.
71 *********************************************************************/
74 * crng_init is protected by base_crng->lock, and only increases
75 * its value (from empty->early->ready).
78 CRNG_EMPTY = 0, /* Little to no entropy collected */
79 CRNG_EARLY = 1, /* At least POOL_EARLY_BITS collected */
80 CRNG_READY = 2 /* Fully initialized with POOL_READY_BITS collected */
81 } crng_init __read_mostly = CRNG_EMPTY;
82 static DEFINE_STATIC_KEY_FALSE(crng_is_ready);
83 #define crng_ready() (static_branch_likely(&crng_is_ready) || crng_init >= CRNG_READY)
84 /* Various types of waiters for crng_init->CRNG_READY transition. */
85 static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
86 static struct fasync_struct *fasync;
87 static DEFINE_SPINLOCK(random_ready_chain_lock);
88 static RAW_NOTIFIER_HEAD(random_ready_chain);
90 /* Control how we warn userspace. */
91 static struct ratelimit_state urandom_warning =
92 RATELIMIT_STATE_INIT("warn_urandom_randomness", HZ, 3);
93 static int ratelimit_disable __read_mostly =
94 IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM);
95 module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
96 MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
99 * Returns whether or not the input pool has been seeded and thus guaranteed
100 * to supply cryptographically secure random numbers. This applies to: the
101 * /dev/urandom device, the get_random_bytes function, and the get_random_{u32,
102 * ,u64,int,long} family of functions.
104 * Returns: true if the input pool has been seeded.
105 * false if the input pool has not been seeded.
107 bool rng_is_initialized(void)
111 EXPORT_SYMBOL(rng_is_initialized);
113 static void __cold crng_set_ready(struct work_struct *work)
115 static_branch_enable(&crng_is_ready);
118 /* Used by wait_for_random_bytes(), and considered an entropy collector, below. */
119 static void try_to_generate_entropy(void);
122 * Wait for the input pool to be seeded and thus guaranteed to supply
123 * cryptographically secure random numbers. This applies to: the /dev/urandom
124 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
125 * family of functions. Using any of these functions without first calling
126 * this function forfeits the guarantee of security.
128 * Returns: 0 if the input pool has been seeded.
129 * -ERESTARTSYS if the function was interrupted by a signal.
131 int wait_for_random_bytes(void)
133 while (!crng_ready()) {
136 try_to_generate_entropy();
137 ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ);
139 return ret > 0 ? 0 : ret;
143 EXPORT_SYMBOL(wait_for_random_bytes);
146 * Add a callback function that will be invoked when the input
147 * pool is initialised.
149 * returns: 0 if callback is successfully added
150 * -EALREADY if pool is already initialised (callback not called)
152 int __cold register_random_ready_notifier(struct notifier_block *nb)
160 spin_lock_irqsave(&random_ready_chain_lock, flags);
162 ret = raw_notifier_chain_register(&random_ready_chain, nb);
163 spin_unlock_irqrestore(&random_ready_chain_lock, flags);
168 * Delete a previously registered readiness callback function.
170 int __cold unregister_random_ready_notifier(struct notifier_block *nb)
175 spin_lock_irqsave(&random_ready_chain_lock, flags);
176 ret = raw_notifier_chain_unregister(&random_ready_chain, nb);
177 spin_unlock_irqrestore(&random_ready_chain_lock, flags);
181 static void __cold process_random_ready_list(void)
185 spin_lock_irqsave(&random_ready_chain_lock, flags);
186 raw_notifier_call_chain(&random_ready_chain, 0, NULL);
187 spin_unlock_irqrestore(&random_ready_chain_lock, flags);
190 #define warn_unseeded_randomness() \
191 if (IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM) && !crng_ready()) \
192 printk_deferred(KERN_NOTICE "random: %s called from %pS with crng_init=%d\n", \
193 __func__, (void *)_RET_IP_, crng_init)
196 /*********************************************************************
198 * Fast key erasure RNG, the "crng".
200 * These functions expand entropy from the entropy extractor into
201 * long streams for external consumption using the "fast key erasure"
202 * RNG described at <https://blog.cr.yp.to/20170723-random.html>.
204 * There are a few exported interfaces for use by other drivers:
206 * void get_random_bytes(void *buf, size_t len)
207 * u32 get_random_u32()
208 * u64 get_random_u64()
209 * unsigned int get_random_int()
210 * unsigned long get_random_long()
212 * These interfaces will return the requested number of random bytes
213 * into the given buffer or as a return value. This is equivalent to
214 * a read from /dev/urandom. The u32, u64, int, and long family of
215 * functions may be higher performance for one-off random integers,
216 * because they do a bit of buffering and do not invoke reseeding
217 * until the buffer is emptied.
219 *********************************************************************/
222 CRNG_RESEED_START_INTERVAL = HZ,
223 CRNG_RESEED_INTERVAL = 60 * HZ
227 u8 key[CHACHA_KEY_SIZE] __aligned(__alignof__(long));
229 unsigned long generation;
232 .lock = __SPIN_LOCK_UNLOCKED(base_crng.lock)
236 u8 key[CHACHA_KEY_SIZE];
237 unsigned long generation;
241 static DEFINE_PER_CPU(struct crng, crngs) = {
242 .generation = ULONG_MAX,
243 .lock = INIT_LOCAL_LOCK(crngs.lock),
246 /* Used by crng_reseed() and crng_make_state() to extract a new seed from the input pool. */
247 static void extract_entropy(void *buf, size_t len);
249 /* This extracts a new crng key from the input pool. */
250 static void crng_reseed(void)
253 unsigned long next_gen;
254 u8 key[CHACHA_KEY_SIZE];
256 extract_entropy(key, sizeof(key));
259 * We copy the new key into the base_crng, overwriting the old one,
260 * and update the generation counter. We avoid hitting ULONG_MAX,
261 * because the per-cpu crngs are initialized to ULONG_MAX, so this
262 * forces new CPUs that come online to always initialize.
264 spin_lock_irqsave(&base_crng.lock, flags);
265 memcpy(base_crng.key, key, sizeof(base_crng.key));
266 next_gen = base_crng.generation + 1;
267 if (next_gen == ULONG_MAX)
269 WRITE_ONCE(base_crng.generation, next_gen);
270 WRITE_ONCE(base_crng.birth, jiffies);
271 if (!static_branch_likely(&crng_is_ready))
272 crng_init = CRNG_READY;
273 spin_unlock_irqrestore(&base_crng.lock, flags);
274 memzero_explicit(key, sizeof(key));
278 * This generates a ChaCha block using the provided key, and then
279 * immediately overwites that key with half the block. It returns
280 * the resultant ChaCha state to the user, along with the second
281 * half of the block containing 32 bytes of random data that may
282 * be used; random_data_len may not be greater than 32.
284 * The returned ChaCha state contains within it a copy of the old
285 * key value, at index 4, so the state should always be zeroed out
286 * immediately after using in order to maintain forward secrecy.
287 * If the state cannot be erased in a timely manner, then it is
288 * safer to set the random_data parameter to &chacha_state[4] so
289 * that this function overwrites it before returning.
291 static void crng_fast_key_erasure(u8 key[CHACHA_KEY_SIZE],
292 u32 chacha_state[CHACHA_STATE_WORDS],
293 u8 *random_data, size_t random_data_len)
295 u8 first_block[CHACHA_BLOCK_SIZE];
297 BUG_ON(random_data_len > 32);
299 chacha_init_consts(chacha_state);
300 memcpy(&chacha_state[4], key, CHACHA_KEY_SIZE);
301 memset(&chacha_state[12], 0, sizeof(u32) * 4);
302 chacha20_block(chacha_state, first_block);
304 memcpy(key, first_block, CHACHA_KEY_SIZE);
305 memcpy(random_data, first_block + CHACHA_KEY_SIZE, random_data_len);
306 memzero_explicit(first_block, sizeof(first_block));
310 * Return whether the crng seed is considered to be sufficiently old
311 * that a reseeding is needed. This happens if the last reseeding
312 * was CRNG_RESEED_INTERVAL ago, or during early boot, at an interval
313 * proportional to the uptime.
315 static bool crng_has_old_seed(void)
317 static bool early_boot = true;
318 unsigned long interval = CRNG_RESEED_INTERVAL;
320 if (unlikely(READ_ONCE(early_boot))) {
321 time64_t uptime = ktime_get_seconds();
322 if (uptime >= CRNG_RESEED_INTERVAL / HZ * 2)
323 WRITE_ONCE(early_boot, false);
325 interval = max_t(unsigned int, CRNG_RESEED_START_INTERVAL,
326 (unsigned int)uptime / 2 * HZ);
328 return time_is_before_jiffies(READ_ONCE(base_crng.birth) + interval);
332 * This function returns a ChaCha state that you may use for generating
333 * random data. It also returns up to 32 bytes on its own of random data
334 * that may be used; random_data_len may not be greater than 32.
336 static void crng_make_state(u32 chacha_state[CHACHA_STATE_WORDS],
337 u8 *random_data, size_t random_data_len)
342 BUG_ON(random_data_len > 32);
345 * For the fast path, we check whether we're ready, unlocked first, and
346 * then re-check once locked later. In the case where we're really not
347 * ready, we do fast key erasure with the base_crng directly, extracting
348 * when crng_init is CRNG_EMPTY.
353 spin_lock_irqsave(&base_crng.lock, flags);
354 ready = crng_ready();
356 if (crng_init == CRNG_EMPTY)
357 extract_entropy(base_crng.key, sizeof(base_crng.key));
358 crng_fast_key_erasure(base_crng.key, chacha_state,
359 random_data, random_data_len);
361 spin_unlock_irqrestore(&base_crng.lock, flags);
367 * If the base_crng is old enough, we reseed, which in turn bumps the
368 * generation counter that we check below.
370 if (unlikely(crng_has_old_seed()))
373 local_lock_irqsave(&crngs.lock, flags);
374 crng = raw_cpu_ptr(&crngs);
377 * If our per-cpu crng is older than the base_crng, then it means
378 * somebody reseeded the base_crng. In that case, we do fast key
379 * erasure on the base_crng, and use its output as the new key
380 * for our per-cpu crng. This brings us up to date with base_crng.
382 if (unlikely(crng->generation != READ_ONCE(base_crng.generation))) {
383 spin_lock(&base_crng.lock);
384 crng_fast_key_erasure(base_crng.key, chacha_state,
385 crng->key, sizeof(crng->key));
386 crng->generation = base_crng.generation;
387 spin_unlock(&base_crng.lock);
391 * Finally, when we've made it this far, our per-cpu crng has an up
392 * to date key, and we can do fast key erasure with it to produce
393 * some random data and a ChaCha state for the caller. All other
394 * branches of this function are "unlikely", so most of the time we
395 * should wind up here immediately.
397 crng_fast_key_erasure(crng->key, chacha_state, random_data, random_data_len);
398 local_unlock_irqrestore(&crngs.lock, flags);
401 static void _get_random_bytes(void *buf, size_t len)
403 u32 chacha_state[CHACHA_STATE_WORDS];
404 u8 tmp[CHACHA_BLOCK_SIZE];
405 size_t first_block_len;
410 first_block_len = min_t(size_t, 32, len);
411 crng_make_state(chacha_state, buf, first_block_len);
412 len -= first_block_len;
413 buf += first_block_len;
416 if (len < CHACHA_BLOCK_SIZE) {
417 chacha20_block(chacha_state, tmp);
418 memcpy(buf, tmp, len);
419 memzero_explicit(tmp, sizeof(tmp));
423 chacha20_block(chacha_state, buf);
424 if (unlikely(chacha_state[12] == 0))
426 len -= CHACHA_BLOCK_SIZE;
427 buf += CHACHA_BLOCK_SIZE;
430 memzero_explicit(chacha_state, sizeof(chacha_state));
434 * This function is the exported kernel interface. It returns some
435 * number of good random numbers, suitable for key generation, seeding
436 * TCP sequence numbers, etc. It does not rely on the hardware random
437 * number generator. For random bytes direct from the hardware RNG
438 * (when available), use get_random_bytes_arch(). In order to ensure
439 * that the randomness provided by this function is okay, the function
440 * wait_for_random_bytes() should be called and return 0 at least once
441 * at any point prior.
443 void get_random_bytes(void *buf, size_t len)
445 warn_unseeded_randomness();
446 _get_random_bytes(buf, len);
448 EXPORT_SYMBOL(get_random_bytes);
450 static ssize_t get_random_bytes_user(struct iov_iter *iter)
452 u32 chacha_state[CHACHA_STATE_WORDS];
453 u8 block[CHACHA_BLOCK_SIZE];
454 size_t ret = 0, copied;
456 if (unlikely(!iov_iter_count(iter)))
460 * Immediately overwrite the ChaCha key at index 4 with random
461 * bytes, in case userspace causes copy_to_user() below to sleep
462 * forever, so that we still retain forward secrecy in that case.
464 crng_make_state(chacha_state, (u8 *)&chacha_state[4], CHACHA_KEY_SIZE);
466 * However, if we're doing a read of len <= 32, we don't need to
467 * use chacha_state after, so we can simply return those bytes to
470 if (iov_iter_count(iter) <= CHACHA_KEY_SIZE) {
471 ret = copy_to_iter(&chacha_state[4], CHACHA_KEY_SIZE, iter);
472 goto out_zero_chacha;
476 chacha20_block(chacha_state, block);
477 if (unlikely(chacha_state[12] == 0))
480 copied = copy_to_iter(block, sizeof(block), iter);
482 if (!iov_iter_count(iter) || copied != sizeof(block))
485 BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0);
486 if (ret % PAGE_SIZE == 0) {
487 if (signal_pending(current))
493 memzero_explicit(block, sizeof(block));
495 memzero_explicit(chacha_state, sizeof(chacha_state));
496 return ret ? ret : -EFAULT;
500 * Batched entropy returns random integers. The quality of the random
501 * number is good as /dev/urandom. In order to ensure that the randomness
502 * provided by this function is okay, the function wait_for_random_bytes()
503 * should be called and return 0 at least once at any point prior.
506 #define DEFINE_BATCHED_ENTROPY(type) \
507 struct batch_ ##type { \
509 * We make this 1.5x a ChaCha block, so that we get the \
510 * remaining 32 bytes from fast key erasure, plus one full \
511 * block from the detached ChaCha state. We can increase \
512 * the size of this later if needed so long as we keep the \
513 * formula of (integer_blocks + 0.5) * CHACHA_BLOCK_SIZE. \
515 type entropy[CHACHA_BLOCK_SIZE * 3 / (2 * sizeof(type))]; \
517 unsigned long generation; \
518 unsigned int position; \
521 static DEFINE_PER_CPU(struct batch_ ##type, batched_entropy_ ##type) = { \
522 .lock = INIT_LOCAL_LOCK(batched_entropy_ ##type.lock), \
523 .position = UINT_MAX \
526 type get_random_ ##type(void) \
529 unsigned long flags; \
530 struct batch_ ##type *batch; \
531 unsigned long next_gen; \
533 warn_unseeded_randomness(); \
535 if (!crng_ready()) { \
536 _get_random_bytes(&ret, sizeof(ret)); \
540 local_lock_irqsave(&batched_entropy_ ##type.lock, flags); \
541 batch = raw_cpu_ptr(&batched_entropy_##type); \
543 next_gen = READ_ONCE(base_crng.generation); \
544 if (batch->position >= ARRAY_SIZE(batch->entropy) || \
545 next_gen != batch->generation) { \
546 _get_random_bytes(batch->entropy, sizeof(batch->entropy)); \
547 batch->position = 0; \
548 batch->generation = next_gen; \
551 ret = batch->entropy[batch->position]; \
552 batch->entropy[batch->position] = 0; \
554 local_unlock_irqrestore(&batched_entropy_ ##type.lock, flags); \
557 EXPORT_SYMBOL(get_random_ ##type);
559 DEFINE_BATCHED_ENTROPY(u64)
560 DEFINE_BATCHED_ENTROPY(u32)
564 * This function is called when the CPU is coming up, with entry
565 * CPUHP_RANDOM_PREPARE, which comes before CPUHP_WORKQUEUE_PREP.
567 int __cold random_prepare_cpu(unsigned int cpu)
570 * When the cpu comes back online, immediately invalidate both
571 * the per-cpu crng and all batches, so that we serve fresh
574 per_cpu_ptr(&crngs, cpu)->generation = ULONG_MAX;
575 per_cpu_ptr(&batched_entropy_u32, cpu)->position = UINT_MAX;
576 per_cpu_ptr(&batched_entropy_u64, cpu)->position = UINT_MAX;
582 * This function will use the architecture-specific hardware random
583 * number generator if it is available. It is not recommended for
584 * use. Use get_random_bytes() instead. It returns the number of
587 size_t __must_check get_random_bytes_arch(void *buf, size_t len)
594 size_t block_len = min_t(size_t, left, sizeof(unsigned long));
596 if (!arch_get_random_long(&v))
599 memcpy(p, &v, block_len);
606 EXPORT_SYMBOL(get_random_bytes_arch);
609 /**********************************************************************
611 * Entropy accumulation and extraction routines.
613 * Callers may add entropy via:
615 * static void mix_pool_bytes(const void *buf, size_t len)
617 * After which, if added entropy should be credited:
619 * static void credit_init_bits(size_t bits)
621 * Finally, extract entropy via:
623 * static void extract_entropy(void *buf, size_t len)
625 **********************************************************************/
628 POOL_BITS = BLAKE2S_HASH_SIZE * 8,
629 POOL_READY_BITS = POOL_BITS, /* When crng_init->CRNG_READY */
630 POOL_EARLY_BITS = POOL_READY_BITS / 2 /* When crng_init->CRNG_EARLY */
634 struct blake2s_state hash;
636 unsigned int init_bits;
638 .hash.h = { BLAKE2S_IV0 ^ (0x01010000 | BLAKE2S_HASH_SIZE),
639 BLAKE2S_IV1, BLAKE2S_IV2, BLAKE2S_IV3, BLAKE2S_IV4,
640 BLAKE2S_IV5, BLAKE2S_IV6, BLAKE2S_IV7 },
641 .hash.outlen = BLAKE2S_HASH_SIZE,
642 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
645 static void _mix_pool_bytes(const void *buf, size_t len)
647 blake2s_update(&input_pool.hash, buf, len);
651 * This function adds bytes into the input pool. It does not
652 * update the initialization bit counter; the caller should call
653 * credit_init_bits if this is appropriate.
655 static void mix_pool_bytes(const void *buf, size_t len)
659 spin_lock_irqsave(&input_pool.lock, flags);
660 _mix_pool_bytes(buf, len);
661 spin_unlock_irqrestore(&input_pool.lock, flags);
665 * This is an HKDF-like construction for using the hashed collected entropy
666 * as a PRF key, that's then expanded block-by-block.
668 static void extract_entropy(void *buf, size_t len)
671 u8 seed[BLAKE2S_HASH_SIZE], next_key[BLAKE2S_HASH_SIZE];
673 unsigned long rdseed[32 / sizeof(long)];
678 for (i = 0; i < ARRAY_SIZE(block.rdseed); ++i) {
679 if (!arch_get_random_seed_long(&block.rdseed[i]) &&
680 !arch_get_random_long(&block.rdseed[i]))
681 block.rdseed[i] = random_get_entropy();
684 spin_lock_irqsave(&input_pool.lock, flags);
686 /* seed = HASHPRF(last_key, entropy_input) */
687 blake2s_final(&input_pool.hash, seed);
689 /* next_key = HASHPRF(seed, RDSEED || 0) */
691 blake2s(next_key, (u8 *)&block, seed, sizeof(next_key), sizeof(block), sizeof(seed));
692 blake2s_init_key(&input_pool.hash, BLAKE2S_HASH_SIZE, next_key, sizeof(next_key));
694 spin_unlock_irqrestore(&input_pool.lock, flags);
695 memzero_explicit(next_key, sizeof(next_key));
698 i = min_t(size_t, len, BLAKE2S_HASH_SIZE);
699 /* output = HASHPRF(seed, RDSEED || ++counter) */
701 blake2s(buf, (u8 *)&block, seed, i, sizeof(block), sizeof(seed));
706 memzero_explicit(seed, sizeof(seed));
707 memzero_explicit(&block, sizeof(block));
710 #define credit_init_bits(bits) if (!crng_ready()) _credit_init_bits(bits)
712 static void __cold _credit_init_bits(size_t bits)
714 static struct execute_work set_ready;
715 unsigned int new, orig, add;
721 add = min_t(size_t, bits, POOL_BITS);
724 orig = READ_ONCE(input_pool.init_bits);
725 new = min_t(unsigned int, POOL_BITS, orig + add);
726 } while (cmpxchg(&input_pool.init_bits, orig, new) != orig);
728 if (orig < POOL_READY_BITS && new >= POOL_READY_BITS) {
729 crng_reseed(); /* Sets crng_init to CRNG_READY under base_crng.lock. */
730 if (static_key_initialized)
731 execute_in_process_context(crng_set_ready, &set_ready);
732 process_random_ready_list();
733 wake_up_interruptible(&crng_init_wait);
734 kill_fasync(&fasync, SIGIO, POLL_IN);
735 pr_notice("crng init done\n");
736 if (urandom_warning.missed)
737 pr_notice("%d urandom warning(s) missed due to ratelimiting\n",
738 urandom_warning.missed);
739 } else if (orig < POOL_EARLY_BITS && new >= POOL_EARLY_BITS) {
740 spin_lock_irqsave(&base_crng.lock, flags);
741 /* Check if crng_init is CRNG_EMPTY, to avoid race with crng_reseed(). */
742 if (crng_init == CRNG_EMPTY) {
743 extract_entropy(base_crng.key, sizeof(base_crng.key));
744 crng_init = CRNG_EARLY;
746 spin_unlock_irqrestore(&base_crng.lock, flags);
751 /**********************************************************************
753 * Entropy collection routines.
755 * The following exported functions are used for pushing entropy into
756 * the above entropy accumulation routines:
758 * void add_device_randomness(const void *buf, size_t len);
759 * void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy);
760 * void add_bootloader_randomness(const void *buf, size_t len);
761 * void add_interrupt_randomness(int irq);
762 * void add_input_randomness(unsigned int type, unsigned int code, unsigned int value);
763 * void add_disk_randomness(struct gendisk *disk);
765 * add_device_randomness() adds data to the input pool that
766 * is likely to differ between two devices (or possibly even per boot).
767 * This would be things like MAC addresses or serial numbers, or the
768 * read-out of the RTC. This does *not* credit any actual entropy to
769 * the pool, but it initializes the pool to different values for devices
770 * that might otherwise be identical and have very little entropy
771 * available to them (particularly common in the embedded world).
773 * add_hwgenerator_randomness() is for true hardware RNGs, and will credit
774 * entropy as specified by the caller. If the entropy pool is full it will
775 * block until more entropy is needed.
777 * add_bootloader_randomness() is called by bootloader drivers, such as EFI
778 * and device tree, and credits its input depending on whether or not the
779 * configuration option CONFIG_RANDOM_TRUST_BOOTLOADER is set.
781 * add_interrupt_randomness() uses the interrupt timing as random
782 * inputs to the entropy pool. Using the cycle counters and the irq source
783 * as inputs, it feeds the input pool roughly once a second or after 64
784 * interrupts, crediting 1 bit of entropy for whichever comes first.
786 * add_input_randomness() uses the input layer interrupt timing, as well
787 * as the event type information from the hardware.
789 * add_disk_randomness() uses what amounts to the seek time of block
790 * layer request events, on a per-disk_devt basis, as input to the
791 * entropy pool. Note that high-speed solid state drives with very low
792 * seek times do not make for good sources of entropy, as their seek
793 * times are usually fairly consistent.
795 * The last two routines try to estimate how many bits of entropy
796 * to credit. They do this by keeping track of the first and second
797 * order deltas of the event timings.
799 **********************************************************************/
801 static bool trust_cpu __ro_after_init = IS_ENABLED(CONFIG_RANDOM_TRUST_CPU);
802 static bool trust_bootloader __ro_after_init = IS_ENABLED(CONFIG_RANDOM_TRUST_BOOTLOADER);
803 static int __init parse_trust_cpu(char *arg)
805 return kstrtobool(arg, &trust_cpu);
807 static int __init parse_trust_bootloader(char *arg)
809 return kstrtobool(arg, &trust_bootloader);
811 early_param("random.trust_cpu", parse_trust_cpu);
812 early_param("random.trust_bootloader", parse_trust_bootloader);
815 * The first collection of entropy occurs at system boot while interrupts
816 * are still turned off. Here we push in latent entropy, RDSEED, a timestamp,
817 * utsname(), and the command line. Depending on the above configuration knob,
818 * RDSEED may be considered sufficient for initialization. Note that much
819 * earlier setup may already have pushed entropy into the input pool by the
822 int __init random_init(const char *command_line)
824 ktime_t now = ktime_get_real();
825 unsigned int i, arch_bytes;
826 unsigned long entropy;
828 #if defined(LATENT_ENTROPY_PLUGIN)
829 static const u8 compiletime_seed[BLAKE2S_BLOCK_SIZE] __initconst __latent_entropy;
830 _mix_pool_bytes(compiletime_seed, sizeof(compiletime_seed));
833 for (i = 0, arch_bytes = BLAKE2S_BLOCK_SIZE;
834 i < BLAKE2S_BLOCK_SIZE; i += sizeof(entropy)) {
835 if (!arch_get_random_seed_long_early(&entropy) &&
836 !arch_get_random_long_early(&entropy)) {
837 entropy = random_get_entropy();
838 arch_bytes -= sizeof(entropy);
840 _mix_pool_bytes(&entropy, sizeof(entropy));
842 _mix_pool_bytes(&now, sizeof(now));
843 _mix_pool_bytes(utsname(), sizeof(*(utsname())));
844 _mix_pool_bytes(command_line, strlen(command_line));
845 add_latent_entropy();
848 * If we were initialized by the bootloader before jump labels are
849 * initialized, then we should enable the static branch here, where
850 * it's guaranteed that jump labels have been initialized.
852 if (!static_branch_likely(&crng_is_ready) && crng_init >= CRNG_READY)
853 crng_set_ready(NULL);
858 credit_init_bits(arch_bytes * 8);
864 * Add device- or boot-specific data to the input pool to help
867 * None of this adds any entropy; it is meant to avoid the problem of
868 * the entropy pool having similar initial state across largely
871 void add_device_randomness(const void *buf, size_t len)
873 unsigned long entropy = random_get_entropy();
876 spin_lock_irqsave(&input_pool.lock, flags);
877 _mix_pool_bytes(&entropy, sizeof(entropy));
878 _mix_pool_bytes(buf, len);
879 spin_unlock_irqrestore(&input_pool.lock, flags);
881 EXPORT_SYMBOL(add_device_randomness);
884 * Interface for in-kernel drivers of true hardware RNGs.
885 * Those devices may produce endless random bits and will be throttled
886 * when our pool is full.
888 void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy)
890 mix_pool_bytes(buf, len);
891 credit_init_bits(entropy);
894 * Throttle writing to once every CRNG_RESEED_INTERVAL, unless
895 * we're not yet initialized.
897 if (!kthread_should_stop() && crng_ready())
898 schedule_timeout_interruptible(CRNG_RESEED_INTERVAL);
900 EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
903 * Handle random seed passed by bootloader, and credit it if
904 * CONFIG_RANDOM_TRUST_BOOTLOADER is set.
906 void __cold add_bootloader_randomness(const void *buf, size_t len)
908 mix_pool_bytes(buf, len);
909 if (trust_bootloader)
910 credit_init_bits(len * 8);
912 EXPORT_SYMBOL_GPL(add_bootloader_randomness);
915 struct work_struct mix;
916 unsigned long pool[4];
921 static DEFINE_PER_CPU(struct fast_pool, irq_randomness) = {
923 #define FASTMIX_PERM SIPHASH_PERMUTATION
924 .pool = { SIPHASH_CONST_0, SIPHASH_CONST_1, SIPHASH_CONST_2, SIPHASH_CONST_3 }
926 #define FASTMIX_PERM HSIPHASH_PERMUTATION
927 .pool = { HSIPHASH_CONST_0, HSIPHASH_CONST_1, HSIPHASH_CONST_2, HSIPHASH_CONST_3 }
932 * This is [Half]SipHash-1-x, starting from an empty key. Because
933 * the key is fixed, it assumes that its inputs are non-malicious,
934 * and therefore this has no security on its own. s represents the
935 * four-word SipHash state, while v represents a two-word input.
937 static void fast_mix(unsigned long s[4], unsigned long v1, unsigned long v2)
940 FASTMIX_PERM(s[0], s[1], s[2], s[3]);
943 FASTMIX_PERM(s[0], s[1], s[2], s[3]);
949 * This function is called when the CPU has just come online, with
950 * entry CPUHP_AP_RANDOM_ONLINE, just after CPUHP_AP_WORKQUEUE_ONLINE.
952 int __cold random_online_cpu(unsigned int cpu)
955 * During CPU shutdown and before CPU onlining, add_interrupt_
956 * randomness() may schedule mix_interrupt_randomness(), and
957 * set the MIX_INFLIGHT flag. However, because the worker can
958 * be scheduled on a different CPU during this period, that
959 * flag will never be cleared. For that reason, we zero out
960 * the flag here, which runs just after workqueues are onlined
961 * for the CPU again. This also has the effect of setting the
962 * irq randomness count to zero so that new accumulated irqs
965 per_cpu_ptr(&irq_randomness, cpu)->count = 0;
970 static void mix_interrupt_randomness(struct work_struct *work)
972 struct fast_pool *fast_pool = container_of(work, struct fast_pool, mix);
974 * The size of the copied stack pool is explicitly 2 longs so that we
975 * only ever ingest half of the siphash output each time, retaining
976 * the other half as the next "key" that carries over. The entropy is
977 * supposed to be sufficiently dispersed between bits so on average
978 * we don't wind up "losing" some.
980 unsigned long pool[2];
983 /* Check to see if we're running on the wrong CPU due to hotplug. */
985 if (fast_pool != this_cpu_ptr(&irq_randomness)) {
991 * Copy the pool to the stack so that the mixer always has a
992 * consistent view, before we reenable irqs again.
994 memcpy(pool, fast_pool->pool, sizeof(pool));
995 count = fast_pool->count;
996 fast_pool->count = 0;
997 fast_pool->last = jiffies;
1000 mix_pool_bytes(pool, sizeof(pool));
1001 credit_init_bits(max(1u, (count & U16_MAX) / 64));
1003 memzero_explicit(pool, sizeof(pool));
1006 void add_interrupt_randomness(int irq)
1008 enum { MIX_INFLIGHT = 1U << 31 };
1009 unsigned long entropy = random_get_entropy();
1010 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1011 struct pt_regs *regs = get_irq_regs();
1012 unsigned int new_count;
1014 fast_mix(fast_pool->pool, entropy,
1015 (regs ? instruction_pointer(regs) : _RET_IP_) ^ swab(irq));
1016 new_count = ++fast_pool->count;
1018 if (new_count & MIX_INFLIGHT)
1021 if (new_count < 64 && !time_is_before_jiffies(fast_pool->last + HZ))
1024 if (unlikely(!fast_pool->mix.func))
1025 INIT_WORK(&fast_pool->mix, mix_interrupt_randomness);
1026 fast_pool->count |= MIX_INFLIGHT;
1027 queue_work_on(raw_smp_processor_id(), system_highpri_wq, &fast_pool->mix);
1029 EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1031 /* There is one of these per entropy source */
1032 struct timer_rand_state {
1033 unsigned long last_time;
1034 long last_delta, last_delta2;
1038 * This function adds entropy to the entropy "pool" by using timing
1039 * delays. It uses the timer_rand_state structure to make an estimate
1040 * of how many bits of entropy this call has added to the pool. The
1041 * value "num" is also added to the pool; it should somehow describe
1042 * the type of event that just happened.
1044 static void add_timer_randomness(struct timer_rand_state *state, unsigned int num)
1046 unsigned long entropy = random_get_entropy(), now = jiffies, flags;
1047 long delta, delta2, delta3;
1051 * If we're in a hard IRQ, add_interrupt_randomness() will be called
1052 * sometime after, so mix into the fast pool.
1055 fast_mix(this_cpu_ptr(&irq_randomness)->pool, entropy, num);
1057 spin_lock_irqsave(&input_pool.lock, flags);
1058 _mix_pool_bytes(&entropy, sizeof(entropy));
1059 _mix_pool_bytes(&num, sizeof(num));
1060 spin_unlock_irqrestore(&input_pool.lock, flags);
1067 * Calculate number of bits of randomness we probably added.
1068 * We take into account the first, second and third-order deltas
1069 * in order to make our estimate.
1071 delta = now - READ_ONCE(state->last_time);
1072 WRITE_ONCE(state->last_time, now);
1074 delta2 = delta - READ_ONCE(state->last_delta);
1075 WRITE_ONCE(state->last_delta, delta);
1077 delta3 = delta2 - READ_ONCE(state->last_delta2);
1078 WRITE_ONCE(state->last_delta2, delta2);
1092 * delta is now minimum absolute delta. Round down by 1 bit
1093 * on general principles, and limit entropy estimate to 11 bits.
1095 bits = min(fls(delta >> 1), 11);
1098 * As mentioned above, if we're in a hard IRQ, add_interrupt_randomness()
1099 * will run after this, which uses a different crediting scheme of 1 bit
1100 * per every 64 interrupts. In order to let that function do accounting
1101 * close to the one in this function, we credit a full 64/64 bit per bit,
1102 * and then subtract one to account for the extra one added.
1105 this_cpu_ptr(&irq_randomness)->count += max(1u, bits * 64) - 1;
1107 _credit_init_bits(bits);
1110 void add_input_randomness(unsigned int type, unsigned int code, unsigned int value)
1112 static unsigned char last_value;
1113 static struct timer_rand_state input_timer_state = { INITIAL_JIFFIES };
1115 /* Ignore autorepeat and the like. */
1116 if (value == last_value)
1120 add_timer_randomness(&input_timer_state,
1121 (type << 4) ^ code ^ (code >> 4) ^ value);
1123 EXPORT_SYMBOL_GPL(add_input_randomness);
1126 void add_disk_randomness(struct gendisk *disk)
1128 if (!disk || !disk->random)
1130 /* First major is 1, so we get >= 0x200 here. */
1131 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1133 EXPORT_SYMBOL_GPL(add_disk_randomness);
1135 void __cold rand_initialize_disk(struct gendisk *disk)
1137 struct timer_rand_state *state;
1140 * If kzalloc returns null, we just won't use that entropy
1143 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1145 state->last_time = INITIAL_JIFFIES;
1146 disk->random = state;
1152 * Each time the timer fires, we expect that we got an unpredictable
1153 * jump in the cycle counter. Even if the timer is running on another
1154 * CPU, the timer activity will be touching the stack of the CPU that is
1155 * generating entropy..
1157 * Note that we don't re-arm the timer in the timer itself - we are
1158 * happy to be scheduled away, since that just makes the load more
1159 * complex, but we do not want the timer to keep ticking unless the
1160 * entropy loop is running.
1162 * So the re-arming always happens in the entropy loop itself.
1164 static void __cold entropy_timer(struct timer_list *t)
1166 credit_init_bits(1);
1170 * If we have an actual cycle counter, see if we can
1171 * generate enough entropy with timing noise
1173 static void __cold try_to_generate_entropy(void)
1176 unsigned long entropy;
1177 struct timer_list timer;
1180 stack.entropy = random_get_entropy();
1182 /* Slow counter - or none. Don't even bother */
1183 if (stack.entropy == random_get_entropy())
1186 timer_setup_on_stack(&stack.timer, entropy_timer, 0);
1187 while (!crng_ready() && !signal_pending(current)) {
1188 if (!timer_pending(&stack.timer))
1189 mod_timer(&stack.timer, jiffies + 1);
1190 mix_pool_bytes(&stack.entropy, sizeof(stack.entropy));
1192 stack.entropy = random_get_entropy();
1195 del_timer_sync(&stack.timer);
1196 destroy_timer_on_stack(&stack.timer);
1197 mix_pool_bytes(&stack.entropy, sizeof(stack.entropy));
1201 /**********************************************************************
1203 * Userspace reader/writer interfaces.
1205 * getrandom(2) is the primary modern interface into the RNG and should
1206 * be used in preference to anything else.
1208 * Reading from /dev/random has the same functionality as calling
1209 * getrandom(2) with flags=0. In earlier versions, however, it had
1210 * vastly different semantics and should therefore be avoided, to
1211 * prevent backwards compatibility issues.
1213 * Reading from /dev/urandom has the same functionality as calling
1214 * getrandom(2) with flags=GRND_INSECURE. Because it does not block
1215 * waiting for the RNG to be ready, it should not be used.
1217 * Writing to either /dev/random or /dev/urandom adds entropy to
1218 * the input pool but does not credit it.
1220 * Polling on /dev/random indicates when the RNG is initialized, on
1221 * the read side, and when it wants new entropy, on the write side.
1223 * Both /dev/random and /dev/urandom have the same set of ioctls for
1224 * adding entropy, getting the entropy count, zeroing the count, and
1225 * reseeding the crng.
1227 **********************************************************************/
1229 SYSCALL_DEFINE3(getrandom, char __user *, ubuf, size_t, len, unsigned int, flags)
1231 struct iov_iter iter;
1235 if (flags & ~(GRND_NONBLOCK | GRND_RANDOM | GRND_INSECURE))
1239 * Requesting insecure and blocking randomness at the same time makes
1242 if ((flags & (GRND_INSECURE | GRND_RANDOM)) == (GRND_INSECURE | GRND_RANDOM))
1245 if (!crng_ready() && !(flags & GRND_INSECURE)) {
1246 if (flags & GRND_NONBLOCK)
1248 ret = wait_for_random_bytes();
1253 ret = import_single_range(READ, ubuf, len, &iov, &iter);
1256 return get_random_bytes_user(&iter);
1259 static __poll_t random_poll(struct file *file, poll_table *wait)
1261 poll_wait(file, &crng_init_wait, wait);
1262 return crng_ready() ? EPOLLIN | EPOLLRDNORM : EPOLLOUT | EPOLLWRNORM;
1265 static ssize_t write_pool_user(struct iov_iter *iter)
1267 u8 block[BLAKE2S_BLOCK_SIZE];
1271 if (unlikely(!iov_iter_count(iter)))
1275 copied = copy_from_iter(block, sizeof(block), iter);
1277 mix_pool_bytes(block, copied);
1278 if (!iov_iter_count(iter) || copied != sizeof(block))
1281 BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0);
1282 if (ret % PAGE_SIZE == 0) {
1283 if (signal_pending(current))
1289 memzero_explicit(block, sizeof(block));
1290 return ret ? ret : -EFAULT;
1293 static ssize_t random_write_iter(struct kiocb *kiocb, struct iov_iter *iter)
1295 return write_pool_user(iter);
1298 static ssize_t urandom_read_iter(struct kiocb *kiocb, struct iov_iter *iter)
1300 static int maxwarn = 10;
1302 if (!crng_ready()) {
1303 if (!ratelimit_disable && maxwarn <= 0)
1304 ++urandom_warning.missed;
1305 else if (ratelimit_disable || __ratelimit(&urandom_warning)) {
1307 pr_notice("%s: uninitialized urandom read (%zu bytes read)\n",
1308 current->comm, iov_iter_count(iter));
1312 return get_random_bytes_user(iter);
1315 static ssize_t random_read_iter(struct kiocb *kiocb, struct iov_iter *iter)
1319 ret = wait_for_random_bytes();
1322 return get_random_bytes_user(iter);
1325 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1327 int __user *p = (int __user *)arg;
1332 /* Inherently racy, no point locking. */
1333 if (put_user(input_pool.init_bits, p))
1336 case RNDADDTOENTCNT:
1337 if (!capable(CAP_SYS_ADMIN))
1339 if (get_user(ent_count, p))
1343 credit_init_bits(ent_count);
1345 case RNDADDENTROPY: {
1346 struct iov_iter iter;
1351 if (!capable(CAP_SYS_ADMIN))
1353 if (get_user(ent_count, p++))
1357 if (get_user(len, p++))
1359 ret = import_single_range(WRITE, p, len, &iov, &iter);
1362 ret = write_pool_user(&iter);
1363 if (unlikely(ret < 0))
1365 /* Since we're crediting, enforce that it was all written into the pool. */
1366 if (unlikely(ret != len))
1368 credit_init_bits(ent_count);
1373 /* No longer has any effect. */
1374 if (!capable(CAP_SYS_ADMIN))
1378 if (!capable(CAP_SYS_ADMIN))
1389 static int random_fasync(int fd, struct file *filp, int on)
1391 return fasync_helper(fd, filp, on, &fasync);
1394 const struct file_operations random_fops = {
1395 .read_iter = random_read_iter,
1396 .write_iter = random_write_iter,
1397 .poll = random_poll,
1398 .unlocked_ioctl = random_ioctl,
1399 .compat_ioctl = compat_ptr_ioctl,
1400 .fasync = random_fasync,
1401 .llseek = noop_llseek,
1402 .splice_read = generic_file_splice_read,
1403 .splice_write = iter_file_splice_write,
1406 const struct file_operations urandom_fops = {
1407 .read_iter = urandom_read_iter,
1408 .write_iter = random_write_iter,
1409 .unlocked_ioctl = random_ioctl,
1410 .compat_ioctl = compat_ptr_ioctl,
1411 .fasync = random_fasync,
1412 .llseek = noop_llseek,
1413 .splice_read = generic_file_splice_read,
1414 .splice_write = iter_file_splice_write,
1418 /********************************************************************
1422 * These are partly unused legacy knobs with dummy values to not break
1423 * userspace and partly still useful things. They are usually accessible
1424 * in /proc/sys/kernel/random/ and are as follows:
1426 * - boot_id - a UUID representing the current boot.
1428 * - uuid - a random UUID, different each time the file is read.
1430 * - poolsize - the number of bits of entropy that the input pool can
1431 * hold, tied to the POOL_BITS constant.
1433 * - entropy_avail - the number of bits of entropy currently in the
1434 * input pool. Always <= poolsize.
1436 * - write_wakeup_threshold - the amount of entropy in the input pool
1437 * below which write polls to /dev/random will unblock, requesting
1438 * more entropy, tied to the POOL_READY_BITS constant. It is writable
1439 * to avoid breaking old userspaces, but writing to it does not
1440 * change any behavior of the RNG.
1442 * - urandom_min_reseed_secs - fixed to the value CRNG_RESEED_INTERVAL.
1443 * It is writable to avoid breaking old userspaces, but writing
1444 * to it does not change any behavior of the RNG.
1446 ********************************************************************/
1448 #ifdef CONFIG_SYSCTL
1450 #include <linux/sysctl.h>
1452 static int sysctl_random_min_urandom_seed = CRNG_RESEED_INTERVAL / HZ;
1453 static int sysctl_random_write_wakeup_bits = POOL_READY_BITS;
1454 static int sysctl_poolsize = POOL_BITS;
1455 static u8 sysctl_bootid[UUID_SIZE];
1458 * This function is used to return both the bootid UUID, and random
1459 * UUID. The difference is in whether table->data is NULL; if it is,
1460 * then a new UUID is generated and returned to the user.
1462 static int proc_do_uuid(struct ctl_table *table, int write, void *buf,
1463 size_t *lenp, loff_t *ppos)
1465 u8 tmp_uuid[UUID_SIZE], *uuid;
1466 char uuid_string[UUID_STRING_LEN + 1];
1467 struct ctl_table fake_table = {
1468 .data = uuid_string,
1469 .maxlen = UUID_STRING_LEN
1478 generate_random_uuid(uuid);
1480 static DEFINE_SPINLOCK(bootid_spinlock);
1482 spin_lock(&bootid_spinlock);
1484 generate_random_uuid(uuid);
1485 spin_unlock(&bootid_spinlock);
1488 snprintf(uuid_string, sizeof(uuid_string), "%pU", uuid);
1489 return proc_dostring(&fake_table, 0, buf, lenp, ppos);
1492 /* The same as proc_dointvec, but writes don't change anything. */
1493 static int proc_do_rointvec(struct ctl_table *table, int write, void *buf,
1494 size_t *lenp, loff_t *ppos)
1496 return write ? 0 : proc_dointvec(table, 0, buf, lenp, ppos);
1499 extern struct ctl_table random_table[];
1500 struct ctl_table random_table[] = {
1502 .procname = "poolsize",
1503 .data = &sysctl_poolsize,
1504 .maxlen = sizeof(int),
1506 .proc_handler = proc_dointvec,
1509 .procname = "entropy_avail",
1510 .data = &input_pool.init_bits,
1511 .maxlen = sizeof(int),
1513 .proc_handler = proc_dointvec,
1516 .procname = "write_wakeup_threshold",
1517 .data = &sysctl_random_write_wakeup_bits,
1518 .maxlen = sizeof(int),
1520 .proc_handler = proc_do_rointvec,
1523 .procname = "urandom_min_reseed_secs",
1524 .data = &sysctl_random_min_urandom_seed,
1525 .maxlen = sizeof(int),
1527 .proc_handler = proc_do_rointvec,
1530 .procname = "boot_id",
1531 .data = &sysctl_bootid,
1533 .proc_handler = proc_do_uuid,
1538 .proc_handler = proc_do_uuid,
1542 #endif /* CONFIG_SYSCTL */