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/kmemcheck.h>
48 #include <linux/workqueue.h>
49 #include <linux/irq.h>
50 #include <linux/ratelimit.h>
51 #include <linux/syscalls.h>
52 #include <linux/completion.h>
53 #include <linux/uuid.h>
54 #include <linux/siphash.h>
55 #include <linux/uio.h>
56 #include <crypto/chacha20.h>
57 #include <crypto/blake2s.h>
58 #include <asm/processor.h>
59 #include <asm/uaccess.h>
61 #include <asm/irq_regs.h>
64 /*********************************************************************
66 * Initialization and readiness waiting.
68 * Much of the RNG infrastructure is devoted to various dependencies
69 * being able to wait until the RNG has collected enough entropy and
70 * is ready for safe consumption.
72 *********************************************************************/
75 * crng_init is protected by base_crng->lock, and only increases
76 * its value (from empty->early->ready).
79 CRNG_EMPTY = 0, /* Little to no entropy collected */
80 CRNG_EARLY = 1, /* At least POOL_EARLY_BITS collected */
81 CRNG_READY = 2 /* Fully initialized with POOL_READY_BITS collected */
82 } crng_init __read_mostly = CRNG_EMPTY;
83 #define crng_ready() (likely(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_FLAGS("urandom_warning", HZ, 3, RATELIMIT_MSG_ON_RELEASE);
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 /* Used by wait_for_random_bytes(), and considered an entropy collector, below. */
114 static void try_to_generate_entropy(void);
117 * Wait for the input pool to be seeded and thus guaranteed to supply
118 * cryptographically secure random numbers. This applies to: the /dev/urandom
119 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
120 * family of functions. Using any of these functions without first calling
121 * this function forfeits the guarantee of security.
123 * Returns: 0 if the input pool has been seeded.
124 * -ERESTARTSYS if the function was interrupted by a signal.
126 int wait_for_random_bytes(void)
128 while (!crng_ready()) {
131 try_to_generate_entropy();
132 ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ);
134 return ret > 0 ? 0 : ret;
138 EXPORT_SYMBOL(wait_for_random_bytes);
141 * Add a callback function that will be invoked when the input
142 * pool is initialised.
144 * returns: 0 if callback is successfully added
145 * -EALREADY if pool is already initialised (callback not called)
147 int __cold register_random_ready_notifier(struct notifier_block *nb)
155 spin_lock_irqsave(&random_ready_chain_lock, flags);
157 ret = raw_notifier_chain_register(&random_ready_chain, nb);
158 spin_unlock_irqrestore(&random_ready_chain_lock, flags);
163 * Delete a previously registered readiness callback function.
165 int __cold 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);
176 static void __cold process_random_ready_list(void)
180 spin_lock_irqsave(&random_ready_chain_lock, flags);
181 raw_notifier_call_chain(&random_ready_chain, 0, NULL);
182 spin_unlock_irqrestore(&random_ready_chain_lock, flags);
185 #define warn_unseeded_randomness() \
186 if (IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM) && !crng_ready()) \
187 printk_deferred(KERN_NOTICE "random: %s called from %pS with crng_init=%d\n", \
188 __func__, (void *)_RET_IP_, crng_init)
191 /*********************************************************************
193 * Fast key erasure RNG, the "crng".
195 * These functions expand entropy from the entropy extractor into
196 * long streams for external consumption using the "fast key erasure"
197 * RNG described at <https://blog.cr.yp.to/20170723-random.html>.
199 * There are a few exported interfaces for use by other drivers:
201 * void get_random_bytes(void *buf, size_t len)
202 * u32 get_random_u32()
203 * u64 get_random_u64()
204 * unsigned int get_random_int()
205 * unsigned long get_random_long()
207 * These interfaces will return the requested number of random bytes
208 * into the given buffer or as a return value. This is equivalent to
209 * a read from /dev/urandom. The u32, u64, int, and long family of
210 * functions may be higher performance for one-off random integers,
211 * because they do a bit of buffering and do not invoke reseeding
212 * until the buffer is emptied.
214 *********************************************************************/
217 CRNG_RESEED_START_INTERVAL = HZ,
218 CRNG_RESEED_INTERVAL = 60 * HZ
222 u8 key[CHACHA20_KEY_SIZE] __aligned(__alignof__(long));
224 unsigned long generation;
227 .lock = __SPIN_LOCK_UNLOCKED(base_crng.lock)
231 u8 key[CHACHA20_KEY_SIZE];
232 unsigned long generation;
235 static DEFINE_PER_CPU(struct crng, crngs) = {
236 .generation = ULONG_MAX
239 /* Used by crng_reseed() and crng_make_state() to extract a new seed from the input pool. */
240 static void extract_entropy(void *buf, size_t len);
242 /* This extracts a new crng key from the input pool. */
243 static void crng_reseed(void)
246 unsigned long next_gen;
247 u8 key[CHACHA20_KEY_SIZE];
249 extract_entropy(key, sizeof(key));
252 * We copy the new key into the base_crng, overwriting the old one,
253 * and update the generation counter. We avoid hitting ULONG_MAX,
254 * because the per-cpu crngs are initialized to ULONG_MAX, so this
255 * forces new CPUs that come online to always initialize.
257 spin_lock_irqsave(&base_crng.lock, flags);
258 memcpy(base_crng.key, key, sizeof(base_crng.key));
259 next_gen = base_crng.generation + 1;
260 if (next_gen == ULONG_MAX)
262 WRITE_ONCE(base_crng.generation, next_gen);
263 WRITE_ONCE(base_crng.birth, jiffies);
265 crng_init = CRNG_READY;
266 spin_unlock_irqrestore(&base_crng.lock, flags);
267 memzero_explicit(key, sizeof(key));
271 * This generates a ChaCha block using the provided key, and then
272 * immediately overwites that key with half the block. It returns
273 * the resultant ChaCha state to the user, along with the second
274 * half of the block containing 32 bytes of random data that may
275 * be used; random_data_len may not be greater than 32.
277 * The returned ChaCha state contains within it a copy of the old
278 * key value, at index 4, so the state should always be zeroed out
279 * immediately after using in order to maintain forward secrecy.
280 * If the state cannot be erased in a timely manner, then it is
281 * safer to set the random_data parameter to &chacha_state[4] so
282 * that this function overwrites it before returning.
284 static void crng_fast_key_erasure(u8 key[CHACHA20_KEY_SIZE],
285 u32 chacha_state[CHACHA20_BLOCK_SIZE / sizeof(u32)],
286 u8 *random_data, size_t random_data_len)
288 u8 first_block[CHACHA20_BLOCK_SIZE];
290 BUG_ON(random_data_len > 32);
292 chacha_init_consts(chacha_state);
293 memcpy(&chacha_state[4], key, CHACHA20_KEY_SIZE);
294 memset(&chacha_state[12], 0, sizeof(u32) * 4);
295 chacha20_block(chacha_state, first_block);
297 memcpy(key, first_block, CHACHA20_KEY_SIZE);
298 memcpy(random_data, first_block + CHACHA20_KEY_SIZE, random_data_len);
299 memzero_explicit(first_block, sizeof(first_block));
303 * Return whether the crng seed is considered to be sufficiently old
304 * that a reseeding is needed. This happens if the last reseeding
305 * was CRNG_RESEED_INTERVAL ago, or during early boot, at an interval
306 * proportional to the uptime.
308 static bool crng_has_old_seed(void)
310 static bool early_boot = true;
311 unsigned long interval = CRNG_RESEED_INTERVAL;
313 if (unlikely(READ_ONCE(early_boot))) {
314 time64_t uptime = ktime_get_seconds();
315 if (uptime >= CRNG_RESEED_INTERVAL / HZ * 2)
316 WRITE_ONCE(early_boot, false);
318 interval = max_t(unsigned int, CRNG_RESEED_START_INTERVAL,
319 (unsigned int)uptime / 2 * HZ);
321 return time_is_before_jiffies(READ_ONCE(base_crng.birth) + interval);
325 * This function returns a ChaCha state that you may use for generating
326 * random data. It also returns up to 32 bytes on its own of random data
327 * that may be used; random_data_len may not be greater than 32.
329 static void crng_make_state(u32 chacha_state[CHACHA20_BLOCK_SIZE / sizeof(u32)],
330 u8 *random_data, size_t random_data_len)
335 BUG_ON(random_data_len > 32);
338 * For the fast path, we check whether we're ready, unlocked first, and
339 * then re-check once locked later. In the case where we're really not
340 * ready, we do fast key erasure with the base_crng directly, extracting
341 * when crng_init is CRNG_EMPTY.
346 spin_lock_irqsave(&base_crng.lock, flags);
347 ready = crng_ready();
349 if (crng_init == CRNG_EMPTY)
350 extract_entropy(base_crng.key, sizeof(base_crng.key));
351 crng_fast_key_erasure(base_crng.key, chacha_state,
352 random_data, random_data_len);
354 spin_unlock_irqrestore(&base_crng.lock, flags);
360 * If the base_crng is old enough, we reseed, which in turn bumps the
361 * generation counter that we check below.
363 if (unlikely(crng_has_old_seed()))
366 local_irq_save(flags);
367 crng = raw_cpu_ptr(&crngs);
370 * If our per-cpu crng is older than the base_crng, then it means
371 * somebody reseeded the base_crng. In that case, we do fast key
372 * erasure on the base_crng, and use its output as the new key
373 * for our per-cpu crng. This brings us up to date with base_crng.
375 if (unlikely(crng->generation != READ_ONCE(base_crng.generation))) {
376 spin_lock(&base_crng.lock);
377 crng_fast_key_erasure(base_crng.key, chacha_state,
378 crng->key, sizeof(crng->key));
379 crng->generation = base_crng.generation;
380 spin_unlock(&base_crng.lock);
384 * Finally, when we've made it this far, our per-cpu crng has an up
385 * to date key, and we can do fast key erasure with it to produce
386 * some random data and a ChaCha state for the caller. All other
387 * branches of this function are "unlikely", so most of the time we
388 * should wind up here immediately.
390 crng_fast_key_erasure(crng->key, chacha_state, random_data, random_data_len);
391 local_irq_restore(flags);
394 static void _get_random_bytes(void *buf, size_t len)
396 u32 chacha_state[CHACHA20_BLOCK_SIZE / sizeof(u32)];
397 u8 tmp[CHACHA20_BLOCK_SIZE];
398 size_t first_block_len;
403 first_block_len = min_t(size_t, 32, len);
404 crng_make_state(chacha_state, buf, first_block_len);
405 len -= first_block_len;
406 buf += first_block_len;
409 if (len < CHACHA20_BLOCK_SIZE) {
410 chacha20_block(chacha_state, tmp);
411 memcpy(buf, tmp, len);
412 memzero_explicit(tmp, sizeof(tmp));
416 chacha20_block(chacha_state, buf);
417 if (unlikely(chacha_state[12] == 0))
419 len -= CHACHA20_BLOCK_SIZE;
420 buf += CHACHA20_BLOCK_SIZE;
423 memzero_explicit(chacha_state, sizeof(chacha_state));
427 * This function is the exported kernel interface. It returns some
428 * number of good random numbers, suitable for key generation, seeding
429 * TCP sequence numbers, etc. It does not rely on the hardware random
430 * number generator. For random bytes direct from the hardware RNG
431 * (when available), use get_random_bytes_arch(). In order to ensure
432 * that the randomness provided by this function is okay, the function
433 * wait_for_random_bytes() should be called and return 0 at least once
434 * at any point prior.
436 void get_random_bytes(void *buf, size_t len)
438 warn_unseeded_randomness();
439 _get_random_bytes(buf, len);
441 EXPORT_SYMBOL(get_random_bytes);
443 static ssize_t get_random_bytes_user(struct iov_iter *iter)
445 u32 chacha_state[CHACHA20_BLOCK_SIZE / sizeof(u32)];
446 u8 block[CHACHA20_BLOCK_SIZE];
447 size_t ret = 0, copied;
449 if (unlikely(!iov_iter_count(iter)))
453 * Immediately overwrite the ChaCha key at index 4 with random
454 * bytes, in case userspace causes copy_to_user() below to sleep
455 * forever, so that we still retain forward secrecy in that case.
457 crng_make_state(chacha_state, (u8 *)&chacha_state[4], CHACHA20_KEY_SIZE);
459 * However, if we're doing a read of len <= 32, we don't need to
460 * use chacha_state after, so we can simply return those bytes to
463 if (iov_iter_count(iter) <= CHACHA20_KEY_SIZE) {
464 ret = copy_to_iter(&chacha_state[4], CHACHA20_KEY_SIZE, iter);
465 goto out_zero_chacha;
469 chacha20_block(chacha_state, block);
470 if (unlikely(chacha_state[12] == 0))
473 copied = copy_to_iter(block, sizeof(block), iter);
475 if (!iov_iter_count(iter) || copied != sizeof(block))
478 BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0);
479 if (ret % PAGE_SIZE == 0) {
480 if (signal_pending(current))
486 memzero_explicit(block, sizeof(block));
488 memzero_explicit(chacha_state, sizeof(chacha_state));
489 return ret ? ret : -EFAULT;
493 * Batched entropy returns random integers. The quality of the random
494 * number is good as /dev/urandom. In order to ensure that the randomness
495 * provided by this function is okay, the function wait_for_random_bytes()
496 * should be called and return 0 at least once at any point prior.
499 #define DEFINE_BATCHED_ENTROPY(type) \
500 struct batch_ ##type { \
502 * We make this 1.5x a ChaCha block, so that we get the \
503 * remaining 32 bytes from fast key erasure, plus one full \
504 * block from the detached ChaCha state. We can increase \
505 * the size of this later if needed so long as we keep the \
506 * formula of (integer_blocks + 0.5) * CHACHA20_BLOCK_SIZE. \
508 type entropy[CHACHA20_BLOCK_SIZE * 3 / (2 * sizeof(type))]; \
509 unsigned long generation; \
510 unsigned int position; \
513 static DEFINE_PER_CPU(struct batch_ ##type, batched_entropy_ ##type) = { \
514 .position = UINT_MAX \
517 type get_random_ ##type(void) \
520 unsigned long flags; \
521 struct batch_ ##type *batch; \
522 unsigned long next_gen; \
524 warn_unseeded_randomness(); \
526 if (!crng_ready()) { \
527 _get_random_bytes(&ret, sizeof(ret)); \
531 local_irq_save(flags); \
532 batch = raw_cpu_ptr(&batched_entropy_##type); \
534 next_gen = READ_ONCE(base_crng.generation); \
535 if (batch->position >= ARRAY_SIZE(batch->entropy) || \
536 next_gen != batch->generation) { \
537 _get_random_bytes(batch->entropy, sizeof(batch->entropy)); \
538 batch->position = 0; \
539 batch->generation = next_gen; \
542 ret = batch->entropy[batch->position]; \
543 batch->entropy[batch->position] = 0; \
545 local_irq_restore(flags); \
548 EXPORT_SYMBOL(get_random_ ##type);
550 DEFINE_BATCHED_ENTROPY(u64)
551 DEFINE_BATCHED_ENTROPY(u32)
555 * This function is called when the CPU is coming up, with entry
556 * CPUHP_RANDOM_PREPARE, which comes before CPUHP_WORKQUEUE_PREP.
558 int __cold random_prepare_cpu(unsigned int cpu)
561 * When the cpu comes back online, immediately invalidate both
562 * the per-cpu crng and all batches, so that we serve fresh
565 per_cpu_ptr(&crngs, cpu)->generation = ULONG_MAX;
566 per_cpu_ptr(&batched_entropy_u32, cpu)->position = UINT_MAX;
567 per_cpu_ptr(&batched_entropy_u64, cpu)->position = UINT_MAX;
573 * This function will use the architecture-specific hardware random
574 * number generator if it is available. It is not recommended for
575 * use. Use get_random_bytes() instead. It returns the number of
578 size_t __must_check get_random_bytes_arch(void *buf, size_t len)
585 size_t block_len = min_t(size_t, left, sizeof(unsigned long));
587 if (!arch_get_random_long(&v))
590 memcpy(p, &v, block_len);
597 EXPORT_SYMBOL(get_random_bytes_arch);
600 /**********************************************************************
602 * Entropy accumulation and extraction routines.
604 * Callers may add entropy via:
606 * static void mix_pool_bytes(const void *buf, size_t len)
608 * After which, if added entropy should be credited:
610 * static void credit_init_bits(size_t bits)
612 * Finally, extract entropy via:
614 * static void extract_entropy(void *buf, size_t len)
616 **********************************************************************/
619 POOL_BITS = BLAKE2S_HASH_SIZE * 8,
620 POOL_READY_BITS = POOL_BITS, /* When crng_init->CRNG_READY */
621 POOL_EARLY_BITS = POOL_READY_BITS / 2 /* When crng_init->CRNG_EARLY */
625 struct blake2s_state hash;
627 unsigned int init_bits;
629 .hash.h = { BLAKE2S_IV0 ^ (0x01010000 | BLAKE2S_HASH_SIZE),
630 BLAKE2S_IV1, BLAKE2S_IV2, BLAKE2S_IV3, BLAKE2S_IV4,
631 BLAKE2S_IV5, BLAKE2S_IV6, BLAKE2S_IV7 },
632 .hash.outlen = BLAKE2S_HASH_SIZE,
633 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
636 static void _mix_pool_bytes(const void *buf, size_t len)
638 blake2s_update(&input_pool.hash, buf, len);
642 * This function adds bytes into the input pool. It does not
643 * update the initialization bit counter; the caller should call
644 * credit_init_bits if this is appropriate.
646 static void mix_pool_bytes(const void *buf, size_t len)
650 spin_lock_irqsave(&input_pool.lock, flags);
651 _mix_pool_bytes(buf, len);
652 spin_unlock_irqrestore(&input_pool.lock, flags);
656 * This is an HKDF-like construction for using the hashed collected entropy
657 * as a PRF key, that's then expanded block-by-block.
659 static void extract_entropy(void *buf, size_t len)
662 u8 seed[BLAKE2S_HASH_SIZE], next_key[BLAKE2S_HASH_SIZE];
664 unsigned long rdseed[32 / sizeof(long)];
669 for (i = 0; i < ARRAY_SIZE(block.rdseed); ++i) {
670 if (!arch_get_random_seed_long(&block.rdseed[i]) &&
671 !arch_get_random_long(&block.rdseed[i]))
672 block.rdseed[i] = random_get_entropy();
675 spin_lock_irqsave(&input_pool.lock, flags);
677 /* seed = HASHPRF(last_key, entropy_input) */
678 blake2s_final(&input_pool.hash, seed);
680 /* next_key = HASHPRF(seed, RDSEED || 0) */
682 blake2s(next_key, (u8 *)&block, seed, sizeof(next_key), sizeof(block), sizeof(seed));
683 blake2s_init_key(&input_pool.hash, BLAKE2S_HASH_SIZE, next_key, sizeof(next_key));
685 spin_unlock_irqrestore(&input_pool.lock, flags);
686 memzero_explicit(next_key, sizeof(next_key));
689 i = min_t(size_t, len, BLAKE2S_HASH_SIZE);
690 /* output = HASHPRF(seed, RDSEED || ++counter) */
692 blake2s(buf, (u8 *)&block, seed, i, sizeof(block), sizeof(seed));
697 memzero_explicit(seed, sizeof(seed));
698 memzero_explicit(&block, sizeof(block));
701 #define credit_init_bits(bits) if (!crng_ready()) _credit_init_bits(bits)
703 static void __cold _credit_init_bits(size_t bits)
705 unsigned int new, orig, add;
711 add = min_t(size_t, bits, POOL_BITS);
714 orig = READ_ONCE(input_pool.init_bits);
715 new = min_t(unsigned int, POOL_BITS, orig + add);
716 } while (cmpxchg(&input_pool.init_bits, orig, new) != orig);
718 if (orig < POOL_READY_BITS && new >= POOL_READY_BITS) {
719 crng_reseed(); /* Sets crng_init to CRNG_READY under base_crng.lock. */
720 process_random_ready_list();
721 wake_up_interruptible(&crng_init_wait);
722 kill_fasync(&fasync, SIGIO, POLL_IN);
723 pr_notice("crng init done\n");
724 if (urandom_warning.missed)
725 pr_notice("%d urandom warning(s) missed due to ratelimiting\n",
726 urandom_warning.missed);
727 } else if (orig < POOL_EARLY_BITS && new >= POOL_EARLY_BITS) {
728 spin_lock_irqsave(&base_crng.lock, flags);
729 /* Check if crng_init is CRNG_EMPTY, to avoid race with crng_reseed(). */
730 if (crng_init == CRNG_EMPTY) {
731 extract_entropy(base_crng.key, sizeof(base_crng.key));
732 crng_init = CRNG_EARLY;
734 spin_unlock_irqrestore(&base_crng.lock, flags);
739 /**********************************************************************
741 * Entropy collection routines.
743 * The following exported functions are used for pushing entropy into
744 * the above entropy accumulation routines:
746 * void add_device_randomness(const void *buf, size_t len);
747 * void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy);
748 * void add_bootloader_randomness(const void *buf, size_t len);
749 * void add_interrupt_randomness(int irq);
750 * void add_input_randomness(unsigned int type, unsigned int code, unsigned int value);
751 * void add_disk_randomness(struct gendisk *disk);
753 * add_device_randomness() adds data to the input pool that
754 * is likely to differ between two devices (or possibly even per boot).
755 * This would be things like MAC addresses or serial numbers, or the
756 * read-out of the RTC. This does *not* credit any actual entropy to
757 * the pool, but it initializes the pool to different values for devices
758 * that might otherwise be identical and have very little entropy
759 * available to them (particularly common in the embedded world).
761 * add_hwgenerator_randomness() is for true hardware RNGs, and will credit
762 * entropy as specified by the caller. If the entropy pool is full it will
763 * block until more entropy is needed.
765 * add_bootloader_randomness() is called by bootloader drivers, such as EFI
766 * and device tree, and credits its input depending on whether or not the
767 * configuration option CONFIG_RANDOM_TRUST_BOOTLOADER is set.
769 * add_interrupt_randomness() uses the interrupt timing as random
770 * inputs to the entropy pool. Using the cycle counters and the irq source
771 * as inputs, it feeds the input pool roughly once a second or after 64
772 * interrupts, crediting 1 bit of entropy for whichever comes first.
774 * add_input_randomness() uses the input layer interrupt timing, as well
775 * as the event type information from the hardware.
777 * add_disk_randomness() uses what amounts to the seek time of block
778 * layer request events, on a per-disk_devt basis, as input to the
779 * entropy pool. Note that high-speed solid state drives with very low
780 * seek times do not make for good sources of entropy, as their seek
781 * times are usually fairly consistent.
783 * The last two routines try to estimate how many bits of entropy
784 * to credit. They do this by keeping track of the first and second
785 * order deltas of the event timings.
787 **********************************************************************/
789 static bool trust_cpu __initdata = IS_ENABLED(CONFIG_RANDOM_TRUST_CPU);
790 static bool trust_bootloader __initdata = IS_ENABLED(CONFIG_RANDOM_TRUST_BOOTLOADER);
791 static int __init parse_trust_cpu(char *arg)
793 return kstrtobool(arg, &trust_cpu);
795 static int __init parse_trust_bootloader(char *arg)
797 return kstrtobool(arg, &trust_bootloader);
799 early_param("random.trust_cpu", parse_trust_cpu);
800 early_param("random.trust_bootloader", parse_trust_bootloader);
803 * The first collection of entropy occurs at system boot while interrupts
804 * are still turned off. Here we push in latent entropy, RDSEED, a timestamp,
805 * utsname(), and the command line. Depending on the above configuration knob,
806 * RDSEED may be considered sufficient for initialization. Note that much
807 * earlier setup may already have pushed entropy into the input pool by the
810 int __init random_init(const char *command_line)
812 ktime_t now = ktime_get_real();
813 unsigned int i, arch_bits;
814 unsigned long entropy;
816 #if defined(LATENT_ENTROPY_PLUGIN)
817 static const u8 compiletime_seed[BLAKE2S_BLOCK_SIZE] __initconst __latent_entropy;
818 _mix_pool_bytes(compiletime_seed, sizeof(compiletime_seed));
821 for (i = 0, arch_bits = BLAKE2S_BLOCK_SIZE * 8;
822 i < BLAKE2S_BLOCK_SIZE; i += sizeof(entropy)) {
823 if (!arch_get_random_seed_long_early(&entropy) &&
824 !arch_get_random_long_early(&entropy)) {
825 entropy = random_get_entropy();
826 arch_bits -= sizeof(entropy) * 8;
828 _mix_pool_bytes(&entropy, sizeof(entropy));
830 _mix_pool_bytes(&now, sizeof(now));
831 _mix_pool_bytes(utsname(), sizeof(*(utsname())));
832 _mix_pool_bytes(command_line, strlen(command_line));
833 add_latent_entropy();
838 _credit_init_bits(arch_bits);
844 * Add device- or boot-specific data to the input pool to help
847 * None of this adds any entropy; it is meant to avoid the problem of
848 * the entropy pool having similar initial state across largely
851 void add_device_randomness(const void *buf, size_t len)
853 unsigned long entropy = random_get_entropy();
856 spin_lock_irqsave(&input_pool.lock, flags);
857 _mix_pool_bytes(&entropy, sizeof(entropy));
858 _mix_pool_bytes(buf, len);
859 spin_unlock_irqrestore(&input_pool.lock, flags);
861 EXPORT_SYMBOL(add_device_randomness);
864 * Interface for in-kernel drivers of true hardware RNGs.
865 * Those devices may produce endless random bits and will be throttled
866 * when our pool is full.
868 void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy)
870 mix_pool_bytes(buf, len);
871 credit_init_bits(entropy);
874 * Throttle writing to once every CRNG_RESEED_INTERVAL, unless
875 * we're not yet initialized.
877 if (!kthread_should_stop() && crng_ready())
878 schedule_timeout_interruptible(CRNG_RESEED_INTERVAL);
880 EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
883 * Handle random seed passed by bootloader, and credit it if
884 * CONFIG_RANDOM_TRUST_BOOTLOADER is set.
886 void __init add_bootloader_randomness(const void *buf, size_t len)
888 mix_pool_bytes(buf, len);
889 if (trust_bootloader)
890 credit_init_bits(len * 8);
894 unsigned long pool[4];
897 struct timer_list mix;
900 static DEFINE_PER_CPU(struct fast_pool, irq_randomness) = {
902 #define FASTMIX_PERM SIPHASH_PERMUTATION
903 .pool = { SIPHASH_CONST_0, SIPHASH_CONST_1, SIPHASH_CONST_2, SIPHASH_CONST_3 }
905 #define FASTMIX_PERM HSIPHASH_PERMUTATION
906 .pool = { HSIPHASH_CONST_0, HSIPHASH_CONST_1, HSIPHASH_CONST_2, HSIPHASH_CONST_3 }
911 * This is [Half]SipHash-1-x, starting from an empty key. Because
912 * the key is fixed, it assumes that its inputs are non-malicious,
913 * and therefore this has no security on its own. s represents the
914 * four-word SipHash state, while v represents a two-word input.
916 static void fast_mix(unsigned long s[4], unsigned long v1, unsigned long v2)
919 FASTMIX_PERM(s[0], s[1], s[2], s[3]);
922 FASTMIX_PERM(s[0], s[1], s[2], s[3]);
928 * This function is called when the CPU has just come online, with
929 * entry CPUHP_AP_RANDOM_ONLINE, just after CPUHP_AP_WORKQUEUE_ONLINE.
931 int __cold random_online_cpu(unsigned int cpu)
934 * During CPU shutdown and before CPU onlining, add_interrupt_
935 * randomness() may schedule mix_interrupt_randomness(), and
936 * set the MIX_INFLIGHT flag. However, because the worker can
937 * be scheduled on a different CPU during this period, that
938 * flag will never be cleared. For that reason, we zero out
939 * the flag here, which runs just after workqueues are onlined
940 * for the CPU again. This also has the effect of setting the
941 * irq randomness count to zero so that new accumulated irqs
944 per_cpu_ptr(&irq_randomness, cpu)->count = 0;
949 static void mix_interrupt_randomness(unsigned long data)
951 struct fast_pool *fast_pool = (struct fast_pool *)data;
953 * The size of the copied stack pool is explicitly 2 longs so that we
954 * only ever ingest half of the siphash output each time, retaining
955 * the other half as the next "key" that carries over. The entropy is
956 * supposed to be sufficiently dispersed between bits so on average
957 * we don't wind up "losing" some.
959 unsigned long pool[2];
962 /* Check to see if we're running on the wrong CPU due to hotplug. */
964 if (fast_pool != this_cpu_ptr(&irq_randomness)) {
970 * Copy the pool to the stack so that the mixer always has a
971 * consistent view, before we reenable irqs again.
973 memcpy(pool, fast_pool->pool, sizeof(pool));
974 count = fast_pool->count;
975 fast_pool->count = 0;
976 fast_pool->last = jiffies;
979 mix_pool_bytes(pool, sizeof(pool));
980 credit_init_bits(clamp_t(unsigned int, (count & U16_MAX) / 64, 1, sizeof(pool) * 8));
982 memzero_explicit(pool, sizeof(pool));
985 void add_interrupt_randomness(int irq)
987 enum { MIX_INFLIGHT = 1U << 31 };
988 unsigned long entropy = random_get_entropy();
989 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
990 struct pt_regs *regs = get_irq_regs();
991 unsigned int new_count;
993 fast_mix(fast_pool->pool, entropy,
994 (regs ? instruction_pointer(regs) : _RET_IP_) ^ swab(irq));
995 new_count = ++fast_pool->count;
997 if (new_count & MIX_INFLIGHT)
1000 if (new_count < 1024 && !time_is_before_jiffies(fast_pool->last + HZ))
1003 if (unlikely(!fast_pool->mix.data))
1004 setup_timer(&fast_pool->mix, mix_interrupt_randomness, (unsigned long)fast_pool);
1006 fast_pool->count |= MIX_INFLIGHT;
1007 if (!timer_pending(&fast_pool->mix)) {
1008 fast_pool->mix.expires = jiffies;
1009 add_timer_on(&fast_pool->mix, raw_smp_processor_id());
1012 EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1014 /* There is one of these per entropy source */
1015 struct timer_rand_state {
1016 unsigned long last_time;
1017 long last_delta, last_delta2;
1021 * This function adds entropy to the entropy "pool" by using timing
1022 * delays. It uses the timer_rand_state structure to make an estimate
1023 * of how many bits of entropy this call has added to the pool. The
1024 * value "num" is also added to the pool; it should somehow describe
1025 * the type of event that just happened.
1027 static void add_timer_randomness(struct timer_rand_state *state, unsigned int num)
1029 unsigned long entropy = random_get_entropy(), now = jiffies, flags;
1030 long delta, delta2, delta3;
1034 * If we're in a hard IRQ, add_interrupt_randomness() will be called
1035 * sometime after, so mix into the fast pool.
1038 fast_mix(this_cpu_ptr(&irq_randomness)->pool, entropy, num);
1040 spin_lock_irqsave(&input_pool.lock, flags);
1041 _mix_pool_bytes(&entropy, sizeof(entropy));
1042 _mix_pool_bytes(&num, sizeof(num));
1043 spin_unlock_irqrestore(&input_pool.lock, flags);
1050 * Calculate number of bits of randomness we probably added.
1051 * We take into account the first, second and third-order deltas
1052 * in order to make our estimate.
1054 delta = now - READ_ONCE(state->last_time);
1055 WRITE_ONCE(state->last_time, now);
1057 delta2 = delta - READ_ONCE(state->last_delta);
1058 WRITE_ONCE(state->last_delta, delta);
1060 delta3 = delta2 - READ_ONCE(state->last_delta2);
1061 WRITE_ONCE(state->last_delta2, delta2);
1075 * delta is now minimum absolute delta. Round down by 1 bit
1076 * on general principles, and limit entropy estimate to 11 bits.
1078 bits = min(fls(delta >> 1), 11);
1081 * As mentioned above, if we're in a hard IRQ, add_interrupt_randomness()
1082 * will run after this, which uses a different crediting scheme of 1 bit
1083 * per every 64 interrupts. In order to let that function do accounting
1084 * close to the one in this function, we credit a full 64/64 bit per bit,
1085 * and then subtract one to account for the extra one added.
1088 this_cpu_ptr(&irq_randomness)->count += max(1u, bits * 64) - 1;
1090 _credit_init_bits(bits);
1093 void add_input_randomness(unsigned int type, unsigned int code, unsigned int value)
1095 static unsigned char last_value;
1096 static struct timer_rand_state input_timer_state = { INITIAL_JIFFIES };
1098 /* Ignore autorepeat and the like. */
1099 if (value == last_value)
1103 add_timer_randomness(&input_timer_state,
1104 (type << 4) ^ code ^ (code >> 4) ^ value);
1106 EXPORT_SYMBOL_GPL(add_input_randomness);
1109 void add_disk_randomness(struct gendisk *disk)
1111 if (!disk || !disk->random)
1113 /* First major is 1, so we get >= 0x200 here. */
1114 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1116 EXPORT_SYMBOL_GPL(add_disk_randomness);
1118 void __cold rand_initialize_disk(struct gendisk *disk)
1120 struct timer_rand_state *state;
1123 * If kzalloc returns null, we just won't use that entropy
1126 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1128 state->last_time = INITIAL_JIFFIES;
1129 disk->random = state;
1135 * Each time the timer fires, we expect that we got an unpredictable
1136 * jump in the cycle counter. Even if the timer is running on another
1137 * CPU, the timer activity will be touching the stack of the CPU that is
1138 * generating entropy..
1140 * Note that we don't re-arm the timer in the timer itself - we are
1141 * happy to be scheduled away, since that just makes the load more
1142 * complex, but we do not want the timer to keep ticking unless the
1143 * entropy loop is running.
1145 * So the re-arming always happens in the entropy loop itself.
1147 static void __cold entropy_timer(unsigned long data)
1149 credit_init_bits(1);
1153 * If we have an actual cycle counter, see if we can
1154 * generate enough entropy with timing noise
1156 static void __cold try_to_generate_entropy(void)
1159 unsigned long entropy;
1160 struct timer_list timer;
1163 stack.entropy = random_get_entropy();
1165 /* Slow counter - or none. Don't even bother */
1166 if (stack.entropy == random_get_entropy())
1169 __setup_timer_on_stack(&stack.timer, entropy_timer, 0, 0);
1170 while (!crng_ready() && !signal_pending(current)) {
1171 if (!timer_pending(&stack.timer))
1172 mod_timer(&stack.timer, jiffies + 1);
1173 mix_pool_bytes(&stack.entropy, sizeof(stack.entropy));
1175 stack.entropy = random_get_entropy();
1178 del_timer_sync(&stack.timer);
1179 destroy_timer_on_stack(&stack.timer);
1180 mix_pool_bytes(&stack.entropy, sizeof(stack.entropy));
1184 /**********************************************************************
1186 * Userspace reader/writer interfaces.
1188 * getrandom(2) is the primary modern interface into the RNG and should
1189 * be used in preference to anything else.
1191 * Reading from /dev/random has the same functionality as calling
1192 * getrandom(2) with flags=0. In earlier versions, however, it had
1193 * vastly different semantics and should therefore be avoided, to
1194 * prevent backwards compatibility issues.
1196 * Reading from /dev/urandom has the same functionality as calling
1197 * getrandom(2) with flags=GRND_INSECURE. Because it does not block
1198 * waiting for the RNG to be ready, it should not be used.
1200 * Writing to either /dev/random or /dev/urandom adds entropy to
1201 * the input pool but does not credit it.
1203 * Polling on /dev/random indicates when the RNG is initialized, on
1204 * the read side, and when it wants new entropy, on the write side.
1206 * Both /dev/random and /dev/urandom have the same set of ioctls for
1207 * adding entropy, getting the entropy count, zeroing the count, and
1208 * reseeding the crng.
1210 **********************************************************************/
1212 SYSCALL_DEFINE3(getrandom, char __user *, ubuf, size_t, len, unsigned int, flags)
1214 struct iov_iter iter;
1218 if (flags & ~(GRND_NONBLOCK | GRND_RANDOM | GRND_INSECURE))
1222 * Requesting insecure and blocking randomness at the same time makes
1225 if ((flags & (GRND_INSECURE | GRND_RANDOM)) == (GRND_INSECURE | GRND_RANDOM))
1228 if (!crng_ready() && !(flags & GRND_INSECURE)) {
1229 if (flags & GRND_NONBLOCK)
1231 ret = wait_for_random_bytes();
1236 ret = import_single_range(READ, ubuf, len, &iov, &iter);
1239 return get_random_bytes_user(&iter);
1242 static unsigned int random_poll(struct file *file, poll_table *wait)
1244 poll_wait(file, &crng_init_wait, wait);
1245 return crng_ready() ? POLLIN | POLLRDNORM : POLLOUT | POLLWRNORM;
1248 static ssize_t write_pool_user(struct iov_iter *iter)
1250 u8 block[BLAKE2S_BLOCK_SIZE];
1254 if (unlikely(!iov_iter_count(iter)))
1258 copied = copy_from_iter(block, sizeof(block), iter);
1260 mix_pool_bytes(block, copied);
1261 if (!iov_iter_count(iter) || copied != sizeof(block))
1264 BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0);
1265 if (ret % PAGE_SIZE == 0) {
1266 if (signal_pending(current))
1272 memzero_explicit(block, sizeof(block));
1273 return ret ? ret : -EFAULT;
1276 static ssize_t random_write_iter(struct kiocb *kiocb, struct iov_iter *iter)
1278 return write_pool_user(iter);
1281 static ssize_t urandom_read_iter(struct kiocb *kiocb, struct iov_iter *iter)
1283 static int maxwarn = 10;
1285 if (!crng_ready()) {
1286 if (!ratelimit_disable && maxwarn <= 0)
1287 ++urandom_warning.missed;
1288 else if (ratelimit_disable || __ratelimit(&urandom_warning)) {
1290 pr_notice("%s: uninitialized urandom read (%zu bytes read)\n",
1291 current->comm, iov_iter_count(iter));
1295 return get_random_bytes_user(iter);
1298 static ssize_t random_read_iter(struct kiocb *kiocb, struct iov_iter *iter)
1302 if (!crng_ready() &&
1303 (kiocb->ki_filp->f_flags & O_NONBLOCK))
1306 ret = wait_for_random_bytes();
1309 return get_random_bytes_user(iter);
1312 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1314 int __user *p = (int __user *)arg;
1319 /* Inherently racy, no point locking. */
1320 if (put_user(input_pool.init_bits, p))
1323 case RNDADDTOENTCNT:
1324 if (!capable(CAP_SYS_ADMIN))
1326 if (get_user(ent_count, p))
1330 credit_init_bits(ent_count);
1332 case RNDADDENTROPY: {
1333 struct iov_iter iter;
1338 if (!capable(CAP_SYS_ADMIN))
1340 if (get_user(ent_count, p++))
1344 if (get_user(len, p++))
1346 ret = import_single_range(WRITE, p, len, &iov, &iter);
1349 ret = write_pool_user(&iter);
1350 if (unlikely(ret < 0))
1352 /* Since we're crediting, enforce that it was all written into the pool. */
1353 if (unlikely(ret != len))
1355 credit_init_bits(ent_count);
1360 /* No longer has any effect. */
1361 if (!capable(CAP_SYS_ADMIN))
1365 if (!capable(CAP_SYS_ADMIN))
1376 static int random_fasync(int fd, struct file *filp, int on)
1378 return fasync_helper(fd, filp, on, &fasync);
1381 const struct file_operations random_fops = {
1382 .read_iter = random_read_iter,
1383 .write_iter = random_write_iter,
1384 .poll = random_poll,
1385 .unlocked_ioctl = random_ioctl,
1386 .fasync = random_fasync,
1387 .llseek = noop_llseek,
1388 .splice_read = generic_file_splice_read,
1389 .splice_write = iter_file_splice_write,
1392 const struct file_operations urandom_fops = {
1393 .read_iter = urandom_read_iter,
1394 .write_iter = random_write_iter,
1395 .unlocked_ioctl = random_ioctl,
1396 .fasync = random_fasync,
1397 .llseek = noop_llseek,
1398 .splice_read = generic_file_splice_read,
1399 .splice_write = iter_file_splice_write,
1403 /********************************************************************
1407 * These are partly unused legacy knobs with dummy values to not break
1408 * userspace and partly still useful things. They are usually accessible
1409 * in /proc/sys/kernel/random/ and are as follows:
1411 * - boot_id - a UUID representing the current boot.
1413 * - uuid - a random UUID, different each time the file is read.
1415 * - poolsize - the number of bits of entropy that the input pool can
1416 * hold, tied to the POOL_BITS constant.
1418 * - entropy_avail - the number of bits of entropy currently in the
1419 * input pool. Always <= poolsize.
1421 * - write_wakeup_threshold - the amount of entropy in the input pool
1422 * below which write polls to /dev/random will unblock, requesting
1423 * more entropy, tied to the POOL_READY_BITS constant. It is writable
1424 * to avoid breaking old userspaces, but writing to it does not
1425 * change any behavior of the RNG.
1427 * - urandom_min_reseed_secs - fixed to the value CRNG_RESEED_INTERVAL.
1428 * It is writable to avoid breaking old userspaces, but writing
1429 * to it does not change any behavior of the RNG.
1431 ********************************************************************/
1433 #ifdef CONFIG_SYSCTL
1435 #include <linux/sysctl.h>
1437 static int sysctl_random_min_urandom_seed = CRNG_RESEED_INTERVAL / HZ;
1438 static int sysctl_random_write_wakeup_bits = POOL_READY_BITS;
1439 static int sysctl_poolsize = POOL_BITS;
1440 static u8 sysctl_bootid[UUID_SIZE];
1443 * This function is used to return both the bootid UUID, and random
1444 * UUID. The difference is in whether table->data is NULL; if it is,
1445 * then a new UUID is generated and returned to the user.
1447 static int proc_do_uuid(struct ctl_table *table, int write, void __user *buf,
1448 size_t *lenp, loff_t *ppos)
1450 u8 tmp_uuid[UUID_SIZE], *uuid;
1451 char uuid_string[UUID_STRING_LEN + 1];
1452 struct ctl_table fake_table = {
1453 .data = uuid_string,
1454 .maxlen = UUID_STRING_LEN
1463 generate_random_uuid(uuid);
1465 static DEFINE_SPINLOCK(bootid_spinlock);
1467 spin_lock(&bootid_spinlock);
1469 generate_random_uuid(uuid);
1470 spin_unlock(&bootid_spinlock);
1473 snprintf(uuid_string, sizeof(uuid_string), "%pU", uuid);
1474 return proc_dostring(&fake_table, 0, buf, lenp, ppos);
1477 /* The same as proc_dointvec, but writes don't change anything. */
1478 static int proc_do_rointvec(struct ctl_table *table, int write, void __user *buf,
1479 size_t *lenp, loff_t *ppos)
1481 return write ? 0 : proc_dointvec(table, 0, buf, lenp, ppos);
1484 extern struct ctl_table random_table[];
1485 struct ctl_table random_table[] = {
1487 .procname = "poolsize",
1488 .data = &sysctl_poolsize,
1489 .maxlen = sizeof(int),
1491 .proc_handler = proc_dointvec,
1494 .procname = "entropy_avail",
1495 .data = &input_pool.init_bits,
1496 .maxlen = sizeof(int),
1498 .proc_handler = proc_dointvec,
1501 .procname = "write_wakeup_threshold",
1502 .data = &sysctl_random_write_wakeup_bits,
1503 .maxlen = sizeof(int),
1505 .proc_handler = proc_do_rointvec,
1508 .procname = "urandom_min_reseed_secs",
1509 .data = &sysctl_random_min_urandom_seed,
1510 .maxlen = sizeof(int),
1512 .proc_handler = proc_do_rointvec,
1515 .procname = "boot_id",
1516 .data = &sysctl_bootid,
1518 .proc_handler = proc_do_uuid,
1523 .proc_handler = proc_do_uuid,
1527 #endif /* CONFIG_SYSCTL */