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 is protected by base_crng->lock, and only increases
74 * its value (from empty->early->ready).
77 CRNG_EMPTY = 0, /* Little to no entropy collected */
78 CRNG_EARLY = 1, /* At least POOL_EARLY_BITS collected */
79 CRNG_READY = 2 /* Fully initialized with POOL_READY_BITS collected */
80 } crng_init __read_mostly = CRNG_EMPTY;
81 static DEFINE_STATIC_KEY_FALSE(crng_is_ready);
82 #define crng_ready() (static_branch_likely(&crng_is_ready) || crng_init >= CRNG_READY)
83 /* Various types of waiters for crng_init->CRNG_READY transition. */
84 static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
85 static struct fasync_struct *fasync;
86 static DEFINE_SPINLOCK(random_ready_chain_lock);
87 static RAW_NOTIFIER_HEAD(random_ready_chain);
89 /* Control how we warn userspace. */
90 static struct ratelimit_state urandom_warning =
91 RATELIMIT_STATE_INIT("warn_urandom_randomness", HZ, 3);
92 static int ratelimit_disable __read_mostly =
93 IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM);
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 static void __cold crng_set_ready(struct work_struct *work)
114 static_branch_enable(&crng_is_ready);
117 /* Used by wait_for_random_bytes(), and considered an entropy collector, below. */
118 static void try_to_generate_entropy(void);
121 * Wait for the input pool to be seeded and thus guaranteed to supply
122 * cryptographically secure random numbers. This applies to: the /dev/urandom
123 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
124 * family of functions. Using any of these functions without first calling
125 * this function forfeits the guarantee of security.
127 * Returns: 0 if the input pool has been seeded.
128 * -ERESTARTSYS if the function was interrupted by a signal.
130 int wait_for_random_bytes(void)
132 while (!crng_ready()) {
135 try_to_generate_entropy();
136 ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ);
138 return ret > 0 ? 0 : ret;
142 EXPORT_SYMBOL(wait_for_random_bytes);
145 * Add a callback function that will be invoked when the input
146 * pool is initialised.
148 * returns: 0 if callback is successfully added
149 * -EALREADY if pool is already initialised (callback not called)
151 int __cold register_random_ready_notifier(struct notifier_block *nb)
159 spin_lock_irqsave(&random_ready_chain_lock, flags);
161 ret = raw_notifier_chain_register(&random_ready_chain, nb);
162 spin_unlock_irqrestore(&random_ready_chain_lock, flags);
165 EXPORT_SYMBOL(register_random_ready_notifier);
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);
180 EXPORT_SYMBOL(unregister_random_ready_notifier);
182 static void __cold process_random_ready_list(void)
186 spin_lock_irqsave(&random_ready_chain_lock, flags);
187 raw_notifier_call_chain(&random_ready_chain, 0, NULL);
188 spin_unlock_irqrestore(&random_ready_chain_lock, flags);
191 #define warn_unseeded_randomness() \
192 if (IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM) && !crng_ready()) \
193 printk_deferred(KERN_NOTICE "random: %s called from %pS with crng_init=%d\n", \
194 __func__, (void *)_RET_IP_, crng_init)
197 /*********************************************************************
199 * Fast key erasure RNG, the "crng".
201 * These functions expand entropy from the entropy extractor into
202 * long streams for external consumption using the "fast key erasure"
203 * RNG described at <https://blog.cr.yp.to/20170723-random.html>.
205 * There are a few exported interfaces for use by other drivers:
207 * void get_random_bytes(void *buf, size_t len)
208 * u32 get_random_u32()
209 * u64 get_random_u64()
210 * unsigned int get_random_int()
211 * unsigned long get_random_long()
213 * These interfaces will return the requested number of random bytes
214 * into the given buffer or as a return value. This is equivalent to
215 * a read from /dev/urandom. The u32, u64, int, and long family of
216 * functions may be higher performance for one-off random integers,
217 * because they do a bit of buffering and do not invoke reseeding
218 * until the buffer is emptied.
220 *********************************************************************/
223 CRNG_RESEED_START_INTERVAL = HZ,
224 CRNG_RESEED_INTERVAL = 60 * HZ
228 u8 key[CHACHA_KEY_SIZE] __aligned(__alignof__(long));
230 unsigned long generation;
233 .lock = __SPIN_LOCK_UNLOCKED(base_crng.lock)
237 u8 key[CHACHA_KEY_SIZE];
238 unsigned long generation;
242 static DEFINE_PER_CPU(struct crng, crngs) = {
243 .generation = ULONG_MAX,
244 .lock = INIT_LOCAL_LOCK(crngs.lock),
247 /* Used by crng_reseed() and crng_make_state() to extract a new seed from the input pool. */
248 static void extract_entropy(void *buf, size_t len);
250 /* This extracts a new crng key from the input pool. */
251 static void crng_reseed(void)
254 unsigned long next_gen;
255 u8 key[CHACHA_KEY_SIZE];
257 extract_entropy(key, sizeof(key));
260 * We copy the new key into the base_crng, overwriting the old one,
261 * and update the generation counter. We avoid hitting ULONG_MAX,
262 * because the per-cpu crngs are initialized to ULONG_MAX, so this
263 * forces new CPUs that come online to always initialize.
265 spin_lock_irqsave(&base_crng.lock, flags);
266 memcpy(base_crng.key, key, sizeof(base_crng.key));
267 next_gen = base_crng.generation + 1;
268 if (next_gen == ULONG_MAX)
270 WRITE_ONCE(base_crng.generation, next_gen);
271 WRITE_ONCE(base_crng.birth, jiffies);
272 if (!static_branch_likely(&crng_is_ready))
273 crng_init = CRNG_READY;
274 spin_unlock_irqrestore(&base_crng.lock, flags);
275 memzero_explicit(key, sizeof(key));
279 * This generates a ChaCha block using the provided key, and then
280 * immediately overwites that key with half the block. It returns
281 * the resultant ChaCha state to the user, along with the second
282 * half of the block containing 32 bytes of random data that may
283 * be used; random_data_len may not be greater than 32.
285 * The returned ChaCha state contains within it a copy of the old
286 * key value, at index 4, so the state should always be zeroed out
287 * immediately after using in order to maintain forward secrecy.
288 * If the state cannot be erased in a timely manner, then it is
289 * safer to set the random_data parameter to &chacha_state[4] so
290 * that this function overwrites it before returning.
292 static void crng_fast_key_erasure(u8 key[CHACHA_KEY_SIZE],
293 u32 chacha_state[CHACHA_STATE_WORDS],
294 u8 *random_data, size_t random_data_len)
296 u8 first_block[CHACHA_BLOCK_SIZE];
298 BUG_ON(random_data_len > 32);
300 chacha_init_consts(chacha_state);
301 memcpy(&chacha_state[4], key, CHACHA_KEY_SIZE);
302 memset(&chacha_state[12], 0, sizeof(u32) * 4);
303 chacha20_block(chacha_state, first_block);
305 memcpy(key, first_block, CHACHA_KEY_SIZE);
306 memcpy(random_data, first_block + CHACHA_KEY_SIZE, random_data_len);
307 memzero_explicit(first_block, sizeof(first_block));
311 * Return whether the crng seed is considered to be sufficiently old
312 * that a reseeding is needed. This happens if the last reseeding
313 * was CRNG_RESEED_INTERVAL ago, or during early boot, at an interval
314 * proportional to the uptime.
316 static bool crng_has_old_seed(void)
318 static bool early_boot = true;
319 unsigned long interval = CRNG_RESEED_INTERVAL;
321 if (unlikely(READ_ONCE(early_boot))) {
322 time64_t uptime = ktime_get_seconds();
323 if (uptime >= CRNG_RESEED_INTERVAL / HZ * 2)
324 WRITE_ONCE(early_boot, false);
326 interval = max_t(unsigned int, CRNG_RESEED_START_INTERVAL,
327 (unsigned int)uptime / 2 * HZ);
329 return time_is_before_jiffies(READ_ONCE(base_crng.birth) + interval);
333 * This function returns a ChaCha state that you may use for generating
334 * random data. It also returns up to 32 bytes on its own of random data
335 * that may be used; random_data_len may not be greater than 32.
337 static void crng_make_state(u32 chacha_state[CHACHA_STATE_WORDS],
338 u8 *random_data, size_t random_data_len)
343 BUG_ON(random_data_len > 32);
346 * For the fast path, we check whether we're ready, unlocked first, and
347 * then re-check once locked later. In the case where we're really not
348 * ready, we do fast key erasure with the base_crng directly, extracting
349 * when crng_init is CRNG_EMPTY.
354 spin_lock_irqsave(&base_crng.lock, flags);
355 ready = crng_ready();
357 if (crng_init == CRNG_EMPTY)
358 extract_entropy(base_crng.key, sizeof(base_crng.key));
359 crng_fast_key_erasure(base_crng.key, chacha_state,
360 random_data, random_data_len);
362 spin_unlock_irqrestore(&base_crng.lock, flags);
368 * If the base_crng is old enough, we reseed, which in turn bumps the
369 * generation counter that we check below.
371 if (unlikely(crng_has_old_seed()))
374 local_lock_irqsave(&crngs.lock, flags);
375 crng = raw_cpu_ptr(&crngs);
378 * If our per-cpu crng is older than the base_crng, then it means
379 * somebody reseeded the base_crng. In that case, we do fast key
380 * erasure on the base_crng, and use its output as the new key
381 * for our per-cpu crng. This brings us up to date with base_crng.
383 if (unlikely(crng->generation != READ_ONCE(base_crng.generation))) {
384 spin_lock(&base_crng.lock);
385 crng_fast_key_erasure(base_crng.key, chacha_state,
386 crng->key, sizeof(crng->key));
387 crng->generation = base_crng.generation;
388 spin_unlock(&base_crng.lock);
392 * Finally, when we've made it this far, our per-cpu crng has an up
393 * to date key, and we can do fast key erasure with it to produce
394 * some random data and a ChaCha state for the caller. All other
395 * branches of this function are "unlikely", so most of the time we
396 * should wind up here immediately.
398 crng_fast_key_erasure(crng->key, chacha_state, random_data, random_data_len);
399 local_unlock_irqrestore(&crngs.lock, flags);
402 static void _get_random_bytes(void *buf, size_t len)
404 u32 chacha_state[CHACHA_STATE_WORDS];
405 u8 tmp[CHACHA_BLOCK_SIZE];
406 size_t first_block_len;
411 first_block_len = min_t(size_t, 32, len);
412 crng_make_state(chacha_state, buf, first_block_len);
413 len -= first_block_len;
414 buf += first_block_len;
417 if (len < CHACHA_BLOCK_SIZE) {
418 chacha20_block(chacha_state, tmp);
419 memcpy(buf, tmp, len);
420 memzero_explicit(tmp, sizeof(tmp));
424 chacha20_block(chacha_state, buf);
425 if (unlikely(chacha_state[12] == 0))
427 len -= CHACHA_BLOCK_SIZE;
428 buf += CHACHA_BLOCK_SIZE;
431 memzero_explicit(chacha_state, sizeof(chacha_state));
435 * This function is the exported kernel interface. It returns some
436 * number of good random numbers, suitable for key generation, seeding
437 * TCP sequence numbers, etc. It does not rely on the hardware random
438 * number generator. For random bytes direct from the hardware RNG
439 * (when available), use get_random_bytes_arch(). In order to ensure
440 * that the randomness provided by this function is okay, the function
441 * wait_for_random_bytes() should be called and return 0 at least once
442 * at any point prior.
444 void get_random_bytes(void *buf, size_t len)
446 warn_unseeded_randomness();
447 _get_random_bytes(buf, len);
449 EXPORT_SYMBOL(get_random_bytes);
451 static ssize_t get_random_bytes_user(void __user *ubuf, size_t len)
453 size_t block_len, left, ret = 0;
454 u32 chacha_state[CHACHA_STATE_WORDS];
455 u8 output[CHACHA_BLOCK_SIZE];
461 * Immediately overwrite the ChaCha key at index 4 with random
462 * bytes, in case userspace causes copy_to_user() below to sleep
463 * forever, so that we still retain forward secrecy in that case.
465 crng_make_state(chacha_state, (u8 *)&chacha_state[4], CHACHA_KEY_SIZE);
467 * However, if we're doing a read of len <= 32, we don't need to
468 * use chacha_state after, so we can simply return those bytes to
471 if (len <= CHACHA_KEY_SIZE) {
472 ret = len - copy_to_user(ubuf, &chacha_state[4], len);
473 goto out_zero_chacha;
477 chacha20_block(chacha_state, output);
478 if (unlikely(chacha_state[12] == 0))
481 block_len = min_t(size_t, len, CHACHA_BLOCK_SIZE);
482 left = copy_to_user(ubuf, output, block_len);
484 ret += block_len - left;
494 BUILD_BUG_ON(PAGE_SIZE % CHACHA_BLOCK_SIZE != 0);
495 if (ret % PAGE_SIZE == 0) {
496 if (signal_pending(current))
502 memzero_explicit(output, sizeof(output));
504 memzero_explicit(chacha_state, sizeof(chacha_state));
505 return ret ? ret : -EFAULT;
509 * Batched entropy returns random integers. The quality of the random
510 * number is good as /dev/urandom. In order to ensure that the randomness
511 * provided by this function is okay, the function wait_for_random_bytes()
512 * should be called and return 0 at least once at any point prior.
514 struct batched_entropy {
517 * We make this 1.5x a ChaCha block, so that we get the
518 * remaining 32 bytes from fast key erasure, plus one full
519 * block from the detached ChaCha state. We can increase
520 * the size of this later if needed so long as we keep the
521 * formula of (integer_blocks + 0.5) * CHACHA_BLOCK_SIZE.
523 u64 entropy_u64[CHACHA_BLOCK_SIZE * 3 / (2 * sizeof(u64))];
524 u32 entropy_u32[CHACHA_BLOCK_SIZE * 3 / (2 * sizeof(u32))];
527 unsigned long generation;
528 unsigned int position;
532 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64) = {
533 .lock = INIT_LOCAL_LOCK(batched_entropy_u64.lock),
537 u64 get_random_u64(void)
541 struct batched_entropy *batch;
542 unsigned long next_gen;
544 warn_unseeded_randomness();
547 _get_random_bytes(&ret, sizeof(ret));
551 local_lock_irqsave(&batched_entropy_u64.lock, flags);
552 batch = raw_cpu_ptr(&batched_entropy_u64);
554 next_gen = READ_ONCE(base_crng.generation);
555 if (batch->position >= ARRAY_SIZE(batch->entropy_u64) ||
556 next_gen != batch->generation) {
557 _get_random_bytes(batch->entropy_u64, sizeof(batch->entropy_u64));
559 batch->generation = next_gen;
562 ret = batch->entropy_u64[batch->position];
563 batch->entropy_u64[batch->position] = 0;
565 local_unlock_irqrestore(&batched_entropy_u64.lock, flags);
568 EXPORT_SYMBOL(get_random_u64);
570 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32) = {
571 .lock = INIT_LOCAL_LOCK(batched_entropy_u32.lock),
575 u32 get_random_u32(void)
579 struct batched_entropy *batch;
580 unsigned long next_gen;
582 warn_unseeded_randomness();
585 _get_random_bytes(&ret, sizeof(ret));
589 local_lock_irqsave(&batched_entropy_u32.lock, flags);
590 batch = raw_cpu_ptr(&batched_entropy_u32);
592 next_gen = READ_ONCE(base_crng.generation);
593 if (batch->position >= ARRAY_SIZE(batch->entropy_u32) ||
594 next_gen != batch->generation) {
595 _get_random_bytes(batch->entropy_u32, sizeof(batch->entropy_u32));
597 batch->generation = next_gen;
600 ret = batch->entropy_u32[batch->position];
601 batch->entropy_u32[batch->position] = 0;
603 local_unlock_irqrestore(&batched_entropy_u32.lock, flags);
606 EXPORT_SYMBOL(get_random_u32);
610 * This function is called when the CPU is coming up, with entry
611 * CPUHP_RANDOM_PREPARE, which comes before CPUHP_WORKQUEUE_PREP.
613 int __cold random_prepare_cpu(unsigned int cpu)
616 * When the cpu comes back online, immediately invalidate both
617 * the per-cpu crng and all batches, so that we serve fresh
620 per_cpu_ptr(&crngs, cpu)->generation = ULONG_MAX;
621 per_cpu_ptr(&batched_entropy_u32, cpu)->position = UINT_MAX;
622 per_cpu_ptr(&batched_entropy_u64, cpu)->position = UINT_MAX;
628 * This function will use the architecture-specific hardware random
629 * number generator if it is available. It is not recommended for
630 * use. Use get_random_bytes() instead. It returns the number of
633 size_t __must_check get_random_bytes_arch(void *buf, size_t len)
640 size_t block_len = min_t(size_t, left, sizeof(unsigned long));
642 if (!arch_get_random_long(&v))
645 memcpy(p, &v, block_len);
652 EXPORT_SYMBOL(get_random_bytes_arch);
655 /**********************************************************************
657 * Entropy accumulation and extraction routines.
659 * Callers may add entropy via:
661 * static void mix_pool_bytes(const void *buf, size_t len)
663 * After which, if added entropy should be credited:
665 * static void credit_init_bits(size_t bits)
667 * Finally, extract entropy via:
669 * static void extract_entropy(void *buf, size_t len)
671 **********************************************************************/
674 POOL_BITS = BLAKE2S_HASH_SIZE * 8,
675 POOL_READY_BITS = POOL_BITS, /* When crng_init->CRNG_READY */
676 POOL_EARLY_BITS = POOL_READY_BITS / 2 /* When crng_init->CRNG_EARLY */
680 struct blake2s_state hash;
682 unsigned int init_bits;
684 .hash.h = { BLAKE2S_IV0 ^ (0x01010000 | BLAKE2S_HASH_SIZE),
685 BLAKE2S_IV1, BLAKE2S_IV2, BLAKE2S_IV3, BLAKE2S_IV4,
686 BLAKE2S_IV5, BLAKE2S_IV6, BLAKE2S_IV7 },
687 .hash.outlen = BLAKE2S_HASH_SIZE,
688 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
691 static void _mix_pool_bytes(const void *buf, size_t len)
693 blake2s_update(&input_pool.hash, buf, len);
697 * This function adds bytes into the input pool. It does not
698 * update the initialization bit counter; the caller should call
699 * credit_init_bits if this is appropriate.
701 static void mix_pool_bytes(const void *buf, size_t len)
705 spin_lock_irqsave(&input_pool.lock, flags);
706 _mix_pool_bytes(buf, len);
707 spin_unlock_irqrestore(&input_pool.lock, flags);
711 * This is an HKDF-like construction for using the hashed collected entropy
712 * as a PRF key, that's then expanded block-by-block.
714 static void extract_entropy(void *buf, size_t len)
717 u8 seed[BLAKE2S_HASH_SIZE], next_key[BLAKE2S_HASH_SIZE];
719 unsigned long rdseed[32 / sizeof(long)];
724 for (i = 0; i < ARRAY_SIZE(block.rdseed); ++i) {
725 if (!arch_get_random_seed_long(&block.rdseed[i]) &&
726 !arch_get_random_long(&block.rdseed[i]))
727 block.rdseed[i] = random_get_entropy();
730 spin_lock_irqsave(&input_pool.lock, flags);
732 /* seed = HASHPRF(last_key, entropy_input) */
733 blake2s_final(&input_pool.hash, seed);
735 /* next_key = HASHPRF(seed, RDSEED || 0) */
737 blake2s(next_key, (u8 *)&block, seed, sizeof(next_key), sizeof(block), sizeof(seed));
738 blake2s_init_key(&input_pool.hash, BLAKE2S_HASH_SIZE, next_key, sizeof(next_key));
740 spin_unlock_irqrestore(&input_pool.lock, flags);
741 memzero_explicit(next_key, sizeof(next_key));
744 i = min_t(size_t, len, BLAKE2S_HASH_SIZE);
745 /* output = HASHPRF(seed, RDSEED || ++counter) */
747 blake2s(buf, (u8 *)&block, seed, i, sizeof(block), sizeof(seed));
752 memzero_explicit(seed, sizeof(seed));
753 memzero_explicit(&block, sizeof(block));
756 #define credit_init_bits(bits) if (!crng_ready()) _credit_init_bits(bits)
758 static void __cold _credit_init_bits(size_t bits)
760 static struct execute_work set_ready;
761 unsigned int new, orig, add;
767 add = min_t(size_t, bits, POOL_BITS);
770 orig = READ_ONCE(input_pool.init_bits);
771 new = min_t(unsigned int, POOL_BITS, orig + add);
772 } while (cmpxchg(&input_pool.init_bits, orig, new) != orig);
774 if (orig < POOL_READY_BITS && new >= POOL_READY_BITS) {
775 crng_reseed(); /* Sets crng_init to CRNG_READY under base_crng.lock. */
776 execute_in_process_context(crng_set_ready, &set_ready);
777 process_random_ready_list();
778 wake_up_interruptible(&crng_init_wait);
779 kill_fasync(&fasync, SIGIO, POLL_IN);
780 pr_notice("crng init done\n");
781 if (urandom_warning.missed)
782 pr_notice("%d urandom warning(s) missed due to ratelimiting\n",
783 urandom_warning.missed);
784 } else if (orig < POOL_EARLY_BITS && new >= POOL_EARLY_BITS) {
785 spin_lock_irqsave(&base_crng.lock, flags);
786 /* Check if crng_init is CRNG_EMPTY, to avoid race with crng_reseed(). */
787 if (crng_init == CRNG_EMPTY) {
788 extract_entropy(base_crng.key, sizeof(base_crng.key));
789 crng_init = CRNG_EARLY;
791 spin_unlock_irqrestore(&base_crng.lock, flags);
796 /**********************************************************************
798 * Entropy collection routines.
800 * The following exported functions are used for pushing entropy into
801 * the above entropy accumulation routines:
803 * void add_device_randomness(const void *buf, size_t len);
804 * void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy);
805 * void add_bootloader_randomness(const void *buf, size_t len);
806 * void add_interrupt_randomness(int irq);
807 * void add_input_randomness(unsigned int type, unsigned int code, unsigned int value);
808 * void add_disk_randomness(struct gendisk *disk);
810 * add_device_randomness() adds data to the input pool that
811 * is likely to differ between two devices (or possibly even per boot).
812 * This would be things like MAC addresses or serial numbers, or the
813 * read-out of the RTC. This does *not* credit any actual entropy to
814 * the pool, but it initializes the pool to different values for devices
815 * that might otherwise be identical and have very little entropy
816 * available to them (particularly common in the embedded world).
818 * add_hwgenerator_randomness() is for true hardware RNGs, and will credit
819 * entropy as specified by the caller. If the entropy pool is full it will
820 * block until more entropy is needed.
822 * add_bootloader_randomness() is called by bootloader drivers, such as EFI
823 * and device tree, and credits its input depending on whether or not the
824 * configuration option CONFIG_RANDOM_TRUST_BOOTLOADER is set.
826 * add_interrupt_randomness() uses the interrupt timing as random
827 * inputs to the entropy pool. Using the cycle counters and the irq source
828 * as inputs, it feeds the input pool roughly once a second or after 64
829 * interrupts, crediting 1 bit of entropy for whichever comes first.
831 * add_input_randomness() uses the input layer interrupt timing, as well
832 * as the event type information from the hardware.
834 * add_disk_randomness() uses what amounts to the seek time of block
835 * layer request events, on a per-disk_devt basis, as input to the
836 * entropy pool. Note that high-speed solid state drives with very low
837 * seek times do not make for good sources of entropy, as their seek
838 * times are usually fairly consistent.
840 * The last two routines try to estimate how many bits of entropy
841 * to credit. They do this by keeping track of the first and second
842 * order deltas of the event timings.
844 **********************************************************************/
846 static bool trust_cpu __ro_after_init = IS_ENABLED(CONFIG_RANDOM_TRUST_CPU);
847 static bool trust_bootloader __ro_after_init = IS_ENABLED(CONFIG_RANDOM_TRUST_BOOTLOADER);
848 static int __init parse_trust_cpu(char *arg)
850 return kstrtobool(arg, &trust_cpu);
852 static int __init parse_trust_bootloader(char *arg)
854 return kstrtobool(arg, &trust_bootloader);
856 early_param("random.trust_cpu", parse_trust_cpu);
857 early_param("random.trust_bootloader", parse_trust_bootloader);
860 * The first collection of entropy occurs at system boot while interrupts
861 * are still turned off. Here we push in latent entropy, RDSEED, a timestamp,
862 * utsname(), and the command line. Depending on the above configuration knob,
863 * RDSEED may be considered sufficient for initialization. Note that much
864 * earlier setup may already have pushed entropy into the input pool by the
867 int __init random_init(const char *command_line)
869 ktime_t now = ktime_get_real();
870 unsigned int i, arch_bytes;
871 unsigned long entropy;
873 #if defined(LATENT_ENTROPY_PLUGIN)
874 static const u8 compiletime_seed[BLAKE2S_BLOCK_SIZE] __initconst __latent_entropy;
875 _mix_pool_bytes(compiletime_seed, sizeof(compiletime_seed));
878 for (i = 0, arch_bytes = BLAKE2S_BLOCK_SIZE;
879 i < BLAKE2S_BLOCK_SIZE; i += sizeof(entropy)) {
880 if (!arch_get_random_seed_long_early(&entropy) &&
881 !arch_get_random_long_early(&entropy)) {
882 entropy = random_get_entropy();
883 arch_bytes -= sizeof(entropy);
885 _mix_pool_bytes(&entropy, sizeof(entropy));
887 _mix_pool_bytes(&now, sizeof(now));
888 _mix_pool_bytes(utsname(), sizeof(*(utsname())));
889 _mix_pool_bytes(command_line, strlen(command_line));
890 add_latent_entropy();
895 credit_init_bits(arch_bytes * 8);
901 * Add device- or boot-specific data to the input pool to help
904 * None of this adds any entropy; it is meant to avoid the problem of
905 * the entropy pool having similar initial state across largely
908 void add_device_randomness(const void *buf, size_t len)
910 unsigned long entropy = random_get_entropy();
913 spin_lock_irqsave(&input_pool.lock, flags);
914 _mix_pool_bytes(&entropy, sizeof(entropy));
915 _mix_pool_bytes(buf, len);
916 spin_unlock_irqrestore(&input_pool.lock, flags);
918 EXPORT_SYMBOL(add_device_randomness);
921 * Interface for in-kernel drivers of true hardware RNGs.
922 * Those devices may produce endless random bits and will be throttled
923 * when our pool is full.
925 void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy)
927 mix_pool_bytes(buf, len);
928 credit_init_bits(entropy);
931 * Throttle writing to once every CRNG_RESEED_INTERVAL, unless
932 * we're not yet initialized.
934 if (!kthread_should_stop() && crng_ready())
935 schedule_timeout_interruptible(CRNG_RESEED_INTERVAL);
937 EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
940 * Handle random seed passed by bootloader, and credit it if
941 * CONFIG_RANDOM_TRUST_BOOTLOADER is set.
943 void __cold add_bootloader_randomness(const void *buf, size_t len)
945 mix_pool_bytes(buf, len);
946 if (trust_bootloader)
947 credit_init_bits(len * 8);
949 EXPORT_SYMBOL_GPL(add_bootloader_randomness);
952 struct work_struct mix;
953 unsigned long pool[4];
958 static DEFINE_PER_CPU(struct fast_pool, irq_randomness) = {
960 #define FASTMIX_PERM SIPHASH_PERMUTATION
961 .pool = { SIPHASH_CONST_0, SIPHASH_CONST_1, SIPHASH_CONST_2, SIPHASH_CONST_3 }
963 #define FASTMIX_PERM HSIPHASH_PERMUTATION
964 .pool = { HSIPHASH_CONST_0, HSIPHASH_CONST_1, HSIPHASH_CONST_2, HSIPHASH_CONST_3 }
969 * This is [Half]SipHash-1-x, starting from an empty key. Because
970 * the key is fixed, it assumes that its inputs are non-malicious,
971 * and therefore this has no security on its own. s represents the
972 * four-word SipHash state, while v represents a two-word input.
974 static void fast_mix(unsigned long s[4], unsigned long v1, unsigned long v2)
977 FASTMIX_PERM(s[0], s[1], s[2], s[3]);
980 FASTMIX_PERM(s[0], s[1], s[2], s[3]);
986 * This function is called when the CPU has just come online, with
987 * entry CPUHP_AP_RANDOM_ONLINE, just after CPUHP_AP_WORKQUEUE_ONLINE.
989 int __cold random_online_cpu(unsigned int cpu)
992 * During CPU shutdown and before CPU onlining, add_interrupt_
993 * randomness() may schedule mix_interrupt_randomness(), and
994 * set the MIX_INFLIGHT flag. However, because the worker can
995 * be scheduled on a different CPU during this period, that
996 * flag will never be cleared. For that reason, we zero out
997 * the flag here, which runs just after workqueues are onlined
998 * for the CPU again. This also has the effect of setting the
999 * irq randomness count to zero so that new accumulated irqs
1002 per_cpu_ptr(&irq_randomness, cpu)->count = 0;
1007 static void mix_interrupt_randomness(struct work_struct *work)
1009 struct fast_pool *fast_pool = container_of(work, struct fast_pool, mix);
1011 * The size of the copied stack pool is explicitly 2 longs so that we
1012 * only ever ingest half of the siphash output each time, retaining
1013 * the other half as the next "key" that carries over. The entropy is
1014 * supposed to be sufficiently dispersed between bits so on average
1015 * we don't wind up "losing" some.
1017 unsigned long pool[2];
1020 /* Check to see if we're running on the wrong CPU due to hotplug. */
1021 local_irq_disable();
1022 if (fast_pool != this_cpu_ptr(&irq_randomness)) {
1028 * Copy the pool to the stack so that the mixer always has a
1029 * consistent view, before we reenable irqs again.
1031 memcpy(pool, fast_pool->pool, sizeof(pool));
1032 count = fast_pool->count;
1033 fast_pool->count = 0;
1034 fast_pool->last = jiffies;
1037 mix_pool_bytes(pool, sizeof(pool));
1038 credit_init_bits(max(1u, (count & U16_MAX) / 64));
1040 memzero_explicit(pool, sizeof(pool));
1043 void add_interrupt_randomness(int irq)
1045 enum { MIX_INFLIGHT = 1U << 31 };
1046 unsigned long entropy = random_get_entropy();
1047 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1048 struct pt_regs *regs = get_irq_regs();
1049 unsigned int new_count;
1051 fast_mix(fast_pool->pool, entropy,
1052 (regs ? instruction_pointer(regs) : _RET_IP_) ^ swab(irq));
1053 new_count = ++fast_pool->count;
1055 if (new_count & MIX_INFLIGHT)
1058 if (new_count < 64 && !time_is_before_jiffies(fast_pool->last + HZ))
1061 if (unlikely(!fast_pool->mix.func))
1062 INIT_WORK(&fast_pool->mix, mix_interrupt_randomness);
1063 fast_pool->count |= MIX_INFLIGHT;
1064 queue_work_on(raw_smp_processor_id(), system_highpri_wq, &fast_pool->mix);
1066 EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1068 /* There is one of these per entropy source */
1069 struct timer_rand_state {
1070 unsigned long last_time;
1071 long last_delta, last_delta2;
1075 * This function adds entropy to the entropy "pool" by using timing
1076 * delays. It uses the timer_rand_state structure to make an estimate
1077 * of how many bits of entropy this call has added to the pool. The
1078 * value "num" is also added to the pool; it should somehow describe
1079 * the type of event that just happened.
1081 static void add_timer_randomness(struct timer_rand_state *state, unsigned int num)
1083 unsigned long entropy = random_get_entropy(), now = jiffies, flags;
1084 long delta, delta2, delta3;
1088 * If we're in a hard IRQ, add_interrupt_randomness() will be called
1089 * sometime after, so mix into the fast pool.
1092 fast_mix(this_cpu_ptr(&irq_randomness)->pool, entropy, num);
1094 spin_lock_irqsave(&input_pool.lock, flags);
1095 _mix_pool_bytes(&entropy, sizeof(entropy));
1096 _mix_pool_bytes(&num, sizeof(num));
1097 spin_unlock_irqrestore(&input_pool.lock, flags);
1104 * Calculate number of bits of randomness we probably added.
1105 * We take into account the first, second and third-order deltas
1106 * in order to make our estimate.
1108 delta = now - READ_ONCE(state->last_time);
1109 WRITE_ONCE(state->last_time, now);
1111 delta2 = delta - READ_ONCE(state->last_delta);
1112 WRITE_ONCE(state->last_delta, delta);
1114 delta3 = delta2 - READ_ONCE(state->last_delta2);
1115 WRITE_ONCE(state->last_delta2, delta2);
1129 * delta is now minimum absolute delta. Round down by 1 bit
1130 * on general principles, and limit entropy estimate to 11 bits.
1132 bits = min(fls(delta >> 1), 11);
1135 * As mentioned above, if we're in a hard IRQ, add_interrupt_randomness()
1136 * will run after this, which uses a different crediting scheme of 1 bit
1137 * per every 64 interrupts. In order to let that function do accounting
1138 * close to the one in this function, we credit a full 64/64 bit per bit,
1139 * and then subtract one to account for the extra one added.
1142 this_cpu_ptr(&irq_randomness)->count += max(1u, bits * 64) - 1;
1144 _credit_init_bits(bits);
1147 void add_input_randomness(unsigned int type, unsigned int code, unsigned int value)
1149 static unsigned char last_value;
1150 static struct timer_rand_state input_timer_state = { INITIAL_JIFFIES };
1152 /* Ignore autorepeat and the like. */
1153 if (value == last_value)
1157 add_timer_randomness(&input_timer_state,
1158 (type << 4) ^ code ^ (code >> 4) ^ value);
1160 EXPORT_SYMBOL_GPL(add_input_randomness);
1163 void add_disk_randomness(struct gendisk *disk)
1165 if (!disk || !disk->random)
1167 /* First major is 1, so we get >= 0x200 here. */
1168 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1170 EXPORT_SYMBOL_GPL(add_disk_randomness);
1172 void __cold rand_initialize_disk(struct gendisk *disk)
1174 struct timer_rand_state *state;
1177 * If kzalloc returns null, we just won't use that entropy
1180 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1182 state->last_time = INITIAL_JIFFIES;
1183 disk->random = state;
1189 * Each time the timer fires, we expect that we got an unpredictable
1190 * jump in the cycle counter. Even if the timer is running on another
1191 * CPU, the timer activity will be touching the stack of the CPU that is
1192 * generating entropy..
1194 * Note that we don't re-arm the timer in the timer itself - we are
1195 * happy to be scheduled away, since that just makes the load more
1196 * complex, but we do not want the timer to keep ticking unless the
1197 * entropy loop is running.
1199 * So the re-arming always happens in the entropy loop itself.
1201 static void __cold entropy_timer(struct timer_list *t)
1203 credit_init_bits(1);
1207 * If we have an actual cycle counter, see if we can
1208 * generate enough entropy with timing noise
1210 static void __cold try_to_generate_entropy(void)
1213 unsigned long entropy;
1214 struct timer_list timer;
1217 stack.entropy = random_get_entropy();
1219 /* Slow counter - or none. Don't even bother */
1220 if (stack.entropy == random_get_entropy())
1223 timer_setup_on_stack(&stack.timer, entropy_timer, 0);
1224 while (!crng_ready() && !signal_pending(current)) {
1225 if (!timer_pending(&stack.timer))
1226 mod_timer(&stack.timer, jiffies + 1);
1227 mix_pool_bytes(&stack.entropy, sizeof(stack.entropy));
1229 stack.entropy = random_get_entropy();
1232 del_timer_sync(&stack.timer);
1233 destroy_timer_on_stack(&stack.timer);
1234 mix_pool_bytes(&stack.entropy, sizeof(stack.entropy));
1238 /**********************************************************************
1240 * Userspace reader/writer interfaces.
1242 * getrandom(2) is the primary modern interface into the RNG and should
1243 * be used in preference to anything else.
1245 * Reading from /dev/random has the same functionality as calling
1246 * getrandom(2) with flags=0. In earlier versions, however, it had
1247 * vastly different semantics and should therefore be avoided, to
1248 * prevent backwards compatibility issues.
1250 * Reading from /dev/urandom has the same functionality as calling
1251 * getrandom(2) with flags=GRND_INSECURE. Because it does not block
1252 * waiting for the RNG to be ready, it should not be used.
1254 * Writing to either /dev/random or /dev/urandom adds entropy to
1255 * the input pool but does not credit it.
1257 * Polling on /dev/random indicates when the RNG is initialized, on
1258 * the read side, and when it wants new entropy, on the write side.
1260 * Both /dev/random and /dev/urandom have the same set of ioctls for
1261 * adding entropy, getting the entropy count, zeroing the count, and
1262 * reseeding the crng.
1264 **********************************************************************/
1266 SYSCALL_DEFINE3(getrandom, char __user *, ubuf, size_t, len, unsigned int, flags)
1268 if (flags & ~(GRND_NONBLOCK | GRND_RANDOM | GRND_INSECURE))
1272 * Requesting insecure and blocking randomness at the same time makes
1275 if ((flags & (GRND_INSECURE | GRND_RANDOM)) == (GRND_INSECURE | GRND_RANDOM))
1281 if (!crng_ready() && !(flags & GRND_INSECURE)) {
1284 if (flags & GRND_NONBLOCK)
1286 ret = wait_for_random_bytes();
1290 return get_random_bytes_user(ubuf, len);
1293 static __poll_t random_poll(struct file *file, poll_table *wait)
1295 poll_wait(file, &crng_init_wait, wait);
1296 return crng_ready() ? EPOLLIN | EPOLLRDNORM : EPOLLOUT | EPOLLWRNORM;
1299 static int write_pool(const char __user *ubuf, size_t len)
1303 u8 block[BLAKE2S_BLOCK_SIZE];
1306 block_len = min(len, sizeof(block));
1307 if (copy_from_user(block, ubuf, block_len)) {
1313 mix_pool_bytes(block, block_len);
1318 memzero_explicit(block, sizeof(block));
1322 static ssize_t random_write(struct file *file, const char __user *ubuf,
1323 size_t len, loff_t *ppos)
1327 ret = write_pool(ubuf, len);
1331 return (ssize_t)len;
1334 static ssize_t urandom_read(struct file *file, char __user *ubuf,
1335 size_t len, loff_t *ppos)
1337 static int maxwarn = 10;
1339 if (!crng_ready()) {
1340 if (!ratelimit_disable && maxwarn <= 0)
1341 ++urandom_warning.missed;
1342 else if (ratelimit_disable || __ratelimit(&urandom_warning)) {
1344 pr_notice("%s: uninitialized urandom read (%zd bytes read)\n",
1345 current->comm, len);
1349 return get_random_bytes_user(ubuf, len);
1352 static ssize_t random_read(struct file *file, char __user *ubuf,
1353 size_t len, loff_t *ppos)
1357 ret = wait_for_random_bytes();
1360 return get_random_bytes_user(ubuf, len);
1363 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1365 int size, ent_count;
1366 int __user *p = (int __user *)arg;
1371 /* Inherently racy, no point locking. */
1372 if (put_user(input_pool.init_bits, p))
1375 case RNDADDTOENTCNT:
1376 if (!capable(CAP_SYS_ADMIN))
1378 if (get_user(ent_count, p))
1382 credit_init_bits(ent_count);
1385 if (!capable(CAP_SYS_ADMIN))
1387 if (get_user(ent_count, p++))
1391 if (get_user(size, p++))
1393 retval = write_pool((const char __user *)p, size);
1396 credit_init_bits(ent_count);
1400 /* No longer has any effect. */
1401 if (!capable(CAP_SYS_ADMIN))
1405 if (!capable(CAP_SYS_ADMIN))
1416 static int random_fasync(int fd, struct file *filp, int on)
1418 return fasync_helper(fd, filp, on, &fasync);
1421 const struct file_operations random_fops = {
1422 .read = random_read,
1423 .write = random_write,
1424 .poll = random_poll,
1425 .unlocked_ioctl = random_ioctl,
1426 .compat_ioctl = compat_ptr_ioctl,
1427 .fasync = random_fasync,
1428 .llseek = noop_llseek,
1431 const struct file_operations urandom_fops = {
1432 .read = urandom_read,
1433 .write = random_write,
1434 .unlocked_ioctl = random_ioctl,
1435 .compat_ioctl = compat_ptr_ioctl,
1436 .fasync = random_fasync,
1437 .llseek = noop_llseek,
1441 /********************************************************************
1445 * These are partly unused legacy knobs with dummy values to not break
1446 * userspace and partly still useful things. They are usually accessible
1447 * in /proc/sys/kernel/random/ and are as follows:
1449 * - boot_id - a UUID representing the current boot.
1451 * - uuid - a random UUID, different each time the file is read.
1453 * - poolsize - the number of bits of entropy that the input pool can
1454 * hold, tied to the POOL_BITS constant.
1456 * - entropy_avail - the number of bits of entropy currently in the
1457 * input pool. Always <= poolsize.
1459 * - write_wakeup_threshold - the amount of entropy in the input pool
1460 * below which write polls to /dev/random will unblock, requesting
1461 * more entropy, tied to the POOL_READY_BITS constant. It is writable
1462 * to avoid breaking old userspaces, but writing to it does not
1463 * change any behavior of the RNG.
1465 * - urandom_min_reseed_secs - fixed to the value CRNG_RESEED_INTERVAL.
1466 * It is writable to avoid breaking old userspaces, but writing
1467 * to it does not change any behavior of the RNG.
1469 ********************************************************************/
1471 #ifdef CONFIG_SYSCTL
1473 #include <linux/sysctl.h>
1475 static int sysctl_random_min_urandom_seed = CRNG_RESEED_INTERVAL / HZ;
1476 static int sysctl_random_write_wakeup_bits = POOL_READY_BITS;
1477 static int sysctl_poolsize = POOL_BITS;
1478 static u8 sysctl_bootid[UUID_SIZE];
1481 * This function is used to return both the bootid UUID, and random
1482 * UUID. The difference is in whether table->data is NULL; if it is,
1483 * then a new UUID is generated and returned to the user.
1485 static int proc_do_uuid(struct ctl_table *table, int write, void *buf,
1486 size_t *lenp, loff_t *ppos)
1488 u8 tmp_uuid[UUID_SIZE], *uuid;
1489 char uuid_string[UUID_STRING_LEN + 1];
1490 struct ctl_table fake_table = {
1491 .data = uuid_string,
1492 .maxlen = UUID_STRING_LEN
1501 generate_random_uuid(uuid);
1503 static DEFINE_SPINLOCK(bootid_spinlock);
1505 spin_lock(&bootid_spinlock);
1507 generate_random_uuid(uuid);
1508 spin_unlock(&bootid_spinlock);
1511 snprintf(uuid_string, sizeof(uuid_string), "%pU", uuid);
1512 return proc_dostring(&fake_table, 0, buf, lenp, ppos);
1515 /* The same as proc_dointvec, but writes don't change anything. */
1516 static int proc_do_rointvec(struct ctl_table *table, int write, void *buf,
1517 size_t *lenp, loff_t *ppos)
1519 return write ? 0 : proc_dointvec(table, 0, buf, lenp, ppos);
1522 extern struct ctl_table random_table[];
1523 struct ctl_table random_table[] = {
1525 .procname = "poolsize",
1526 .data = &sysctl_poolsize,
1527 .maxlen = sizeof(int),
1529 .proc_handler = proc_dointvec,
1532 .procname = "entropy_avail",
1533 .data = &input_pool.init_bits,
1534 .maxlen = sizeof(int),
1536 .proc_handler = proc_dointvec,
1539 .procname = "write_wakeup_threshold",
1540 .data = &sysctl_random_write_wakeup_bits,
1541 .maxlen = sizeof(int),
1543 .proc_handler = proc_do_rointvec,
1546 .procname = "urandom_min_reseed_secs",
1547 .data = &sysctl_random_min_urandom_seed,
1548 .maxlen = sizeof(int),
1550 .proc_handler = proc_do_rointvec,
1553 .procname = "boot_id",
1554 .data = &sysctl_bootid,
1556 .proc_handler = proc_do_uuid,
1561 .proc_handler = proc_do_uuid,
1565 #endif /* CONFIG_SYSCTL */