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. Some of that data is then "credited" as
19 * having a certain number of bits of entropy. When enough bits of entropy are
20 * available, the hash is finalized and handed as a key to a stream cipher that
21 * expands it indefinitely for various consumers. This key is periodically
22 * refreshed as the various entropy collectors, described below, add data to the
23 * input pool and credit it. There is currently no Fortuna-like scheduler
24 * involved, which can lead to malicious entropy sources causing a premature
25 * reseed, and the entropy estimates are, at best, conservative guesses.
28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30 #include <linux/utsname.h>
31 #include <linux/module.h>
32 #include <linux/kernel.h>
33 #include <linux/major.h>
34 #include <linux/string.h>
35 #include <linux/fcntl.h>
36 #include <linux/slab.h>
37 #include <linux/random.h>
38 #include <linux/poll.h>
39 #include <linux/init.h>
41 #include <linux/blkdev.h>
42 #include <linux/interrupt.h>
44 #include <linux/nodemask.h>
45 #include <linux/spinlock.h>
46 #include <linux/kthread.h>
47 #include <linux/percpu.h>
48 #include <linux/ptrace.h>
49 #include <linux/workqueue.h>
50 #include <linux/irq.h>
51 #include <linux/ratelimit.h>
52 #include <linux/syscalls.h>
53 #include <linux/completion.h>
54 #include <linux/uuid.h>
55 #include <linux/uaccess.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 = 0 --> Uninitialized
76 * 2 --> Initialized from input_pool
78 * crng_init is protected by base_crng->lock, and only increases
79 * its value (from 0->1->2).
81 static int crng_init = 0;
82 #define crng_ready() (likely(crng_init > 1))
83 /* Various types of waiters for crng_init->2 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 unseeded_warning =
91 RATELIMIT_STATE_INIT("warn_unseeded_randomness", HZ, 3);
92 static struct ratelimit_state urandom_warning =
93 RATELIMIT_STATE_INIT("warn_urandom_randomness", HZ, 3);
94 static int ratelimit_disable __read_mostly;
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 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 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 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(previous) \
186 _warn_unseeded_randomness(__func__, (void *)_RET_IP_, (previous))
188 static void _warn_unseeded_randomness(const char *func_name, void *caller, void **previous)
190 #ifdef CONFIG_WARN_ALL_UNSEEDED_RANDOM
191 const bool print_once = false;
193 static bool print_once __read_mostly;
196 if (print_once || crng_ready() ||
197 (previous && (caller == READ_ONCE(*previous))))
199 WRITE_ONCE(*previous, caller);
200 #ifndef CONFIG_WARN_ALL_UNSEEDED_RANDOM
203 if (__ratelimit(&unseeded_warning))
204 printk_deferred(KERN_NOTICE "random: %s called from %pS with crng_init=%d\n",
205 func_name, caller, crng_init);
209 /*********************************************************************
211 * Fast key erasure RNG, the "crng".
213 * These functions expand entropy from the entropy extractor into
214 * long streams for external consumption using the "fast key erasure"
215 * RNG described at <https://blog.cr.yp.to/20170723-random.html>.
217 * There are a few exported interfaces for use by other drivers:
219 * void get_random_bytes(void *buf, size_t nbytes)
220 * u32 get_random_u32()
221 * u64 get_random_u64()
222 * unsigned int get_random_int()
223 * unsigned long get_random_long()
225 * These interfaces will return the requested number of random bytes
226 * into the given buffer or as a return value. This is equivalent to
227 * a read from /dev/urandom. The u32, u64, int, and long family of
228 * functions may be higher performance for one-off random integers,
229 * because they do a bit of buffering and do not invoke reseeding
230 * until the buffer is emptied.
232 *********************************************************************/
235 CRNG_RESEED_INTERVAL = 300 * HZ,
236 CRNG_INIT_CNT_THRESH = 2 * CHACHA_KEY_SIZE
240 u8 key[CHACHA_KEY_SIZE] __aligned(__alignof__(long));
242 unsigned long generation;
245 .lock = __SPIN_LOCK_UNLOCKED(base_crng.lock)
249 u8 key[CHACHA_KEY_SIZE];
250 unsigned long generation;
254 static DEFINE_PER_CPU(struct crng, crngs) = {
255 .generation = ULONG_MAX,
256 .lock = INIT_LOCAL_LOCK(crngs.lock),
259 /* Used by crng_reseed() to extract a new seed from the input pool. */
260 static bool drain_entropy(void *buf, size_t nbytes, bool force);
263 * This extracts a new crng key from the input pool, but only if there is a
264 * sufficient amount of entropy available or force is true, in order to
265 * mitigate bruteforcing of newly added bits.
267 static void crng_reseed(bool force)
270 unsigned long next_gen;
271 u8 key[CHACHA_KEY_SIZE];
272 bool finalize_init = false;
274 /* Only reseed if we can, to prevent brute forcing a small amount of new bits. */
275 if (!drain_entropy(key, sizeof(key), force))
279 * We copy the new key into the base_crng, overwriting the old one,
280 * and update the generation counter. We avoid hitting ULONG_MAX,
281 * because the per-cpu crngs are initialized to ULONG_MAX, so this
282 * forces new CPUs that come online to always initialize.
284 spin_lock_irqsave(&base_crng.lock, flags);
285 memcpy(base_crng.key, key, sizeof(base_crng.key));
286 next_gen = base_crng.generation + 1;
287 if (next_gen == ULONG_MAX)
289 WRITE_ONCE(base_crng.generation, next_gen);
290 WRITE_ONCE(base_crng.birth, jiffies);
293 finalize_init = true;
295 spin_unlock_irqrestore(&base_crng.lock, flags);
296 memzero_explicit(key, sizeof(key));
298 process_random_ready_list();
299 wake_up_interruptible(&crng_init_wait);
300 kill_fasync(&fasync, SIGIO, POLL_IN);
301 pr_notice("crng init done\n");
302 if (unseeded_warning.missed) {
303 pr_notice("%d get_random_xx warning(s) missed due to ratelimiting\n",
304 unseeded_warning.missed);
305 unseeded_warning.missed = 0;
307 if (urandom_warning.missed) {
308 pr_notice("%d urandom warning(s) missed due to ratelimiting\n",
309 urandom_warning.missed);
310 urandom_warning.missed = 0;
316 * This generates a ChaCha block using the provided key, and then
317 * immediately overwites that key with half the block. It returns
318 * the resultant ChaCha state to the user, along with the second
319 * half of the block containing 32 bytes of random data that may
320 * be used; random_data_len may not be greater than 32.
322 static void crng_fast_key_erasure(u8 key[CHACHA_KEY_SIZE],
323 u32 chacha_state[CHACHA_STATE_WORDS],
324 u8 *random_data, size_t random_data_len)
326 u8 first_block[CHACHA_BLOCK_SIZE];
328 BUG_ON(random_data_len > 32);
330 chacha_init_consts(chacha_state);
331 memcpy(&chacha_state[4], key, CHACHA_KEY_SIZE);
332 memset(&chacha_state[12], 0, sizeof(u32) * 4);
333 chacha20_block(chacha_state, first_block);
335 memcpy(key, first_block, CHACHA_KEY_SIZE);
336 memcpy(random_data, first_block + CHACHA_KEY_SIZE, random_data_len);
337 memzero_explicit(first_block, sizeof(first_block));
341 * Return whether the crng seed is considered to be sufficiently
342 * old that a reseeding might be attempted. This happens if the last
343 * reseeding was CRNG_RESEED_INTERVAL ago, or during early boot, at
344 * an interval proportional to the uptime.
346 static bool crng_has_old_seed(void)
348 static bool early_boot = true;
349 unsigned long interval = CRNG_RESEED_INTERVAL;
351 if (unlikely(READ_ONCE(early_boot))) {
352 time64_t uptime = ktime_get_seconds();
353 if (uptime >= CRNG_RESEED_INTERVAL / HZ * 2)
354 WRITE_ONCE(early_boot, false);
356 interval = max_t(unsigned int, 5 * HZ,
357 (unsigned int)uptime / 2 * HZ);
359 return time_after(jiffies, READ_ONCE(base_crng.birth) + interval);
363 * This function returns a ChaCha state that you may use for generating
364 * random data. It also returns up to 32 bytes on its own of random data
365 * that may be used; random_data_len may not be greater than 32.
367 static void crng_make_state(u32 chacha_state[CHACHA_STATE_WORDS],
368 u8 *random_data, size_t random_data_len)
373 BUG_ON(random_data_len > 32);
376 * For the fast path, we check whether we're ready, unlocked first, and
377 * then re-check once locked later. In the case where we're really not
378 * ready, we do fast key erasure with the base_crng directly, because
379 * this is what crng_pre_init_inject() mutates during early init.
384 spin_lock_irqsave(&base_crng.lock, flags);
385 ready = crng_ready();
387 crng_fast_key_erasure(base_crng.key, chacha_state,
388 random_data, random_data_len);
389 spin_unlock_irqrestore(&base_crng.lock, flags);
395 * If the base_crng is old enough, we try to reseed, which in turn
396 * bumps the generation counter that we check below.
398 if (unlikely(crng_has_old_seed()))
401 local_lock_irqsave(&crngs.lock, flags);
402 crng = raw_cpu_ptr(&crngs);
405 * If our per-cpu crng is older than the base_crng, then it means
406 * somebody reseeded the base_crng. In that case, we do fast key
407 * erasure on the base_crng, and use its output as the new key
408 * for our per-cpu crng. This brings us up to date with base_crng.
410 if (unlikely(crng->generation != READ_ONCE(base_crng.generation))) {
411 spin_lock(&base_crng.lock);
412 crng_fast_key_erasure(base_crng.key, chacha_state,
413 crng->key, sizeof(crng->key));
414 crng->generation = base_crng.generation;
415 spin_unlock(&base_crng.lock);
419 * Finally, when we've made it this far, our per-cpu crng has an up
420 * to date key, and we can do fast key erasure with it to produce
421 * some random data and a ChaCha state for the caller. All other
422 * branches of this function are "unlikely", so most of the time we
423 * should wind up here immediately.
425 crng_fast_key_erasure(crng->key, chacha_state, random_data, random_data_len);
426 local_unlock_irqrestore(&crngs.lock, flags);
430 * This function is for crng_init == 0 only. It loads entropy directly
431 * into the crng's key, without going through the input pool. It is,
432 * generally speaking, not very safe, but we use this only at early
433 * boot time when it's better to have something there rather than
436 * If account is set, then the crng_init_cnt counter is incremented.
437 * This shouldn't be set by functions like add_device_randomness(),
438 * where we can't trust the buffer passed to it is guaranteed to be
439 * unpredictable (so it might not have any entropy at all).
441 static void crng_pre_init_inject(const void *input, size_t len, bool account)
443 static int crng_init_cnt = 0;
444 struct blake2s_state hash;
447 blake2s_init(&hash, sizeof(base_crng.key));
449 spin_lock_irqsave(&base_crng.lock, flags);
450 if (crng_init != 0) {
451 spin_unlock_irqrestore(&base_crng.lock, flags);
455 blake2s_update(&hash, base_crng.key, sizeof(base_crng.key));
456 blake2s_update(&hash, input, len);
457 blake2s_final(&hash, base_crng.key);
460 crng_init_cnt += min_t(size_t, len, CRNG_INIT_CNT_THRESH - crng_init_cnt);
461 if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
462 ++base_crng.generation;
467 spin_unlock_irqrestore(&base_crng.lock, flags);
470 pr_notice("fast init done\n");
473 static void _get_random_bytes(void *buf, size_t nbytes)
475 u32 chacha_state[CHACHA_STATE_WORDS];
476 u8 tmp[CHACHA_BLOCK_SIZE];
482 len = min_t(size_t, 32, nbytes);
483 crng_make_state(chacha_state, buf, len);
488 if (nbytes < CHACHA_BLOCK_SIZE) {
489 chacha20_block(chacha_state, tmp);
490 memcpy(buf, tmp, nbytes);
491 memzero_explicit(tmp, sizeof(tmp));
495 chacha20_block(chacha_state, buf);
496 if (unlikely(chacha_state[12] == 0))
498 nbytes -= CHACHA_BLOCK_SIZE;
499 buf += CHACHA_BLOCK_SIZE;
502 memzero_explicit(chacha_state, sizeof(chacha_state));
506 * This function is the exported kernel interface. It returns some
507 * number of good random numbers, suitable for key generation, seeding
508 * TCP sequence numbers, etc. It does not rely on the hardware random
509 * number generator. For random bytes direct from the hardware RNG
510 * (when available), use get_random_bytes_arch(). In order to ensure
511 * that the randomness provided by this function is okay, the function
512 * wait_for_random_bytes() should be called and return 0 at least once
513 * at any point prior.
515 void get_random_bytes(void *buf, size_t nbytes)
517 static void *previous;
519 warn_unseeded_randomness(&previous);
520 _get_random_bytes(buf, nbytes);
522 EXPORT_SYMBOL(get_random_bytes);
524 static ssize_t get_random_bytes_user(void __user *buf, size_t nbytes)
528 u32 chacha_state[CHACHA_STATE_WORDS];
529 u8 output[CHACHA_BLOCK_SIZE];
535 * Immediately overwrite the ChaCha key at index 4 with random
536 * bytes, in case userspace causes copy_to_user() below to sleep
537 * forever, so that we still retain forward secrecy in that case.
539 crng_make_state(chacha_state, (u8 *)&chacha_state[4], CHACHA_KEY_SIZE);
541 * However, if we're doing a read of len <= 32, we don't need to
542 * use chacha_state after, so we can simply return those bytes to
545 if (nbytes <= CHACHA_KEY_SIZE) {
546 ret = copy_to_user(buf, &chacha_state[4], nbytes) ? -EFAULT : nbytes;
547 goto out_zero_chacha;
551 chacha20_block(chacha_state, output);
552 if (unlikely(chacha_state[12] == 0))
555 len = min_t(size_t, nbytes, CHACHA_BLOCK_SIZE);
556 if (copy_to_user(buf, output, len)) {
565 BUILD_BUG_ON(PAGE_SIZE % CHACHA_BLOCK_SIZE != 0);
566 if (!(ret % PAGE_SIZE) && nbytes) {
567 if (signal_pending(current))
573 memzero_explicit(output, sizeof(output));
575 memzero_explicit(chacha_state, sizeof(chacha_state));
580 * Batched entropy returns random integers. The quality of the random
581 * number is good as /dev/urandom. In order to ensure that the randomness
582 * provided by this function is okay, the function wait_for_random_bytes()
583 * should be called and return 0 at least once at any point prior.
585 struct batched_entropy {
588 * We make this 1.5x a ChaCha block, so that we get the
589 * remaining 32 bytes from fast key erasure, plus one full
590 * block from the detached ChaCha state. We can increase
591 * the size of this later if needed so long as we keep the
592 * formula of (integer_blocks + 0.5) * CHACHA_BLOCK_SIZE.
594 u64 entropy_u64[CHACHA_BLOCK_SIZE * 3 / (2 * sizeof(u64))];
595 u32 entropy_u32[CHACHA_BLOCK_SIZE * 3 / (2 * sizeof(u32))];
598 unsigned long generation;
599 unsigned int position;
603 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64) = {
604 .lock = INIT_LOCAL_LOCK(batched_entropy_u64.lock),
608 u64 get_random_u64(void)
612 struct batched_entropy *batch;
613 static void *previous;
614 unsigned long next_gen;
616 warn_unseeded_randomness(&previous);
618 local_lock_irqsave(&batched_entropy_u64.lock, flags);
619 batch = raw_cpu_ptr(&batched_entropy_u64);
621 next_gen = READ_ONCE(base_crng.generation);
622 if (batch->position >= ARRAY_SIZE(batch->entropy_u64) ||
623 next_gen != batch->generation) {
624 _get_random_bytes(batch->entropy_u64, sizeof(batch->entropy_u64));
626 batch->generation = next_gen;
629 ret = batch->entropy_u64[batch->position];
630 batch->entropy_u64[batch->position] = 0;
632 local_unlock_irqrestore(&batched_entropy_u64.lock, flags);
635 EXPORT_SYMBOL(get_random_u64);
637 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32) = {
638 .lock = INIT_LOCAL_LOCK(batched_entropy_u32.lock),
642 u32 get_random_u32(void)
646 struct batched_entropy *batch;
647 static void *previous;
648 unsigned long next_gen;
650 warn_unseeded_randomness(&previous);
652 local_lock_irqsave(&batched_entropy_u32.lock, flags);
653 batch = raw_cpu_ptr(&batched_entropy_u32);
655 next_gen = READ_ONCE(base_crng.generation);
656 if (batch->position >= ARRAY_SIZE(batch->entropy_u32) ||
657 next_gen != batch->generation) {
658 _get_random_bytes(batch->entropy_u32, sizeof(batch->entropy_u32));
660 batch->generation = next_gen;
663 ret = batch->entropy_u32[batch->position];
664 batch->entropy_u32[batch->position] = 0;
666 local_unlock_irqrestore(&batched_entropy_u32.lock, flags);
669 EXPORT_SYMBOL(get_random_u32);
673 * This function is called when the CPU is coming up, with entry
674 * CPUHP_RANDOM_PREPARE, which comes before CPUHP_WORKQUEUE_PREP.
676 int random_prepare_cpu(unsigned int cpu)
679 * When the cpu comes back online, immediately invalidate both
680 * the per-cpu crng and all batches, so that we serve fresh
683 per_cpu_ptr(&crngs, cpu)->generation = ULONG_MAX;
684 per_cpu_ptr(&batched_entropy_u32, cpu)->position = UINT_MAX;
685 per_cpu_ptr(&batched_entropy_u64, cpu)->position = UINT_MAX;
691 * randomize_page - Generate a random, page aligned address
692 * @start: The smallest acceptable address the caller will take.
693 * @range: The size of the area, starting at @start, within which the
694 * random address must fall.
696 * If @start + @range would overflow, @range is capped.
698 * NOTE: Historical use of randomize_range, which this replaces, presumed that
699 * @start was already page aligned. We now align it regardless.
701 * Return: A page aligned address within [start, start + range). On error,
702 * @start is returned.
704 unsigned long randomize_page(unsigned long start, unsigned long range)
706 if (!PAGE_ALIGNED(start)) {
707 range -= PAGE_ALIGN(start) - start;
708 start = PAGE_ALIGN(start);
711 if (start > ULONG_MAX - range)
712 range = ULONG_MAX - start;
714 range >>= PAGE_SHIFT;
719 return start + (get_random_long() % range << PAGE_SHIFT);
723 * This function will use the architecture-specific hardware random
724 * number generator if it is available. It is not recommended for
725 * use. Use get_random_bytes() instead. It returns the number of
728 size_t __must_check get_random_bytes_arch(void *buf, size_t nbytes)
730 size_t left = nbytes;
735 size_t chunk = min_t(size_t, left, sizeof(unsigned long));
737 if (!arch_get_random_long(&v))
740 memcpy(p, &v, chunk);
745 return nbytes - left;
747 EXPORT_SYMBOL(get_random_bytes_arch);
750 /**********************************************************************
752 * Entropy accumulation and extraction routines.
754 * Callers may add entropy via:
756 * static void mix_pool_bytes(const void *in, size_t nbytes)
758 * After which, if added entropy should be credited:
760 * static void credit_entropy_bits(size_t nbits)
762 * Finally, extract entropy via these two, with the latter one
763 * setting the entropy count to zero and extracting only if there
764 * is POOL_MIN_BITS entropy credited prior or force is true:
766 * static void extract_entropy(void *buf, size_t nbytes)
767 * static bool drain_entropy(void *buf, size_t nbytes, bool force)
769 **********************************************************************/
772 POOL_BITS = BLAKE2S_HASH_SIZE * 8,
773 POOL_MIN_BITS = POOL_BITS /* No point in settling for less. */
776 /* For notifying userspace should write into /dev/random. */
777 static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
780 struct blake2s_state hash;
782 unsigned int entropy_count;
784 .hash.h = { BLAKE2S_IV0 ^ (0x01010000 | BLAKE2S_HASH_SIZE),
785 BLAKE2S_IV1, BLAKE2S_IV2, BLAKE2S_IV3, BLAKE2S_IV4,
786 BLAKE2S_IV5, BLAKE2S_IV6, BLAKE2S_IV7 },
787 .hash.outlen = BLAKE2S_HASH_SIZE,
788 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
791 static void _mix_pool_bytes(const void *in, size_t nbytes)
793 blake2s_update(&input_pool.hash, in, nbytes);
797 * This function adds bytes into the entropy "pool". It does not
798 * update the entropy estimate. The caller should call
799 * credit_entropy_bits if this is appropriate.
801 static void mix_pool_bytes(const void *in, size_t nbytes)
805 spin_lock_irqsave(&input_pool.lock, flags);
806 _mix_pool_bytes(in, nbytes);
807 spin_unlock_irqrestore(&input_pool.lock, flags);
810 static void credit_entropy_bits(size_t nbits)
812 unsigned int entropy_count, orig, add;
817 add = min_t(size_t, nbits, POOL_BITS);
820 orig = READ_ONCE(input_pool.entropy_count);
821 entropy_count = min_t(unsigned int, POOL_BITS, orig + add);
822 } while (cmpxchg(&input_pool.entropy_count, orig, entropy_count) != orig);
824 if (!crng_ready() && entropy_count >= POOL_MIN_BITS)
829 * This is an HKDF-like construction for using the hashed collected entropy
830 * as a PRF key, that's then expanded block-by-block.
832 static void extract_entropy(void *buf, size_t nbytes)
835 u8 seed[BLAKE2S_HASH_SIZE], next_key[BLAKE2S_HASH_SIZE];
837 unsigned long rdseed[32 / sizeof(long)];
842 for (i = 0; i < ARRAY_SIZE(block.rdseed); ++i) {
843 if (!arch_get_random_seed_long(&block.rdseed[i]) &&
844 !arch_get_random_long(&block.rdseed[i]))
845 block.rdseed[i] = random_get_entropy();
848 spin_lock_irqsave(&input_pool.lock, flags);
850 /* seed = HASHPRF(last_key, entropy_input) */
851 blake2s_final(&input_pool.hash, seed);
853 /* next_key = HASHPRF(seed, RDSEED || 0) */
855 blake2s(next_key, (u8 *)&block, seed, sizeof(next_key), sizeof(block), sizeof(seed));
856 blake2s_init_key(&input_pool.hash, BLAKE2S_HASH_SIZE, next_key, sizeof(next_key));
858 spin_unlock_irqrestore(&input_pool.lock, flags);
859 memzero_explicit(next_key, sizeof(next_key));
862 i = min_t(size_t, nbytes, BLAKE2S_HASH_SIZE);
863 /* output = HASHPRF(seed, RDSEED || ++counter) */
865 blake2s(buf, (u8 *)&block, seed, i, sizeof(block), sizeof(seed));
870 memzero_explicit(seed, sizeof(seed));
871 memzero_explicit(&block, sizeof(block));
875 * First we make sure we have POOL_MIN_BITS of entropy in the pool unless force
876 * is true, and then we set the entropy count to zero (but don't actually touch
877 * any data). Only then can we extract a new key with extract_entropy().
879 static bool drain_entropy(void *buf, size_t nbytes, bool force)
881 unsigned int entropy_count;
883 entropy_count = READ_ONCE(input_pool.entropy_count);
884 if (!force && entropy_count < POOL_MIN_BITS)
886 } while (cmpxchg(&input_pool.entropy_count, entropy_count, 0) != entropy_count);
887 extract_entropy(buf, nbytes);
888 wake_up_interruptible(&random_write_wait);
889 kill_fasync(&fasync, SIGIO, POLL_OUT);
894 /**********************************************************************
896 * Entropy collection routines.
898 * The following exported functions are used for pushing entropy into
899 * the above entropy accumulation routines:
901 * void add_device_randomness(const void *buf, size_t size);
902 * void add_input_randomness(unsigned int type, unsigned int code,
903 * unsigned int value);
904 * void add_disk_randomness(struct gendisk *disk);
905 * void add_hwgenerator_randomness(const void *buffer, size_t count,
907 * void add_bootloader_randomness(const void *buf, size_t size);
908 * void add_vmfork_randomness(const void *unique_vm_id, size_t size);
909 * void add_interrupt_randomness(int irq);
911 * add_device_randomness() adds data to the input pool that
912 * is likely to differ between two devices (or possibly even per boot).
913 * This would be things like MAC addresses or serial numbers, or the
914 * read-out of the RTC. This does *not* credit any actual entropy to
915 * the pool, but it initializes the pool to different values for devices
916 * that might otherwise be identical and have very little entropy
917 * available to them (particularly common in the embedded world).
919 * add_input_randomness() uses the input layer interrupt timing, as well
920 * as the event type information from the hardware.
922 * add_disk_randomness() uses what amounts to the seek time of block
923 * layer request events, on a per-disk_devt basis, as input to the
924 * entropy pool. Note that high-speed solid state drives with very low
925 * seek times do not make for good sources of entropy, as their seek
926 * times are usually fairly consistent.
928 * The above two routines try to estimate how many bits of entropy
929 * to credit. They do this by keeping track of the first and second
930 * order deltas of the event timings.
932 * add_hwgenerator_randomness() is for true hardware RNGs, and will credit
933 * entropy as specified by the caller. If the entropy pool is full it will
934 * block until more entropy is needed.
936 * add_bootloader_randomness() is the same as add_hwgenerator_randomness() or
937 * add_device_randomness(), depending on whether or not the configuration
938 * option CONFIG_RANDOM_TRUST_BOOTLOADER is set.
940 * add_vmfork_randomness() adds a unique (but not necessarily secret) ID
941 * representing the current instance of a VM to the pool, without crediting,
942 * and then force-reseeds the crng so that it takes effect immediately.
944 * add_interrupt_randomness() uses the interrupt timing as random
945 * inputs to the entropy pool. Using the cycle counters and the irq source
946 * as inputs, it feeds the input pool roughly once a second or after 64
947 * interrupts, crediting 1 bit of entropy for whichever comes first.
949 **********************************************************************/
951 static bool trust_cpu __ro_after_init = IS_ENABLED(CONFIG_RANDOM_TRUST_CPU);
952 static bool trust_bootloader __ro_after_init = IS_ENABLED(CONFIG_RANDOM_TRUST_BOOTLOADER);
953 static int __init parse_trust_cpu(char *arg)
955 return kstrtobool(arg, &trust_cpu);
957 static int __init parse_trust_bootloader(char *arg)
959 return kstrtobool(arg, &trust_bootloader);
961 early_param("random.trust_cpu", parse_trust_cpu);
962 early_param("random.trust_bootloader", parse_trust_bootloader);
965 * The first collection of entropy occurs at system boot while interrupts
966 * are still turned off. Here we push in RDSEED, a timestamp, and utsname().
967 * Depending on the above configuration knob, RDSEED may be considered
968 * sufficient for initialization. Note that much earlier setup may already
969 * have pushed entropy into the input pool by the time we get here.
971 int __init rand_initialize(void)
974 ktime_t now = ktime_get_real();
975 bool arch_init = true;
978 #if defined(LATENT_ENTROPY_PLUGIN)
979 static const u8 compiletime_seed[BLAKE2S_BLOCK_SIZE] __initconst __latent_entropy;
980 _mix_pool_bytes(compiletime_seed, sizeof(compiletime_seed));
983 for (i = 0; i < BLAKE2S_BLOCK_SIZE; i += sizeof(rv)) {
984 if (!arch_get_random_seed_long_early(&rv) &&
985 !arch_get_random_long_early(&rv)) {
986 rv = random_get_entropy();
989 _mix_pool_bytes(&rv, sizeof(rv));
991 _mix_pool_bytes(&now, sizeof(now));
992 _mix_pool_bytes(utsname(), sizeof(*(utsname())));
994 extract_entropy(base_crng.key, sizeof(base_crng.key));
995 ++base_crng.generation;
997 if (arch_init && trust_cpu && !crng_ready()) {
999 pr_notice("crng init done (trusting CPU's manufacturer)\n");
1002 if (ratelimit_disable) {
1003 urandom_warning.interval = 0;
1004 unseeded_warning.interval = 0;
1010 * Add device- or boot-specific data to the input pool to help
1013 * None of this adds any entropy; it is meant to avoid the problem of
1014 * the entropy pool having similar initial state across largely
1015 * identical devices.
1017 void add_device_randomness(const void *buf, size_t size)
1019 cycles_t cycles = random_get_entropy();
1020 unsigned long flags, now = jiffies;
1022 if (crng_init == 0 && size)
1023 crng_pre_init_inject(buf, size, false);
1025 spin_lock_irqsave(&input_pool.lock, flags);
1026 _mix_pool_bytes(&cycles, sizeof(cycles));
1027 _mix_pool_bytes(&now, sizeof(now));
1028 _mix_pool_bytes(buf, size);
1029 spin_unlock_irqrestore(&input_pool.lock, flags);
1031 EXPORT_SYMBOL(add_device_randomness);
1033 /* There is one of these per entropy source */
1034 struct timer_rand_state {
1035 unsigned long last_time;
1036 long last_delta, last_delta2;
1040 * This function adds entropy to the entropy "pool" by using timing
1041 * delays. It uses the timer_rand_state structure to make an estimate
1042 * of how many bits of entropy this call has added to the pool.
1044 * The number "num" is also added to the pool - it should somehow describe
1045 * the type of event which just happened. This is currently 0-255 for
1046 * keyboard scan codes, and 256 upwards for interrupts.
1048 static void add_timer_randomness(struct timer_rand_state *state, unsigned int num)
1050 cycles_t cycles = random_get_entropy();
1051 unsigned long flags, now = jiffies;
1052 long delta, delta2, delta3;
1054 spin_lock_irqsave(&input_pool.lock, flags);
1055 _mix_pool_bytes(&cycles, sizeof(cycles));
1056 _mix_pool_bytes(&now, sizeof(now));
1057 _mix_pool_bytes(&num, sizeof(num));
1058 spin_unlock_irqrestore(&input_pool.lock, flags);
1061 * Calculate number of bits of randomness we probably added.
1062 * We take into account the first, second and third-order deltas
1063 * in order to make our estimate.
1065 delta = now - READ_ONCE(state->last_time);
1066 WRITE_ONCE(state->last_time, now);
1068 delta2 = delta - READ_ONCE(state->last_delta);
1069 WRITE_ONCE(state->last_delta, delta);
1071 delta3 = delta2 - READ_ONCE(state->last_delta2);
1072 WRITE_ONCE(state->last_delta2, delta2);
1086 * delta is now minimum absolute delta.
1087 * Round down by 1 bit on general principles,
1088 * and limit entropy estimate to 12 bits.
1090 credit_entropy_bits(min_t(unsigned int, fls(delta >> 1), 11));
1093 void add_input_randomness(unsigned int type, unsigned int code,
1096 static unsigned char last_value;
1097 static struct timer_rand_state input_timer_state = { INITIAL_JIFFIES };
1099 /* Ignore autorepeat and the like. */
1100 if (value == last_value)
1104 add_timer_randomness(&input_timer_state,
1105 (type << 4) ^ code ^ (code >> 4) ^ value);
1107 EXPORT_SYMBOL_GPL(add_input_randomness);
1110 void add_disk_randomness(struct gendisk *disk)
1112 if (!disk || !disk->random)
1114 /* First major is 1, so we get >= 0x200 here. */
1115 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1117 EXPORT_SYMBOL_GPL(add_disk_randomness);
1119 void rand_initialize_disk(struct gendisk *disk)
1121 struct timer_rand_state *state;
1124 * If kzalloc returns null, we just won't use that entropy
1127 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1129 state->last_time = INITIAL_JIFFIES;
1130 disk->random = state;
1136 * Interface for in-kernel drivers of true hardware RNGs.
1137 * Those devices may produce endless random bits and will be throttled
1138 * when our pool is full.
1140 void add_hwgenerator_randomness(const void *buffer, size_t count,
1143 if (unlikely(crng_init == 0 && entropy < POOL_MIN_BITS)) {
1144 crng_pre_init_inject(buffer, count, true);
1145 mix_pool_bytes(buffer, count);
1150 * Throttle writing if we're above the trickle threshold.
1151 * We'll be woken up again once below POOL_MIN_BITS, when
1152 * the calling thread is about to terminate, or once
1153 * CRNG_RESEED_INTERVAL has elapsed.
1155 wait_event_interruptible_timeout(random_write_wait,
1156 !system_wq || kthread_should_stop() ||
1157 input_pool.entropy_count < POOL_MIN_BITS,
1158 CRNG_RESEED_INTERVAL);
1159 mix_pool_bytes(buffer, count);
1160 credit_entropy_bits(entropy);
1162 EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
1165 * Handle random seed passed by bootloader.
1166 * If the seed is trustworthy, it would be regarded as hardware RNGs. Otherwise
1167 * it would be regarded as device data.
1168 * The decision is controlled by CONFIG_RANDOM_TRUST_BOOTLOADER.
1170 void add_bootloader_randomness(const void *buf, size_t size)
1172 if (trust_bootloader)
1173 add_hwgenerator_randomness(buf, size, size * 8);
1175 add_device_randomness(buf, size);
1177 EXPORT_SYMBOL_GPL(add_bootloader_randomness);
1179 #if IS_ENABLED(CONFIG_VMGENID)
1180 static BLOCKING_NOTIFIER_HEAD(vmfork_chain);
1183 * Handle a new unique VM ID, which is unique, not secret, so we
1184 * don't credit it, but we do immediately force a reseed after so
1185 * that it's used by the crng posthaste.
1187 void add_vmfork_randomness(const void *unique_vm_id, size_t size)
1189 add_device_randomness(unique_vm_id, size);
1192 pr_notice("crng reseeded due to virtual machine fork\n");
1194 blocking_notifier_call_chain(&vmfork_chain, 0, NULL);
1196 #if IS_MODULE(CONFIG_VMGENID)
1197 EXPORT_SYMBOL_GPL(add_vmfork_randomness);
1200 int register_random_vmfork_notifier(struct notifier_block *nb)
1202 return blocking_notifier_chain_register(&vmfork_chain, nb);
1204 EXPORT_SYMBOL_GPL(register_random_vmfork_notifier);
1206 int unregister_random_vmfork_notifier(struct notifier_block *nb)
1208 return blocking_notifier_chain_unregister(&vmfork_chain, nb);
1210 EXPORT_SYMBOL_GPL(unregister_random_vmfork_notifier);
1214 struct work_struct mix;
1215 unsigned long pool[4];
1221 static DEFINE_PER_CPU(struct fast_pool, irq_randomness) = {
1223 /* SipHash constants */
1224 .pool = { 0x736f6d6570736575UL, 0x646f72616e646f6dUL,
1225 0x6c7967656e657261UL, 0x7465646279746573UL }
1227 /* HalfSipHash constants */
1228 .pool = { 0, 0, 0x6c796765U, 0x74656462U }
1233 * This is [Half]SipHash-1-x, starting from an empty key. Because
1234 * the key is fixed, it assumes that its inputs are non-malicious,
1235 * and therefore this has no security on its own. s represents the
1236 * 128 or 256-bit SipHash state, while v represents a 128-bit input.
1238 static void fast_mix(unsigned long s[4], const unsigned long *v)
1242 for (i = 0; i < 16 / sizeof(long); ++i) {
1245 s[0] += s[1]; s[1] = rol64(s[1], 13); s[1] ^= s[0]; s[0] = rol64(s[0], 32);
1246 s[2] += s[3]; s[3] = rol64(s[3], 16); s[3] ^= s[2];
1247 s[0] += s[3]; s[3] = rol64(s[3], 21); s[3] ^= s[0];
1248 s[2] += s[1]; s[1] = rol64(s[1], 17); s[1] ^= s[2]; s[2] = rol64(s[2], 32);
1250 s[0] += s[1]; s[1] = rol32(s[1], 5); s[1] ^= s[0]; s[0] = rol32(s[0], 16);
1251 s[2] += s[3]; s[3] = rol32(s[3], 8); s[3] ^= s[2];
1252 s[0] += s[3]; s[3] = rol32(s[3], 7); s[3] ^= s[0];
1253 s[2] += s[1]; s[1] = rol32(s[1], 13); s[1] ^= s[2]; s[2] = rol32(s[2], 16);
1261 * This function is called when the CPU has just come online, with
1262 * entry CPUHP_AP_RANDOM_ONLINE, just after CPUHP_AP_WORKQUEUE_ONLINE.
1264 int random_online_cpu(unsigned int cpu)
1267 * During CPU shutdown and before CPU onlining, add_interrupt_
1268 * randomness() may schedule mix_interrupt_randomness(), and
1269 * set the MIX_INFLIGHT flag. However, because the worker can
1270 * be scheduled on a different CPU during this period, that
1271 * flag will never be cleared. For that reason, we zero out
1272 * the flag here, which runs just after workqueues are onlined
1273 * for the CPU again. This also has the effect of setting the
1274 * irq randomness count to zero so that new accumulated irqs
1277 per_cpu_ptr(&irq_randomness, cpu)->count = 0;
1282 static unsigned long get_reg(struct fast_pool *f, struct pt_regs *regs)
1284 unsigned long *ptr = (unsigned long *)regs;
1289 idx = READ_ONCE(f->reg_idx);
1290 if (idx >= sizeof(struct pt_regs) / sizeof(unsigned long))
1293 WRITE_ONCE(f->reg_idx, idx);
1297 static void mix_interrupt_randomness(struct work_struct *work)
1299 struct fast_pool *fast_pool = container_of(work, struct fast_pool, mix);
1301 * The size of the copied stack pool is explicitly 16 bytes so that we
1302 * tax mix_pool_byte()'s compression function the same amount on all
1303 * platforms. This means on 64-bit we copy half the pool into this,
1304 * while on 32-bit we copy all of it. The entropy is supposed to be
1305 * sufficiently dispersed between bits that in the sponge-like
1306 * half case, on average we don't wind up "losing" some.
1310 /* Check to see if we're running on the wrong CPU due to hotplug. */
1311 local_irq_disable();
1312 if (fast_pool != this_cpu_ptr(&irq_randomness)) {
1318 * Copy the pool to the stack so that the mixer always has a
1319 * consistent view, before we reenable irqs again.
1321 memcpy(pool, fast_pool->pool, sizeof(pool));
1322 fast_pool->count = 0;
1323 fast_pool->last = jiffies;
1326 if (unlikely(crng_init == 0)) {
1327 crng_pre_init_inject(pool, sizeof(pool), true);
1328 mix_pool_bytes(pool, sizeof(pool));
1330 mix_pool_bytes(pool, sizeof(pool));
1331 credit_entropy_bits(1);
1334 memzero_explicit(pool, sizeof(pool));
1337 void add_interrupt_randomness(int irq)
1339 enum { MIX_INFLIGHT = 1U << 31 };
1340 cycles_t cycles = random_get_entropy();
1341 unsigned long now = jiffies;
1342 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1343 struct pt_regs *regs = get_irq_regs();
1344 unsigned int new_count;
1348 unsigned long longs[16 / sizeof(long)];
1352 cycles = get_reg(fast_pool, regs);
1354 if (sizeof(cycles) == 8)
1355 irq_data.u64[0] = cycles ^ rol64(now, 32) ^ irq;
1357 irq_data.u32[0] = cycles ^ irq;
1358 irq_data.u32[1] = now;
1361 if (sizeof(unsigned long) == 8)
1362 irq_data.u64[1] = regs ? instruction_pointer(regs) : _RET_IP_;
1364 irq_data.u32[2] = regs ? instruction_pointer(regs) : _RET_IP_;
1365 irq_data.u32[3] = get_reg(fast_pool, regs);
1368 fast_mix(fast_pool->pool, irq_data.longs);
1369 new_count = ++fast_pool->count;
1371 if (new_count & MIX_INFLIGHT)
1374 if (new_count < 64 && (!time_after(now, fast_pool->last + HZ) ||
1375 unlikely(crng_init == 0)))
1378 if (unlikely(!fast_pool->mix.func))
1379 INIT_WORK(&fast_pool->mix, mix_interrupt_randomness);
1380 fast_pool->count |= MIX_INFLIGHT;
1381 queue_work_on(raw_smp_processor_id(), system_highpri_wq, &fast_pool->mix);
1383 EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1386 * Each time the timer fires, we expect that we got an unpredictable
1387 * jump in the cycle counter. Even if the timer is running on another
1388 * CPU, the timer activity will be touching the stack of the CPU that is
1389 * generating entropy..
1391 * Note that we don't re-arm the timer in the timer itself - we are
1392 * happy to be scheduled away, since that just makes the load more
1393 * complex, but we do not want the timer to keep ticking unless the
1394 * entropy loop is running.
1396 * So the re-arming always happens in the entropy loop itself.
1398 static void entropy_timer(struct timer_list *t)
1400 credit_entropy_bits(1);
1404 * If we have an actual cycle counter, see if we can
1405 * generate enough entropy with timing noise
1407 static void try_to_generate_entropy(void)
1411 struct timer_list timer;
1414 stack.cycles = random_get_entropy();
1416 /* Slow counter - or none. Don't even bother */
1417 if (stack.cycles == random_get_entropy())
1420 timer_setup_on_stack(&stack.timer, entropy_timer, 0);
1421 while (!crng_ready() && !signal_pending(current)) {
1422 if (!timer_pending(&stack.timer))
1423 mod_timer(&stack.timer, jiffies + 1);
1424 mix_pool_bytes(&stack.cycles, sizeof(stack.cycles));
1426 stack.cycles = random_get_entropy();
1429 del_timer_sync(&stack.timer);
1430 destroy_timer_on_stack(&stack.timer);
1431 mix_pool_bytes(&stack.cycles, sizeof(stack.cycles));
1435 /**********************************************************************
1437 * Userspace reader/writer interfaces.
1439 * getrandom(2) is the primary modern interface into the RNG and should
1440 * be used in preference to anything else.
1442 * Reading from /dev/random has the same functionality as calling
1443 * getrandom(2) with flags=0. In earlier versions, however, it had
1444 * vastly different semantics and should therefore be avoided, to
1445 * prevent backwards compatibility issues.
1447 * Reading from /dev/urandom has the same functionality as calling
1448 * getrandom(2) with flags=GRND_INSECURE. Because it does not block
1449 * waiting for the RNG to be ready, it should not be used.
1451 * Writing to either /dev/random or /dev/urandom adds entropy to
1452 * the input pool but does not credit it.
1454 * Polling on /dev/random indicates when the RNG is initialized, on
1455 * the read side, and when it wants new entropy, on the write side.
1457 * Both /dev/random and /dev/urandom have the same set of ioctls for
1458 * adding entropy, getting the entropy count, zeroing the count, and
1459 * reseeding the crng.
1461 **********************************************************************/
1463 SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count, unsigned int,
1466 if (flags & ~(GRND_NONBLOCK | GRND_RANDOM | GRND_INSECURE))
1470 * Requesting insecure and blocking randomness at the same time makes
1473 if ((flags & (GRND_INSECURE | GRND_RANDOM)) == (GRND_INSECURE | GRND_RANDOM))
1476 if (count > INT_MAX)
1479 if (!(flags & GRND_INSECURE) && !crng_ready()) {
1482 if (flags & GRND_NONBLOCK)
1484 ret = wait_for_random_bytes();
1488 return get_random_bytes_user(buf, count);
1491 static __poll_t random_poll(struct file *file, poll_table *wait)
1495 poll_wait(file, &crng_init_wait, wait);
1496 poll_wait(file, &random_write_wait, wait);
1499 mask |= EPOLLIN | EPOLLRDNORM;
1500 if (input_pool.entropy_count < POOL_MIN_BITS)
1501 mask |= EPOLLOUT | EPOLLWRNORM;
1505 static int write_pool(const char __user *ubuf, size_t count)
1509 u8 block[BLAKE2S_BLOCK_SIZE];
1512 len = min(count, sizeof(block));
1513 if (copy_from_user(block, ubuf, len)) {
1519 mix_pool_bytes(block, len);
1524 memzero_explicit(block, sizeof(block));
1528 static ssize_t random_write(struct file *file, const char __user *buffer,
1529 size_t count, loff_t *ppos)
1533 ret = write_pool(buffer, count);
1537 return (ssize_t)count;
1540 static ssize_t urandom_read(struct file *file, char __user *buf, size_t nbytes,
1543 static int maxwarn = 10;
1546 * Opportunistically attempt to initialize the RNG on platforms that
1547 * have fast cycle counters, but don't (for now) require it to succeed.
1550 try_to_generate_entropy();
1552 if (!crng_ready() && maxwarn > 0) {
1554 if (__ratelimit(&urandom_warning))
1555 pr_notice("%s: uninitialized urandom read (%zd bytes read)\n",
1556 current->comm, nbytes);
1559 return get_random_bytes_user(buf, nbytes);
1562 static ssize_t random_read(struct file *file, char __user *buf, size_t nbytes,
1567 ret = wait_for_random_bytes();
1570 return get_random_bytes_user(buf, nbytes);
1573 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1575 int size, ent_count;
1576 int __user *p = (int __user *)arg;
1581 /* Inherently racy, no point locking. */
1582 if (put_user(input_pool.entropy_count, p))
1585 case RNDADDTOENTCNT:
1586 if (!capable(CAP_SYS_ADMIN))
1588 if (get_user(ent_count, p))
1592 credit_entropy_bits(ent_count);
1595 if (!capable(CAP_SYS_ADMIN))
1597 if (get_user(ent_count, p++))
1601 if (get_user(size, p++))
1603 retval = write_pool((const char __user *)p, size);
1606 credit_entropy_bits(ent_count);
1611 * Clear the entropy pool counters. We no longer clear
1612 * the entropy pool, as that's silly.
1614 if (!capable(CAP_SYS_ADMIN))
1616 if (xchg(&input_pool.entropy_count, 0) >= POOL_MIN_BITS) {
1617 wake_up_interruptible(&random_write_wait);
1618 kill_fasync(&fasync, SIGIO, POLL_OUT);
1622 if (!capable(CAP_SYS_ADMIN))
1633 static int random_fasync(int fd, struct file *filp, int on)
1635 return fasync_helper(fd, filp, on, &fasync);
1638 const struct file_operations random_fops = {
1639 .read = random_read,
1640 .write = random_write,
1641 .poll = random_poll,
1642 .unlocked_ioctl = random_ioctl,
1643 .compat_ioctl = compat_ptr_ioctl,
1644 .fasync = random_fasync,
1645 .llseek = noop_llseek,
1648 const struct file_operations urandom_fops = {
1649 .read = urandom_read,
1650 .write = random_write,
1651 .unlocked_ioctl = random_ioctl,
1652 .compat_ioctl = compat_ptr_ioctl,
1653 .fasync = random_fasync,
1654 .llseek = noop_llseek,
1658 /********************************************************************
1662 * These are partly unused legacy knobs with dummy values to not break
1663 * userspace and partly still useful things. They are usually accessible
1664 * in /proc/sys/kernel/random/ and are as follows:
1666 * - boot_id - a UUID representing the current boot.
1668 * - uuid - a random UUID, different each time the file is read.
1670 * - poolsize - the number of bits of entropy that the input pool can
1671 * hold, tied to the POOL_BITS constant.
1673 * - entropy_avail - the number of bits of entropy currently in the
1674 * input pool. Always <= poolsize.
1676 * - write_wakeup_threshold - the amount of entropy in the input pool
1677 * below which write polls to /dev/random will unblock, requesting
1678 * more entropy, tied to the POOL_MIN_BITS constant. It is writable
1679 * to avoid breaking old userspaces, but writing to it does not
1680 * change any behavior of the RNG.
1682 * - urandom_min_reseed_secs - fixed to the value CRNG_RESEED_INTERVAL.
1683 * It is writable to avoid breaking old userspaces, but writing
1684 * to it does not change any behavior of the RNG.
1686 ********************************************************************/
1688 #ifdef CONFIG_SYSCTL
1690 #include <linux/sysctl.h>
1692 static int sysctl_random_min_urandom_seed = CRNG_RESEED_INTERVAL / HZ;
1693 static int sysctl_random_write_wakeup_bits = POOL_MIN_BITS;
1694 static int sysctl_poolsize = POOL_BITS;
1695 static u8 sysctl_bootid[UUID_SIZE];
1698 * This function is used to return both the bootid UUID, and random
1699 * UUID. The difference is in whether table->data is NULL; if it is,
1700 * then a new UUID is generated and returned to the user.
1702 static int proc_do_uuid(struct ctl_table *table, int write, void *buffer,
1703 size_t *lenp, loff_t *ppos)
1705 u8 tmp_uuid[UUID_SIZE], *uuid;
1706 char uuid_string[UUID_STRING_LEN + 1];
1707 struct ctl_table fake_table = {
1708 .data = uuid_string,
1709 .maxlen = UUID_STRING_LEN
1718 generate_random_uuid(uuid);
1720 static DEFINE_SPINLOCK(bootid_spinlock);
1722 spin_lock(&bootid_spinlock);
1724 generate_random_uuid(uuid);
1725 spin_unlock(&bootid_spinlock);
1728 snprintf(uuid_string, sizeof(uuid_string), "%pU", uuid);
1729 return proc_dostring(&fake_table, 0, buffer, lenp, ppos);
1732 /* The same as proc_dointvec, but writes don't change anything. */
1733 static int proc_do_rointvec(struct ctl_table *table, int write, void *buffer,
1734 size_t *lenp, loff_t *ppos)
1736 return write ? 0 : proc_dointvec(table, 0, buffer, lenp, ppos);
1739 static struct ctl_table random_table[] = {
1741 .procname = "poolsize",
1742 .data = &sysctl_poolsize,
1743 .maxlen = sizeof(int),
1745 .proc_handler = proc_dointvec,
1748 .procname = "entropy_avail",
1749 .data = &input_pool.entropy_count,
1750 .maxlen = sizeof(int),
1752 .proc_handler = proc_dointvec,
1755 .procname = "write_wakeup_threshold",
1756 .data = &sysctl_random_write_wakeup_bits,
1757 .maxlen = sizeof(int),
1759 .proc_handler = proc_do_rointvec,
1762 .procname = "urandom_min_reseed_secs",
1763 .data = &sysctl_random_min_urandom_seed,
1764 .maxlen = sizeof(int),
1766 .proc_handler = proc_do_rointvec,
1769 .procname = "boot_id",
1770 .data = &sysctl_bootid,
1772 .proc_handler = proc_do_uuid,
1777 .proc_handler = proc_do_uuid,
1783 * rand_initialize() is called before sysctl_init(),
1784 * so we cannot call register_sysctl_init() in rand_initialize()
1786 static int __init random_sysctls_init(void)
1788 register_sysctl_init("kernel/random", random_table);
1791 device_initcall(random_sysctls_init);