xref: /linux/drivers/char/random.c (revision 31354121bf03dac6498a4236928a38490745d601)
1 // SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause)
2 /*
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.
6  *
7  * This driver produces cryptographically secure pseudorandom data. It is divided
8  * into roughly six sections, each with a section header:
9  *
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.
15  *   - Sysctl interface.
16  *
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.
24  */
25 
26 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
27 
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>
38 #include <linux/fs.h>
39 #include <linux/blkdev.h>
40 #include <linux/interrupt.h>
41 #include <linux/mm.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/suspend.h>
55 #include <linux/siphash.h>
56 #include <crypto/chacha.h>
57 #include <crypto/blake2s.h>
58 #include <asm/processor.h>
59 #include <asm/irq.h>
60 #include <asm/irq_regs.h>
61 #include <asm/io.h>
62 
63 /*********************************************************************
64  *
65  * Initialization and readiness waiting.
66  *
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.
70  *
71  *********************************************************************/
72 
73 /*
74  * crng_init is protected by base_crng->lock, and only increases
75  * its value (from empty->early->ready).
76  */
77 static enum {
78 	CRNG_EMPTY = 0, /* Little to no entropy collected */
79 	CRNG_EARLY = 1, /* At least POOL_EARLY_BITS collected */
80 	CRNG_READY = 2  /* Fully initialized with POOL_READY_BITS collected */
81 } crng_init __read_mostly = CRNG_EMPTY;
82 static DEFINE_STATIC_KEY_FALSE(crng_is_ready);
83 #define crng_ready() (static_branch_likely(&crng_is_ready) || crng_init >= CRNG_READY)
84 /* Various types of waiters for crng_init->CRNG_READY transition. */
85 static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
86 static struct fasync_struct *fasync;
87 
88 /* Control how we warn userspace. */
89 static struct ratelimit_state urandom_warning =
90 	RATELIMIT_STATE_INIT_FLAGS("urandom_warning", HZ, 3, RATELIMIT_MSG_ON_RELEASE);
91 static int ratelimit_disable __read_mostly =
92 	IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM);
93 module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
94 MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
95 
96 /*
97  * Returns whether or not the input pool has been seeded and thus guaranteed
98  * to supply cryptographically secure random numbers. This applies to: the
99  * /dev/urandom device, the get_random_bytes function, and the get_random_{u8,
100  * u16,u32,u64,long} family of functions.
101  *
102  * Returns: true if the input pool has been seeded.
103  *          false if the input pool has not been seeded.
104  */
105 bool rng_is_initialized(void)
106 {
107 	return crng_ready();
108 }
109 EXPORT_SYMBOL(rng_is_initialized);
110 
111 static void __cold crng_set_ready(struct work_struct *work)
112 {
113 	static_branch_enable(&crng_is_ready);
114 }
115 
116 /* Used by wait_for_random_bytes(), and considered an entropy collector, below. */
117 static void try_to_generate_entropy(void);
118 
119 /*
120  * Wait for the input pool to be seeded and thus guaranteed to supply
121  * cryptographically secure random numbers. This applies to: the /dev/urandom
122  * device, the get_random_bytes function, and the get_random_{u8,u16,u32,u64,
123  * int,long} family of functions. Using any of these functions without first
124  * calling this function forfeits the guarantee of security.
125  *
126  * Returns: 0 if the input pool has been seeded.
127  *          -ERESTARTSYS if the function was interrupted by a signal.
128  */
129 int wait_for_random_bytes(void)
130 {
131 	while (!crng_ready()) {
132 		int ret;
133 
134 		try_to_generate_entropy();
135 		ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ);
136 		if (ret)
137 			return ret > 0 ? 0 : ret;
138 	}
139 	return 0;
140 }
141 EXPORT_SYMBOL(wait_for_random_bytes);
142 
143 #define warn_unseeded_randomness() \
144 	if (IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM) && !crng_ready()) \
145 		printk_deferred(KERN_NOTICE "random: %s called from %pS with crng_init=%d\n", \
146 				__func__, (void *)_RET_IP_, crng_init)
147 
148 
149 /*********************************************************************
150  *
151  * Fast key erasure RNG, the "crng".
152  *
153  * These functions expand entropy from the entropy extractor into
154  * long streams for external consumption using the "fast key erasure"
155  * RNG described at <https://blog.cr.yp.to/20170723-random.html>.
156  *
157  * There are a few exported interfaces for use by other drivers:
158  *
159  *	void get_random_bytes(void *buf, size_t len)
160  *	u8 get_random_u8()
161  *	u16 get_random_u16()
162  *	u32 get_random_u32()
163  *	u64 get_random_u64()
164  *	unsigned long get_random_long()
165  *
166  * These interfaces will return the requested number of random bytes
167  * into the given buffer or as a return value. This is equivalent to
168  * a read from /dev/urandom. The u8, u16, u32, u64, long family of
169  * functions may be higher performance for one-off random integers,
170  * because they do a bit of buffering and do not invoke reseeding
171  * until the buffer is emptied.
172  *
173  *********************************************************************/
174 
175 enum {
176 	CRNG_RESEED_START_INTERVAL = HZ,
177 	CRNG_RESEED_INTERVAL = 60 * HZ
178 };
179 
180 static struct {
181 	u8 key[CHACHA_KEY_SIZE] __aligned(__alignof__(long));
182 	unsigned long birth;
183 	unsigned long generation;
184 	spinlock_t lock;
185 } base_crng = {
186 	.lock = __SPIN_LOCK_UNLOCKED(base_crng.lock)
187 };
188 
189 struct crng {
190 	u8 key[CHACHA_KEY_SIZE];
191 	unsigned long generation;
192 	local_lock_t lock;
193 };
194 
195 static DEFINE_PER_CPU(struct crng, crngs) = {
196 	.generation = ULONG_MAX,
197 	.lock = INIT_LOCAL_LOCK(crngs.lock),
198 };
199 
200 /* Used by crng_reseed() and crng_make_state() to extract a new seed from the input pool. */
201 static void extract_entropy(void *buf, size_t len);
202 
203 /* This extracts a new crng key from the input pool. */
204 static void crng_reseed(void)
205 {
206 	unsigned long flags;
207 	unsigned long next_gen;
208 	u8 key[CHACHA_KEY_SIZE];
209 
210 	extract_entropy(key, sizeof(key));
211 
212 	/*
213 	 * We copy the new key into the base_crng, overwriting the old one,
214 	 * and update the generation counter. We avoid hitting ULONG_MAX,
215 	 * because the per-cpu crngs are initialized to ULONG_MAX, so this
216 	 * forces new CPUs that come online to always initialize.
217 	 */
218 	spin_lock_irqsave(&base_crng.lock, flags);
219 	memcpy(base_crng.key, key, sizeof(base_crng.key));
220 	next_gen = base_crng.generation + 1;
221 	if (next_gen == ULONG_MAX)
222 		++next_gen;
223 	WRITE_ONCE(base_crng.generation, next_gen);
224 	WRITE_ONCE(base_crng.birth, jiffies);
225 	if (!static_branch_likely(&crng_is_ready))
226 		crng_init = CRNG_READY;
227 	spin_unlock_irqrestore(&base_crng.lock, flags);
228 	memzero_explicit(key, sizeof(key));
229 }
230 
231 /*
232  * This generates a ChaCha block using the provided key, and then
233  * immediately overwrites that key with half the block. It returns
234  * the resultant ChaCha state to the user, along with the second
235  * half of the block containing 32 bytes of random data that may
236  * be used; random_data_len may not be greater than 32.
237  *
238  * The returned ChaCha state contains within it a copy of the old
239  * key value, at index 4, so the state should always be zeroed out
240  * immediately after using in order to maintain forward secrecy.
241  * If the state cannot be erased in a timely manner, then it is
242  * safer to set the random_data parameter to &chacha_state[4] so
243  * that this function overwrites it before returning.
244  */
245 static void crng_fast_key_erasure(u8 key[CHACHA_KEY_SIZE],
246 				  u32 chacha_state[CHACHA_STATE_WORDS],
247 				  u8 *random_data, size_t random_data_len)
248 {
249 	u8 first_block[CHACHA_BLOCK_SIZE];
250 
251 	BUG_ON(random_data_len > 32);
252 
253 	chacha_init_consts(chacha_state);
254 	memcpy(&chacha_state[4], key, CHACHA_KEY_SIZE);
255 	memset(&chacha_state[12], 0, sizeof(u32) * 4);
256 	chacha20_block(chacha_state, first_block);
257 
258 	memcpy(key, first_block, CHACHA_KEY_SIZE);
259 	memcpy(random_data, first_block + CHACHA_KEY_SIZE, random_data_len);
260 	memzero_explicit(first_block, sizeof(first_block));
261 }
262 
263 /*
264  * Return the interval until the next reseeding, which is normally
265  * CRNG_RESEED_INTERVAL, but during early boot, it is at an interval
266  * proportional to the uptime.
267  */
268 static unsigned int crng_reseed_interval(void)
269 {
270 	static bool early_boot = true;
271 
272 	if (unlikely(READ_ONCE(early_boot))) {
273 		time64_t uptime = ktime_get_seconds();
274 		if (uptime >= CRNG_RESEED_INTERVAL / HZ * 2)
275 			WRITE_ONCE(early_boot, false);
276 		else
277 			return max_t(unsigned int, CRNG_RESEED_START_INTERVAL,
278 				     (unsigned int)uptime / 2 * HZ);
279 	}
280 	return CRNG_RESEED_INTERVAL;
281 }
282 
283 /*
284  * This function returns a ChaCha state that you may use for generating
285  * random data. It also returns up to 32 bytes on its own of random data
286  * that may be used; random_data_len may not be greater than 32.
287  */
288 static void crng_make_state(u32 chacha_state[CHACHA_STATE_WORDS],
289 			    u8 *random_data, size_t random_data_len)
290 {
291 	unsigned long flags;
292 	struct crng *crng;
293 
294 	BUG_ON(random_data_len > 32);
295 
296 	/*
297 	 * For the fast path, we check whether we're ready, unlocked first, and
298 	 * then re-check once locked later. In the case where we're really not
299 	 * ready, we do fast key erasure with the base_crng directly, extracting
300 	 * when crng_init is CRNG_EMPTY.
301 	 */
302 	if (!crng_ready()) {
303 		bool ready;
304 
305 		spin_lock_irqsave(&base_crng.lock, flags);
306 		ready = crng_ready();
307 		if (!ready) {
308 			if (crng_init == CRNG_EMPTY)
309 				extract_entropy(base_crng.key, sizeof(base_crng.key));
310 			crng_fast_key_erasure(base_crng.key, chacha_state,
311 					      random_data, random_data_len);
312 		}
313 		spin_unlock_irqrestore(&base_crng.lock, flags);
314 		if (!ready)
315 			return;
316 	}
317 
318 	/*
319 	 * If the base_crng is old enough, we reseed, which in turn bumps the
320 	 * generation counter that we check below.
321 	 */
322 	if (unlikely(time_is_before_jiffies(READ_ONCE(base_crng.birth) + crng_reseed_interval())))
323 		crng_reseed();
324 
325 	local_lock_irqsave(&crngs.lock, flags);
326 	crng = raw_cpu_ptr(&crngs);
327 
328 	/*
329 	 * If our per-cpu crng is older than the base_crng, then it means
330 	 * somebody reseeded the base_crng. In that case, we do fast key
331 	 * erasure on the base_crng, and use its output as the new key
332 	 * for our per-cpu crng. This brings us up to date with base_crng.
333 	 */
334 	if (unlikely(crng->generation != READ_ONCE(base_crng.generation))) {
335 		spin_lock(&base_crng.lock);
336 		crng_fast_key_erasure(base_crng.key, chacha_state,
337 				      crng->key, sizeof(crng->key));
338 		crng->generation = base_crng.generation;
339 		spin_unlock(&base_crng.lock);
340 	}
341 
342 	/*
343 	 * Finally, when we've made it this far, our per-cpu crng has an up
344 	 * to date key, and we can do fast key erasure with it to produce
345 	 * some random data and a ChaCha state for the caller. All other
346 	 * branches of this function are "unlikely", so most of the time we
347 	 * should wind up here immediately.
348 	 */
349 	crng_fast_key_erasure(crng->key, chacha_state, random_data, random_data_len);
350 	local_unlock_irqrestore(&crngs.lock, flags);
351 }
352 
353 static void _get_random_bytes(void *buf, size_t len)
354 {
355 	u32 chacha_state[CHACHA_STATE_WORDS];
356 	u8 tmp[CHACHA_BLOCK_SIZE];
357 	size_t first_block_len;
358 
359 	if (!len)
360 		return;
361 
362 	first_block_len = min_t(size_t, 32, len);
363 	crng_make_state(chacha_state, buf, first_block_len);
364 	len -= first_block_len;
365 	buf += first_block_len;
366 
367 	while (len) {
368 		if (len < CHACHA_BLOCK_SIZE) {
369 			chacha20_block(chacha_state, tmp);
370 			memcpy(buf, tmp, len);
371 			memzero_explicit(tmp, sizeof(tmp));
372 			break;
373 		}
374 
375 		chacha20_block(chacha_state, buf);
376 		if (unlikely(chacha_state[12] == 0))
377 			++chacha_state[13];
378 		len -= CHACHA_BLOCK_SIZE;
379 		buf += CHACHA_BLOCK_SIZE;
380 	}
381 
382 	memzero_explicit(chacha_state, sizeof(chacha_state));
383 }
384 
385 /*
386  * This function is the exported kernel interface. It returns some number of
387  * good random numbers, suitable for key generation, seeding TCP sequence
388  * numbers, etc. In order to ensure that the randomness returned by this
389  * function is okay, the function wait_for_random_bytes() should be called and
390  * return 0 at least once at any point prior.
391  */
392 void get_random_bytes(void *buf, size_t len)
393 {
394 	warn_unseeded_randomness();
395 	_get_random_bytes(buf, len);
396 }
397 EXPORT_SYMBOL(get_random_bytes);
398 
399 static ssize_t get_random_bytes_user(struct iov_iter *iter)
400 {
401 	u32 chacha_state[CHACHA_STATE_WORDS];
402 	u8 block[CHACHA_BLOCK_SIZE];
403 	size_t ret = 0, copied;
404 
405 	if (unlikely(!iov_iter_count(iter)))
406 		return 0;
407 
408 	/*
409 	 * Immediately overwrite the ChaCha key at index 4 with random
410 	 * bytes, in case userspace causes copy_to_iter() below to sleep
411 	 * forever, so that we still retain forward secrecy in that case.
412 	 */
413 	crng_make_state(chacha_state, (u8 *)&chacha_state[4], CHACHA_KEY_SIZE);
414 	/*
415 	 * However, if we're doing a read of len <= 32, we don't need to
416 	 * use chacha_state after, so we can simply return those bytes to
417 	 * the user directly.
418 	 */
419 	if (iov_iter_count(iter) <= CHACHA_KEY_SIZE) {
420 		ret = copy_to_iter(&chacha_state[4], CHACHA_KEY_SIZE, iter);
421 		goto out_zero_chacha;
422 	}
423 
424 	for (;;) {
425 		chacha20_block(chacha_state, block);
426 		if (unlikely(chacha_state[12] == 0))
427 			++chacha_state[13];
428 
429 		copied = copy_to_iter(block, sizeof(block), iter);
430 		ret += copied;
431 		if (!iov_iter_count(iter) || copied != sizeof(block))
432 			break;
433 
434 		BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0);
435 		if (ret % PAGE_SIZE == 0) {
436 			if (signal_pending(current))
437 				break;
438 			cond_resched();
439 		}
440 	}
441 
442 	memzero_explicit(block, sizeof(block));
443 out_zero_chacha:
444 	memzero_explicit(chacha_state, sizeof(chacha_state));
445 	return ret ? ret : -EFAULT;
446 }
447 
448 /*
449  * Batched entropy returns random integers. The quality of the random
450  * number is good as /dev/urandom. In order to ensure that the randomness
451  * provided by this function is okay, the function wait_for_random_bytes()
452  * should be called and return 0 at least once at any point prior.
453  */
454 
455 #define DEFINE_BATCHED_ENTROPY(type)						\
456 struct batch_ ##type {								\
457 	/*									\
458 	 * We make this 1.5x a ChaCha block, so that we get the			\
459 	 * remaining 32 bytes from fast key erasure, plus one full		\
460 	 * block from the detached ChaCha state. We can increase		\
461 	 * the size of this later if needed so long as we keep the		\
462 	 * formula of (integer_blocks + 0.5) * CHACHA_BLOCK_SIZE.		\
463 	 */									\
464 	type entropy[CHACHA_BLOCK_SIZE * 3 / (2 * sizeof(type))];		\
465 	local_lock_t lock;							\
466 	unsigned long generation;						\
467 	unsigned int position;							\
468 };										\
469 										\
470 static DEFINE_PER_CPU(struct batch_ ##type, batched_entropy_ ##type) = {	\
471 	.lock = INIT_LOCAL_LOCK(batched_entropy_ ##type.lock),			\
472 	.position = UINT_MAX							\
473 };										\
474 										\
475 type get_random_ ##type(void)							\
476 {										\
477 	type ret;								\
478 	unsigned long flags;							\
479 	struct batch_ ##type *batch;						\
480 	unsigned long next_gen;							\
481 										\
482 	warn_unseeded_randomness();						\
483 										\
484 	if  (!crng_ready()) {							\
485 		_get_random_bytes(&ret, sizeof(ret));				\
486 		return ret;							\
487 	}									\
488 										\
489 	local_lock_irqsave(&batched_entropy_ ##type.lock, flags);		\
490 	batch = raw_cpu_ptr(&batched_entropy_##type);				\
491 										\
492 	next_gen = READ_ONCE(base_crng.generation);				\
493 	if (batch->position >= ARRAY_SIZE(batch->entropy) ||			\
494 	    next_gen != batch->generation) {					\
495 		_get_random_bytes(batch->entropy, sizeof(batch->entropy));	\
496 		batch->position = 0;						\
497 		batch->generation = next_gen;					\
498 	}									\
499 										\
500 	ret = batch->entropy[batch->position];					\
501 	batch->entropy[batch->position] = 0;					\
502 	++batch->position;							\
503 	local_unlock_irqrestore(&batched_entropy_ ##type.lock, flags);		\
504 	return ret;								\
505 }										\
506 EXPORT_SYMBOL(get_random_ ##type);
507 
508 DEFINE_BATCHED_ENTROPY(u8)
509 DEFINE_BATCHED_ENTROPY(u16)
510 DEFINE_BATCHED_ENTROPY(u32)
511 DEFINE_BATCHED_ENTROPY(u64)
512 
513 #ifdef CONFIG_SMP
514 /*
515  * This function is called when the CPU is coming up, with entry
516  * CPUHP_RANDOM_PREPARE, which comes before CPUHP_WORKQUEUE_PREP.
517  */
518 int __cold random_prepare_cpu(unsigned int cpu)
519 {
520 	/*
521 	 * When the cpu comes back online, immediately invalidate both
522 	 * the per-cpu crng and all batches, so that we serve fresh
523 	 * randomness.
524 	 */
525 	per_cpu_ptr(&crngs, cpu)->generation = ULONG_MAX;
526 	per_cpu_ptr(&batched_entropy_u8, cpu)->position = UINT_MAX;
527 	per_cpu_ptr(&batched_entropy_u16, cpu)->position = UINT_MAX;
528 	per_cpu_ptr(&batched_entropy_u32, cpu)->position = UINT_MAX;
529 	per_cpu_ptr(&batched_entropy_u64, cpu)->position = UINT_MAX;
530 	return 0;
531 }
532 #endif
533 
534 
535 /**********************************************************************
536  *
537  * Entropy accumulation and extraction routines.
538  *
539  * Callers may add entropy via:
540  *
541  *     static void mix_pool_bytes(const void *buf, size_t len)
542  *
543  * After which, if added entropy should be credited:
544  *
545  *     static void credit_init_bits(size_t bits)
546  *
547  * Finally, extract entropy via:
548  *
549  *     static void extract_entropy(void *buf, size_t len)
550  *
551  **********************************************************************/
552 
553 enum {
554 	POOL_BITS = BLAKE2S_HASH_SIZE * 8,
555 	POOL_READY_BITS = POOL_BITS, /* When crng_init->CRNG_READY */
556 	POOL_EARLY_BITS = POOL_READY_BITS / 2 /* When crng_init->CRNG_EARLY */
557 };
558 
559 static struct {
560 	struct blake2s_state hash;
561 	spinlock_t lock;
562 	unsigned int init_bits;
563 } input_pool = {
564 	.hash.h = { BLAKE2S_IV0 ^ (0x01010000 | BLAKE2S_HASH_SIZE),
565 		    BLAKE2S_IV1, BLAKE2S_IV2, BLAKE2S_IV3, BLAKE2S_IV4,
566 		    BLAKE2S_IV5, BLAKE2S_IV6, BLAKE2S_IV7 },
567 	.hash.outlen = BLAKE2S_HASH_SIZE,
568 	.lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
569 };
570 
571 static void _mix_pool_bytes(const void *buf, size_t len)
572 {
573 	blake2s_update(&input_pool.hash, buf, len);
574 }
575 
576 /*
577  * This function adds bytes into the input pool. It does not
578  * update the initialization bit counter; the caller should call
579  * credit_init_bits if this is appropriate.
580  */
581 static void mix_pool_bytes(const void *buf, size_t len)
582 {
583 	unsigned long flags;
584 
585 	spin_lock_irqsave(&input_pool.lock, flags);
586 	_mix_pool_bytes(buf, len);
587 	spin_unlock_irqrestore(&input_pool.lock, flags);
588 }
589 
590 /*
591  * This is an HKDF-like construction for using the hashed collected entropy
592  * as a PRF key, that's then expanded block-by-block.
593  */
594 static void extract_entropy(void *buf, size_t len)
595 {
596 	unsigned long flags;
597 	u8 seed[BLAKE2S_HASH_SIZE], next_key[BLAKE2S_HASH_SIZE];
598 	struct {
599 		unsigned long rdseed[32 / sizeof(long)];
600 		size_t counter;
601 	} block;
602 	size_t i, longs;
603 
604 	for (i = 0; i < ARRAY_SIZE(block.rdseed);) {
605 		longs = arch_get_random_seed_longs(&block.rdseed[i], ARRAY_SIZE(block.rdseed) - i);
606 		if (longs) {
607 			i += longs;
608 			continue;
609 		}
610 		longs = arch_get_random_longs(&block.rdseed[i], ARRAY_SIZE(block.rdseed) - i);
611 		if (longs) {
612 			i += longs;
613 			continue;
614 		}
615 		block.rdseed[i++] = random_get_entropy();
616 	}
617 
618 	spin_lock_irqsave(&input_pool.lock, flags);
619 
620 	/* seed = HASHPRF(last_key, entropy_input) */
621 	blake2s_final(&input_pool.hash, seed);
622 
623 	/* next_key = HASHPRF(seed, RDSEED || 0) */
624 	block.counter = 0;
625 	blake2s(next_key, (u8 *)&block, seed, sizeof(next_key), sizeof(block), sizeof(seed));
626 	blake2s_init_key(&input_pool.hash, BLAKE2S_HASH_SIZE, next_key, sizeof(next_key));
627 
628 	spin_unlock_irqrestore(&input_pool.lock, flags);
629 	memzero_explicit(next_key, sizeof(next_key));
630 
631 	while (len) {
632 		i = min_t(size_t, len, BLAKE2S_HASH_SIZE);
633 		/* output = HASHPRF(seed, RDSEED || ++counter) */
634 		++block.counter;
635 		blake2s(buf, (u8 *)&block, seed, i, sizeof(block), sizeof(seed));
636 		len -= i;
637 		buf += i;
638 	}
639 
640 	memzero_explicit(seed, sizeof(seed));
641 	memzero_explicit(&block, sizeof(block));
642 }
643 
644 #define credit_init_bits(bits) if (!crng_ready()) _credit_init_bits(bits)
645 
646 static void __cold _credit_init_bits(size_t bits)
647 {
648 	static struct execute_work set_ready;
649 	unsigned int new, orig, add;
650 	unsigned long flags;
651 
652 	if (!bits)
653 		return;
654 
655 	add = min_t(size_t, bits, POOL_BITS);
656 
657 	orig = READ_ONCE(input_pool.init_bits);
658 	do {
659 		new = min_t(unsigned int, POOL_BITS, orig + add);
660 	} while (!try_cmpxchg(&input_pool.init_bits, &orig, new));
661 
662 	if (orig < POOL_READY_BITS && new >= POOL_READY_BITS) {
663 		crng_reseed(); /* Sets crng_init to CRNG_READY under base_crng.lock. */
664 		if (static_key_initialized)
665 			execute_in_process_context(crng_set_ready, &set_ready);
666 		wake_up_interruptible(&crng_init_wait);
667 		kill_fasync(&fasync, SIGIO, POLL_IN);
668 		pr_notice("crng init done\n");
669 		if (urandom_warning.missed)
670 			pr_notice("%d urandom warning(s) missed due to ratelimiting\n",
671 				  urandom_warning.missed);
672 	} else if (orig < POOL_EARLY_BITS && new >= POOL_EARLY_BITS) {
673 		spin_lock_irqsave(&base_crng.lock, flags);
674 		/* Check if crng_init is CRNG_EMPTY, to avoid race with crng_reseed(). */
675 		if (crng_init == CRNG_EMPTY) {
676 			extract_entropy(base_crng.key, sizeof(base_crng.key));
677 			crng_init = CRNG_EARLY;
678 		}
679 		spin_unlock_irqrestore(&base_crng.lock, flags);
680 	}
681 }
682 
683 
684 /**********************************************************************
685  *
686  * Entropy collection routines.
687  *
688  * The following exported functions are used for pushing entropy into
689  * the above entropy accumulation routines:
690  *
691  *	void add_device_randomness(const void *buf, size_t len);
692  *	void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy);
693  *	void add_bootloader_randomness(const void *buf, size_t len);
694  *	void add_vmfork_randomness(const void *unique_vm_id, size_t len);
695  *	void add_interrupt_randomness(int irq);
696  *	void add_input_randomness(unsigned int type, unsigned int code, unsigned int value);
697  *	void add_disk_randomness(struct gendisk *disk);
698  *
699  * add_device_randomness() adds data to the input pool that
700  * is likely to differ between two devices (or possibly even per boot).
701  * This would be things like MAC addresses or serial numbers, or the
702  * read-out of the RTC. This does *not* credit any actual entropy to
703  * the pool, but it initializes the pool to different values for devices
704  * that might otherwise be identical and have very little entropy
705  * available to them (particularly common in the embedded world).
706  *
707  * add_hwgenerator_randomness() is for true hardware RNGs, and will credit
708  * entropy as specified by the caller. If the entropy pool is full it will
709  * block until more entropy is needed.
710  *
711  * add_bootloader_randomness() is called by bootloader drivers, such as EFI
712  * and device tree, and credits its input depending on whether or not the
713  * configuration option CONFIG_RANDOM_TRUST_BOOTLOADER is set.
714  *
715  * add_vmfork_randomness() adds a unique (but not necessarily secret) ID
716  * representing the current instance of a VM to the pool, without crediting,
717  * and then force-reseeds the crng so that it takes effect immediately.
718  *
719  * add_interrupt_randomness() uses the interrupt timing as random
720  * inputs to the entropy pool. Using the cycle counters and the irq source
721  * as inputs, it feeds the input pool roughly once a second or after 64
722  * interrupts, crediting 1 bit of entropy for whichever comes first.
723  *
724  * add_input_randomness() uses the input layer interrupt timing, as well
725  * as the event type information from the hardware.
726  *
727  * add_disk_randomness() uses what amounts to the seek time of block
728  * layer request events, on a per-disk_devt basis, as input to the
729  * entropy pool. Note that high-speed solid state drives with very low
730  * seek times do not make for good sources of entropy, as their seek
731  * times are usually fairly consistent.
732  *
733  * The last two routines try to estimate how many bits of entropy
734  * to credit. They do this by keeping track of the first and second
735  * order deltas of the event timings.
736  *
737  **********************************************************************/
738 
739 static bool trust_cpu __initdata = IS_ENABLED(CONFIG_RANDOM_TRUST_CPU);
740 static bool trust_bootloader __initdata = IS_ENABLED(CONFIG_RANDOM_TRUST_BOOTLOADER);
741 static int __init parse_trust_cpu(char *arg)
742 {
743 	return kstrtobool(arg, &trust_cpu);
744 }
745 static int __init parse_trust_bootloader(char *arg)
746 {
747 	return kstrtobool(arg, &trust_bootloader);
748 }
749 early_param("random.trust_cpu", parse_trust_cpu);
750 early_param("random.trust_bootloader", parse_trust_bootloader);
751 
752 static int random_pm_notification(struct notifier_block *nb, unsigned long action, void *data)
753 {
754 	unsigned long flags, entropy = random_get_entropy();
755 
756 	/*
757 	 * Encode a representation of how long the system has been suspended,
758 	 * in a way that is distinct from prior system suspends.
759 	 */
760 	ktime_t stamps[] = { ktime_get(), ktime_get_boottime(), ktime_get_real() };
761 
762 	spin_lock_irqsave(&input_pool.lock, flags);
763 	_mix_pool_bytes(&action, sizeof(action));
764 	_mix_pool_bytes(stamps, sizeof(stamps));
765 	_mix_pool_bytes(&entropy, sizeof(entropy));
766 	spin_unlock_irqrestore(&input_pool.lock, flags);
767 
768 	if (crng_ready() && (action == PM_RESTORE_PREPARE ||
769 	    (action == PM_POST_SUSPEND && !IS_ENABLED(CONFIG_PM_AUTOSLEEP) &&
770 	     !IS_ENABLED(CONFIG_PM_USERSPACE_AUTOSLEEP)))) {
771 		crng_reseed();
772 		pr_notice("crng reseeded on system resumption\n");
773 	}
774 	return 0;
775 }
776 
777 static struct notifier_block pm_notifier = { .notifier_call = random_pm_notification };
778 
779 /*
780  * This is called extremely early, before time keeping functionality is
781  * available, but arch randomness is. Interrupts are not yet enabled.
782  */
783 void __init random_init_early(const char *command_line)
784 {
785 	unsigned long entropy[BLAKE2S_BLOCK_SIZE / sizeof(long)];
786 	size_t i, longs, arch_bits;
787 
788 #if defined(LATENT_ENTROPY_PLUGIN)
789 	static const u8 compiletime_seed[BLAKE2S_BLOCK_SIZE] __initconst __latent_entropy;
790 	_mix_pool_bytes(compiletime_seed, sizeof(compiletime_seed));
791 #endif
792 
793 	for (i = 0, arch_bits = sizeof(entropy) * 8; i < ARRAY_SIZE(entropy);) {
794 		longs = arch_get_random_seed_longs(entropy, ARRAY_SIZE(entropy) - i);
795 		if (longs) {
796 			_mix_pool_bytes(entropy, sizeof(*entropy) * longs);
797 			i += longs;
798 			continue;
799 		}
800 		longs = arch_get_random_longs(entropy, ARRAY_SIZE(entropy) - i);
801 		if (longs) {
802 			_mix_pool_bytes(entropy, sizeof(*entropy) * longs);
803 			i += longs;
804 			continue;
805 		}
806 		arch_bits -= sizeof(*entropy) * 8;
807 		++i;
808 	}
809 
810 	_mix_pool_bytes(init_utsname(), sizeof(*(init_utsname())));
811 	_mix_pool_bytes(command_line, strlen(command_line));
812 
813 	/* Reseed if already seeded by earlier phases. */
814 	if (crng_ready())
815 		crng_reseed();
816 	else if (trust_cpu)
817 		_credit_init_bits(arch_bits);
818 }
819 
820 /*
821  * This is called a little bit after the prior function, and now there is
822  * access to timestamps counters. Interrupts are not yet enabled.
823  */
824 void __init random_init(void)
825 {
826 	unsigned long entropy = random_get_entropy();
827 	ktime_t now = ktime_get_real();
828 
829 	_mix_pool_bytes(&now, sizeof(now));
830 	_mix_pool_bytes(&entropy, sizeof(entropy));
831 	add_latent_entropy();
832 
833 	/*
834 	 * If we were initialized by the cpu or bootloader before jump labels
835 	 * are initialized, then we should enable the static branch here, where
836 	 * it's guaranteed that jump labels have been initialized.
837 	 */
838 	if (!static_branch_likely(&crng_is_ready) && crng_init >= CRNG_READY)
839 		crng_set_ready(NULL);
840 
841 	/* Reseed if already seeded by earlier phases. */
842 	if (crng_ready())
843 		crng_reseed();
844 
845 	WARN_ON(register_pm_notifier(&pm_notifier));
846 
847 	WARN(!entropy, "Missing cycle counter and fallback timer; RNG "
848 		       "entropy collection will consequently suffer.");
849 }
850 
851 /*
852  * Add device- or boot-specific data to the input pool to help
853  * initialize it.
854  *
855  * None of this adds any entropy; it is meant to avoid the problem of
856  * the entropy pool having similar initial state across largely
857  * identical devices.
858  */
859 void add_device_randomness(const void *buf, size_t len)
860 {
861 	unsigned long entropy = random_get_entropy();
862 	unsigned long flags;
863 
864 	spin_lock_irqsave(&input_pool.lock, flags);
865 	_mix_pool_bytes(&entropy, sizeof(entropy));
866 	_mix_pool_bytes(buf, len);
867 	spin_unlock_irqrestore(&input_pool.lock, flags);
868 }
869 EXPORT_SYMBOL(add_device_randomness);
870 
871 /*
872  * Interface for in-kernel drivers of true hardware RNGs.
873  * Those devices may produce endless random bits and will be throttled
874  * when our pool is full.
875  */
876 void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy)
877 {
878 	mix_pool_bytes(buf, len);
879 	credit_init_bits(entropy);
880 
881 	/*
882 	 * Throttle writing to once every reseed interval, unless we're not yet
883 	 * initialized or no entropy is credited.
884 	 */
885 	if (!kthread_should_stop() && (crng_ready() || !entropy))
886 		schedule_timeout_interruptible(crng_reseed_interval());
887 }
888 EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
889 
890 /*
891  * Handle random seed passed by bootloader, and credit it if
892  * CONFIG_RANDOM_TRUST_BOOTLOADER is set.
893  */
894 void __init add_bootloader_randomness(const void *buf, size_t len)
895 {
896 	mix_pool_bytes(buf, len);
897 	if (trust_bootloader)
898 		credit_init_bits(len * 8);
899 }
900 
901 #if IS_ENABLED(CONFIG_VMGENID)
902 static BLOCKING_NOTIFIER_HEAD(vmfork_chain);
903 
904 /*
905  * Handle a new unique VM ID, which is unique, not secret, so we
906  * don't credit it, but we do immediately force a reseed after so
907  * that it's used by the crng posthaste.
908  */
909 void __cold add_vmfork_randomness(const void *unique_vm_id, size_t len)
910 {
911 	add_device_randomness(unique_vm_id, len);
912 	if (crng_ready()) {
913 		crng_reseed();
914 		pr_notice("crng reseeded due to virtual machine fork\n");
915 	}
916 	blocking_notifier_call_chain(&vmfork_chain, 0, NULL);
917 }
918 #if IS_MODULE(CONFIG_VMGENID)
919 EXPORT_SYMBOL_GPL(add_vmfork_randomness);
920 #endif
921 
922 int __cold register_random_vmfork_notifier(struct notifier_block *nb)
923 {
924 	return blocking_notifier_chain_register(&vmfork_chain, nb);
925 }
926 EXPORT_SYMBOL_GPL(register_random_vmfork_notifier);
927 
928 int __cold unregister_random_vmfork_notifier(struct notifier_block *nb)
929 {
930 	return blocking_notifier_chain_unregister(&vmfork_chain, nb);
931 }
932 EXPORT_SYMBOL_GPL(unregister_random_vmfork_notifier);
933 #endif
934 
935 struct fast_pool {
936 	unsigned long pool[4];
937 	unsigned long last;
938 	unsigned int count;
939 	struct timer_list mix;
940 };
941 
942 static void mix_interrupt_randomness(struct timer_list *work);
943 
944 static DEFINE_PER_CPU(struct fast_pool, irq_randomness) = {
945 #ifdef CONFIG_64BIT
946 #define FASTMIX_PERM SIPHASH_PERMUTATION
947 	.pool = { SIPHASH_CONST_0, SIPHASH_CONST_1, SIPHASH_CONST_2, SIPHASH_CONST_3 },
948 #else
949 #define FASTMIX_PERM HSIPHASH_PERMUTATION
950 	.pool = { HSIPHASH_CONST_0, HSIPHASH_CONST_1, HSIPHASH_CONST_2, HSIPHASH_CONST_3 },
951 #endif
952 	.mix = __TIMER_INITIALIZER(mix_interrupt_randomness, 0)
953 };
954 
955 /*
956  * This is [Half]SipHash-1-x, starting from an empty key. Because
957  * the key is fixed, it assumes that its inputs are non-malicious,
958  * and therefore this has no security on its own. s represents the
959  * four-word SipHash state, while v represents a two-word input.
960  */
961 static void fast_mix(unsigned long s[4], unsigned long v1, unsigned long v2)
962 {
963 	s[3] ^= v1;
964 	FASTMIX_PERM(s[0], s[1], s[2], s[3]);
965 	s[0] ^= v1;
966 	s[3] ^= v2;
967 	FASTMIX_PERM(s[0], s[1], s[2], s[3]);
968 	s[0] ^= v2;
969 }
970 
971 #ifdef CONFIG_SMP
972 /*
973  * This function is called when the CPU has just come online, with
974  * entry CPUHP_AP_RANDOM_ONLINE, just after CPUHP_AP_WORKQUEUE_ONLINE.
975  */
976 int __cold random_online_cpu(unsigned int cpu)
977 {
978 	/*
979 	 * During CPU shutdown and before CPU onlining, add_interrupt_
980 	 * randomness() may schedule mix_interrupt_randomness(), and
981 	 * set the MIX_INFLIGHT flag. However, because the worker can
982 	 * be scheduled on a different CPU during this period, that
983 	 * flag will never be cleared. For that reason, we zero out
984 	 * the flag here, which runs just after workqueues are onlined
985 	 * for the CPU again. This also has the effect of setting the
986 	 * irq randomness count to zero so that new accumulated irqs
987 	 * are fresh.
988 	 */
989 	per_cpu_ptr(&irq_randomness, cpu)->count = 0;
990 	return 0;
991 }
992 #endif
993 
994 static void mix_interrupt_randomness(struct timer_list *work)
995 {
996 	struct fast_pool *fast_pool = container_of(work, struct fast_pool, mix);
997 	/*
998 	 * The size of the copied stack pool is explicitly 2 longs so that we
999 	 * only ever ingest half of the siphash output each time, retaining
1000 	 * the other half as the next "key" that carries over. The entropy is
1001 	 * supposed to be sufficiently dispersed between bits so on average
1002 	 * we don't wind up "losing" some.
1003 	 */
1004 	unsigned long pool[2];
1005 	unsigned int count;
1006 
1007 	/* Check to see if we're running on the wrong CPU due to hotplug. */
1008 	local_irq_disable();
1009 	if (fast_pool != this_cpu_ptr(&irq_randomness)) {
1010 		local_irq_enable();
1011 		return;
1012 	}
1013 
1014 	/*
1015 	 * Copy the pool to the stack so that the mixer always has a
1016 	 * consistent view, before we reenable irqs again.
1017 	 */
1018 	memcpy(pool, fast_pool->pool, sizeof(pool));
1019 	count = fast_pool->count;
1020 	fast_pool->count = 0;
1021 	fast_pool->last = jiffies;
1022 	local_irq_enable();
1023 
1024 	mix_pool_bytes(pool, sizeof(pool));
1025 	credit_init_bits(clamp_t(unsigned int, (count & U16_MAX) / 64, 1, sizeof(pool) * 8));
1026 
1027 	memzero_explicit(pool, sizeof(pool));
1028 }
1029 
1030 void add_interrupt_randomness(int irq)
1031 {
1032 	enum { MIX_INFLIGHT = 1U << 31 };
1033 	unsigned long entropy = random_get_entropy();
1034 	struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1035 	struct pt_regs *regs = get_irq_regs();
1036 	unsigned int new_count;
1037 
1038 	fast_mix(fast_pool->pool, entropy,
1039 		 (regs ? instruction_pointer(regs) : _RET_IP_) ^ swab(irq));
1040 	new_count = ++fast_pool->count;
1041 
1042 	if (new_count & MIX_INFLIGHT)
1043 		return;
1044 
1045 	if (new_count < 1024 && !time_is_before_jiffies(fast_pool->last + HZ))
1046 		return;
1047 
1048 	fast_pool->count |= MIX_INFLIGHT;
1049 	if (!timer_pending(&fast_pool->mix)) {
1050 		fast_pool->mix.expires = jiffies;
1051 		add_timer_on(&fast_pool->mix, raw_smp_processor_id());
1052 	}
1053 }
1054 EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1055 
1056 /* There is one of these per entropy source */
1057 struct timer_rand_state {
1058 	unsigned long last_time;
1059 	long last_delta, last_delta2;
1060 };
1061 
1062 /*
1063  * This function adds entropy to the entropy "pool" by using timing
1064  * delays. It uses the timer_rand_state structure to make an estimate
1065  * of how many bits of entropy this call has added to the pool. The
1066  * value "num" is also added to the pool; it should somehow describe
1067  * the type of event that just happened.
1068  */
1069 static void add_timer_randomness(struct timer_rand_state *state, unsigned int num)
1070 {
1071 	unsigned long entropy = random_get_entropy(), now = jiffies, flags;
1072 	long delta, delta2, delta3;
1073 	unsigned int bits;
1074 
1075 	/*
1076 	 * If we're in a hard IRQ, add_interrupt_randomness() will be called
1077 	 * sometime after, so mix into the fast pool.
1078 	 */
1079 	if (in_hardirq()) {
1080 		fast_mix(this_cpu_ptr(&irq_randomness)->pool, entropy, num);
1081 	} else {
1082 		spin_lock_irqsave(&input_pool.lock, flags);
1083 		_mix_pool_bytes(&entropy, sizeof(entropy));
1084 		_mix_pool_bytes(&num, sizeof(num));
1085 		spin_unlock_irqrestore(&input_pool.lock, flags);
1086 	}
1087 
1088 	if (crng_ready())
1089 		return;
1090 
1091 	/*
1092 	 * Calculate number of bits of randomness we probably added.
1093 	 * We take into account the first, second and third-order deltas
1094 	 * in order to make our estimate.
1095 	 */
1096 	delta = now - READ_ONCE(state->last_time);
1097 	WRITE_ONCE(state->last_time, now);
1098 
1099 	delta2 = delta - READ_ONCE(state->last_delta);
1100 	WRITE_ONCE(state->last_delta, delta);
1101 
1102 	delta3 = delta2 - READ_ONCE(state->last_delta2);
1103 	WRITE_ONCE(state->last_delta2, delta2);
1104 
1105 	if (delta < 0)
1106 		delta = -delta;
1107 	if (delta2 < 0)
1108 		delta2 = -delta2;
1109 	if (delta3 < 0)
1110 		delta3 = -delta3;
1111 	if (delta > delta2)
1112 		delta = delta2;
1113 	if (delta > delta3)
1114 		delta = delta3;
1115 
1116 	/*
1117 	 * delta is now minimum absolute delta. Round down by 1 bit
1118 	 * on general principles, and limit entropy estimate to 11 bits.
1119 	 */
1120 	bits = min(fls(delta >> 1), 11);
1121 
1122 	/*
1123 	 * As mentioned above, if we're in a hard IRQ, add_interrupt_randomness()
1124 	 * will run after this, which uses a different crediting scheme of 1 bit
1125 	 * per every 64 interrupts. In order to let that function do accounting
1126 	 * close to the one in this function, we credit a full 64/64 bit per bit,
1127 	 * and then subtract one to account for the extra one added.
1128 	 */
1129 	if (in_hardirq())
1130 		this_cpu_ptr(&irq_randomness)->count += max(1u, bits * 64) - 1;
1131 	else
1132 		_credit_init_bits(bits);
1133 }
1134 
1135 void add_input_randomness(unsigned int type, unsigned int code, unsigned int value)
1136 {
1137 	static unsigned char last_value;
1138 	static struct timer_rand_state input_timer_state = { INITIAL_JIFFIES };
1139 
1140 	/* Ignore autorepeat and the like. */
1141 	if (value == last_value)
1142 		return;
1143 
1144 	last_value = value;
1145 	add_timer_randomness(&input_timer_state,
1146 			     (type << 4) ^ code ^ (code >> 4) ^ value);
1147 }
1148 EXPORT_SYMBOL_GPL(add_input_randomness);
1149 
1150 #ifdef CONFIG_BLOCK
1151 void add_disk_randomness(struct gendisk *disk)
1152 {
1153 	if (!disk || !disk->random)
1154 		return;
1155 	/* First major is 1, so we get >= 0x200 here. */
1156 	add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1157 }
1158 EXPORT_SYMBOL_GPL(add_disk_randomness);
1159 
1160 void __cold rand_initialize_disk(struct gendisk *disk)
1161 {
1162 	struct timer_rand_state *state;
1163 
1164 	/*
1165 	 * If kzalloc returns null, we just won't use that entropy
1166 	 * source.
1167 	 */
1168 	state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1169 	if (state) {
1170 		state->last_time = INITIAL_JIFFIES;
1171 		disk->random = state;
1172 	}
1173 }
1174 #endif
1175 
1176 struct entropy_timer_state {
1177 	unsigned long entropy;
1178 	struct timer_list timer;
1179 	unsigned int samples, samples_per_bit;
1180 };
1181 
1182 /*
1183  * Each time the timer fires, we expect that we got an unpredictable
1184  * jump in the cycle counter. Even if the timer is running on another
1185  * CPU, the timer activity will be touching the stack of the CPU that is
1186  * generating entropy..
1187  *
1188  * Note that we don't re-arm the timer in the timer itself - we are
1189  * happy to be scheduled away, since that just makes the load more
1190  * complex, but we do not want the timer to keep ticking unless the
1191  * entropy loop is running.
1192  *
1193  * So the re-arming always happens in the entropy loop itself.
1194  */
1195 static void __cold entropy_timer(struct timer_list *timer)
1196 {
1197 	struct entropy_timer_state *state = container_of(timer, struct entropy_timer_state, timer);
1198 
1199 	if (++state->samples == state->samples_per_bit) {
1200 		credit_init_bits(1);
1201 		state->samples = 0;
1202 	}
1203 }
1204 
1205 /*
1206  * If we have an actual cycle counter, see if we can
1207  * generate enough entropy with timing noise
1208  */
1209 static void __cold try_to_generate_entropy(void)
1210 {
1211 	enum { NUM_TRIAL_SAMPLES = 8192, MAX_SAMPLES_PER_BIT = HZ / 15 };
1212 	struct entropy_timer_state stack;
1213 	unsigned int i, num_different = 0;
1214 	unsigned long last = random_get_entropy();
1215 
1216 	for (i = 0; i < NUM_TRIAL_SAMPLES - 1; ++i) {
1217 		stack.entropy = random_get_entropy();
1218 		if (stack.entropy != last)
1219 			++num_different;
1220 		last = stack.entropy;
1221 	}
1222 	stack.samples_per_bit = DIV_ROUND_UP(NUM_TRIAL_SAMPLES, num_different + 1);
1223 	if (stack.samples_per_bit > MAX_SAMPLES_PER_BIT)
1224 		return;
1225 
1226 	stack.samples = 0;
1227 	timer_setup_on_stack(&stack.timer, entropy_timer, 0);
1228 	while (!crng_ready() && !signal_pending(current)) {
1229 		if (!timer_pending(&stack.timer))
1230 			mod_timer(&stack.timer, jiffies);
1231 		mix_pool_bytes(&stack.entropy, sizeof(stack.entropy));
1232 		schedule();
1233 		stack.entropy = random_get_entropy();
1234 	}
1235 
1236 	del_timer_sync(&stack.timer);
1237 	destroy_timer_on_stack(&stack.timer);
1238 	mix_pool_bytes(&stack.entropy, sizeof(stack.entropy));
1239 }
1240 
1241 
1242 /**********************************************************************
1243  *
1244  * Userspace reader/writer interfaces.
1245  *
1246  * getrandom(2) is the primary modern interface into the RNG and should
1247  * be used in preference to anything else.
1248  *
1249  * Reading from /dev/random has the same functionality as calling
1250  * getrandom(2) with flags=0. In earlier versions, however, it had
1251  * vastly different semantics and should therefore be avoided, to
1252  * prevent backwards compatibility issues.
1253  *
1254  * Reading from /dev/urandom has the same functionality as calling
1255  * getrandom(2) with flags=GRND_INSECURE. Because it does not block
1256  * waiting for the RNG to be ready, it should not be used.
1257  *
1258  * Writing to either /dev/random or /dev/urandom adds entropy to
1259  * the input pool but does not credit it.
1260  *
1261  * Polling on /dev/random indicates when the RNG is initialized, on
1262  * the read side, and when it wants new entropy, on the write side.
1263  *
1264  * Both /dev/random and /dev/urandom have the same set of ioctls for
1265  * adding entropy, getting the entropy count, zeroing the count, and
1266  * reseeding the crng.
1267  *
1268  **********************************************************************/
1269 
1270 SYSCALL_DEFINE3(getrandom, char __user *, ubuf, size_t, len, unsigned int, flags)
1271 {
1272 	struct iov_iter iter;
1273 	struct iovec iov;
1274 	int ret;
1275 
1276 	if (flags & ~(GRND_NONBLOCK | GRND_RANDOM | GRND_INSECURE))
1277 		return -EINVAL;
1278 
1279 	/*
1280 	 * Requesting insecure and blocking randomness at the same time makes
1281 	 * no sense.
1282 	 */
1283 	if ((flags & (GRND_INSECURE | GRND_RANDOM)) == (GRND_INSECURE | GRND_RANDOM))
1284 		return -EINVAL;
1285 
1286 	if (!crng_ready() && !(flags & GRND_INSECURE)) {
1287 		if (flags & GRND_NONBLOCK)
1288 			return -EAGAIN;
1289 		ret = wait_for_random_bytes();
1290 		if (unlikely(ret))
1291 			return ret;
1292 	}
1293 
1294 	ret = import_single_range(READ, ubuf, len, &iov, &iter);
1295 	if (unlikely(ret))
1296 		return ret;
1297 	return get_random_bytes_user(&iter);
1298 }
1299 
1300 static __poll_t random_poll(struct file *file, poll_table *wait)
1301 {
1302 	poll_wait(file, &crng_init_wait, wait);
1303 	return crng_ready() ? EPOLLIN | EPOLLRDNORM : EPOLLOUT | EPOLLWRNORM;
1304 }
1305 
1306 static ssize_t write_pool_user(struct iov_iter *iter)
1307 {
1308 	u8 block[BLAKE2S_BLOCK_SIZE];
1309 	ssize_t ret = 0;
1310 	size_t copied;
1311 
1312 	if (unlikely(!iov_iter_count(iter)))
1313 		return 0;
1314 
1315 	for (;;) {
1316 		copied = copy_from_iter(block, sizeof(block), iter);
1317 		ret += copied;
1318 		mix_pool_bytes(block, copied);
1319 		if (!iov_iter_count(iter) || copied != sizeof(block))
1320 			break;
1321 
1322 		BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0);
1323 		if (ret % PAGE_SIZE == 0) {
1324 			if (signal_pending(current))
1325 				break;
1326 			cond_resched();
1327 		}
1328 	}
1329 
1330 	memzero_explicit(block, sizeof(block));
1331 	return ret ? ret : -EFAULT;
1332 }
1333 
1334 static ssize_t random_write_iter(struct kiocb *kiocb, struct iov_iter *iter)
1335 {
1336 	return write_pool_user(iter);
1337 }
1338 
1339 static ssize_t urandom_read_iter(struct kiocb *kiocb, struct iov_iter *iter)
1340 {
1341 	static int maxwarn = 10;
1342 
1343 	/*
1344 	 * Opportunistically attempt to initialize the RNG on platforms that
1345 	 * have fast cycle counters, but don't (for now) require it to succeed.
1346 	 */
1347 	if (!crng_ready())
1348 		try_to_generate_entropy();
1349 
1350 	if (!crng_ready()) {
1351 		if (!ratelimit_disable && maxwarn <= 0)
1352 			++urandom_warning.missed;
1353 		else if (ratelimit_disable || __ratelimit(&urandom_warning)) {
1354 			--maxwarn;
1355 			pr_notice("%s: uninitialized urandom read (%zu bytes read)\n",
1356 				  current->comm, iov_iter_count(iter));
1357 		}
1358 	}
1359 
1360 	return get_random_bytes_user(iter);
1361 }
1362 
1363 static ssize_t random_read_iter(struct kiocb *kiocb, struct iov_iter *iter)
1364 {
1365 	int ret;
1366 
1367 	if (!crng_ready() &&
1368 	    ((kiocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO)) ||
1369 	     (kiocb->ki_filp->f_flags & O_NONBLOCK)))
1370 		return -EAGAIN;
1371 
1372 	ret = wait_for_random_bytes();
1373 	if (ret != 0)
1374 		return ret;
1375 	return get_random_bytes_user(iter);
1376 }
1377 
1378 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1379 {
1380 	int __user *p = (int __user *)arg;
1381 	int ent_count;
1382 
1383 	switch (cmd) {
1384 	case RNDGETENTCNT:
1385 		/* Inherently racy, no point locking. */
1386 		if (put_user(input_pool.init_bits, p))
1387 			return -EFAULT;
1388 		return 0;
1389 	case RNDADDTOENTCNT:
1390 		if (!capable(CAP_SYS_ADMIN))
1391 			return -EPERM;
1392 		if (get_user(ent_count, p))
1393 			return -EFAULT;
1394 		if (ent_count < 0)
1395 			return -EINVAL;
1396 		credit_init_bits(ent_count);
1397 		return 0;
1398 	case RNDADDENTROPY: {
1399 		struct iov_iter iter;
1400 		struct iovec iov;
1401 		ssize_t ret;
1402 		int len;
1403 
1404 		if (!capable(CAP_SYS_ADMIN))
1405 			return -EPERM;
1406 		if (get_user(ent_count, p++))
1407 			return -EFAULT;
1408 		if (ent_count < 0)
1409 			return -EINVAL;
1410 		if (get_user(len, p++))
1411 			return -EFAULT;
1412 		ret = import_single_range(WRITE, p, len, &iov, &iter);
1413 		if (unlikely(ret))
1414 			return ret;
1415 		ret = write_pool_user(&iter);
1416 		if (unlikely(ret < 0))
1417 			return ret;
1418 		/* Since we're crediting, enforce that it was all written into the pool. */
1419 		if (unlikely(ret != len))
1420 			return -EFAULT;
1421 		credit_init_bits(ent_count);
1422 		return 0;
1423 	}
1424 	case RNDZAPENTCNT:
1425 	case RNDCLEARPOOL:
1426 		/* No longer has any effect. */
1427 		if (!capable(CAP_SYS_ADMIN))
1428 			return -EPERM;
1429 		return 0;
1430 	case RNDRESEEDCRNG:
1431 		if (!capable(CAP_SYS_ADMIN))
1432 			return -EPERM;
1433 		if (!crng_ready())
1434 			return -ENODATA;
1435 		crng_reseed();
1436 		return 0;
1437 	default:
1438 		return -EINVAL;
1439 	}
1440 }
1441 
1442 static int random_fasync(int fd, struct file *filp, int on)
1443 {
1444 	return fasync_helper(fd, filp, on, &fasync);
1445 }
1446 
1447 const struct file_operations random_fops = {
1448 	.read_iter = random_read_iter,
1449 	.write_iter = random_write_iter,
1450 	.poll = random_poll,
1451 	.unlocked_ioctl = random_ioctl,
1452 	.compat_ioctl = compat_ptr_ioctl,
1453 	.fasync = random_fasync,
1454 	.llseek = noop_llseek,
1455 	.splice_read = generic_file_splice_read,
1456 	.splice_write = iter_file_splice_write,
1457 };
1458 
1459 const struct file_operations urandom_fops = {
1460 	.read_iter = urandom_read_iter,
1461 	.write_iter = random_write_iter,
1462 	.unlocked_ioctl = random_ioctl,
1463 	.compat_ioctl = compat_ptr_ioctl,
1464 	.fasync = random_fasync,
1465 	.llseek = noop_llseek,
1466 	.splice_read = generic_file_splice_read,
1467 	.splice_write = iter_file_splice_write,
1468 };
1469 
1470 
1471 /********************************************************************
1472  *
1473  * Sysctl interface.
1474  *
1475  * These are partly unused legacy knobs with dummy values to not break
1476  * userspace and partly still useful things. They are usually accessible
1477  * in /proc/sys/kernel/random/ and are as follows:
1478  *
1479  * - boot_id - a UUID representing the current boot.
1480  *
1481  * - uuid - a random UUID, different each time the file is read.
1482  *
1483  * - poolsize - the number of bits of entropy that the input pool can
1484  *   hold, tied to the POOL_BITS constant.
1485  *
1486  * - entropy_avail - the number of bits of entropy currently in the
1487  *   input pool. Always <= poolsize.
1488  *
1489  * - write_wakeup_threshold - the amount of entropy in the input pool
1490  *   below which write polls to /dev/random will unblock, requesting
1491  *   more entropy, tied to the POOL_READY_BITS constant. It is writable
1492  *   to avoid breaking old userspaces, but writing to it does not
1493  *   change any behavior of the RNG.
1494  *
1495  * - urandom_min_reseed_secs - fixed to the value CRNG_RESEED_INTERVAL.
1496  *   It is writable to avoid breaking old userspaces, but writing
1497  *   to it does not change any behavior of the RNG.
1498  *
1499  ********************************************************************/
1500 
1501 #ifdef CONFIG_SYSCTL
1502 
1503 #include <linux/sysctl.h>
1504 
1505 static int sysctl_random_min_urandom_seed = CRNG_RESEED_INTERVAL / HZ;
1506 static int sysctl_random_write_wakeup_bits = POOL_READY_BITS;
1507 static int sysctl_poolsize = POOL_BITS;
1508 static u8 sysctl_bootid[UUID_SIZE];
1509 
1510 /*
1511  * This function is used to return both the bootid UUID, and random
1512  * UUID. The difference is in whether table->data is NULL; if it is,
1513  * then a new UUID is generated and returned to the user.
1514  */
1515 static int proc_do_uuid(struct ctl_table *table, int write, void *buf,
1516 			size_t *lenp, loff_t *ppos)
1517 {
1518 	u8 tmp_uuid[UUID_SIZE], *uuid;
1519 	char uuid_string[UUID_STRING_LEN + 1];
1520 	struct ctl_table fake_table = {
1521 		.data = uuid_string,
1522 		.maxlen = UUID_STRING_LEN
1523 	};
1524 
1525 	if (write)
1526 		return -EPERM;
1527 
1528 	uuid = table->data;
1529 	if (!uuid) {
1530 		uuid = tmp_uuid;
1531 		generate_random_uuid(uuid);
1532 	} else {
1533 		static DEFINE_SPINLOCK(bootid_spinlock);
1534 
1535 		spin_lock(&bootid_spinlock);
1536 		if (!uuid[8])
1537 			generate_random_uuid(uuid);
1538 		spin_unlock(&bootid_spinlock);
1539 	}
1540 
1541 	snprintf(uuid_string, sizeof(uuid_string), "%pU", uuid);
1542 	return proc_dostring(&fake_table, 0, buf, lenp, ppos);
1543 }
1544 
1545 /* The same as proc_dointvec, but writes don't change anything. */
1546 static int proc_do_rointvec(struct ctl_table *table, int write, void *buf,
1547 			    size_t *lenp, loff_t *ppos)
1548 {
1549 	return write ? 0 : proc_dointvec(table, 0, buf, lenp, ppos);
1550 }
1551 
1552 static struct ctl_table random_table[] = {
1553 	{
1554 		.procname	= "poolsize",
1555 		.data		= &sysctl_poolsize,
1556 		.maxlen		= sizeof(int),
1557 		.mode		= 0444,
1558 		.proc_handler	= proc_dointvec,
1559 	},
1560 	{
1561 		.procname	= "entropy_avail",
1562 		.data		= &input_pool.init_bits,
1563 		.maxlen		= sizeof(int),
1564 		.mode		= 0444,
1565 		.proc_handler	= proc_dointvec,
1566 	},
1567 	{
1568 		.procname	= "write_wakeup_threshold",
1569 		.data		= &sysctl_random_write_wakeup_bits,
1570 		.maxlen		= sizeof(int),
1571 		.mode		= 0644,
1572 		.proc_handler	= proc_do_rointvec,
1573 	},
1574 	{
1575 		.procname	= "urandom_min_reseed_secs",
1576 		.data		= &sysctl_random_min_urandom_seed,
1577 		.maxlen		= sizeof(int),
1578 		.mode		= 0644,
1579 		.proc_handler	= proc_do_rointvec,
1580 	},
1581 	{
1582 		.procname	= "boot_id",
1583 		.data		= &sysctl_bootid,
1584 		.mode		= 0444,
1585 		.proc_handler	= proc_do_uuid,
1586 	},
1587 	{
1588 		.procname	= "uuid",
1589 		.mode		= 0444,
1590 		.proc_handler	= proc_do_uuid,
1591 	},
1592 	{ }
1593 };
1594 
1595 /*
1596  * random_init() is called before sysctl_init(),
1597  * so we cannot call register_sysctl_init() in random_init()
1598  */
1599 static int __init random_sysctls_init(void)
1600 {
1601 	register_sysctl_init("kernel/random", random_table);
1602 	return 0;
1603 }
1604 device_initcall(random_sysctls_init);
1605 #endif
1606