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_{u32,
100  * ,u64,int,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  */
rng_is_initialized(void)105 bool rng_is_initialized(void)
106 {
107 	return crng_ready();
108 }
109 EXPORT_SYMBOL(rng_is_initialized);
110 
crng_set_ready(struct work_struct * work)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_{u32,u64,int,long}
123  * family of functions. Using any of these functions without first calling
124  * 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  */
wait_for_random_bytes(void)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  *	u32 get_random_u32()
161  *	u64 get_random_u64()
162  *	unsigned int get_random_int()
163  *	unsigned long get_random_long()
164  *
165  * These interfaces will return the requested number of random bytes
166  * into the given buffer or as a return value. This is equivalent to
167  * a read from /dev/urandom. The u32, u64, int, and long family of
168  * functions may be higher performance for one-off random integers,
169  * because they do a bit of buffering and do not invoke reseeding
170  * until the buffer is emptied.
171  *
172  *********************************************************************/
173 
174 enum {
175 	CRNG_RESEED_START_INTERVAL = HZ,
176 	CRNG_RESEED_INTERVAL = 60 * HZ
177 };
178 
179 static struct {
180 	u8 key[CHACHA_KEY_SIZE] __aligned(__alignof__(long));
181 	unsigned long birth;
182 	unsigned long generation;
183 	spinlock_t lock;
184 } base_crng = {
185 	.lock = __SPIN_LOCK_UNLOCKED(base_crng.lock)
186 };
187 
188 struct crng {
189 	u8 key[CHACHA_KEY_SIZE];
190 	unsigned long generation;
191 	local_lock_t lock;
192 };
193 
194 static DEFINE_PER_CPU(struct crng, crngs) = {
195 	.generation = ULONG_MAX,
196 	.lock = INIT_LOCAL_LOCK(crngs.lock),
197 };
198 
199 /* Used by crng_reseed() and crng_make_state() to extract a new seed from the input pool. */
200 static void extract_entropy(void *buf, size_t len);
201 
202 /* This extracts a new crng key from the input pool. */
crng_reseed(void)203 static void crng_reseed(void)
204 {
205 	unsigned long flags;
206 	unsigned long next_gen;
207 	u8 key[CHACHA_KEY_SIZE];
208 
209 	extract_entropy(key, sizeof(key));
210 
211 	/*
212 	 * We copy the new key into the base_crng, overwriting the old one,
213 	 * and update the generation counter. We avoid hitting ULONG_MAX,
214 	 * because the per-cpu crngs are initialized to ULONG_MAX, so this
215 	 * forces new CPUs that come online to always initialize.
216 	 */
217 	spin_lock_irqsave(&base_crng.lock, flags);
218 	memcpy(base_crng.key, key, sizeof(base_crng.key));
219 	next_gen = base_crng.generation + 1;
220 	if (next_gen == ULONG_MAX)
221 		++next_gen;
222 	WRITE_ONCE(base_crng.generation, next_gen);
223 	WRITE_ONCE(base_crng.birth, jiffies);
224 	if (!static_branch_likely(&crng_is_ready))
225 		crng_init = CRNG_READY;
226 	spin_unlock_irqrestore(&base_crng.lock, flags);
227 	memzero_explicit(key, sizeof(key));
228 }
229 
230 /*
231  * This generates a ChaCha block using the provided key, and then
232  * immediately overwites that key with half the block. It returns
233  * the resultant ChaCha state to the user, along with the second
234  * half of the block containing 32 bytes of random data that may
235  * be used; random_data_len may not be greater than 32.
236  *
237  * The returned ChaCha state contains within it a copy of the old
238  * key value, at index 4, so the state should always be zeroed out
239  * immediately after using in order to maintain forward secrecy.
240  * If the state cannot be erased in a timely manner, then it is
241  * safer to set the random_data parameter to &chacha_state[4] so
242  * that this function overwrites it before returning.
243  */
crng_fast_key_erasure(u8 key[CHACHA_KEY_SIZE],u32 chacha_state[CHACHA_STATE_WORDS],u8 * random_data,size_t random_data_len)244 static void crng_fast_key_erasure(u8 key[CHACHA_KEY_SIZE],
245 				  u32 chacha_state[CHACHA_STATE_WORDS],
246 				  u8 *random_data, size_t random_data_len)
247 {
248 	u8 first_block[CHACHA_BLOCK_SIZE];
249 
250 	BUG_ON(random_data_len > 32);
251 
252 	chacha_init_consts(chacha_state);
253 	memcpy(&chacha_state[4], key, CHACHA_KEY_SIZE);
254 	memset(&chacha_state[12], 0, sizeof(u32) * 4);
255 	chacha20_block(chacha_state, first_block);
256 
257 	memcpy(key, first_block, CHACHA_KEY_SIZE);
258 	memcpy(random_data, first_block + CHACHA_KEY_SIZE, random_data_len);
259 	memzero_explicit(first_block, sizeof(first_block));
260 }
261 
262 /*
263  * Return whether the crng seed is considered to be sufficiently old
264  * that a reseeding is needed. This happens if the last reseeding
265  * was CRNG_RESEED_INTERVAL ago, or during early boot, at an interval
266  * proportional to the uptime.
267  */
crng_has_old_seed(void)268 static bool crng_has_old_seed(void)
269 {
270 	static bool early_boot = true;
271 	unsigned long interval = CRNG_RESEED_INTERVAL;
272 
273 	if (unlikely(READ_ONCE(early_boot))) {
274 		time64_t uptime = ktime_get_seconds();
275 		if (uptime >= CRNG_RESEED_INTERVAL / HZ * 2)
276 			WRITE_ONCE(early_boot, false);
277 		else
278 			interval = max_t(unsigned int, CRNG_RESEED_START_INTERVAL,
279 					 (unsigned int)uptime / 2 * HZ);
280 	}
281 	return time_is_before_jiffies(READ_ONCE(base_crng.birth) + interval);
282 }
283 
284 /*
285  * This function returns a ChaCha state that you may use for generating
286  * random data. It also returns up to 32 bytes on its own of random data
287  * that may be used; random_data_len may not be greater than 32.
288  */
crng_make_state(u32 chacha_state[CHACHA_STATE_WORDS],u8 * random_data,size_t random_data_len)289 static void crng_make_state(u32 chacha_state[CHACHA_STATE_WORDS],
290 			    u8 *random_data, size_t random_data_len)
291 {
292 	unsigned long flags;
293 	struct crng *crng;
294 
295 	BUG_ON(random_data_len > 32);
296 
297 	/*
298 	 * For the fast path, we check whether we're ready, unlocked first, and
299 	 * then re-check once locked later. In the case where we're really not
300 	 * ready, we do fast key erasure with the base_crng directly, extracting
301 	 * when crng_init is CRNG_EMPTY.
302 	 */
303 	if (!crng_ready()) {
304 		bool ready;
305 
306 		spin_lock_irqsave(&base_crng.lock, flags);
307 		ready = crng_ready();
308 		if (!ready) {
309 			if (crng_init == CRNG_EMPTY)
310 				extract_entropy(base_crng.key, sizeof(base_crng.key));
311 			crng_fast_key_erasure(base_crng.key, chacha_state,
312 					      random_data, random_data_len);
313 		}
314 		spin_unlock_irqrestore(&base_crng.lock, flags);
315 		if (!ready)
316 			return;
317 	}
318 
319 	/*
320 	 * If the base_crng is old enough, we reseed, which in turn bumps the
321 	 * generation counter that we check below.
322 	 */
323 	if (unlikely(crng_has_old_seed()))
324 		crng_reseed();
325 
326 	local_lock_irqsave(&crngs.lock, flags);
327 	crng = raw_cpu_ptr(&crngs);
328 
329 	/*
330 	 * If our per-cpu crng is older than the base_crng, then it means
331 	 * somebody reseeded the base_crng. In that case, we do fast key
332 	 * erasure on the base_crng, and use its output as the new key
333 	 * for our per-cpu crng. This brings us up to date with base_crng.
334 	 */
335 	if (unlikely(crng->generation != READ_ONCE(base_crng.generation))) {
336 		spin_lock(&base_crng.lock);
337 		crng_fast_key_erasure(base_crng.key, chacha_state,
338 				      crng->key, sizeof(crng->key));
339 		crng->generation = base_crng.generation;
340 		spin_unlock(&base_crng.lock);
341 	}
342 
343 	/*
344 	 * Finally, when we've made it this far, our per-cpu crng has an up
345 	 * to date key, and we can do fast key erasure with it to produce
346 	 * some random data and a ChaCha state for the caller. All other
347 	 * branches of this function are "unlikely", so most of the time we
348 	 * should wind up here immediately.
349 	 */
350 	crng_fast_key_erasure(crng->key, chacha_state, random_data, random_data_len);
351 	local_unlock_irqrestore(&crngs.lock, flags);
352 }
353 
_get_random_bytes(void * buf,size_t len)354 static void _get_random_bytes(void *buf, size_t len)
355 {
356 	u32 chacha_state[CHACHA_STATE_WORDS];
357 	u8 tmp[CHACHA_BLOCK_SIZE];
358 	size_t first_block_len;
359 
360 	if (!len)
361 		return;
362 
363 	first_block_len = min_t(size_t, 32, len);
364 	crng_make_state(chacha_state, buf, first_block_len);
365 	len -= first_block_len;
366 	buf += first_block_len;
367 
368 	while (len) {
369 		if (len < CHACHA_BLOCK_SIZE) {
370 			chacha20_block(chacha_state, tmp);
371 			memcpy(buf, tmp, len);
372 			memzero_explicit(tmp, sizeof(tmp));
373 			break;
374 		}
375 
376 		chacha20_block(chacha_state, buf);
377 		if (unlikely(chacha_state[12] == 0))
378 			++chacha_state[13];
379 		len -= CHACHA_BLOCK_SIZE;
380 		buf += CHACHA_BLOCK_SIZE;
381 	}
382 
383 	memzero_explicit(chacha_state, sizeof(chacha_state));
384 }
385 
386 /*
387  * This function is the exported kernel interface.  It returns some
388  * number of good random numbers, suitable for key generation, seeding
389  * TCP sequence numbers, etc. In order to ensure that the randomness
390  * by this function is okay, the function wait_for_random_bytes()
391  * should be called and return 0 at least once at any point prior.
392  */
get_random_bytes(void * buf,size_t len)393 void get_random_bytes(void *buf, size_t len)
394 {
395 	warn_unseeded_randomness();
396 	_get_random_bytes(buf, len);
397 }
398 EXPORT_SYMBOL(get_random_bytes);
399 
get_random_bytes_user(struct iov_iter * iter)400 static ssize_t get_random_bytes_user(struct iov_iter *iter)
401 {
402 	u32 chacha_state[CHACHA_STATE_WORDS];
403 	u8 block[CHACHA_BLOCK_SIZE];
404 	size_t ret = 0, copied;
405 
406 	if (unlikely(!iov_iter_count(iter)))
407 		return 0;
408 
409 	/*
410 	 * Immediately overwrite the ChaCha key at index 4 with random
411 	 * bytes, in case userspace causes copy_to_iter() below to sleep
412 	 * forever, so that we still retain forward secrecy in that case.
413 	 */
414 	crng_make_state(chacha_state, (u8 *)&chacha_state[4], CHACHA_KEY_SIZE);
415 	/*
416 	 * However, if we're doing a read of len <= 32, we don't need to
417 	 * use chacha_state after, so we can simply return those bytes to
418 	 * the user directly.
419 	 */
420 	if (iov_iter_count(iter) <= CHACHA_KEY_SIZE) {
421 		ret = copy_to_iter(&chacha_state[4], CHACHA_KEY_SIZE, iter);
422 		goto out_zero_chacha;
423 	}
424 
425 	for (;;) {
426 		chacha20_block(chacha_state, block);
427 		if (unlikely(chacha_state[12] == 0))
428 			++chacha_state[13];
429 
430 		copied = copy_to_iter(block, sizeof(block), iter);
431 		ret += copied;
432 		if (!iov_iter_count(iter) || copied != sizeof(block))
433 			break;
434 
435 		BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0);
436 		if (ret % PAGE_SIZE == 0) {
437 			if (signal_pending(current))
438 				break;
439 			cond_resched();
440 		}
441 	}
442 
443 	memzero_explicit(block, sizeof(block));
444 out_zero_chacha:
445 	memzero_explicit(chacha_state, sizeof(chacha_state));
446 	return ret ? ret : -EFAULT;
447 }
448 
449 /*
450  * Batched entropy returns random integers. The quality of the random
451  * number is good as /dev/urandom. In order to ensure that the randomness
452  * provided by this function is okay, the function wait_for_random_bytes()
453  * should be called and return 0 at least once at any point prior.
454  */
455 
456 #define DEFINE_BATCHED_ENTROPY(type)						\
457 struct batch_ ##type {								\
458 	/*									\
459 	 * We make this 1.5x a ChaCha block, so that we get the			\
460 	 * remaining 32 bytes from fast key erasure, plus one full		\
461 	 * block from the detached ChaCha state. We can increase		\
462 	 * the size of this later if needed so long as we keep the		\
463 	 * formula of (integer_blocks + 0.5) * CHACHA_BLOCK_SIZE.		\
464 	 */									\
465 	type entropy[CHACHA_BLOCK_SIZE * 3 / (2 * sizeof(type))];		\
466 	local_lock_t lock;							\
467 	unsigned long generation;						\
468 	unsigned int position;							\
469 };										\
470 										\
471 static DEFINE_PER_CPU(struct batch_ ##type, batched_entropy_ ##type) = {	\
472 	.lock = INIT_LOCAL_LOCK(batched_entropy_ ##type.lock),			\
473 	.position = UINT_MAX							\
474 };										\
475 										\
476 type get_random_ ##type(void)							\
477 {										\
478 	type ret;								\
479 	unsigned long flags;							\
480 	struct batch_ ##type *batch;						\
481 	unsigned long next_gen;							\
482 										\
483 	warn_unseeded_randomness();						\
484 										\
485 	if  (!crng_ready()) {							\
486 		_get_random_bytes(&ret, sizeof(ret));				\
487 		return ret;							\
488 	}									\
489 										\
490 	local_lock_irqsave(&batched_entropy_ ##type.lock, flags);		\
491 	batch = raw_cpu_ptr(&batched_entropy_##type);				\
492 										\
493 	next_gen = READ_ONCE(base_crng.generation);				\
494 	if (batch->position >= ARRAY_SIZE(batch->entropy) ||			\
495 	    next_gen != batch->generation) {					\
496 		_get_random_bytes(batch->entropy, sizeof(batch->entropy));	\
497 		batch->position = 0;						\
498 		batch->generation = next_gen;					\
499 	}									\
500 										\
501 	ret = batch->entropy[batch->position];					\
502 	batch->entropy[batch->position] = 0;					\
503 	++batch->position;							\
504 	local_unlock_irqrestore(&batched_entropy_ ##type.lock, flags);		\
505 	return ret;								\
506 }										\
507 EXPORT_SYMBOL(get_random_ ##type);
508 
509 DEFINE_BATCHED_ENTROPY(u64)
DEFINE_BATCHED_ENTROPY(u32)510 DEFINE_BATCHED_ENTROPY(u32)
511 
512 #ifdef CONFIG_SMP
513 /*
514  * This function is called when the CPU is coming up, with entry
515  * CPUHP_RANDOM_PREPARE, which comes before CPUHP_WORKQUEUE_PREP.
516  */
517 int __cold random_prepare_cpu(unsigned int cpu)
518 {
519 	/*
520 	 * When the cpu comes back online, immediately invalidate both
521 	 * the per-cpu crng and all batches, so that we serve fresh
522 	 * randomness.
523 	 */
524 	per_cpu_ptr(&crngs, cpu)->generation = ULONG_MAX;
525 	per_cpu_ptr(&batched_entropy_u32, cpu)->position = UINT_MAX;
526 	per_cpu_ptr(&batched_entropy_u64, cpu)->position = UINT_MAX;
527 	return 0;
528 }
529 #endif
530 
531 
532 /**********************************************************************
533  *
534  * Entropy accumulation and extraction routines.
535  *
536  * Callers may add entropy via:
537  *
538  *     static void mix_pool_bytes(const void *buf, size_t len)
539  *
540  * After which, if added entropy should be credited:
541  *
542  *     static void credit_init_bits(size_t bits)
543  *
544  * Finally, extract entropy via:
545  *
546  *     static void extract_entropy(void *buf, size_t len)
547  *
548  **********************************************************************/
549 
550 enum {
551 	POOL_BITS = BLAKE2S_HASH_SIZE * 8,
552 	POOL_READY_BITS = POOL_BITS, /* When crng_init->CRNG_READY */
553 	POOL_EARLY_BITS = POOL_READY_BITS / 2 /* When crng_init->CRNG_EARLY */
554 };
555 
556 static struct {
557 	struct blake2s_state hash;
558 	spinlock_t lock;
559 	unsigned int init_bits;
560 } input_pool = {
561 	.hash.h = { BLAKE2S_IV0 ^ (0x01010000 | BLAKE2S_HASH_SIZE),
562 		    BLAKE2S_IV1, BLAKE2S_IV2, BLAKE2S_IV3, BLAKE2S_IV4,
563 		    BLAKE2S_IV5, BLAKE2S_IV6, BLAKE2S_IV7 },
564 	.hash.outlen = BLAKE2S_HASH_SIZE,
565 	.lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
566 };
567 
_mix_pool_bytes(const void * buf,size_t len)568 static void _mix_pool_bytes(const void *buf, size_t len)
569 {
570 	blake2s_update(&input_pool.hash, buf, len);
571 }
572 
573 /*
574  * This function adds bytes into the input pool. It does not
575  * update the initialization bit counter; the caller should call
576  * credit_init_bits if this is appropriate.
577  */
mix_pool_bytes(const void * buf,size_t len)578 static void mix_pool_bytes(const void *buf, size_t len)
579 {
580 	unsigned long flags;
581 
582 	spin_lock_irqsave(&input_pool.lock, flags);
583 	_mix_pool_bytes(buf, len);
584 	spin_unlock_irqrestore(&input_pool.lock, flags);
585 }
586 
587 /*
588  * This is an HKDF-like construction for using the hashed collected entropy
589  * as a PRF key, that's then expanded block-by-block.
590  */
extract_entropy(void * buf,size_t len)591 static void extract_entropy(void *buf, size_t len)
592 {
593 	unsigned long flags;
594 	u8 seed[BLAKE2S_HASH_SIZE], next_key[BLAKE2S_HASH_SIZE];
595 	struct {
596 		unsigned long rdseed[32 / sizeof(long)];
597 		size_t counter;
598 	} block;
599 	size_t i;
600 
601 	for (i = 0; i < ARRAY_SIZE(block.rdseed); ++i) {
602 		if (!arch_get_random_seed_long(&block.rdseed[i]) &&
603 		    !arch_get_random_long(&block.rdseed[i]))
604 			block.rdseed[i] = random_get_entropy();
605 	}
606 
607 	spin_lock_irqsave(&input_pool.lock, flags);
608 
609 	/* seed = HASHPRF(last_key, entropy_input) */
610 	blake2s_final(&input_pool.hash, seed);
611 
612 	/* next_key = HASHPRF(seed, RDSEED || 0) */
613 	block.counter = 0;
614 	blake2s(next_key, (u8 *)&block, seed, sizeof(next_key), sizeof(block), sizeof(seed));
615 	blake2s_init_key(&input_pool.hash, BLAKE2S_HASH_SIZE, next_key, sizeof(next_key));
616 
617 	spin_unlock_irqrestore(&input_pool.lock, flags);
618 	memzero_explicit(next_key, sizeof(next_key));
619 
620 	while (len) {
621 		i = min_t(size_t, len, BLAKE2S_HASH_SIZE);
622 		/* output = HASHPRF(seed, RDSEED || ++counter) */
623 		++block.counter;
624 		blake2s(buf, (u8 *)&block, seed, i, sizeof(block), sizeof(seed));
625 		len -= i;
626 		buf += i;
627 	}
628 
629 	memzero_explicit(seed, sizeof(seed));
630 	memzero_explicit(&block, sizeof(block));
631 }
632 
633 #define credit_init_bits(bits) if (!crng_ready()) _credit_init_bits(bits)
634 
_credit_init_bits(size_t bits)635 static void __cold _credit_init_bits(size_t bits)
636 {
637 	static struct execute_work set_ready;
638 	unsigned int new, orig, add;
639 	unsigned long flags;
640 
641 	if (!bits)
642 		return;
643 
644 	add = min_t(size_t, bits, POOL_BITS);
645 
646 	do {
647 		orig = READ_ONCE(input_pool.init_bits);
648 		new = min_t(unsigned int, POOL_BITS, orig + add);
649 	} while (cmpxchg(&input_pool.init_bits, orig, new) != orig);
650 
651 	if (orig < POOL_READY_BITS && new >= POOL_READY_BITS) {
652 		crng_reseed(); /* Sets crng_init to CRNG_READY under base_crng.lock. */
653 		if (static_key_initialized)
654 			execute_in_process_context(crng_set_ready, &set_ready);
655 		wake_up_interruptible(&crng_init_wait);
656 		kill_fasync(&fasync, SIGIO, POLL_IN);
657 		pr_notice("crng init done\n");
658 		if (urandom_warning.missed)
659 			pr_notice("%d urandom warning(s) missed due to ratelimiting\n",
660 				  urandom_warning.missed);
661 	} else if (orig < POOL_EARLY_BITS && new >= POOL_EARLY_BITS) {
662 		spin_lock_irqsave(&base_crng.lock, flags);
663 		/* Check if crng_init is CRNG_EMPTY, to avoid race with crng_reseed(). */
664 		if (crng_init == CRNG_EMPTY) {
665 			extract_entropy(base_crng.key, sizeof(base_crng.key));
666 			crng_init = CRNG_EARLY;
667 		}
668 		spin_unlock_irqrestore(&base_crng.lock, flags);
669 	}
670 }
671 
672 
673 /**********************************************************************
674  *
675  * Entropy collection routines.
676  *
677  * The following exported functions are used for pushing entropy into
678  * the above entropy accumulation routines:
679  *
680  *	void add_device_randomness(const void *buf, size_t len);
681  *	void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy);
682  *	void add_bootloader_randomness(const void *buf, size_t len);
683  *	void add_vmfork_randomness(const void *unique_vm_id, size_t len);
684  *	void add_interrupt_randomness(int irq);
685  *	void add_input_randomness(unsigned int type, unsigned int code, unsigned int value);
686  *	void add_disk_randomness(struct gendisk *disk);
687  *
688  * add_device_randomness() adds data to the input pool that
689  * is likely to differ between two devices (or possibly even per boot).
690  * This would be things like MAC addresses or serial numbers, or the
691  * read-out of the RTC. This does *not* credit any actual entropy to
692  * the pool, but it initializes the pool to different values for devices
693  * that might otherwise be identical and have very little entropy
694  * available to them (particularly common in the embedded world).
695  *
696  * add_hwgenerator_randomness() is for true hardware RNGs, and will credit
697  * entropy as specified by the caller. If the entropy pool is full it will
698  * block until more entropy is needed.
699  *
700  * add_bootloader_randomness() is called by bootloader drivers, such as EFI
701  * and device tree, and credits its input depending on whether or not the
702  * configuration option CONFIG_RANDOM_TRUST_BOOTLOADER is set.
703  *
704  * add_vmfork_randomness() adds a unique (but not necessarily secret) ID
705  * representing the current instance of a VM to the pool, without crediting,
706  * and then force-reseeds the crng so that it takes effect immediately.
707  *
708  * add_interrupt_randomness() uses the interrupt timing as random
709  * inputs to the entropy pool. Using the cycle counters and the irq source
710  * as inputs, it feeds the input pool roughly once a second or after 64
711  * interrupts, crediting 1 bit of entropy for whichever comes first.
712  *
713  * add_input_randomness() uses the input layer interrupt timing, as well
714  * as the event type information from the hardware.
715  *
716  * add_disk_randomness() uses what amounts to the seek time of block
717  * layer request events, on a per-disk_devt basis, as input to the
718  * entropy pool. Note that high-speed solid state drives with very low
719  * seek times do not make for good sources of entropy, as their seek
720  * times are usually fairly consistent.
721  *
722  * The last two routines try to estimate how many bits of entropy
723  * to credit. They do this by keeping track of the first and second
724  * order deltas of the event timings.
725  *
726  **********************************************************************/
727 
728 static bool trust_cpu __initdata = IS_ENABLED(CONFIG_RANDOM_TRUST_CPU);
729 static bool trust_bootloader __initdata = IS_ENABLED(CONFIG_RANDOM_TRUST_BOOTLOADER);
parse_trust_cpu(char * arg)730 static int __init parse_trust_cpu(char *arg)
731 {
732 	return kstrtobool(arg, &trust_cpu);
733 }
parse_trust_bootloader(char * arg)734 static int __init parse_trust_bootloader(char *arg)
735 {
736 	return kstrtobool(arg, &trust_bootloader);
737 }
738 early_param("random.trust_cpu", parse_trust_cpu);
739 early_param("random.trust_bootloader", parse_trust_bootloader);
740 
random_pm_notification(struct notifier_block * nb,unsigned long action,void * data)741 static int random_pm_notification(struct notifier_block *nb, unsigned long action, void *data)
742 {
743 	unsigned long flags, entropy = random_get_entropy();
744 
745 	/*
746 	 * Encode a representation of how long the system has been suspended,
747 	 * in a way that is distinct from prior system suspends.
748 	 */
749 	ktime_t stamps[] = { ktime_get(), ktime_get_boottime(), ktime_get_real() };
750 
751 	spin_lock_irqsave(&input_pool.lock, flags);
752 	_mix_pool_bytes(&action, sizeof(action));
753 	_mix_pool_bytes(stamps, sizeof(stamps));
754 	_mix_pool_bytes(&entropy, sizeof(entropy));
755 	spin_unlock_irqrestore(&input_pool.lock, flags);
756 
757 	if (crng_ready() && (action == PM_RESTORE_PREPARE ||
758 	    (action == PM_POST_SUSPEND &&
759 	     !IS_ENABLED(CONFIG_PM_AUTOSLEEP) && !IS_ENABLED(CONFIG_ANDROID)))) {
760 		crng_reseed();
761 		pr_notice("crng reseeded on system resumption\n");
762 	}
763 	return 0;
764 }
765 
766 static struct notifier_block pm_notifier = { .notifier_call = random_pm_notification };
767 
768 /*
769  * The first collection of entropy occurs at system boot while interrupts
770  * are still turned off. Here we push in latent entropy, RDSEED, a timestamp,
771  * utsname(), and the command line. Depending on the above configuration knob,
772  * RDSEED may be considered sufficient for initialization. Note that much
773  * earlier setup may already have pushed entropy into the input pool by the
774  * time we get here.
775  */
random_init(const char * command_line)776 int __init random_init(const char *command_line)
777 {
778 	ktime_t now = ktime_get_real();
779 	unsigned int i, arch_bits;
780 	unsigned long entropy;
781 
782 #if defined(LATENT_ENTROPY_PLUGIN)
783 	static const u8 compiletime_seed[BLAKE2S_BLOCK_SIZE] __initconst __latent_entropy;
784 	_mix_pool_bytes(compiletime_seed, sizeof(compiletime_seed));
785 #endif
786 
787 	for (i = 0, arch_bits = BLAKE2S_BLOCK_SIZE * 8;
788 	     i < BLAKE2S_BLOCK_SIZE; i += sizeof(entropy)) {
789 		if (!arch_get_random_seed_long_early(&entropy) &&
790 		    !arch_get_random_long_early(&entropy)) {
791 			entropy = random_get_entropy();
792 			arch_bits -= sizeof(entropy) * 8;
793 		}
794 		_mix_pool_bytes(&entropy, sizeof(entropy));
795 	}
796 	_mix_pool_bytes(&now, sizeof(now));
797 	_mix_pool_bytes(utsname(), sizeof(*(utsname())));
798 	_mix_pool_bytes(command_line, strlen(command_line));
799 	add_latent_entropy();
800 
801 	/*
802 	 * If we were initialized by the bootloader before jump labels are
803 	 * initialized, then we should enable the static branch here, where
804 	 * it's guaranteed that jump labels have been initialized.
805 	 */
806 	if (!static_branch_likely(&crng_is_ready) && crng_init >= CRNG_READY)
807 		crng_set_ready(NULL);
808 
809 	if (crng_ready())
810 		crng_reseed();
811 	else if (trust_cpu)
812 		_credit_init_bits(arch_bits);
813 
814 	WARN_ON(register_pm_notifier(&pm_notifier));
815 
816 	WARN(!random_get_entropy(), "Missing cycle counter and fallback timer; RNG "
817 				    "entropy collection will consequently suffer.");
818 	return 0;
819 }
820 
821 /*
822  * Add device- or boot-specific data to the input pool to help
823  * initialize it.
824  *
825  * None of this adds any entropy; it is meant to avoid the problem of
826  * the entropy pool having similar initial state across largely
827  * identical devices.
828  */
add_device_randomness(const void * buf,size_t len)829 void add_device_randomness(const void *buf, size_t len)
830 {
831 	unsigned long entropy = random_get_entropy();
832 	unsigned long flags;
833 
834 	spin_lock_irqsave(&input_pool.lock, flags);
835 	_mix_pool_bytes(&entropy, sizeof(entropy));
836 	_mix_pool_bytes(buf, len);
837 	spin_unlock_irqrestore(&input_pool.lock, flags);
838 }
839 EXPORT_SYMBOL(add_device_randomness);
840 
841 /*
842  * Interface for in-kernel drivers of true hardware RNGs.
843  * Those devices may produce endless random bits and will be throttled
844  * when our pool is full.
845  */
add_hwgenerator_randomness(const void * buf,size_t len,size_t entropy)846 void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy)
847 {
848 	mix_pool_bytes(buf, len);
849 	credit_init_bits(entropy);
850 
851 	/*
852 	 * Throttle writing to once every CRNG_RESEED_INTERVAL, unless
853 	 * we're not yet initialized.
854 	 */
855 	if (!kthread_should_stop() && crng_ready())
856 		schedule_timeout_interruptible(CRNG_RESEED_INTERVAL);
857 }
858 EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
859 
860 /*
861  * Handle random seed passed by bootloader, and credit it if
862  * CONFIG_RANDOM_TRUST_BOOTLOADER is set.
863  */
add_bootloader_randomness(const void * buf,size_t len)864 void __init add_bootloader_randomness(const void *buf, size_t len)
865 {
866 	mix_pool_bytes(buf, len);
867 	if (trust_bootloader)
868 		credit_init_bits(len * 8);
869 }
870 
871 #if IS_ENABLED(CONFIG_VMGENID)
872 static BLOCKING_NOTIFIER_HEAD(vmfork_chain);
873 
874 /*
875  * Handle a new unique VM ID, which is unique, not secret, so we
876  * don't credit it, but we do immediately force a reseed after so
877  * that it's used by the crng posthaste.
878  */
add_vmfork_randomness(const void * unique_vm_id,size_t len)879 void __cold add_vmfork_randomness(const void *unique_vm_id, size_t len)
880 {
881 	add_device_randomness(unique_vm_id, len);
882 	if (crng_ready()) {
883 		crng_reseed();
884 		pr_notice("crng reseeded due to virtual machine fork\n");
885 	}
886 	blocking_notifier_call_chain(&vmfork_chain, 0, NULL);
887 }
888 #if IS_MODULE(CONFIG_VMGENID)
889 EXPORT_SYMBOL_GPL(add_vmfork_randomness);
890 #endif
891 
register_random_vmfork_notifier(struct notifier_block * nb)892 int __cold register_random_vmfork_notifier(struct notifier_block *nb)
893 {
894 	return blocking_notifier_chain_register(&vmfork_chain, nb);
895 }
896 EXPORT_SYMBOL_GPL(register_random_vmfork_notifier);
897 
unregister_random_vmfork_notifier(struct notifier_block * nb)898 int __cold unregister_random_vmfork_notifier(struct notifier_block *nb)
899 {
900 	return blocking_notifier_chain_unregister(&vmfork_chain, nb);
901 }
902 EXPORT_SYMBOL_GPL(unregister_random_vmfork_notifier);
903 #endif
904 
905 struct fast_pool {
906 	struct work_struct mix;
907 	unsigned long pool[4];
908 	unsigned long last;
909 	unsigned int count;
910 };
911 
912 static DEFINE_PER_CPU(struct fast_pool, irq_randomness) = {
913 #ifdef CONFIG_64BIT
914 #define FASTMIX_PERM SIPHASH_PERMUTATION
915 	.pool = { SIPHASH_CONST_0, SIPHASH_CONST_1, SIPHASH_CONST_2, SIPHASH_CONST_3 }
916 #else
917 #define FASTMIX_PERM HSIPHASH_PERMUTATION
918 	.pool = { HSIPHASH_CONST_0, HSIPHASH_CONST_1, HSIPHASH_CONST_2, HSIPHASH_CONST_3 }
919 #endif
920 };
921 
922 /*
923  * This is [Half]SipHash-1-x, starting from an empty key. Because
924  * the key is fixed, it assumes that its inputs are non-malicious,
925  * and therefore this has no security on its own. s represents the
926  * four-word SipHash state, while v represents a two-word input.
927  */
fast_mix(unsigned long s[4],unsigned long v1,unsigned long v2)928 static void fast_mix(unsigned long s[4], unsigned long v1, unsigned long v2)
929 {
930 	s[3] ^= v1;
931 	FASTMIX_PERM(s[0], s[1], s[2], s[3]);
932 	s[0] ^= v1;
933 	s[3] ^= v2;
934 	FASTMIX_PERM(s[0], s[1], s[2], s[3]);
935 	s[0] ^= v2;
936 }
937 
938 #ifdef CONFIG_SMP
939 /*
940  * This function is called when the CPU has just come online, with
941  * entry CPUHP_AP_RANDOM_ONLINE, just after CPUHP_AP_WORKQUEUE_ONLINE.
942  */
random_online_cpu(unsigned int cpu)943 int __cold random_online_cpu(unsigned int cpu)
944 {
945 	/*
946 	 * During CPU shutdown and before CPU onlining, add_interrupt_
947 	 * randomness() may schedule mix_interrupt_randomness(), and
948 	 * set the MIX_INFLIGHT flag. However, because the worker can
949 	 * be scheduled on a different CPU during this period, that
950 	 * flag will never be cleared. For that reason, we zero out
951 	 * the flag here, which runs just after workqueues are onlined
952 	 * for the CPU again. This also has the effect of setting the
953 	 * irq randomness count to zero so that new accumulated irqs
954 	 * are fresh.
955 	 */
956 	per_cpu_ptr(&irq_randomness, cpu)->count = 0;
957 	return 0;
958 }
959 #endif
960 
mix_interrupt_randomness(struct work_struct * work)961 static void mix_interrupt_randomness(struct work_struct *work)
962 {
963 	struct fast_pool *fast_pool = container_of(work, struct fast_pool, mix);
964 	/*
965 	 * The size of the copied stack pool is explicitly 2 longs so that we
966 	 * only ever ingest half of the siphash output each time, retaining
967 	 * the other half as the next "key" that carries over. The entropy is
968 	 * supposed to be sufficiently dispersed between bits so on average
969 	 * we don't wind up "losing" some.
970 	 */
971 	unsigned long pool[2];
972 	unsigned int count;
973 
974 	/* Check to see if we're running on the wrong CPU due to hotplug. */
975 	local_irq_disable();
976 	if (fast_pool != this_cpu_ptr(&irq_randomness)) {
977 		local_irq_enable();
978 		return;
979 	}
980 
981 	/*
982 	 * Copy the pool to the stack so that the mixer always has a
983 	 * consistent view, before we reenable irqs again.
984 	 */
985 	memcpy(pool, fast_pool->pool, sizeof(pool));
986 	count = fast_pool->count;
987 	fast_pool->count = 0;
988 	fast_pool->last = jiffies;
989 	local_irq_enable();
990 
991 	mix_pool_bytes(pool, sizeof(pool));
992 	credit_init_bits(max(1u, (count & U16_MAX) / 64));
993 
994 	memzero_explicit(pool, sizeof(pool));
995 }
996 
add_interrupt_randomness(int irq)997 void add_interrupt_randomness(int irq)
998 {
999 	enum { MIX_INFLIGHT = 1U << 31 };
1000 	unsigned long entropy = random_get_entropy();
1001 	struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1002 	struct pt_regs *regs = get_irq_regs();
1003 	unsigned int new_count;
1004 
1005 	fast_mix(fast_pool->pool, entropy,
1006 		 (regs ? instruction_pointer(regs) : _RET_IP_) ^ swab(irq));
1007 	new_count = ++fast_pool->count;
1008 
1009 	if (new_count & MIX_INFLIGHT)
1010 		return;
1011 
1012 	if (new_count < 1024 && !time_is_before_jiffies(fast_pool->last + HZ))
1013 		return;
1014 
1015 	if (unlikely(!fast_pool->mix.func))
1016 		INIT_WORK(&fast_pool->mix, mix_interrupt_randomness);
1017 	fast_pool->count |= MIX_INFLIGHT;
1018 	queue_work_on(raw_smp_processor_id(), system_highpri_wq, &fast_pool->mix);
1019 }
1020 EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1021 
1022 /* There is one of these per entropy source */
1023 struct timer_rand_state {
1024 	unsigned long last_time;
1025 	long last_delta, last_delta2;
1026 };
1027 
1028 /*
1029  * This function adds entropy to the entropy "pool" by using timing
1030  * delays. It uses the timer_rand_state structure to make an estimate
1031  * of how many bits of entropy this call has added to the pool. The
1032  * value "num" is also added to the pool; it should somehow describe
1033  * the type of event that just happened.
1034  */
add_timer_randomness(struct timer_rand_state * state,unsigned int num)1035 static void add_timer_randomness(struct timer_rand_state *state, unsigned int num)
1036 {
1037 	unsigned long entropy = random_get_entropy(), now = jiffies, flags;
1038 	long delta, delta2, delta3;
1039 	unsigned int bits;
1040 
1041 	/*
1042 	 * If we're in a hard IRQ, add_interrupt_randomness() will be called
1043 	 * sometime after, so mix into the fast pool.
1044 	 */
1045 	if (in_hardirq()) {
1046 		fast_mix(this_cpu_ptr(&irq_randomness)->pool, entropy, num);
1047 	} else {
1048 		spin_lock_irqsave(&input_pool.lock, flags);
1049 		_mix_pool_bytes(&entropy, sizeof(entropy));
1050 		_mix_pool_bytes(&num, sizeof(num));
1051 		spin_unlock_irqrestore(&input_pool.lock, flags);
1052 	}
1053 
1054 	if (crng_ready())
1055 		return;
1056 
1057 	/*
1058 	 * Calculate number of bits of randomness we probably added.
1059 	 * We take into account the first, second and third-order deltas
1060 	 * in order to make our estimate.
1061 	 */
1062 	delta = now - READ_ONCE(state->last_time);
1063 	WRITE_ONCE(state->last_time, now);
1064 
1065 	delta2 = delta - READ_ONCE(state->last_delta);
1066 	WRITE_ONCE(state->last_delta, delta);
1067 
1068 	delta3 = delta2 - READ_ONCE(state->last_delta2);
1069 	WRITE_ONCE(state->last_delta2, delta2);
1070 
1071 	if (delta < 0)
1072 		delta = -delta;
1073 	if (delta2 < 0)
1074 		delta2 = -delta2;
1075 	if (delta3 < 0)
1076 		delta3 = -delta3;
1077 	if (delta > delta2)
1078 		delta = delta2;
1079 	if (delta > delta3)
1080 		delta = delta3;
1081 
1082 	/*
1083 	 * delta is now minimum absolute delta. Round down by 1 bit
1084 	 * on general principles, and limit entropy estimate to 11 bits.
1085 	 */
1086 	bits = min(fls(delta >> 1), 11);
1087 
1088 	/*
1089 	 * As mentioned above, if we're in a hard IRQ, add_interrupt_randomness()
1090 	 * will run after this, which uses a different crediting scheme of 1 bit
1091 	 * per every 64 interrupts. In order to let that function do accounting
1092 	 * close to the one in this function, we credit a full 64/64 bit per bit,
1093 	 * and then subtract one to account for the extra one added.
1094 	 */
1095 	if (in_hardirq())
1096 		this_cpu_ptr(&irq_randomness)->count += max(1u, bits * 64) - 1;
1097 	else
1098 		_credit_init_bits(bits);
1099 }
1100 
add_input_randomness(unsigned int type,unsigned int code,unsigned int value)1101 void add_input_randomness(unsigned int type, unsigned int code, unsigned int value)
1102 {
1103 	static unsigned char last_value;
1104 	static struct timer_rand_state input_timer_state = { INITIAL_JIFFIES };
1105 
1106 	/* Ignore autorepeat and the like. */
1107 	if (value == last_value)
1108 		return;
1109 
1110 	last_value = value;
1111 	add_timer_randomness(&input_timer_state,
1112 			     (type << 4) ^ code ^ (code >> 4) ^ value);
1113 }
1114 EXPORT_SYMBOL_GPL(add_input_randomness);
1115 
1116 #ifdef CONFIG_BLOCK
add_disk_randomness(struct gendisk * disk)1117 void add_disk_randomness(struct gendisk *disk)
1118 {
1119 	if (!disk || !disk->random)
1120 		return;
1121 	/* First major is 1, so we get >= 0x200 here. */
1122 	add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1123 }
1124 EXPORT_SYMBOL_GPL(add_disk_randomness);
1125 
rand_initialize_disk(struct gendisk * disk)1126 void __cold rand_initialize_disk(struct gendisk *disk)
1127 {
1128 	struct timer_rand_state *state;
1129 
1130 	/*
1131 	 * If kzalloc returns null, we just won't use that entropy
1132 	 * source.
1133 	 */
1134 	state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1135 	if (state) {
1136 		state->last_time = INITIAL_JIFFIES;
1137 		disk->random = state;
1138 	}
1139 }
1140 #endif
1141 
1142 struct entropy_timer_state {
1143 	unsigned long entropy;
1144 	struct timer_list timer;
1145 	unsigned int samples, samples_per_bit;
1146 };
1147 
1148 /*
1149  * Each time the timer fires, we expect that we got an unpredictable
1150  * jump in the cycle counter. Even if the timer is running on another
1151  * CPU, the timer activity will be touching the stack of the CPU that is
1152  * generating entropy..
1153  *
1154  * Note that we don't re-arm the timer in the timer itself - we are
1155  * happy to be scheduled away, since that just makes the load more
1156  * complex, but we do not want the timer to keep ticking unless the
1157  * entropy loop is running.
1158  *
1159  * So the re-arming always happens in the entropy loop itself.
1160  */
entropy_timer(struct timer_list * timer)1161 static void __cold entropy_timer(struct timer_list *timer)
1162 {
1163 	struct entropy_timer_state *state = container_of(timer, struct entropy_timer_state, timer);
1164 
1165 	if (++state->samples == state->samples_per_bit) {
1166 		credit_init_bits(1);
1167 		state->samples = 0;
1168 	}
1169 }
1170 
1171 /*
1172  * If we have an actual cycle counter, see if we can
1173  * generate enough entropy with timing noise
1174  */
try_to_generate_entropy(void)1175 static void __cold try_to_generate_entropy(void)
1176 {
1177 	enum { NUM_TRIAL_SAMPLES = 8192, MAX_SAMPLES_PER_BIT = HZ / 30 };
1178 	struct entropy_timer_state stack;
1179 	unsigned int i, num_different = 0;
1180 	unsigned long last = random_get_entropy();
1181 
1182 	for (i = 0; i < NUM_TRIAL_SAMPLES - 1; ++i) {
1183 		stack.entropy = random_get_entropy();
1184 		if (stack.entropy != last)
1185 			++num_different;
1186 		last = stack.entropy;
1187 	}
1188 	stack.samples_per_bit = DIV_ROUND_UP(NUM_TRIAL_SAMPLES, num_different + 1);
1189 	if (stack.samples_per_bit > MAX_SAMPLES_PER_BIT)
1190 		return;
1191 
1192 	stack.samples = 0;
1193 	timer_setup_on_stack(&stack.timer, entropy_timer, 0);
1194 	while (!crng_ready() && !signal_pending(current)) {
1195 		if (!timer_pending(&stack.timer))
1196 			mod_timer(&stack.timer, jiffies + 1);
1197 		mix_pool_bytes(&stack.entropy, sizeof(stack.entropy));
1198 		schedule();
1199 		stack.entropy = random_get_entropy();
1200 	}
1201 
1202 	del_timer_sync(&stack.timer);
1203 	destroy_timer_on_stack(&stack.timer);
1204 	mix_pool_bytes(&stack.entropy, sizeof(stack.entropy));
1205 }
1206 
1207 
1208 /**********************************************************************
1209  *
1210  * Userspace reader/writer interfaces.
1211  *
1212  * getrandom(2) is the primary modern interface into the RNG and should
1213  * be used in preference to anything else.
1214  *
1215  * Reading from /dev/random has the same functionality as calling
1216  * getrandom(2) with flags=0. In earlier versions, however, it had
1217  * vastly different semantics and should therefore be avoided, to
1218  * prevent backwards compatibility issues.
1219  *
1220  * Reading from /dev/urandom has the same functionality as calling
1221  * getrandom(2) with flags=GRND_INSECURE. Because it does not block
1222  * waiting for the RNG to be ready, it should not be used.
1223  *
1224  * Writing to either /dev/random or /dev/urandom adds entropy to
1225  * the input pool but does not credit it.
1226  *
1227  * Polling on /dev/random indicates when the RNG is initialized, on
1228  * the read side, and when it wants new entropy, on the write side.
1229  *
1230  * Both /dev/random and /dev/urandom have the same set of ioctls for
1231  * adding entropy, getting the entropy count, zeroing the count, and
1232  * reseeding the crng.
1233  *
1234  **********************************************************************/
1235 
SYSCALL_DEFINE3(getrandom,char __user *,ubuf,size_t,len,unsigned int,flags)1236 SYSCALL_DEFINE3(getrandom, char __user *, ubuf, size_t, len, unsigned int, flags)
1237 {
1238 	struct iov_iter iter;
1239 	struct iovec iov;
1240 	int ret;
1241 
1242 	if (flags & ~(GRND_NONBLOCK | GRND_RANDOM | GRND_INSECURE))
1243 		return -EINVAL;
1244 
1245 	/*
1246 	 * Requesting insecure and blocking randomness at the same time makes
1247 	 * no sense.
1248 	 */
1249 	if ((flags & (GRND_INSECURE | GRND_RANDOM)) == (GRND_INSECURE | GRND_RANDOM))
1250 		return -EINVAL;
1251 
1252 	if (!crng_ready() && !(flags & GRND_INSECURE)) {
1253 		if (flags & GRND_NONBLOCK)
1254 			return -EAGAIN;
1255 		ret = wait_for_random_bytes();
1256 		if (unlikely(ret))
1257 			return ret;
1258 	}
1259 
1260 	ret = import_single_range(READ, ubuf, len, &iov, &iter);
1261 	if (unlikely(ret))
1262 		return ret;
1263 	return get_random_bytes_user(&iter);
1264 }
1265 
random_poll(struct file * file,poll_table * wait)1266 static __poll_t random_poll(struct file *file, poll_table *wait)
1267 {
1268 	poll_wait(file, &crng_init_wait, wait);
1269 	return crng_ready() ? EPOLLIN | EPOLLRDNORM : EPOLLOUT | EPOLLWRNORM;
1270 }
1271 
write_pool_user(struct iov_iter * iter)1272 static ssize_t write_pool_user(struct iov_iter *iter)
1273 {
1274 	u8 block[BLAKE2S_BLOCK_SIZE];
1275 	ssize_t ret = 0;
1276 	size_t copied;
1277 
1278 	if (unlikely(!iov_iter_count(iter)))
1279 		return 0;
1280 
1281 	for (;;) {
1282 		copied = copy_from_iter(block, sizeof(block), iter);
1283 		ret += copied;
1284 		mix_pool_bytes(block, copied);
1285 		if (!iov_iter_count(iter) || copied != sizeof(block))
1286 			break;
1287 
1288 		BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0);
1289 		if (ret % PAGE_SIZE == 0) {
1290 			if (signal_pending(current))
1291 				break;
1292 			cond_resched();
1293 		}
1294 	}
1295 
1296 	memzero_explicit(block, sizeof(block));
1297 	return ret ? ret : -EFAULT;
1298 }
1299 
random_write_iter(struct kiocb * kiocb,struct iov_iter * iter)1300 static ssize_t random_write_iter(struct kiocb *kiocb, struct iov_iter *iter)
1301 {
1302 	return write_pool_user(iter);
1303 }
1304 
urandom_read_iter(struct kiocb * kiocb,struct iov_iter * iter)1305 static ssize_t urandom_read_iter(struct kiocb *kiocb, struct iov_iter *iter)
1306 {
1307 	static int maxwarn = 10;
1308 
1309 	/*
1310 	 * Opportunistically attempt to initialize the RNG on platforms that
1311 	 * have fast cycle counters, but don't (for now) require it to succeed.
1312 	 */
1313 	if (!crng_ready())
1314 		try_to_generate_entropy();
1315 
1316 	if (!crng_ready()) {
1317 		if (!ratelimit_disable && maxwarn <= 0)
1318 			++urandom_warning.missed;
1319 		else if (ratelimit_disable || __ratelimit(&urandom_warning)) {
1320 			--maxwarn;
1321 			pr_notice("%s: uninitialized urandom read (%zu bytes read)\n",
1322 				  current->comm, iov_iter_count(iter));
1323 		}
1324 	}
1325 
1326 	return get_random_bytes_user(iter);
1327 }
1328 
random_read_iter(struct kiocb * kiocb,struct iov_iter * iter)1329 static ssize_t random_read_iter(struct kiocb *kiocb, struct iov_iter *iter)
1330 {
1331 	int ret;
1332 
1333 	ret = wait_for_random_bytes();
1334 	if (ret != 0)
1335 		return ret;
1336 	return get_random_bytes_user(iter);
1337 }
1338 
random_ioctl(struct file * f,unsigned int cmd,unsigned long arg)1339 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1340 {
1341 	int __user *p = (int __user *)arg;
1342 	int ent_count;
1343 
1344 	switch (cmd) {
1345 	case RNDGETENTCNT:
1346 		/* Inherently racy, no point locking. */
1347 		if (put_user(input_pool.init_bits, p))
1348 			return -EFAULT;
1349 		return 0;
1350 	case RNDADDTOENTCNT:
1351 		if (!capable(CAP_SYS_ADMIN))
1352 			return -EPERM;
1353 		if (get_user(ent_count, p))
1354 			return -EFAULT;
1355 		if (ent_count < 0)
1356 			return -EINVAL;
1357 		credit_init_bits(ent_count);
1358 		return 0;
1359 	case RNDADDENTROPY: {
1360 		struct iov_iter iter;
1361 		struct iovec iov;
1362 		ssize_t ret;
1363 		int len;
1364 
1365 		if (!capable(CAP_SYS_ADMIN))
1366 			return -EPERM;
1367 		if (get_user(ent_count, p++))
1368 			return -EFAULT;
1369 		if (ent_count < 0)
1370 			return -EINVAL;
1371 		if (get_user(len, p++))
1372 			return -EFAULT;
1373 		ret = import_single_range(WRITE, p, len, &iov, &iter);
1374 		if (unlikely(ret))
1375 			return ret;
1376 		ret = write_pool_user(&iter);
1377 		if (unlikely(ret < 0))
1378 			return ret;
1379 		/* Since we're crediting, enforce that it was all written into the pool. */
1380 		if (unlikely(ret != len))
1381 			return -EFAULT;
1382 		credit_init_bits(ent_count);
1383 		return 0;
1384 	}
1385 	case RNDZAPENTCNT:
1386 	case RNDCLEARPOOL:
1387 		/* No longer has any effect. */
1388 		if (!capable(CAP_SYS_ADMIN))
1389 			return -EPERM;
1390 		return 0;
1391 	case RNDRESEEDCRNG:
1392 		if (!capable(CAP_SYS_ADMIN))
1393 			return -EPERM;
1394 		if (!crng_ready())
1395 			return -ENODATA;
1396 		crng_reseed();
1397 		return 0;
1398 	default:
1399 		return -EINVAL;
1400 	}
1401 }
1402 
random_fasync(int fd,struct file * filp,int on)1403 static int random_fasync(int fd, struct file *filp, int on)
1404 {
1405 	return fasync_helper(fd, filp, on, &fasync);
1406 }
1407 
1408 const struct file_operations random_fops = {
1409 	.read_iter = random_read_iter,
1410 	.write_iter = random_write_iter,
1411 	.poll = random_poll,
1412 	.unlocked_ioctl = random_ioctl,
1413 	.compat_ioctl = compat_ptr_ioctl,
1414 	.fasync = random_fasync,
1415 	.llseek = noop_llseek,
1416 	.splice_read = generic_file_splice_read,
1417 	.splice_write = iter_file_splice_write,
1418 };
1419 
1420 const struct file_operations urandom_fops = {
1421 	.read_iter = urandom_read_iter,
1422 	.write_iter = random_write_iter,
1423 	.unlocked_ioctl = random_ioctl,
1424 	.compat_ioctl = compat_ptr_ioctl,
1425 	.fasync = random_fasync,
1426 	.llseek = noop_llseek,
1427 	.splice_read = generic_file_splice_read,
1428 	.splice_write = iter_file_splice_write,
1429 };
1430 
1431 
1432 /********************************************************************
1433  *
1434  * Sysctl interface.
1435  *
1436  * These are partly unused legacy knobs with dummy values to not break
1437  * userspace and partly still useful things. They are usually accessible
1438  * in /proc/sys/kernel/random/ and are as follows:
1439  *
1440  * - boot_id - a UUID representing the current boot.
1441  *
1442  * - uuid - a random UUID, different each time the file is read.
1443  *
1444  * - poolsize - the number of bits of entropy that the input pool can
1445  *   hold, tied to the POOL_BITS constant.
1446  *
1447  * - entropy_avail - the number of bits of entropy currently in the
1448  *   input pool. Always <= poolsize.
1449  *
1450  * - write_wakeup_threshold - the amount of entropy in the input pool
1451  *   below which write polls to /dev/random will unblock, requesting
1452  *   more entropy, tied to the POOL_READY_BITS constant. It is writable
1453  *   to avoid breaking old userspaces, but writing to it does not
1454  *   change any behavior of the RNG.
1455  *
1456  * - urandom_min_reseed_secs - fixed to the value CRNG_RESEED_INTERVAL.
1457  *   It is writable to avoid breaking old userspaces, but writing
1458  *   to it does not change any behavior of the RNG.
1459  *
1460  ********************************************************************/
1461 
1462 #ifdef CONFIG_SYSCTL
1463 
1464 #include <linux/sysctl.h>
1465 
1466 static int sysctl_random_min_urandom_seed = CRNG_RESEED_INTERVAL / HZ;
1467 static int sysctl_random_write_wakeup_bits = POOL_READY_BITS;
1468 static int sysctl_poolsize = POOL_BITS;
1469 static u8 sysctl_bootid[UUID_SIZE];
1470 
1471 /*
1472  * This function is used to return both the bootid UUID, and random
1473  * UUID. The difference is in whether table->data is NULL; if it is,
1474  * then a new UUID is generated and returned to the user.
1475  */
proc_do_uuid(struct ctl_table * table,int write,void * buf,size_t * lenp,loff_t * ppos)1476 static int proc_do_uuid(struct ctl_table *table, int write, void *buf,
1477 			size_t *lenp, loff_t *ppos)
1478 {
1479 	u8 tmp_uuid[UUID_SIZE], *uuid;
1480 	char uuid_string[UUID_STRING_LEN + 1];
1481 	struct ctl_table fake_table = {
1482 		.data = uuid_string,
1483 		.maxlen = UUID_STRING_LEN
1484 	};
1485 
1486 	if (write)
1487 		return -EPERM;
1488 
1489 	uuid = table->data;
1490 	if (!uuid) {
1491 		uuid = tmp_uuid;
1492 		generate_random_uuid(uuid);
1493 	} else {
1494 		static DEFINE_SPINLOCK(bootid_spinlock);
1495 
1496 		spin_lock(&bootid_spinlock);
1497 		if (!uuid[8])
1498 			generate_random_uuid(uuid);
1499 		spin_unlock(&bootid_spinlock);
1500 	}
1501 
1502 	snprintf(uuid_string, sizeof(uuid_string), "%pU", uuid);
1503 	return proc_dostring(&fake_table, 0, buf, lenp, ppos);
1504 }
1505 
1506 /* The same as proc_dointvec, but writes don't change anything. */
proc_do_rointvec(struct ctl_table * table,int write,void * buf,size_t * lenp,loff_t * ppos)1507 static int proc_do_rointvec(struct ctl_table *table, int write, void *buf,
1508 			    size_t *lenp, loff_t *ppos)
1509 {
1510 	return write ? 0 : proc_dointvec(table, 0, buf, lenp, ppos);
1511 }
1512 
1513 static struct ctl_table random_table[] = {
1514 	{
1515 		.procname	= "poolsize",
1516 		.data		= &sysctl_poolsize,
1517 		.maxlen		= sizeof(int),
1518 		.mode		= 0444,
1519 		.proc_handler	= proc_dointvec,
1520 	},
1521 	{
1522 		.procname	= "entropy_avail",
1523 		.data		= &input_pool.init_bits,
1524 		.maxlen		= sizeof(int),
1525 		.mode		= 0444,
1526 		.proc_handler	= proc_dointvec,
1527 	},
1528 	{
1529 		.procname	= "write_wakeup_threshold",
1530 		.data		= &sysctl_random_write_wakeup_bits,
1531 		.maxlen		= sizeof(int),
1532 		.mode		= 0644,
1533 		.proc_handler	= proc_do_rointvec,
1534 	},
1535 	{
1536 		.procname	= "urandom_min_reseed_secs",
1537 		.data		= &sysctl_random_min_urandom_seed,
1538 		.maxlen		= sizeof(int),
1539 		.mode		= 0644,
1540 		.proc_handler	= proc_do_rointvec,
1541 	},
1542 	{
1543 		.procname	= "boot_id",
1544 		.data		= &sysctl_bootid,
1545 		.mode		= 0444,
1546 		.proc_handler	= proc_do_uuid,
1547 	},
1548 	{
1549 		.procname	= "uuid",
1550 		.mode		= 0444,
1551 		.proc_handler	= proc_do_uuid,
1552 	},
1553 	{ }
1554 };
1555 
1556 /*
1557  * random_init() is called before sysctl_init(),
1558  * so we cannot call register_sysctl_init() in random_init()
1559  */
random_sysctls_init(void)1560 static int __init random_sysctls_init(void)
1561 {
1562 	register_sysctl_init("kernel/random", random_table);
1563 	return 0;
1564 }
1565 device_initcall(random_sysctls_init);
1566 #endif
1567