Loading...
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 <linux/sched/isolation.h>
57#include <crypto/chacha.h>
58#include <crypto/blake2s.h>
59#include <asm/archrandom.h>
60#include <asm/processor.h>
61#include <asm/irq.h>
62#include <asm/irq_regs.h>
63#include <asm/io.h>
64
65/*********************************************************************
66 *
67 * Initialization and readiness waiting.
68 *
69 * Much of the RNG infrastructure is devoted to various dependencies
70 * being able to wait until the RNG has collected enough entropy and
71 * is ready for safe consumption.
72 *
73 *********************************************************************/
74
75/*
76 * crng_init is protected by base_crng->lock, and only increases
77 * its value (from empty->early->ready).
78 */
79static enum {
80 CRNG_EMPTY = 0, /* Little to no entropy collected */
81 CRNG_EARLY = 1, /* At least POOL_EARLY_BITS collected */
82 CRNG_READY = 2 /* Fully initialized with POOL_READY_BITS collected */
83} crng_init __read_mostly = CRNG_EMPTY;
84static DEFINE_STATIC_KEY_FALSE(crng_is_ready);
85#define crng_ready() (static_branch_likely(&crng_is_ready) || crng_init >= CRNG_READY)
86/* Various types of waiters for crng_init->CRNG_READY transition. */
87static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
88static struct fasync_struct *fasync;
89static ATOMIC_NOTIFIER_HEAD(random_ready_notifier);
90
91/* Control how we warn userspace. */
92static struct ratelimit_state urandom_warning =
93 RATELIMIT_STATE_INIT_FLAGS("urandom_warning", HZ, 3, RATELIMIT_MSG_ON_RELEASE);
94static int ratelimit_disable __read_mostly =
95 IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM);
96module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
97MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
98
99/*
100 * Returns whether or not the input pool has been seeded and thus guaranteed
101 * to supply cryptographically secure random numbers. This applies to: the
102 * /dev/urandom device, the get_random_bytes function, and the get_random_{u8,
103 * u16,u32,u64,long} family of functions.
104 *
105 * Returns: true if the input pool has been seeded.
106 * false if the input pool has not been seeded.
107 */
108bool rng_is_initialized(void)
109{
110 return crng_ready();
111}
112EXPORT_SYMBOL(rng_is_initialized);
113
114static void __cold crng_set_ready(struct work_struct *work)
115{
116 static_branch_enable(&crng_is_ready);
117}
118
119/* Used by wait_for_random_bytes(), and considered an entropy collector, below. */
120static void try_to_generate_entropy(void);
121
122/*
123 * Wait for the input pool to be seeded and thus guaranteed to supply
124 * cryptographically secure random numbers. This applies to: the /dev/urandom
125 * device, the get_random_bytes function, and the get_random_{u8,u16,u32,u64,
126 * long} family of functions. Using any of these functions without first
127 * calling this function forfeits the guarantee of security.
128 *
129 * Returns: 0 if the input pool has been seeded.
130 * -ERESTARTSYS if the function was interrupted by a signal.
131 */
132int wait_for_random_bytes(void)
133{
134 while (!crng_ready()) {
135 int ret;
136
137 try_to_generate_entropy();
138 ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ);
139 if (ret)
140 return ret > 0 ? 0 : ret;
141 }
142 return 0;
143}
144EXPORT_SYMBOL(wait_for_random_bytes);
145
146/*
147 * Add a callback function that will be invoked when the crng is initialised,
148 * or immediately if it already has been. Only use this is you are absolutely
149 * sure it is required. Most users should instead be able to test
150 * `rng_is_initialized()` on demand, or make use of `get_random_bytes_wait()`.
151 */
152int __cold execute_with_initialized_rng(struct notifier_block *nb)
153{
154 unsigned long flags;
155 int ret = 0;
156
157 spin_lock_irqsave(&random_ready_notifier.lock, flags);
158 if (crng_ready())
159 nb->notifier_call(nb, 0, NULL);
160 else
161 ret = raw_notifier_chain_register((struct raw_notifier_head *)&random_ready_notifier.head, nb);
162 spin_unlock_irqrestore(&random_ready_notifier.lock, flags);
163 return ret;
164}
165
166#define warn_unseeded_randomness() \
167 if (IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM) && !crng_ready()) \
168 printk_deferred(KERN_NOTICE "random: %s called from %pS with crng_init=%d\n", \
169 __func__, (void *)_RET_IP_, crng_init)
170
171
172/*********************************************************************
173 *
174 * Fast key erasure RNG, the "crng".
175 *
176 * These functions expand entropy from the entropy extractor into
177 * long streams for external consumption using the "fast key erasure"
178 * RNG described at <https://blog.cr.yp.to/20170723-random.html>.
179 *
180 * There are a few exported interfaces for use by other drivers:
181 *
182 * void get_random_bytes(void *buf, size_t len)
183 * u8 get_random_u8()
184 * u16 get_random_u16()
185 * u32 get_random_u32()
186 * u32 get_random_u32_below(u32 ceil)
187 * u32 get_random_u32_above(u32 floor)
188 * u32 get_random_u32_inclusive(u32 floor, u32 ceil)
189 * u64 get_random_u64()
190 * unsigned long get_random_long()
191 *
192 * These interfaces will return the requested number of random bytes
193 * into the given buffer or as a return value. This is equivalent to
194 * a read from /dev/urandom. The u8, u16, u32, u64, long family of
195 * functions may be higher performance for one-off random integers,
196 * because they do a bit of buffering and do not invoke reseeding
197 * until the buffer is emptied.
198 *
199 *********************************************************************/
200
201enum {
202 CRNG_RESEED_START_INTERVAL = HZ,
203 CRNG_RESEED_INTERVAL = 60 * HZ
204};
205
206static struct {
207 u8 key[CHACHA_KEY_SIZE] __aligned(__alignof__(long));
208 unsigned long generation;
209 spinlock_t lock;
210} base_crng = {
211 .lock = __SPIN_LOCK_UNLOCKED(base_crng.lock)
212};
213
214struct crng {
215 u8 key[CHACHA_KEY_SIZE];
216 unsigned long generation;
217 local_lock_t lock;
218};
219
220static DEFINE_PER_CPU(struct crng, crngs) = {
221 .generation = ULONG_MAX,
222 .lock = INIT_LOCAL_LOCK(crngs.lock),
223};
224
225/*
226 * Return the interval until the next reseeding, which is normally
227 * CRNG_RESEED_INTERVAL, but during early boot, it is at an interval
228 * proportional to the uptime.
229 */
230static unsigned int crng_reseed_interval(void)
231{
232 static bool early_boot = true;
233
234 if (unlikely(READ_ONCE(early_boot))) {
235 time64_t uptime = ktime_get_seconds();
236 if (uptime >= CRNG_RESEED_INTERVAL / HZ * 2)
237 WRITE_ONCE(early_boot, false);
238 else
239 return max_t(unsigned int, CRNG_RESEED_START_INTERVAL,
240 (unsigned int)uptime / 2 * HZ);
241 }
242 return CRNG_RESEED_INTERVAL;
243}
244
245/* Used by crng_reseed() and crng_make_state() to extract a new seed from the input pool. */
246static void extract_entropy(void *buf, size_t len);
247
248/* This extracts a new crng key from the input pool. */
249static void crng_reseed(struct work_struct *work)
250{
251 static DECLARE_DELAYED_WORK(next_reseed, crng_reseed);
252 unsigned long flags;
253 unsigned long next_gen;
254 u8 key[CHACHA_KEY_SIZE];
255
256 /* Immediately schedule the next reseeding, so that it fires sooner rather than later. */
257 if (likely(system_unbound_wq))
258 queue_delayed_work(system_unbound_wq, &next_reseed, crng_reseed_interval());
259
260 extract_entropy(key, sizeof(key));
261
262 /*
263 * We copy the new key into the base_crng, overwriting the old one,
264 * and update the generation counter. We avoid hitting ULONG_MAX,
265 * because the per-cpu crngs are initialized to ULONG_MAX, so this
266 * forces new CPUs that come online to always initialize.
267 */
268 spin_lock_irqsave(&base_crng.lock, flags);
269 memcpy(base_crng.key, key, sizeof(base_crng.key));
270 next_gen = base_crng.generation + 1;
271 if (next_gen == ULONG_MAX)
272 ++next_gen;
273 WRITE_ONCE(base_crng.generation, next_gen);
274 if (!static_branch_likely(&crng_is_ready))
275 crng_init = CRNG_READY;
276 spin_unlock_irqrestore(&base_crng.lock, flags);
277 memzero_explicit(key, sizeof(key));
278}
279
280/*
281 * This generates a ChaCha block using the provided key, and then
282 * immediately overwrites that key with half the block. It returns
283 * the resultant ChaCha state to the user, along with the second
284 * half of the block containing 32 bytes of random data that may
285 * be used; random_data_len may not be greater than 32.
286 *
287 * The returned ChaCha state contains within it a copy of the old
288 * key value, at index 4, so the state should always be zeroed out
289 * immediately after using in order to maintain forward secrecy.
290 * If the state cannot be erased in a timely manner, then it is
291 * safer to set the random_data parameter to &chacha_state[4] so
292 * that this function overwrites it before returning.
293 */
294static void crng_fast_key_erasure(u8 key[CHACHA_KEY_SIZE],
295 u32 chacha_state[CHACHA_STATE_WORDS],
296 u8 *random_data, size_t random_data_len)
297{
298 u8 first_block[CHACHA_BLOCK_SIZE];
299
300 BUG_ON(random_data_len > 32);
301
302 chacha_init_consts(chacha_state);
303 memcpy(&chacha_state[4], key, CHACHA_KEY_SIZE);
304 memset(&chacha_state[12], 0, sizeof(u32) * 4);
305 chacha20_block(chacha_state, first_block);
306
307 memcpy(key, first_block, CHACHA_KEY_SIZE);
308 memcpy(random_data, first_block + CHACHA_KEY_SIZE, random_data_len);
309 memzero_explicit(first_block, sizeof(first_block));
310}
311
312/*
313 * This function returns a ChaCha state that you may use for generating
314 * random data. It also returns up to 32 bytes on its own of random data
315 * that may be used; random_data_len may not be greater than 32.
316 */
317static void crng_make_state(u32 chacha_state[CHACHA_STATE_WORDS],
318 u8 *random_data, size_t random_data_len)
319{
320 unsigned long flags;
321 struct crng *crng;
322
323 BUG_ON(random_data_len > 32);
324
325 /*
326 * For the fast path, we check whether we're ready, unlocked first, and
327 * then re-check once locked later. In the case where we're really not
328 * ready, we do fast key erasure with the base_crng directly, extracting
329 * when crng_init is CRNG_EMPTY.
330 */
331 if (!crng_ready()) {
332 bool ready;
333
334 spin_lock_irqsave(&base_crng.lock, flags);
335 ready = crng_ready();
336 if (!ready) {
337 if (crng_init == CRNG_EMPTY)
338 extract_entropy(base_crng.key, sizeof(base_crng.key));
339 crng_fast_key_erasure(base_crng.key, chacha_state,
340 random_data, random_data_len);
341 }
342 spin_unlock_irqrestore(&base_crng.lock, flags);
343 if (!ready)
344 return;
345 }
346
347 local_lock_irqsave(&crngs.lock, flags);
348 crng = raw_cpu_ptr(&crngs);
349
350 /*
351 * If our per-cpu crng is older than the base_crng, then it means
352 * somebody reseeded the base_crng. In that case, we do fast key
353 * erasure on the base_crng, and use its output as the new key
354 * for our per-cpu crng. This brings us up to date with base_crng.
355 */
356 if (unlikely(crng->generation != READ_ONCE(base_crng.generation))) {
357 spin_lock(&base_crng.lock);
358 crng_fast_key_erasure(base_crng.key, chacha_state,
359 crng->key, sizeof(crng->key));
360 crng->generation = base_crng.generation;
361 spin_unlock(&base_crng.lock);
362 }
363
364 /*
365 * Finally, when we've made it this far, our per-cpu crng has an up
366 * to date key, and we can do fast key erasure with it to produce
367 * some random data and a ChaCha state for the caller. All other
368 * branches of this function are "unlikely", so most of the time we
369 * should wind up here immediately.
370 */
371 crng_fast_key_erasure(crng->key, chacha_state, random_data, random_data_len);
372 local_unlock_irqrestore(&crngs.lock, flags);
373}
374
375static void _get_random_bytes(void *buf, size_t len)
376{
377 u32 chacha_state[CHACHA_STATE_WORDS];
378 u8 tmp[CHACHA_BLOCK_SIZE];
379 size_t first_block_len;
380
381 if (!len)
382 return;
383
384 first_block_len = min_t(size_t, 32, len);
385 crng_make_state(chacha_state, buf, first_block_len);
386 len -= first_block_len;
387 buf += first_block_len;
388
389 while (len) {
390 if (len < CHACHA_BLOCK_SIZE) {
391 chacha20_block(chacha_state, tmp);
392 memcpy(buf, tmp, len);
393 memzero_explicit(tmp, sizeof(tmp));
394 break;
395 }
396
397 chacha20_block(chacha_state, buf);
398 if (unlikely(chacha_state[12] == 0))
399 ++chacha_state[13];
400 len -= CHACHA_BLOCK_SIZE;
401 buf += CHACHA_BLOCK_SIZE;
402 }
403
404 memzero_explicit(chacha_state, sizeof(chacha_state));
405}
406
407/*
408 * This returns random bytes in arbitrary quantities. The quality of the
409 * random bytes is good as /dev/urandom. In order to ensure that the
410 * randomness provided by this function is okay, the function
411 * wait_for_random_bytes() should be called and return 0 at least once
412 * at any point prior.
413 */
414void get_random_bytes(void *buf, size_t len)
415{
416 warn_unseeded_randomness();
417 _get_random_bytes(buf, len);
418}
419EXPORT_SYMBOL(get_random_bytes);
420
421static ssize_t get_random_bytes_user(struct iov_iter *iter)
422{
423 u32 chacha_state[CHACHA_STATE_WORDS];
424 u8 block[CHACHA_BLOCK_SIZE];
425 size_t ret = 0, copied;
426
427 if (unlikely(!iov_iter_count(iter)))
428 return 0;
429
430 /*
431 * Immediately overwrite the ChaCha key at index 4 with random
432 * bytes, in case userspace causes copy_to_iter() below to sleep
433 * forever, so that we still retain forward secrecy in that case.
434 */
435 crng_make_state(chacha_state, (u8 *)&chacha_state[4], CHACHA_KEY_SIZE);
436 /*
437 * However, if we're doing a read of len <= 32, we don't need to
438 * use chacha_state after, so we can simply return those bytes to
439 * the user directly.
440 */
441 if (iov_iter_count(iter) <= CHACHA_KEY_SIZE) {
442 ret = copy_to_iter(&chacha_state[4], CHACHA_KEY_SIZE, iter);
443 goto out_zero_chacha;
444 }
445
446 for (;;) {
447 chacha20_block(chacha_state, block);
448 if (unlikely(chacha_state[12] == 0))
449 ++chacha_state[13];
450
451 copied = copy_to_iter(block, sizeof(block), iter);
452 ret += copied;
453 if (!iov_iter_count(iter) || copied != sizeof(block))
454 break;
455
456 BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0);
457 if (ret % PAGE_SIZE == 0) {
458 if (signal_pending(current))
459 break;
460 cond_resched();
461 }
462 }
463
464 memzero_explicit(block, sizeof(block));
465out_zero_chacha:
466 memzero_explicit(chacha_state, sizeof(chacha_state));
467 return ret ? ret : -EFAULT;
468}
469
470/*
471 * Batched entropy returns random integers. The quality of the random
472 * number is good as /dev/urandom. In order to ensure that the randomness
473 * provided by this function is okay, the function wait_for_random_bytes()
474 * should be called and return 0 at least once at any point prior.
475 */
476
477#define DEFINE_BATCHED_ENTROPY(type) \
478struct batch_ ##type { \
479 /* \
480 * We make this 1.5x a ChaCha block, so that we get the \
481 * remaining 32 bytes from fast key erasure, plus one full \
482 * block from the detached ChaCha state. We can increase \
483 * the size of this later if needed so long as we keep the \
484 * formula of (integer_blocks + 0.5) * CHACHA_BLOCK_SIZE. \
485 */ \
486 type entropy[CHACHA_BLOCK_SIZE * 3 / (2 * sizeof(type))]; \
487 local_lock_t lock; \
488 unsigned long generation; \
489 unsigned int position; \
490}; \
491 \
492static DEFINE_PER_CPU(struct batch_ ##type, batched_entropy_ ##type) = { \
493 .lock = INIT_LOCAL_LOCK(batched_entropy_ ##type.lock), \
494 .position = UINT_MAX \
495}; \
496 \
497type get_random_ ##type(void) \
498{ \
499 type ret; \
500 unsigned long flags; \
501 struct batch_ ##type *batch; \
502 unsigned long next_gen; \
503 \
504 warn_unseeded_randomness(); \
505 \
506 if (!crng_ready()) { \
507 _get_random_bytes(&ret, sizeof(ret)); \
508 return ret; \
509 } \
510 \
511 local_lock_irqsave(&batched_entropy_ ##type.lock, flags); \
512 batch = raw_cpu_ptr(&batched_entropy_##type); \
513 \
514 next_gen = READ_ONCE(base_crng.generation); \
515 if (batch->position >= ARRAY_SIZE(batch->entropy) || \
516 next_gen != batch->generation) { \
517 _get_random_bytes(batch->entropy, sizeof(batch->entropy)); \
518 batch->position = 0; \
519 batch->generation = next_gen; \
520 } \
521 \
522 ret = batch->entropy[batch->position]; \
523 batch->entropy[batch->position] = 0; \
524 ++batch->position; \
525 local_unlock_irqrestore(&batched_entropy_ ##type.lock, flags); \
526 return ret; \
527} \
528EXPORT_SYMBOL(get_random_ ##type);
529
530DEFINE_BATCHED_ENTROPY(u8)
531DEFINE_BATCHED_ENTROPY(u16)
532DEFINE_BATCHED_ENTROPY(u32)
533DEFINE_BATCHED_ENTROPY(u64)
534
535u32 __get_random_u32_below(u32 ceil)
536{
537 /*
538 * This is the slow path for variable ceil. It is still fast, most of
539 * the time, by doing traditional reciprocal multiplication and
540 * opportunistically comparing the lower half to ceil itself, before
541 * falling back to computing a larger bound, and then rejecting samples
542 * whose lower half would indicate a range indivisible by ceil. The use
543 * of `-ceil % ceil` is analogous to `2^32 % ceil`, but is computable
544 * in 32-bits.
545 */
546 u32 rand = get_random_u32();
547 u64 mult;
548
549 /*
550 * This function is technically undefined for ceil == 0, and in fact
551 * for the non-underscored constant version in the header, we build bug
552 * on that. But for the non-constant case, it's convenient to have that
553 * evaluate to being a straight call to get_random_u32(), so that
554 * get_random_u32_inclusive() can work over its whole range without
555 * undefined behavior.
556 */
557 if (unlikely(!ceil))
558 return rand;
559
560 mult = (u64)ceil * rand;
561 if (unlikely((u32)mult < ceil)) {
562 u32 bound = -ceil % ceil;
563 while (unlikely((u32)mult < bound))
564 mult = (u64)ceil * get_random_u32();
565 }
566 return mult >> 32;
567}
568EXPORT_SYMBOL(__get_random_u32_below);
569
570#ifdef CONFIG_SMP
571/*
572 * This function is called when the CPU is coming up, with entry
573 * CPUHP_RANDOM_PREPARE, which comes before CPUHP_WORKQUEUE_PREP.
574 */
575int __cold random_prepare_cpu(unsigned int cpu)
576{
577 /*
578 * When the cpu comes back online, immediately invalidate both
579 * the per-cpu crng and all batches, so that we serve fresh
580 * randomness.
581 */
582 per_cpu_ptr(&crngs, cpu)->generation = ULONG_MAX;
583 per_cpu_ptr(&batched_entropy_u8, cpu)->position = UINT_MAX;
584 per_cpu_ptr(&batched_entropy_u16, cpu)->position = UINT_MAX;
585 per_cpu_ptr(&batched_entropy_u32, cpu)->position = UINT_MAX;
586 per_cpu_ptr(&batched_entropy_u64, cpu)->position = UINT_MAX;
587 return 0;
588}
589#endif
590
591
592/**********************************************************************
593 *
594 * Entropy accumulation and extraction routines.
595 *
596 * Callers may add entropy via:
597 *
598 * static void mix_pool_bytes(const void *buf, size_t len)
599 *
600 * After which, if added entropy should be credited:
601 *
602 * static void credit_init_bits(size_t bits)
603 *
604 * Finally, extract entropy via:
605 *
606 * static void extract_entropy(void *buf, size_t len)
607 *
608 **********************************************************************/
609
610enum {
611 POOL_BITS = BLAKE2S_HASH_SIZE * 8,
612 POOL_READY_BITS = POOL_BITS, /* When crng_init->CRNG_READY */
613 POOL_EARLY_BITS = POOL_READY_BITS / 2 /* When crng_init->CRNG_EARLY */
614};
615
616static struct {
617 struct blake2s_state hash;
618 spinlock_t lock;
619 unsigned int init_bits;
620} input_pool = {
621 .hash.h = { BLAKE2S_IV0 ^ (0x01010000 | BLAKE2S_HASH_SIZE),
622 BLAKE2S_IV1, BLAKE2S_IV2, BLAKE2S_IV3, BLAKE2S_IV4,
623 BLAKE2S_IV5, BLAKE2S_IV6, BLAKE2S_IV7 },
624 .hash.outlen = BLAKE2S_HASH_SIZE,
625 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
626};
627
628static void _mix_pool_bytes(const void *buf, size_t len)
629{
630 blake2s_update(&input_pool.hash, buf, len);
631}
632
633/*
634 * This function adds bytes into the input pool. It does not
635 * update the initialization bit counter; the caller should call
636 * credit_init_bits if this is appropriate.
637 */
638static void mix_pool_bytes(const void *buf, size_t len)
639{
640 unsigned long flags;
641
642 spin_lock_irqsave(&input_pool.lock, flags);
643 _mix_pool_bytes(buf, len);
644 spin_unlock_irqrestore(&input_pool.lock, flags);
645}
646
647/*
648 * This is an HKDF-like construction for using the hashed collected entropy
649 * as a PRF key, that's then expanded block-by-block.
650 */
651static void extract_entropy(void *buf, size_t len)
652{
653 unsigned long flags;
654 u8 seed[BLAKE2S_HASH_SIZE], next_key[BLAKE2S_HASH_SIZE];
655 struct {
656 unsigned long rdseed[32 / sizeof(long)];
657 size_t counter;
658 } block;
659 size_t i, longs;
660
661 for (i = 0; i < ARRAY_SIZE(block.rdseed);) {
662 longs = arch_get_random_seed_longs(&block.rdseed[i], ARRAY_SIZE(block.rdseed) - i);
663 if (longs) {
664 i += longs;
665 continue;
666 }
667 longs = arch_get_random_longs(&block.rdseed[i], ARRAY_SIZE(block.rdseed) - i);
668 if (longs) {
669 i += longs;
670 continue;
671 }
672 block.rdseed[i++] = random_get_entropy();
673 }
674
675 spin_lock_irqsave(&input_pool.lock, flags);
676
677 /* seed = HASHPRF(last_key, entropy_input) */
678 blake2s_final(&input_pool.hash, seed);
679
680 /* next_key = HASHPRF(seed, RDSEED || 0) */
681 block.counter = 0;
682 blake2s(next_key, (u8 *)&block, seed, sizeof(next_key), sizeof(block), sizeof(seed));
683 blake2s_init_key(&input_pool.hash, BLAKE2S_HASH_SIZE, next_key, sizeof(next_key));
684
685 spin_unlock_irqrestore(&input_pool.lock, flags);
686 memzero_explicit(next_key, sizeof(next_key));
687
688 while (len) {
689 i = min_t(size_t, len, BLAKE2S_HASH_SIZE);
690 /* output = HASHPRF(seed, RDSEED || ++counter) */
691 ++block.counter;
692 blake2s(buf, (u8 *)&block, seed, i, sizeof(block), sizeof(seed));
693 len -= i;
694 buf += i;
695 }
696
697 memzero_explicit(seed, sizeof(seed));
698 memzero_explicit(&block, sizeof(block));
699}
700
701#define credit_init_bits(bits) if (!crng_ready()) _credit_init_bits(bits)
702
703static void __cold _credit_init_bits(size_t bits)
704{
705 static struct execute_work set_ready;
706 unsigned int new, orig, add;
707 unsigned long flags;
708
709 if (!bits)
710 return;
711
712 add = min_t(size_t, bits, POOL_BITS);
713
714 orig = READ_ONCE(input_pool.init_bits);
715 do {
716 new = min_t(unsigned int, POOL_BITS, orig + add);
717 } while (!try_cmpxchg(&input_pool.init_bits, &orig, new));
718
719 if (orig < POOL_READY_BITS && new >= POOL_READY_BITS) {
720 crng_reseed(NULL); /* Sets crng_init to CRNG_READY under base_crng.lock. */
721 if (static_key_initialized)
722 execute_in_process_context(crng_set_ready, &set_ready);
723 atomic_notifier_call_chain(&random_ready_notifier, 0, NULL);
724 wake_up_interruptible(&crng_init_wait);
725 kill_fasync(&fasync, SIGIO, POLL_IN);
726 pr_notice("crng init done\n");
727 if (urandom_warning.missed)
728 pr_notice("%d urandom warning(s) missed due to ratelimiting\n",
729 urandom_warning.missed);
730 } else if (orig < POOL_EARLY_BITS && new >= POOL_EARLY_BITS) {
731 spin_lock_irqsave(&base_crng.lock, flags);
732 /* Check if crng_init is CRNG_EMPTY, to avoid race with crng_reseed(). */
733 if (crng_init == CRNG_EMPTY) {
734 extract_entropy(base_crng.key, sizeof(base_crng.key));
735 crng_init = CRNG_EARLY;
736 }
737 spin_unlock_irqrestore(&base_crng.lock, flags);
738 }
739}
740
741
742/**********************************************************************
743 *
744 * Entropy collection routines.
745 *
746 * The following exported functions are used for pushing entropy into
747 * the above entropy accumulation routines:
748 *
749 * void add_device_randomness(const void *buf, size_t len);
750 * void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy, bool sleep_after);
751 * void add_bootloader_randomness(const void *buf, size_t len);
752 * void add_vmfork_randomness(const void *unique_vm_id, size_t len);
753 * void add_interrupt_randomness(int irq);
754 * void add_input_randomness(unsigned int type, unsigned int code, unsigned int value);
755 * void add_disk_randomness(struct gendisk *disk);
756 *
757 * add_device_randomness() adds data to the input pool that
758 * is likely to differ between two devices (or possibly even per boot).
759 * This would be things like MAC addresses or serial numbers, or the
760 * read-out of the RTC. This does *not* credit any actual entropy to
761 * the pool, but it initializes the pool to different values for devices
762 * that might otherwise be identical and have very little entropy
763 * available to them (particularly common in the embedded world).
764 *
765 * add_hwgenerator_randomness() is for true hardware RNGs, and will credit
766 * entropy as specified by the caller. If the entropy pool is full it will
767 * block until more entropy is needed.
768 *
769 * add_bootloader_randomness() is called by bootloader drivers, such as EFI
770 * and device tree, and credits its input depending on whether or not the
771 * command line option 'random.trust_bootloader'.
772 *
773 * add_vmfork_randomness() adds a unique (but not necessarily secret) ID
774 * representing the current instance of a VM to the pool, without crediting,
775 * and then force-reseeds the crng so that it takes effect immediately.
776 *
777 * add_interrupt_randomness() uses the interrupt timing as random
778 * inputs to the entropy pool. Using the cycle counters and the irq source
779 * as inputs, it feeds the input pool roughly once a second or after 64
780 * interrupts, crediting 1 bit of entropy for whichever comes first.
781 *
782 * add_input_randomness() uses the input layer interrupt timing, as well
783 * as the event type information from the hardware.
784 *
785 * add_disk_randomness() uses what amounts to the seek time of block
786 * layer request events, on a per-disk_devt basis, as input to the
787 * entropy pool. Note that high-speed solid state drives with very low
788 * seek times do not make for good sources of entropy, as their seek
789 * times are usually fairly consistent.
790 *
791 * The last two routines try to estimate how many bits of entropy
792 * to credit. They do this by keeping track of the first and second
793 * order deltas of the event timings.
794 *
795 **********************************************************************/
796
797static bool trust_cpu __initdata = true;
798static bool trust_bootloader __initdata = true;
799static int __init parse_trust_cpu(char *arg)
800{
801 return kstrtobool(arg, &trust_cpu);
802}
803static int __init parse_trust_bootloader(char *arg)
804{
805 return kstrtobool(arg, &trust_bootloader);
806}
807early_param("random.trust_cpu", parse_trust_cpu);
808early_param("random.trust_bootloader", parse_trust_bootloader);
809
810static int random_pm_notification(struct notifier_block *nb, unsigned long action, void *data)
811{
812 unsigned long flags, entropy = random_get_entropy();
813
814 /*
815 * Encode a representation of how long the system has been suspended,
816 * in a way that is distinct from prior system suspends.
817 */
818 ktime_t stamps[] = { ktime_get(), ktime_get_boottime(), ktime_get_real() };
819
820 spin_lock_irqsave(&input_pool.lock, flags);
821 _mix_pool_bytes(&action, sizeof(action));
822 _mix_pool_bytes(stamps, sizeof(stamps));
823 _mix_pool_bytes(&entropy, sizeof(entropy));
824 spin_unlock_irqrestore(&input_pool.lock, flags);
825
826 if (crng_ready() && (action == PM_RESTORE_PREPARE ||
827 (action == PM_POST_SUSPEND && !IS_ENABLED(CONFIG_PM_AUTOSLEEP) &&
828 !IS_ENABLED(CONFIG_PM_USERSPACE_AUTOSLEEP)))) {
829 crng_reseed(NULL);
830 pr_notice("crng reseeded on system resumption\n");
831 }
832 return 0;
833}
834
835static struct notifier_block pm_notifier = { .notifier_call = random_pm_notification };
836
837/*
838 * This is called extremely early, before time keeping functionality is
839 * available, but arch randomness is. Interrupts are not yet enabled.
840 */
841void __init random_init_early(const char *command_line)
842{
843 unsigned long entropy[BLAKE2S_BLOCK_SIZE / sizeof(long)];
844 size_t i, longs, arch_bits;
845
846#if defined(LATENT_ENTROPY_PLUGIN)
847 static const u8 compiletime_seed[BLAKE2S_BLOCK_SIZE] __initconst __latent_entropy;
848 _mix_pool_bytes(compiletime_seed, sizeof(compiletime_seed));
849#endif
850
851 for (i = 0, arch_bits = sizeof(entropy) * 8; i < ARRAY_SIZE(entropy);) {
852 longs = arch_get_random_seed_longs(entropy, ARRAY_SIZE(entropy) - i);
853 if (longs) {
854 _mix_pool_bytes(entropy, sizeof(*entropy) * longs);
855 i += longs;
856 continue;
857 }
858 longs = arch_get_random_longs(entropy, ARRAY_SIZE(entropy) - i);
859 if (longs) {
860 _mix_pool_bytes(entropy, sizeof(*entropy) * longs);
861 i += longs;
862 continue;
863 }
864 arch_bits -= sizeof(*entropy) * 8;
865 ++i;
866 }
867
868 _mix_pool_bytes(init_utsname(), sizeof(*(init_utsname())));
869 _mix_pool_bytes(command_line, strlen(command_line));
870
871 /* Reseed if already seeded by earlier phases. */
872 if (crng_ready())
873 crng_reseed(NULL);
874 else if (trust_cpu)
875 _credit_init_bits(arch_bits);
876}
877
878/*
879 * This is called a little bit after the prior function, and now there is
880 * access to timestamps counters. Interrupts are not yet enabled.
881 */
882void __init random_init(void)
883{
884 unsigned long entropy = random_get_entropy();
885 ktime_t now = ktime_get_real();
886
887 _mix_pool_bytes(&now, sizeof(now));
888 _mix_pool_bytes(&entropy, sizeof(entropy));
889 add_latent_entropy();
890
891 /*
892 * If we were initialized by the cpu or bootloader before jump labels
893 * are initialized, then we should enable the static branch here, where
894 * it's guaranteed that jump labels have been initialized.
895 */
896 if (!static_branch_likely(&crng_is_ready) && crng_init >= CRNG_READY)
897 crng_set_ready(NULL);
898
899 /* Reseed if already seeded by earlier phases. */
900 if (crng_ready())
901 crng_reseed(NULL);
902
903 WARN_ON(register_pm_notifier(&pm_notifier));
904
905 WARN(!entropy, "Missing cycle counter and fallback timer; RNG "
906 "entropy collection will consequently suffer.");
907}
908
909/*
910 * Add device- or boot-specific data to the input pool to help
911 * initialize it.
912 *
913 * None of this adds any entropy; it is meant to avoid the problem of
914 * the entropy pool having similar initial state across largely
915 * identical devices.
916 */
917void add_device_randomness(const void *buf, size_t len)
918{
919 unsigned long entropy = random_get_entropy();
920 unsigned long flags;
921
922 spin_lock_irqsave(&input_pool.lock, flags);
923 _mix_pool_bytes(&entropy, sizeof(entropy));
924 _mix_pool_bytes(buf, len);
925 spin_unlock_irqrestore(&input_pool.lock, flags);
926}
927EXPORT_SYMBOL(add_device_randomness);
928
929/*
930 * Interface for in-kernel drivers of true hardware RNGs. Those devices
931 * may produce endless random bits, so this function will sleep for
932 * some amount of time after, if the sleep_after parameter is true.
933 */
934void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy, bool sleep_after)
935{
936 mix_pool_bytes(buf, len);
937 credit_init_bits(entropy);
938
939 /*
940 * Throttle writing to once every reseed interval, unless we're not yet
941 * initialized or no entropy is credited.
942 */
943 if (sleep_after && !kthread_should_stop() && (crng_ready() || !entropy))
944 schedule_timeout_interruptible(crng_reseed_interval());
945}
946EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
947
948/*
949 * Handle random seed passed by bootloader, and credit it depending
950 * on the command line option 'random.trust_bootloader'.
951 */
952void __init add_bootloader_randomness(const void *buf, size_t len)
953{
954 mix_pool_bytes(buf, len);
955 if (trust_bootloader)
956 credit_init_bits(len * 8);
957}
958
959#if IS_ENABLED(CONFIG_VMGENID)
960static BLOCKING_NOTIFIER_HEAD(vmfork_chain);
961
962/*
963 * Handle a new unique VM ID, which is unique, not secret, so we
964 * don't credit it, but we do immediately force a reseed after so
965 * that it's used by the crng posthaste.
966 */
967void __cold add_vmfork_randomness(const void *unique_vm_id, size_t len)
968{
969 add_device_randomness(unique_vm_id, len);
970 if (crng_ready()) {
971 crng_reseed(NULL);
972 pr_notice("crng reseeded due to virtual machine fork\n");
973 }
974 blocking_notifier_call_chain(&vmfork_chain, 0, NULL);
975}
976#if IS_MODULE(CONFIG_VMGENID)
977EXPORT_SYMBOL_GPL(add_vmfork_randomness);
978#endif
979
980int __cold register_random_vmfork_notifier(struct notifier_block *nb)
981{
982 return blocking_notifier_chain_register(&vmfork_chain, nb);
983}
984EXPORT_SYMBOL_GPL(register_random_vmfork_notifier);
985
986int __cold unregister_random_vmfork_notifier(struct notifier_block *nb)
987{
988 return blocking_notifier_chain_unregister(&vmfork_chain, nb);
989}
990EXPORT_SYMBOL_GPL(unregister_random_vmfork_notifier);
991#endif
992
993struct fast_pool {
994 unsigned long pool[4];
995 unsigned long last;
996 unsigned int count;
997 struct timer_list mix;
998};
999
1000static void mix_interrupt_randomness(struct timer_list *work);
1001
1002static DEFINE_PER_CPU(struct fast_pool, irq_randomness) = {
1003#ifdef CONFIG_64BIT
1004#define FASTMIX_PERM SIPHASH_PERMUTATION
1005 .pool = { SIPHASH_CONST_0, SIPHASH_CONST_1, SIPHASH_CONST_2, SIPHASH_CONST_3 },
1006#else
1007#define FASTMIX_PERM HSIPHASH_PERMUTATION
1008 .pool = { HSIPHASH_CONST_0, HSIPHASH_CONST_1, HSIPHASH_CONST_2, HSIPHASH_CONST_3 },
1009#endif
1010 .mix = __TIMER_INITIALIZER(mix_interrupt_randomness, 0)
1011};
1012
1013/*
1014 * This is [Half]SipHash-1-x, starting from an empty key. Because
1015 * the key is fixed, it assumes that its inputs are non-malicious,
1016 * and therefore this has no security on its own. s represents the
1017 * four-word SipHash state, while v represents a two-word input.
1018 */
1019static void fast_mix(unsigned long s[4], unsigned long v1, unsigned long v2)
1020{
1021 s[3] ^= v1;
1022 FASTMIX_PERM(s[0], s[1], s[2], s[3]);
1023 s[0] ^= v1;
1024 s[3] ^= v2;
1025 FASTMIX_PERM(s[0], s[1], s[2], s[3]);
1026 s[0] ^= v2;
1027}
1028
1029#ifdef CONFIG_SMP
1030/*
1031 * This function is called when the CPU has just come online, with
1032 * entry CPUHP_AP_RANDOM_ONLINE, just after CPUHP_AP_WORKQUEUE_ONLINE.
1033 */
1034int __cold random_online_cpu(unsigned int cpu)
1035{
1036 /*
1037 * During CPU shutdown and before CPU onlining, add_interrupt_
1038 * randomness() may schedule mix_interrupt_randomness(), and
1039 * set the MIX_INFLIGHT flag. However, because the worker can
1040 * be scheduled on a different CPU during this period, that
1041 * flag will never be cleared. For that reason, we zero out
1042 * the flag here, which runs just after workqueues are onlined
1043 * for the CPU again. This also has the effect of setting the
1044 * irq randomness count to zero so that new accumulated irqs
1045 * are fresh.
1046 */
1047 per_cpu_ptr(&irq_randomness, cpu)->count = 0;
1048 return 0;
1049}
1050#endif
1051
1052static void mix_interrupt_randomness(struct timer_list *work)
1053{
1054 struct fast_pool *fast_pool = container_of(work, struct fast_pool, mix);
1055 /*
1056 * The size of the copied stack pool is explicitly 2 longs so that we
1057 * only ever ingest half of the siphash output each time, retaining
1058 * the other half as the next "key" that carries over. The entropy is
1059 * supposed to be sufficiently dispersed between bits so on average
1060 * we don't wind up "losing" some.
1061 */
1062 unsigned long pool[2];
1063 unsigned int count;
1064
1065 /* Check to see if we're running on the wrong CPU due to hotplug. */
1066 local_irq_disable();
1067 if (fast_pool != this_cpu_ptr(&irq_randomness)) {
1068 local_irq_enable();
1069 return;
1070 }
1071
1072 /*
1073 * Copy the pool to the stack so that the mixer always has a
1074 * consistent view, before we reenable irqs again.
1075 */
1076 memcpy(pool, fast_pool->pool, sizeof(pool));
1077 count = fast_pool->count;
1078 fast_pool->count = 0;
1079 fast_pool->last = jiffies;
1080 local_irq_enable();
1081
1082 mix_pool_bytes(pool, sizeof(pool));
1083 credit_init_bits(clamp_t(unsigned int, (count & U16_MAX) / 64, 1, sizeof(pool) * 8));
1084
1085 memzero_explicit(pool, sizeof(pool));
1086}
1087
1088void add_interrupt_randomness(int irq)
1089{
1090 enum { MIX_INFLIGHT = 1U << 31 };
1091 unsigned long entropy = random_get_entropy();
1092 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1093 struct pt_regs *regs = get_irq_regs();
1094 unsigned int new_count;
1095
1096 fast_mix(fast_pool->pool, entropy,
1097 (regs ? instruction_pointer(regs) : _RET_IP_) ^ swab(irq));
1098 new_count = ++fast_pool->count;
1099
1100 if (new_count & MIX_INFLIGHT)
1101 return;
1102
1103 if (new_count < 1024 && !time_is_before_jiffies(fast_pool->last + HZ))
1104 return;
1105
1106 fast_pool->count |= MIX_INFLIGHT;
1107 if (!timer_pending(&fast_pool->mix)) {
1108 fast_pool->mix.expires = jiffies;
1109 add_timer_on(&fast_pool->mix, raw_smp_processor_id());
1110 }
1111}
1112EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1113
1114/* There is one of these per entropy source */
1115struct timer_rand_state {
1116 unsigned long last_time;
1117 long last_delta, last_delta2;
1118};
1119
1120/*
1121 * This function adds entropy to the entropy "pool" by using timing
1122 * delays. It uses the timer_rand_state structure to make an estimate
1123 * of how many bits of entropy this call has added to the pool. The
1124 * value "num" is also added to the pool; it should somehow describe
1125 * the type of event that just happened.
1126 */
1127static void add_timer_randomness(struct timer_rand_state *state, unsigned int num)
1128{
1129 unsigned long entropy = random_get_entropy(), now = jiffies, flags;
1130 long delta, delta2, delta3;
1131 unsigned int bits;
1132
1133 /*
1134 * If we're in a hard IRQ, add_interrupt_randomness() will be called
1135 * sometime after, so mix into the fast pool.
1136 */
1137 if (in_hardirq()) {
1138 fast_mix(this_cpu_ptr(&irq_randomness)->pool, entropy, num);
1139 } else {
1140 spin_lock_irqsave(&input_pool.lock, flags);
1141 _mix_pool_bytes(&entropy, sizeof(entropy));
1142 _mix_pool_bytes(&num, sizeof(num));
1143 spin_unlock_irqrestore(&input_pool.lock, flags);
1144 }
1145
1146 if (crng_ready())
1147 return;
1148
1149 /*
1150 * Calculate number of bits of randomness we probably added.
1151 * We take into account the first, second and third-order deltas
1152 * in order to make our estimate.
1153 */
1154 delta = now - READ_ONCE(state->last_time);
1155 WRITE_ONCE(state->last_time, now);
1156
1157 delta2 = delta - READ_ONCE(state->last_delta);
1158 WRITE_ONCE(state->last_delta, delta);
1159
1160 delta3 = delta2 - READ_ONCE(state->last_delta2);
1161 WRITE_ONCE(state->last_delta2, delta2);
1162
1163 if (delta < 0)
1164 delta = -delta;
1165 if (delta2 < 0)
1166 delta2 = -delta2;
1167 if (delta3 < 0)
1168 delta3 = -delta3;
1169 if (delta > delta2)
1170 delta = delta2;
1171 if (delta > delta3)
1172 delta = delta3;
1173
1174 /*
1175 * delta is now minimum absolute delta. Round down by 1 bit
1176 * on general principles, and limit entropy estimate to 11 bits.
1177 */
1178 bits = min(fls(delta >> 1), 11);
1179
1180 /*
1181 * As mentioned above, if we're in a hard IRQ, add_interrupt_randomness()
1182 * will run after this, which uses a different crediting scheme of 1 bit
1183 * per every 64 interrupts. In order to let that function do accounting
1184 * close to the one in this function, we credit a full 64/64 bit per bit,
1185 * and then subtract one to account for the extra one added.
1186 */
1187 if (in_hardirq())
1188 this_cpu_ptr(&irq_randomness)->count += max(1u, bits * 64) - 1;
1189 else
1190 _credit_init_bits(bits);
1191}
1192
1193void add_input_randomness(unsigned int type, unsigned int code, unsigned int value)
1194{
1195 static unsigned char last_value;
1196 static struct timer_rand_state input_timer_state = { INITIAL_JIFFIES };
1197
1198 /* Ignore autorepeat and the like. */
1199 if (value == last_value)
1200 return;
1201
1202 last_value = value;
1203 add_timer_randomness(&input_timer_state,
1204 (type << 4) ^ code ^ (code >> 4) ^ value);
1205}
1206EXPORT_SYMBOL_GPL(add_input_randomness);
1207
1208#ifdef CONFIG_BLOCK
1209void add_disk_randomness(struct gendisk *disk)
1210{
1211 if (!disk || !disk->random)
1212 return;
1213 /* First major is 1, so we get >= 0x200 here. */
1214 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1215}
1216EXPORT_SYMBOL_GPL(add_disk_randomness);
1217
1218void __cold rand_initialize_disk(struct gendisk *disk)
1219{
1220 struct timer_rand_state *state;
1221
1222 /*
1223 * If kzalloc returns null, we just won't use that entropy
1224 * source.
1225 */
1226 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1227 if (state) {
1228 state->last_time = INITIAL_JIFFIES;
1229 disk->random = state;
1230 }
1231}
1232#endif
1233
1234struct entropy_timer_state {
1235 unsigned long entropy;
1236 struct timer_list timer;
1237 atomic_t samples;
1238 unsigned int samples_per_bit;
1239};
1240
1241/*
1242 * Each time the timer fires, we expect that we got an unpredictable jump in
1243 * the cycle counter. Even if the timer is running on another CPU, the timer
1244 * activity will be touching the stack of the CPU that is generating entropy.
1245 *
1246 * Note that we don't re-arm the timer in the timer itself - we are happy to be
1247 * scheduled away, since that just makes the load more complex, but we do not
1248 * want the timer to keep ticking unless the entropy loop is running.
1249 *
1250 * So the re-arming always happens in the entropy loop itself.
1251 */
1252static void __cold entropy_timer(struct timer_list *timer)
1253{
1254 struct entropy_timer_state *state = container_of(timer, struct entropy_timer_state, timer);
1255 unsigned long entropy = random_get_entropy();
1256
1257 mix_pool_bytes(&entropy, sizeof(entropy));
1258 if (atomic_inc_return(&state->samples) % state->samples_per_bit == 0)
1259 credit_init_bits(1);
1260}
1261
1262/*
1263 * If we have an actual cycle counter, see if we can generate enough entropy
1264 * with timing noise.
1265 */
1266static void __cold try_to_generate_entropy(void)
1267{
1268 enum { NUM_TRIAL_SAMPLES = 8192, MAX_SAMPLES_PER_BIT = HZ / 15 };
1269 u8 stack_bytes[sizeof(struct entropy_timer_state) + SMP_CACHE_BYTES - 1];
1270 struct entropy_timer_state *stack = PTR_ALIGN((void *)stack_bytes, SMP_CACHE_BYTES);
1271 unsigned int i, num_different = 0;
1272 unsigned long last = random_get_entropy();
1273 int cpu = -1;
1274
1275 for (i = 0; i < NUM_TRIAL_SAMPLES - 1; ++i) {
1276 stack->entropy = random_get_entropy();
1277 if (stack->entropy != last)
1278 ++num_different;
1279 last = stack->entropy;
1280 }
1281 stack->samples_per_bit = DIV_ROUND_UP(NUM_TRIAL_SAMPLES, num_different + 1);
1282 if (stack->samples_per_bit > MAX_SAMPLES_PER_BIT)
1283 return;
1284
1285 atomic_set(&stack->samples, 0);
1286 timer_setup_on_stack(&stack->timer, entropy_timer, 0);
1287 while (!crng_ready() && !signal_pending(current)) {
1288 /*
1289 * Check !timer_pending() and then ensure that any previous callback has finished
1290 * executing by checking try_to_del_timer_sync(), before queueing the next one.
1291 */
1292 if (!timer_pending(&stack->timer) && try_to_del_timer_sync(&stack->timer) >= 0) {
1293 struct cpumask timer_cpus;
1294 unsigned int num_cpus;
1295
1296 /*
1297 * Preemption must be disabled here, both to read the current CPU number
1298 * and to avoid scheduling a timer on a dead CPU.
1299 */
1300 preempt_disable();
1301
1302 /* Only schedule callbacks on timer CPUs that are online. */
1303 cpumask_and(&timer_cpus, housekeeping_cpumask(HK_TYPE_TIMER), cpu_online_mask);
1304 num_cpus = cpumask_weight(&timer_cpus);
1305 /* In very bizarre case of misconfiguration, fallback to all online. */
1306 if (unlikely(num_cpus == 0)) {
1307 timer_cpus = *cpu_online_mask;
1308 num_cpus = cpumask_weight(&timer_cpus);
1309 }
1310
1311 /* Basic CPU round-robin, which avoids the current CPU. */
1312 do {
1313 cpu = cpumask_next(cpu, &timer_cpus);
1314 if (cpu >= nr_cpu_ids)
1315 cpu = cpumask_first(&timer_cpus);
1316 } while (cpu == smp_processor_id() && num_cpus > 1);
1317
1318 /* Expiring the timer at `jiffies` means it's the next tick. */
1319 stack->timer.expires = jiffies;
1320
1321 add_timer_on(&stack->timer, cpu);
1322
1323 preempt_enable();
1324 }
1325 mix_pool_bytes(&stack->entropy, sizeof(stack->entropy));
1326 schedule();
1327 stack->entropy = random_get_entropy();
1328 }
1329 mix_pool_bytes(&stack->entropy, sizeof(stack->entropy));
1330
1331 del_timer_sync(&stack->timer);
1332 destroy_timer_on_stack(&stack->timer);
1333}
1334
1335
1336/**********************************************************************
1337 *
1338 * Userspace reader/writer interfaces.
1339 *
1340 * getrandom(2) is the primary modern interface into the RNG and should
1341 * be used in preference to anything else.
1342 *
1343 * Reading from /dev/random has the same functionality as calling
1344 * getrandom(2) with flags=0. In earlier versions, however, it had
1345 * vastly different semantics and should therefore be avoided, to
1346 * prevent backwards compatibility issues.
1347 *
1348 * Reading from /dev/urandom has the same functionality as calling
1349 * getrandom(2) with flags=GRND_INSECURE. Because it does not block
1350 * waiting for the RNG to be ready, it should not be used.
1351 *
1352 * Writing to either /dev/random or /dev/urandom adds entropy to
1353 * the input pool but does not credit it.
1354 *
1355 * Polling on /dev/random indicates when the RNG is initialized, on
1356 * the read side, and when it wants new entropy, on the write side.
1357 *
1358 * Both /dev/random and /dev/urandom have the same set of ioctls for
1359 * adding entropy, getting the entropy count, zeroing the count, and
1360 * reseeding the crng.
1361 *
1362 **********************************************************************/
1363
1364SYSCALL_DEFINE3(getrandom, char __user *, ubuf, size_t, len, unsigned int, flags)
1365{
1366 struct iov_iter iter;
1367 int ret;
1368
1369 if (flags & ~(GRND_NONBLOCK | GRND_RANDOM | GRND_INSECURE))
1370 return -EINVAL;
1371
1372 /*
1373 * Requesting insecure and blocking randomness at the same time makes
1374 * no sense.
1375 */
1376 if ((flags & (GRND_INSECURE | GRND_RANDOM)) == (GRND_INSECURE | GRND_RANDOM))
1377 return -EINVAL;
1378
1379 if (!crng_ready() && !(flags & GRND_INSECURE)) {
1380 if (flags & GRND_NONBLOCK)
1381 return -EAGAIN;
1382 ret = wait_for_random_bytes();
1383 if (unlikely(ret))
1384 return ret;
1385 }
1386
1387 ret = import_ubuf(ITER_DEST, ubuf, len, &iter);
1388 if (unlikely(ret))
1389 return ret;
1390 return get_random_bytes_user(&iter);
1391}
1392
1393static __poll_t random_poll(struct file *file, poll_table *wait)
1394{
1395 poll_wait(file, &crng_init_wait, wait);
1396 return crng_ready() ? EPOLLIN | EPOLLRDNORM : EPOLLOUT | EPOLLWRNORM;
1397}
1398
1399static ssize_t write_pool_user(struct iov_iter *iter)
1400{
1401 u8 block[BLAKE2S_BLOCK_SIZE];
1402 ssize_t ret = 0;
1403 size_t copied;
1404
1405 if (unlikely(!iov_iter_count(iter)))
1406 return 0;
1407
1408 for (;;) {
1409 copied = copy_from_iter(block, sizeof(block), iter);
1410 ret += copied;
1411 mix_pool_bytes(block, copied);
1412 if (!iov_iter_count(iter) || copied != sizeof(block))
1413 break;
1414
1415 BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0);
1416 if (ret % PAGE_SIZE == 0) {
1417 if (signal_pending(current))
1418 break;
1419 cond_resched();
1420 }
1421 }
1422
1423 memzero_explicit(block, sizeof(block));
1424 return ret ? ret : -EFAULT;
1425}
1426
1427static ssize_t random_write_iter(struct kiocb *kiocb, struct iov_iter *iter)
1428{
1429 return write_pool_user(iter);
1430}
1431
1432static ssize_t urandom_read_iter(struct kiocb *kiocb, struct iov_iter *iter)
1433{
1434 static int maxwarn = 10;
1435
1436 /*
1437 * Opportunistically attempt to initialize the RNG on platforms that
1438 * have fast cycle counters, but don't (for now) require it to succeed.
1439 */
1440 if (!crng_ready())
1441 try_to_generate_entropy();
1442
1443 if (!crng_ready()) {
1444 if (!ratelimit_disable && maxwarn <= 0)
1445 ++urandom_warning.missed;
1446 else if (ratelimit_disable || __ratelimit(&urandom_warning)) {
1447 --maxwarn;
1448 pr_notice("%s: uninitialized urandom read (%zu bytes read)\n",
1449 current->comm, iov_iter_count(iter));
1450 }
1451 }
1452
1453 return get_random_bytes_user(iter);
1454}
1455
1456static ssize_t random_read_iter(struct kiocb *kiocb, struct iov_iter *iter)
1457{
1458 int ret;
1459
1460 if (!crng_ready() &&
1461 ((kiocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO)) ||
1462 (kiocb->ki_filp->f_flags & O_NONBLOCK)))
1463 return -EAGAIN;
1464
1465 ret = wait_for_random_bytes();
1466 if (ret != 0)
1467 return ret;
1468 return get_random_bytes_user(iter);
1469}
1470
1471static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1472{
1473 int __user *p = (int __user *)arg;
1474 int ent_count;
1475
1476 switch (cmd) {
1477 case RNDGETENTCNT:
1478 /* Inherently racy, no point locking. */
1479 if (put_user(input_pool.init_bits, p))
1480 return -EFAULT;
1481 return 0;
1482 case RNDADDTOENTCNT:
1483 if (!capable(CAP_SYS_ADMIN))
1484 return -EPERM;
1485 if (get_user(ent_count, p))
1486 return -EFAULT;
1487 if (ent_count < 0)
1488 return -EINVAL;
1489 credit_init_bits(ent_count);
1490 return 0;
1491 case RNDADDENTROPY: {
1492 struct iov_iter iter;
1493 ssize_t ret;
1494 int len;
1495
1496 if (!capable(CAP_SYS_ADMIN))
1497 return -EPERM;
1498 if (get_user(ent_count, p++))
1499 return -EFAULT;
1500 if (ent_count < 0)
1501 return -EINVAL;
1502 if (get_user(len, p++))
1503 return -EFAULT;
1504 ret = import_ubuf(ITER_SOURCE, p, len, &iter);
1505 if (unlikely(ret))
1506 return ret;
1507 ret = write_pool_user(&iter);
1508 if (unlikely(ret < 0))
1509 return ret;
1510 /* Since we're crediting, enforce that it was all written into the pool. */
1511 if (unlikely(ret != len))
1512 return -EFAULT;
1513 credit_init_bits(ent_count);
1514 return 0;
1515 }
1516 case RNDZAPENTCNT:
1517 case RNDCLEARPOOL:
1518 /* No longer has any effect. */
1519 if (!capable(CAP_SYS_ADMIN))
1520 return -EPERM;
1521 return 0;
1522 case RNDRESEEDCRNG:
1523 if (!capable(CAP_SYS_ADMIN))
1524 return -EPERM;
1525 if (!crng_ready())
1526 return -ENODATA;
1527 crng_reseed(NULL);
1528 return 0;
1529 default:
1530 return -EINVAL;
1531 }
1532}
1533
1534static int random_fasync(int fd, struct file *filp, int on)
1535{
1536 return fasync_helper(fd, filp, on, &fasync);
1537}
1538
1539const struct file_operations random_fops = {
1540 .read_iter = random_read_iter,
1541 .write_iter = random_write_iter,
1542 .poll = random_poll,
1543 .unlocked_ioctl = random_ioctl,
1544 .compat_ioctl = compat_ptr_ioctl,
1545 .fasync = random_fasync,
1546 .llseek = noop_llseek,
1547 .splice_read = copy_splice_read,
1548 .splice_write = iter_file_splice_write,
1549};
1550
1551const struct file_operations urandom_fops = {
1552 .read_iter = urandom_read_iter,
1553 .write_iter = random_write_iter,
1554 .unlocked_ioctl = random_ioctl,
1555 .compat_ioctl = compat_ptr_ioctl,
1556 .fasync = random_fasync,
1557 .llseek = noop_llseek,
1558 .splice_read = copy_splice_read,
1559 .splice_write = iter_file_splice_write,
1560};
1561
1562
1563/********************************************************************
1564 *
1565 * Sysctl interface.
1566 *
1567 * These are partly unused legacy knobs with dummy values to not break
1568 * userspace and partly still useful things. They are usually accessible
1569 * in /proc/sys/kernel/random/ and are as follows:
1570 *
1571 * - boot_id - a UUID representing the current boot.
1572 *
1573 * - uuid - a random UUID, different each time the file is read.
1574 *
1575 * - poolsize - the number of bits of entropy that the input pool can
1576 * hold, tied to the POOL_BITS constant.
1577 *
1578 * - entropy_avail - the number of bits of entropy currently in the
1579 * input pool. Always <= poolsize.
1580 *
1581 * - write_wakeup_threshold - the amount of entropy in the input pool
1582 * below which write polls to /dev/random will unblock, requesting
1583 * more entropy, tied to the POOL_READY_BITS constant. It is writable
1584 * to avoid breaking old userspaces, but writing to it does not
1585 * change any behavior of the RNG.
1586 *
1587 * - urandom_min_reseed_secs - fixed to the value CRNG_RESEED_INTERVAL.
1588 * It is writable to avoid breaking old userspaces, but writing
1589 * to it does not change any behavior of the RNG.
1590 *
1591 ********************************************************************/
1592
1593#ifdef CONFIG_SYSCTL
1594
1595#include <linux/sysctl.h>
1596
1597static int sysctl_random_min_urandom_seed = CRNG_RESEED_INTERVAL / HZ;
1598static int sysctl_random_write_wakeup_bits = POOL_READY_BITS;
1599static int sysctl_poolsize = POOL_BITS;
1600static u8 sysctl_bootid[UUID_SIZE];
1601
1602/*
1603 * This function is used to return both the bootid UUID, and random
1604 * UUID. The difference is in whether table->data is NULL; if it is,
1605 * then a new UUID is generated and returned to the user.
1606 */
1607static int proc_do_uuid(struct ctl_table *table, int write, void *buf,
1608 size_t *lenp, loff_t *ppos)
1609{
1610 u8 tmp_uuid[UUID_SIZE], *uuid;
1611 char uuid_string[UUID_STRING_LEN + 1];
1612 struct ctl_table fake_table = {
1613 .data = uuid_string,
1614 .maxlen = UUID_STRING_LEN
1615 };
1616
1617 if (write)
1618 return -EPERM;
1619
1620 uuid = table->data;
1621 if (!uuid) {
1622 uuid = tmp_uuid;
1623 generate_random_uuid(uuid);
1624 } else {
1625 static DEFINE_SPINLOCK(bootid_spinlock);
1626
1627 spin_lock(&bootid_spinlock);
1628 if (!uuid[8])
1629 generate_random_uuid(uuid);
1630 spin_unlock(&bootid_spinlock);
1631 }
1632
1633 snprintf(uuid_string, sizeof(uuid_string), "%pU", uuid);
1634 return proc_dostring(&fake_table, 0, buf, lenp, ppos);
1635}
1636
1637/* The same as proc_dointvec, but writes don't change anything. */
1638static int proc_do_rointvec(struct ctl_table *table, int write, void *buf,
1639 size_t *lenp, loff_t *ppos)
1640{
1641 return write ? 0 : proc_dointvec(table, 0, buf, lenp, ppos);
1642}
1643
1644static struct ctl_table random_table[] = {
1645 {
1646 .procname = "poolsize",
1647 .data = &sysctl_poolsize,
1648 .maxlen = sizeof(int),
1649 .mode = 0444,
1650 .proc_handler = proc_dointvec,
1651 },
1652 {
1653 .procname = "entropy_avail",
1654 .data = &input_pool.init_bits,
1655 .maxlen = sizeof(int),
1656 .mode = 0444,
1657 .proc_handler = proc_dointvec,
1658 },
1659 {
1660 .procname = "write_wakeup_threshold",
1661 .data = &sysctl_random_write_wakeup_bits,
1662 .maxlen = sizeof(int),
1663 .mode = 0644,
1664 .proc_handler = proc_do_rointvec,
1665 },
1666 {
1667 .procname = "urandom_min_reseed_secs",
1668 .data = &sysctl_random_min_urandom_seed,
1669 .maxlen = sizeof(int),
1670 .mode = 0644,
1671 .proc_handler = proc_do_rointvec,
1672 },
1673 {
1674 .procname = "boot_id",
1675 .data = &sysctl_bootid,
1676 .mode = 0444,
1677 .proc_handler = proc_do_uuid,
1678 },
1679 {
1680 .procname = "uuid",
1681 .mode = 0444,
1682 .proc_handler = proc_do_uuid,
1683 },
1684};
1685
1686/*
1687 * random_init() is called before sysctl_init(),
1688 * so we cannot call register_sysctl_init() in random_init()
1689 */
1690static int __init random_sysctls_init(void)
1691{
1692 register_sysctl_init("kernel/random", random_table);
1693 return 0;
1694}
1695device_initcall(random_sysctls_init);
1696#endif
1/*
2 * random.c -- A strong random number generator
3 *
4 * Copyright (C) 2017 Jason A. Donenfeld <Jason@zx2c4.com>. All
5 * Rights Reserved.
6 *
7 * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005
8 *
9 * Copyright Theodore Ts'o, 1994, 1995, 1996, 1997, 1998, 1999. All
10 * rights reserved.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, and the entire permission notice in its entirety,
17 * including the disclaimer of warranties.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. The name of the author may not be used to endorse or promote
22 * products derived from this software without specific prior
23 * written permission.
24 *
25 * ALTERNATIVELY, this product may be distributed under the terms of
26 * the GNU General Public License, in which case the provisions of the GPL are
27 * required INSTEAD OF the above restrictions. (This clause is
28 * necessary due to a potential bad interaction between the GPL and
29 * the restrictions contained in a BSD-style copyright.)
30 *
31 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
32 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
33 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ALL OF
34 * WHICH ARE HEREBY DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE
35 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
36 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT
37 * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
38 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
39 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
40 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
41 * USE OF THIS SOFTWARE, EVEN IF NOT ADVISED OF THE POSSIBILITY OF SUCH
42 * DAMAGE.
43 */
44
45/*
46 * (now, with legal B.S. out of the way.....)
47 *
48 * This routine gathers environmental noise from device drivers, etc.,
49 * and returns good random numbers, suitable for cryptographic use.
50 * Besides the obvious cryptographic uses, these numbers are also good
51 * for seeding TCP sequence numbers, and other places where it is
52 * desirable to have numbers which are not only random, but hard to
53 * predict by an attacker.
54 *
55 * Theory of operation
56 * ===================
57 *
58 * Computers are very predictable devices. Hence it is extremely hard
59 * to produce truly random numbers on a computer --- as opposed to
60 * pseudo-random numbers, which can easily generated by using a
61 * algorithm. Unfortunately, it is very easy for attackers to guess
62 * the sequence of pseudo-random number generators, and for some
63 * applications this is not acceptable. So instead, we must try to
64 * gather "environmental noise" from the computer's environment, which
65 * must be hard for outside attackers to observe, and use that to
66 * generate random numbers. In a Unix environment, this is best done
67 * from inside the kernel.
68 *
69 * Sources of randomness from the environment include inter-keyboard
70 * timings, inter-interrupt timings from some interrupts, and other
71 * events which are both (a) non-deterministic and (b) hard for an
72 * outside observer to measure. Randomness from these sources are
73 * added to an "entropy pool", which is mixed using a CRC-like function.
74 * This is not cryptographically strong, but it is adequate assuming
75 * the randomness is not chosen maliciously, and it is fast enough that
76 * the overhead of doing it on every interrupt is very reasonable.
77 * As random bytes are mixed into the entropy pool, the routines keep
78 * an *estimate* of how many bits of randomness have been stored into
79 * the random number generator's internal state.
80 *
81 * When random bytes are desired, they are obtained by taking the SHA
82 * hash of the contents of the "entropy pool". The SHA hash avoids
83 * exposing the internal state of the entropy pool. It is believed to
84 * be computationally infeasible to derive any useful information
85 * about the input of SHA from its output. Even if it is possible to
86 * analyze SHA in some clever way, as long as the amount of data
87 * returned from the generator is less than the inherent entropy in
88 * the pool, the output data is totally unpredictable. For this
89 * reason, the routine decreases its internal estimate of how many
90 * bits of "true randomness" are contained in the entropy pool as it
91 * outputs random numbers.
92 *
93 * If this estimate goes to zero, the routine can still generate
94 * random numbers; however, an attacker may (at least in theory) be
95 * able to infer the future output of the generator from prior
96 * outputs. This requires successful cryptanalysis of SHA, which is
97 * not believed to be feasible, but there is a remote possibility.
98 * Nonetheless, these numbers should be useful for the vast majority
99 * of purposes.
100 *
101 * Exported interfaces ---- output
102 * ===============================
103 *
104 * There are four exported interfaces; two for use within the kernel,
105 * and two or use from userspace.
106 *
107 * Exported interfaces ---- userspace output
108 * -----------------------------------------
109 *
110 * The userspace interfaces are two character devices /dev/random and
111 * /dev/urandom. /dev/random is suitable for use when very high
112 * quality randomness is desired (for example, for key generation or
113 * one-time pads), as it will only return a maximum of the number of
114 * bits of randomness (as estimated by the random number generator)
115 * contained in the entropy pool.
116 *
117 * The /dev/urandom device does not have this limit, and will return
118 * as many bytes as are requested. As more and more random bytes are
119 * requested without giving time for the entropy pool to recharge,
120 * this will result in random numbers that are merely cryptographically
121 * strong. For many applications, however, this is acceptable.
122 *
123 * Exported interfaces ---- kernel output
124 * --------------------------------------
125 *
126 * The primary kernel interface is
127 *
128 * void get_random_bytes(void *buf, int nbytes);
129 *
130 * This interface will return the requested number of random bytes,
131 * and place it in the requested buffer. This is equivalent to a
132 * read from /dev/urandom.
133 *
134 * For less critical applications, there are the functions:
135 *
136 * u32 get_random_u32()
137 * u64 get_random_u64()
138 * unsigned int get_random_int()
139 * unsigned long get_random_long()
140 *
141 * These are produced by a cryptographic RNG seeded from get_random_bytes,
142 * and so do not deplete the entropy pool as much. These are recommended
143 * for most in-kernel operations *if the result is going to be stored in
144 * the kernel*.
145 *
146 * Specifically, the get_random_int() family do not attempt to do
147 * "anti-backtracking". If you capture the state of the kernel (e.g.
148 * by snapshotting the VM), you can figure out previous get_random_int()
149 * return values. But if the value is stored in the kernel anyway,
150 * this is not a problem.
151 *
152 * It *is* safe to expose get_random_int() output to attackers (e.g. as
153 * network cookies); given outputs 1..n, it's not feasible to predict
154 * outputs 0 or n+1. The only concern is an attacker who breaks into
155 * the kernel later; the get_random_int() engine is not reseeded as
156 * often as the get_random_bytes() one.
157 *
158 * get_random_bytes() is needed for keys that need to stay secret after
159 * they are erased from the kernel. For example, any key that will
160 * be wrapped and stored encrypted. And session encryption keys: we'd
161 * like to know that after the session is closed and the keys erased,
162 * the plaintext is unrecoverable to someone who recorded the ciphertext.
163 *
164 * But for network ports/cookies, stack canaries, PRNG seeds, address
165 * space layout randomization, session *authentication* keys, or other
166 * applications where the sensitive data is stored in the kernel in
167 * plaintext for as long as it's sensitive, the get_random_int() family
168 * is just fine.
169 *
170 * Consider ASLR. We want to keep the address space secret from an
171 * outside attacker while the process is running, but once the address
172 * space is torn down, it's of no use to an attacker any more. And it's
173 * stored in kernel data structures as long as it's alive, so worrying
174 * about an attacker's ability to extrapolate it from the get_random_int()
175 * CRNG is silly.
176 *
177 * Even some cryptographic keys are safe to generate with get_random_int().
178 * In particular, keys for SipHash are generally fine. Here, knowledge
179 * of the key authorizes you to do something to a kernel object (inject
180 * packets to a network connection, or flood a hash table), and the
181 * key is stored with the object being protected. Once it goes away,
182 * we no longer care if anyone knows the key.
183 *
184 * prandom_u32()
185 * -------------
186 *
187 * For even weaker applications, see the pseudorandom generator
188 * prandom_u32(), prandom_max(), and prandom_bytes(). If the random
189 * numbers aren't security-critical at all, these are *far* cheaper.
190 * Useful for self-tests, random error simulation, randomized backoffs,
191 * and any other application where you trust that nobody is trying to
192 * maliciously mess with you by guessing the "random" numbers.
193 *
194 * Exported interfaces ---- input
195 * ==============================
196 *
197 * The current exported interfaces for gathering environmental noise
198 * from the devices are:
199 *
200 * void add_device_randomness(const void *buf, unsigned int size);
201 * void add_input_randomness(unsigned int type, unsigned int code,
202 * unsigned int value);
203 * void add_interrupt_randomness(int irq, int irq_flags);
204 * void add_disk_randomness(struct gendisk *disk);
205 *
206 * add_device_randomness() is for adding data to the random pool that
207 * is likely to differ between two devices (or possibly even per boot).
208 * This would be things like MAC addresses or serial numbers, or the
209 * read-out of the RTC. This does *not* add any actual entropy to the
210 * pool, but it initializes the pool to different values for devices
211 * that might otherwise be identical and have very little entropy
212 * available to them (particularly common in the embedded world).
213 *
214 * add_input_randomness() uses the input layer interrupt timing, as well as
215 * the event type information from the hardware.
216 *
217 * add_interrupt_randomness() uses the interrupt timing as random
218 * inputs to the entropy pool. Using the cycle counters and the irq source
219 * as inputs, it feeds the randomness roughly once a second.
220 *
221 * add_disk_randomness() uses what amounts to the seek time of block
222 * layer request events, on a per-disk_devt basis, as input to the
223 * entropy pool. Note that high-speed solid state drives with very low
224 * seek times do not make for good sources of entropy, as their seek
225 * times are usually fairly consistent.
226 *
227 * All of these routines try to estimate how many bits of randomness a
228 * particular randomness source. They do this by keeping track of the
229 * first and second order deltas of the event timings.
230 *
231 * Ensuring unpredictability at system startup
232 * ============================================
233 *
234 * When any operating system starts up, it will go through a sequence
235 * of actions that are fairly predictable by an adversary, especially
236 * if the start-up does not involve interaction with a human operator.
237 * This reduces the actual number of bits of unpredictability in the
238 * entropy pool below the value in entropy_count. In order to
239 * counteract this effect, it helps to carry information in the
240 * entropy pool across shut-downs and start-ups. To do this, put the
241 * following lines an appropriate script which is run during the boot
242 * sequence:
243 *
244 * echo "Initializing random number generator..."
245 * random_seed=/var/run/random-seed
246 * # Carry a random seed from start-up to start-up
247 * # Load and then save the whole entropy pool
248 * if [ -f $random_seed ]; then
249 * cat $random_seed >/dev/urandom
250 * else
251 * touch $random_seed
252 * fi
253 * chmod 600 $random_seed
254 * dd if=/dev/urandom of=$random_seed count=1 bs=512
255 *
256 * and the following lines in an appropriate script which is run as
257 * the system is shutdown:
258 *
259 * # Carry a random seed from shut-down to start-up
260 * # Save the whole entropy pool
261 * echo "Saving random seed..."
262 * random_seed=/var/run/random-seed
263 * touch $random_seed
264 * chmod 600 $random_seed
265 * dd if=/dev/urandom of=$random_seed count=1 bs=512
266 *
267 * For example, on most modern systems using the System V init
268 * scripts, such code fragments would be found in
269 * /etc/rc.d/init.d/random. On older Linux systems, the correct script
270 * location might be in /etc/rcb.d/rc.local or /etc/rc.d/rc.0.
271 *
272 * Effectively, these commands cause the contents of the entropy pool
273 * to be saved at shut-down time and reloaded into the entropy pool at
274 * start-up. (The 'dd' in the addition to the bootup script is to
275 * make sure that /etc/random-seed is different for every start-up,
276 * even if the system crashes without executing rc.0.) Even with
277 * complete knowledge of the start-up activities, predicting the state
278 * of the entropy pool requires knowledge of the previous history of
279 * the system.
280 *
281 * Configuring the /dev/random driver under Linux
282 * ==============================================
283 *
284 * The /dev/random driver under Linux uses minor numbers 8 and 9 of
285 * the /dev/mem major number (#1). So if your system does not have
286 * /dev/random and /dev/urandom created already, they can be created
287 * by using the commands:
288 *
289 * mknod /dev/random c 1 8
290 * mknod /dev/urandom c 1 9
291 *
292 * Acknowledgements:
293 * =================
294 *
295 * Ideas for constructing this random number generator were derived
296 * from Pretty Good Privacy's random number generator, and from private
297 * discussions with Phil Karn. Colin Plumb provided a faster random
298 * number generator, which speed up the mixing function of the entropy
299 * pool, taken from PGPfone. Dale Worley has also contributed many
300 * useful ideas and suggestions to improve this driver.
301 *
302 * Any flaws in the design are solely my responsibility, and should
303 * not be attributed to the Phil, Colin, or any of authors of PGP.
304 *
305 * Further background information on this topic may be obtained from
306 * RFC 1750, "Randomness Recommendations for Security", by Donald
307 * Eastlake, Steve Crocker, and Jeff Schiller.
308 */
309
310#include <linux/utsname.h>
311#include <linux/module.h>
312#include <linux/kernel.h>
313#include <linux/major.h>
314#include <linux/string.h>
315#include <linux/fcntl.h>
316#include <linux/slab.h>
317#include <linux/random.h>
318#include <linux/poll.h>
319#include <linux/init.h>
320#include <linux/fs.h>
321#include <linux/genhd.h>
322#include <linux/interrupt.h>
323#include <linux/mm.h>
324#include <linux/nodemask.h>
325#include <linux/spinlock.h>
326#include <linux/kthread.h>
327#include <linux/percpu.h>
328#include <linux/cryptohash.h>
329#include <linux/fips.h>
330#include <linux/ptrace.h>
331#include <linux/workqueue.h>
332#include <linux/irq.h>
333#include <linux/ratelimit.h>
334#include <linux/syscalls.h>
335#include <linux/completion.h>
336#include <linux/uuid.h>
337#include <crypto/chacha.h>
338
339#include <asm/processor.h>
340#include <linux/uaccess.h>
341#include <asm/irq.h>
342#include <asm/irq_regs.h>
343#include <asm/io.h>
344
345#define CREATE_TRACE_POINTS
346#include <trace/events/random.h>
347
348/* #define ADD_INTERRUPT_BENCH */
349
350/*
351 * Configuration information
352 */
353#define INPUT_POOL_SHIFT 12
354#define INPUT_POOL_WORDS (1 << (INPUT_POOL_SHIFT-5))
355#define OUTPUT_POOL_SHIFT 10
356#define OUTPUT_POOL_WORDS (1 << (OUTPUT_POOL_SHIFT-5))
357#define SEC_XFER_SIZE 512
358#define EXTRACT_SIZE 10
359
360
361#define LONGS(x) (((x) + sizeof(unsigned long) - 1)/sizeof(unsigned long))
362
363/*
364 * To allow fractional bits to be tracked, the entropy_count field is
365 * denominated in units of 1/8th bits.
366 *
367 * 2*(ENTROPY_SHIFT + poolbitshift) must <= 31, or the multiply in
368 * credit_entropy_bits() needs to be 64 bits wide.
369 */
370#define ENTROPY_SHIFT 3
371#define ENTROPY_BITS(r) ((r)->entropy_count >> ENTROPY_SHIFT)
372
373/*
374 * The minimum number of bits of entropy before we wake up a read on
375 * /dev/random. Should be enough to do a significant reseed.
376 */
377static int random_read_wakeup_bits = 64;
378
379/*
380 * If the entropy count falls under this number of bits, then we
381 * should wake up processes which are selecting or polling on write
382 * access to /dev/random.
383 */
384static int random_write_wakeup_bits = 28 * OUTPUT_POOL_WORDS;
385
386/*
387 * Originally, we used a primitive polynomial of degree .poolwords
388 * over GF(2). The taps for various sizes are defined below. They
389 * were chosen to be evenly spaced except for the last tap, which is 1
390 * to get the twisting happening as fast as possible.
391 *
392 * For the purposes of better mixing, we use the CRC-32 polynomial as
393 * well to make a (modified) twisted Generalized Feedback Shift
394 * Register. (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR
395 * generators. ACM Transactions on Modeling and Computer Simulation
396 * 2(3):179-194. Also see M. Matsumoto & Y. Kurita, 1994. Twisted
397 * GFSR generators II. ACM Transactions on Modeling and Computer
398 * Simulation 4:254-266)
399 *
400 * Thanks to Colin Plumb for suggesting this.
401 *
402 * The mixing operation is much less sensitive than the output hash,
403 * where we use SHA-1. All that we want of mixing operation is that
404 * it be a good non-cryptographic hash; i.e. it not produce collisions
405 * when fed "random" data of the sort we expect to see. As long as
406 * the pool state differs for different inputs, we have preserved the
407 * input entropy and done a good job. The fact that an intelligent
408 * attacker can construct inputs that will produce controlled
409 * alterations to the pool's state is not important because we don't
410 * consider such inputs to contribute any randomness. The only
411 * property we need with respect to them is that the attacker can't
412 * increase his/her knowledge of the pool's state. Since all
413 * additions are reversible (knowing the final state and the input,
414 * you can reconstruct the initial state), if an attacker has any
415 * uncertainty about the initial state, he/she can only shuffle that
416 * uncertainty about, but never cause any collisions (which would
417 * decrease the uncertainty).
418 *
419 * Our mixing functions were analyzed by Lacharme, Roeck, Strubel, and
420 * Videau in their paper, "The Linux Pseudorandom Number Generator
421 * Revisited" (see: http://eprint.iacr.org/2012/251.pdf). In their
422 * paper, they point out that we are not using a true Twisted GFSR,
423 * since Matsumoto & Kurita used a trinomial feedback polynomial (that
424 * is, with only three taps, instead of the six that we are using).
425 * As a result, the resulting polynomial is neither primitive nor
426 * irreducible, and hence does not have a maximal period over
427 * GF(2**32). They suggest a slight change to the generator
428 * polynomial which improves the resulting TGFSR polynomial to be
429 * irreducible, which we have made here.
430 */
431static const struct poolinfo {
432 int poolbitshift, poolwords, poolbytes, poolfracbits;
433#define S(x) ilog2(x)+5, (x), (x)*4, (x) << (ENTROPY_SHIFT+5)
434 int tap1, tap2, tap3, tap4, tap5;
435} poolinfo_table[] = {
436 /* was: x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 */
437 /* x^128 + x^104 + x^76 + x^51 +x^25 + x + 1 */
438 { S(128), 104, 76, 51, 25, 1 },
439 /* was: x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 */
440 /* x^32 + x^26 + x^19 + x^14 + x^7 + x + 1 */
441 { S(32), 26, 19, 14, 7, 1 },
442#if 0
443 /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1 -- 115 */
444 { S(2048), 1638, 1231, 819, 411, 1 },
445
446 /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
447 { S(1024), 817, 615, 412, 204, 1 },
448
449 /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
450 { S(1024), 819, 616, 410, 207, 2 },
451
452 /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
453 { S(512), 411, 308, 208, 104, 1 },
454
455 /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
456 { S(512), 409, 307, 206, 102, 2 },
457 /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
458 { S(512), 409, 309, 205, 103, 2 },
459
460 /* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
461 { S(256), 205, 155, 101, 52, 1 },
462
463 /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
464 { S(128), 103, 78, 51, 27, 2 },
465
466 /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
467 { S(64), 52, 39, 26, 14, 1 },
468#endif
469};
470
471/*
472 * Static global variables
473 */
474static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
475static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
476static struct fasync_struct *fasync;
477
478static DEFINE_SPINLOCK(random_ready_list_lock);
479static LIST_HEAD(random_ready_list);
480
481struct crng_state {
482 __u32 state[16];
483 unsigned long init_time;
484 spinlock_t lock;
485};
486
487static struct crng_state primary_crng = {
488 .lock = __SPIN_LOCK_UNLOCKED(primary_crng.lock),
489};
490
491/*
492 * crng_init = 0 --> Uninitialized
493 * 1 --> Initialized
494 * 2 --> Initialized from input_pool
495 *
496 * crng_init is protected by primary_crng->lock, and only increases
497 * its value (from 0->1->2).
498 */
499static int crng_init = 0;
500#define crng_ready() (likely(crng_init > 1))
501static int crng_init_cnt = 0;
502static unsigned long crng_global_init_time = 0;
503#define CRNG_INIT_CNT_THRESH (2*CHACHA_KEY_SIZE)
504static void _extract_crng(struct crng_state *crng, __u8 out[CHACHA_BLOCK_SIZE]);
505static void _crng_backtrack_protect(struct crng_state *crng,
506 __u8 tmp[CHACHA_BLOCK_SIZE], int used);
507static void process_random_ready_list(void);
508static void _get_random_bytes(void *buf, int nbytes);
509
510static struct ratelimit_state unseeded_warning =
511 RATELIMIT_STATE_INIT("warn_unseeded_randomness", HZ, 3);
512static struct ratelimit_state urandom_warning =
513 RATELIMIT_STATE_INIT("warn_urandom_randomness", HZ, 3);
514
515static int ratelimit_disable __read_mostly;
516
517module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
518MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
519
520/**********************************************************************
521 *
522 * OS independent entropy store. Here are the functions which handle
523 * storing entropy in an entropy pool.
524 *
525 **********************************************************************/
526
527struct entropy_store;
528struct entropy_store {
529 /* read-only data: */
530 const struct poolinfo *poolinfo;
531 __u32 *pool;
532 const char *name;
533 struct entropy_store *pull;
534 struct work_struct push_work;
535
536 /* read-write data: */
537 unsigned long last_pulled;
538 spinlock_t lock;
539 unsigned short add_ptr;
540 unsigned short input_rotate;
541 int entropy_count;
542 unsigned int initialized:1;
543 unsigned int last_data_init:1;
544 __u8 last_data[EXTRACT_SIZE];
545};
546
547static ssize_t extract_entropy(struct entropy_store *r, void *buf,
548 size_t nbytes, int min, int rsvd);
549static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
550 size_t nbytes, int fips);
551
552static void crng_reseed(struct crng_state *crng, struct entropy_store *r);
553static void push_to_pool(struct work_struct *work);
554static __u32 input_pool_data[INPUT_POOL_WORDS] __latent_entropy;
555static __u32 blocking_pool_data[OUTPUT_POOL_WORDS] __latent_entropy;
556
557static struct entropy_store input_pool = {
558 .poolinfo = &poolinfo_table[0],
559 .name = "input",
560 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
561 .pool = input_pool_data
562};
563
564static struct entropy_store blocking_pool = {
565 .poolinfo = &poolinfo_table[1],
566 .name = "blocking",
567 .pull = &input_pool,
568 .lock = __SPIN_LOCK_UNLOCKED(blocking_pool.lock),
569 .pool = blocking_pool_data,
570 .push_work = __WORK_INITIALIZER(blocking_pool.push_work,
571 push_to_pool),
572};
573
574static __u32 const twist_table[8] = {
575 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
576 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
577
578/*
579 * This function adds bytes into the entropy "pool". It does not
580 * update the entropy estimate. The caller should call
581 * credit_entropy_bits if this is appropriate.
582 *
583 * The pool is stirred with a primitive polynomial of the appropriate
584 * degree, and then twisted. We twist by three bits at a time because
585 * it's cheap to do so and helps slightly in the expected case where
586 * the entropy is concentrated in the low-order bits.
587 */
588static void _mix_pool_bytes(struct entropy_store *r, const void *in,
589 int nbytes)
590{
591 unsigned long i, tap1, tap2, tap3, tap4, tap5;
592 int input_rotate;
593 int wordmask = r->poolinfo->poolwords - 1;
594 const char *bytes = in;
595 __u32 w;
596
597 tap1 = r->poolinfo->tap1;
598 tap2 = r->poolinfo->tap2;
599 tap3 = r->poolinfo->tap3;
600 tap4 = r->poolinfo->tap4;
601 tap5 = r->poolinfo->tap5;
602
603 input_rotate = r->input_rotate;
604 i = r->add_ptr;
605
606 /* mix one byte at a time to simplify size handling and churn faster */
607 while (nbytes--) {
608 w = rol32(*bytes++, input_rotate);
609 i = (i - 1) & wordmask;
610
611 /* XOR in the various taps */
612 w ^= r->pool[i];
613 w ^= r->pool[(i + tap1) & wordmask];
614 w ^= r->pool[(i + tap2) & wordmask];
615 w ^= r->pool[(i + tap3) & wordmask];
616 w ^= r->pool[(i + tap4) & wordmask];
617 w ^= r->pool[(i + tap5) & wordmask];
618
619 /* Mix the result back in with a twist */
620 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
621
622 /*
623 * Normally, we add 7 bits of rotation to the pool.
624 * At the beginning of the pool, add an extra 7 bits
625 * rotation, so that successive passes spread the
626 * input bits across the pool evenly.
627 */
628 input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
629 }
630
631 r->input_rotate = input_rotate;
632 r->add_ptr = i;
633}
634
635static void __mix_pool_bytes(struct entropy_store *r, const void *in,
636 int nbytes)
637{
638 trace_mix_pool_bytes_nolock(r->name, nbytes, _RET_IP_);
639 _mix_pool_bytes(r, in, nbytes);
640}
641
642static void mix_pool_bytes(struct entropy_store *r, const void *in,
643 int nbytes)
644{
645 unsigned long flags;
646
647 trace_mix_pool_bytes(r->name, nbytes, _RET_IP_);
648 spin_lock_irqsave(&r->lock, flags);
649 _mix_pool_bytes(r, in, nbytes);
650 spin_unlock_irqrestore(&r->lock, flags);
651}
652
653struct fast_pool {
654 __u32 pool[4];
655 unsigned long last;
656 unsigned short reg_idx;
657 unsigned char count;
658};
659
660/*
661 * This is a fast mixing routine used by the interrupt randomness
662 * collector. It's hardcoded for an 128 bit pool and assumes that any
663 * locks that might be needed are taken by the caller.
664 */
665static void fast_mix(struct fast_pool *f)
666{
667 __u32 a = f->pool[0], b = f->pool[1];
668 __u32 c = f->pool[2], d = f->pool[3];
669
670 a += b; c += d;
671 b = rol32(b, 6); d = rol32(d, 27);
672 d ^= a; b ^= c;
673
674 a += b; c += d;
675 b = rol32(b, 16); d = rol32(d, 14);
676 d ^= a; b ^= c;
677
678 a += b; c += d;
679 b = rol32(b, 6); d = rol32(d, 27);
680 d ^= a; b ^= c;
681
682 a += b; c += d;
683 b = rol32(b, 16); d = rol32(d, 14);
684 d ^= a; b ^= c;
685
686 f->pool[0] = a; f->pool[1] = b;
687 f->pool[2] = c; f->pool[3] = d;
688 f->count++;
689}
690
691static void process_random_ready_list(void)
692{
693 unsigned long flags;
694 struct random_ready_callback *rdy, *tmp;
695
696 spin_lock_irqsave(&random_ready_list_lock, flags);
697 list_for_each_entry_safe(rdy, tmp, &random_ready_list, list) {
698 struct module *owner = rdy->owner;
699
700 list_del_init(&rdy->list);
701 rdy->func(rdy);
702 module_put(owner);
703 }
704 spin_unlock_irqrestore(&random_ready_list_lock, flags);
705}
706
707/*
708 * Credit (or debit) the entropy store with n bits of entropy.
709 * Use credit_entropy_bits_safe() if the value comes from userspace
710 * or otherwise should be checked for extreme values.
711 */
712static void credit_entropy_bits(struct entropy_store *r, int nbits)
713{
714 int entropy_count, orig, has_initialized = 0;
715 const int pool_size = r->poolinfo->poolfracbits;
716 int nfrac = nbits << ENTROPY_SHIFT;
717
718 if (!nbits)
719 return;
720
721retry:
722 entropy_count = orig = READ_ONCE(r->entropy_count);
723 if (nfrac < 0) {
724 /* Debit */
725 entropy_count += nfrac;
726 } else {
727 /*
728 * Credit: we have to account for the possibility of
729 * overwriting already present entropy. Even in the
730 * ideal case of pure Shannon entropy, new contributions
731 * approach the full value asymptotically:
732 *
733 * entropy <- entropy + (pool_size - entropy) *
734 * (1 - exp(-add_entropy/pool_size))
735 *
736 * For add_entropy <= pool_size/2 then
737 * (1 - exp(-add_entropy/pool_size)) >=
738 * (add_entropy/pool_size)*0.7869...
739 * so we can approximate the exponential with
740 * 3/4*add_entropy/pool_size and still be on the
741 * safe side by adding at most pool_size/2 at a time.
742 *
743 * The use of pool_size-2 in the while statement is to
744 * prevent rounding artifacts from making the loop
745 * arbitrarily long; this limits the loop to log2(pool_size)*2
746 * turns no matter how large nbits is.
747 */
748 int pnfrac = nfrac;
749 const int s = r->poolinfo->poolbitshift + ENTROPY_SHIFT + 2;
750 /* The +2 corresponds to the /4 in the denominator */
751
752 do {
753 unsigned int anfrac = min(pnfrac, pool_size/2);
754 unsigned int add =
755 ((pool_size - entropy_count)*anfrac*3) >> s;
756
757 entropy_count += add;
758 pnfrac -= anfrac;
759 } while (unlikely(entropy_count < pool_size-2 && pnfrac));
760 }
761
762 if (unlikely(entropy_count < 0)) {
763 pr_warn("random: negative entropy/overflow: pool %s count %d\n",
764 r->name, entropy_count);
765 WARN_ON(1);
766 entropy_count = 0;
767 } else if (entropy_count > pool_size)
768 entropy_count = pool_size;
769 if ((r == &blocking_pool) && !r->initialized &&
770 (entropy_count >> ENTROPY_SHIFT) > 128)
771 has_initialized = 1;
772 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
773 goto retry;
774
775 if (has_initialized) {
776 r->initialized = 1;
777 wake_up_interruptible(&random_read_wait);
778 kill_fasync(&fasync, SIGIO, POLL_IN);
779 }
780
781 trace_credit_entropy_bits(r->name, nbits,
782 entropy_count >> ENTROPY_SHIFT, _RET_IP_);
783
784 if (r == &input_pool) {
785 int entropy_bits = entropy_count >> ENTROPY_SHIFT;
786 struct entropy_store *other = &blocking_pool;
787
788 if (crng_init < 2) {
789 if (entropy_bits < 128)
790 return;
791 crng_reseed(&primary_crng, r);
792 entropy_bits = r->entropy_count >> ENTROPY_SHIFT;
793 }
794
795 /* initialize the blocking pool if necessary */
796 if (entropy_bits >= random_read_wakeup_bits &&
797 !other->initialized) {
798 schedule_work(&other->push_work);
799 return;
800 }
801
802 /* should we wake readers? */
803 if (entropy_bits >= random_read_wakeup_bits &&
804 wq_has_sleeper(&random_read_wait)) {
805 wake_up_interruptible(&random_read_wait);
806 kill_fasync(&fasync, SIGIO, POLL_IN);
807 }
808 /* If the input pool is getting full, and the blocking
809 * pool has room, send some entropy to the blocking
810 * pool.
811 */
812 if (!work_pending(&other->push_work) &&
813 (ENTROPY_BITS(r) > 6 * r->poolinfo->poolbytes) &&
814 (ENTROPY_BITS(other) <= 6 * other->poolinfo->poolbytes))
815 schedule_work(&other->push_work);
816 }
817}
818
819static int credit_entropy_bits_safe(struct entropy_store *r, int nbits)
820{
821 const int nbits_max = r->poolinfo->poolwords * 32;
822
823 if (nbits < 0)
824 return -EINVAL;
825
826 /* Cap the value to avoid overflows */
827 nbits = min(nbits, nbits_max);
828
829 credit_entropy_bits(r, nbits);
830 return 0;
831}
832
833/*********************************************************************
834 *
835 * CRNG using CHACHA20
836 *
837 *********************************************************************/
838
839#define CRNG_RESEED_INTERVAL (300*HZ)
840
841static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
842
843#ifdef CONFIG_NUMA
844/*
845 * Hack to deal with crazy userspace progams when they are all trying
846 * to access /dev/urandom in parallel. The programs are almost
847 * certainly doing something terribly wrong, but we'll work around
848 * their brain damage.
849 */
850static struct crng_state **crng_node_pool __read_mostly;
851#endif
852
853static void invalidate_batched_entropy(void);
854static void numa_crng_init(void);
855
856static bool trust_cpu __ro_after_init = IS_ENABLED(CONFIG_RANDOM_TRUST_CPU);
857static int __init parse_trust_cpu(char *arg)
858{
859 return kstrtobool(arg, &trust_cpu);
860}
861early_param("random.trust_cpu", parse_trust_cpu);
862
863static void crng_initialize(struct crng_state *crng)
864{
865 int i;
866 int arch_init = 1;
867 unsigned long rv;
868
869 memcpy(&crng->state[0], "expand 32-byte k", 16);
870 if (crng == &primary_crng)
871 _extract_entropy(&input_pool, &crng->state[4],
872 sizeof(__u32) * 12, 0);
873 else
874 _get_random_bytes(&crng->state[4], sizeof(__u32) * 12);
875 for (i = 4; i < 16; i++) {
876 if (!arch_get_random_seed_long(&rv) &&
877 !arch_get_random_long(&rv)) {
878 rv = random_get_entropy();
879 arch_init = 0;
880 }
881 crng->state[i] ^= rv;
882 }
883 if (trust_cpu && arch_init && crng == &primary_crng) {
884 invalidate_batched_entropy();
885 numa_crng_init();
886 crng_init = 2;
887 pr_notice("random: crng done (trusting CPU's manufacturer)\n");
888 }
889 crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
890}
891
892#ifdef CONFIG_NUMA
893static void do_numa_crng_init(struct work_struct *work)
894{
895 int i;
896 struct crng_state *crng;
897 struct crng_state **pool;
898
899 pool = kcalloc(nr_node_ids, sizeof(*pool), GFP_KERNEL|__GFP_NOFAIL);
900 for_each_online_node(i) {
901 crng = kmalloc_node(sizeof(struct crng_state),
902 GFP_KERNEL | __GFP_NOFAIL, i);
903 spin_lock_init(&crng->lock);
904 crng_initialize(crng);
905 pool[i] = crng;
906 }
907 mb();
908 if (cmpxchg(&crng_node_pool, NULL, pool)) {
909 for_each_node(i)
910 kfree(pool[i]);
911 kfree(pool);
912 }
913}
914
915static DECLARE_WORK(numa_crng_init_work, do_numa_crng_init);
916
917static void numa_crng_init(void)
918{
919 schedule_work(&numa_crng_init_work);
920}
921#else
922static void numa_crng_init(void) {}
923#endif
924
925/*
926 * crng_fast_load() can be called by code in the interrupt service
927 * path. So we can't afford to dilly-dally.
928 */
929static int crng_fast_load(const char *cp, size_t len)
930{
931 unsigned long flags;
932 char *p;
933
934 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
935 return 0;
936 if (crng_init != 0) {
937 spin_unlock_irqrestore(&primary_crng.lock, flags);
938 return 0;
939 }
940 p = (unsigned char *) &primary_crng.state[4];
941 while (len > 0 && crng_init_cnt < CRNG_INIT_CNT_THRESH) {
942 p[crng_init_cnt % CHACHA_KEY_SIZE] ^= *cp;
943 cp++; crng_init_cnt++; len--;
944 }
945 spin_unlock_irqrestore(&primary_crng.lock, flags);
946 if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
947 invalidate_batched_entropy();
948 crng_init = 1;
949 wake_up_interruptible(&crng_init_wait);
950 pr_notice("random: fast init done\n");
951 }
952 return 1;
953}
954
955/*
956 * crng_slow_load() is called by add_device_randomness, which has two
957 * attributes. (1) We can't trust the buffer passed to it is
958 * guaranteed to be unpredictable (so it might not have any entropy at
959 * all), and (2) it doesn't have the performance constraints of
960 * crng_fast_load().
961 *
962 * So we do something more comprehensive which is guaranteed to touch
963 * all of the primary_crng's state, and which uses a LFSR with a
964 * period of 255 as part of the mixing algorithm. Finally, we do
965 * *not* advance crng_init_cnt since buffer we may get may be something
966 * like a fixed DMI table (for example), which might very well be
967 * unique to the machine, but is otherwise unvarying.
968 */
969static int crng_slow_load(const char *cp, size_t len)
970{
971 unsigned long flags;
972 static unsigned char lfsr = 1;
973 unsigned char tmp;
974 unsigned i, max = CHACHA_KEY_SIZE;
975 const char * src_buf = cp;
976 char * dest_buf = (char *) &primary_crng.state[4];
977
978 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
979 return 0;
980 if (crng_init != 0) {
981 spin_unlock_irqrestore(&primary_crng.lock, flags);
982 return 0;
983 }
984 if (len > max)
985 max = len;
986
987 for (i = 0; i < max ; i++) {
988 tmp = lfsr;
989 lfsr >>= 1;
990 if (tmp & 1)
991 lfsr ^= 0xE1;
992 tmp = dest_buf[i % CHACHA_KEY_SIZE];
993 dest_buf[i % CHACHA_KEY_SIZE] ^= src_buf[i % len] ^ lfsr;
994 lfsr += (tmp << 3) | (tmp >> 5);
995 }
996 spin_unlock_irqrestore(&primary_crng.lock, flags);
997 return 1;
998}
999
1000static void crng_reseed(struct crng_state *crng, struct entropy_store *r)
1001{
1002 unsigned long flags;
1003 int i, num;
1004 union {
1005 __u8 block[CHACHA_BLOCK_SIZE];
1006 __u32 key[8];
1007 } buf;
1008
1009 if (r) {
1010 num = extract_entropy(r, &buf, 32, 16, 0);
1011 if (num == 0)
1012 return;
1013 } else {
1014 _extract_crng(&primary_crng, buf.block);
1015 _crng_backtrack_protect(&primary_crng, buf.block,
1016 CHACHA_KEY_SIZE);
1017 }
1018 spin_lock_irqsave(&crng->lock, flags);
1019 for (i = 0; i < 8; i++) {
1020 unsigned long rv;
1021 if (!arch_get_random_seed_long(&rv) &&
1022 !arch_get_random_long(&rv))
1023 rv = random_get_entropy();
1024 crng->state[i+4] ^= buf.key[i] ^ rv;
1025 }
1026 memzero_explicit(&buf, sizeof(buf));
1027 crng->init_time = jiffies;
1028 spin_unlock_irqrestore(&crng->lock, flags);
1029 if (crng == &primary_crng && crng_init < 2) {
1030 invalidate_batched_entropy();
1031 numa_crng_init();
1032 crng_init = 2;
1033 process_random_ready_list();
1034 wake_up_interruptible(&crng_init_wait);
1035 pr_notice("random: crng init done\n");
1036 if (unseeded_warning.missed) {
1037 pr_notice("random: %d get_random_xx warning(s) missed "
1038 "due to ratelimiting\n",
1039 unseeded_warning.missed);
1040 unseeded_warning.missed = 0;
1041 }
1042 if (urandom_warning.missed) {
1043 pr_notice("random: %d urandom warning(s) missed "
1044 "due to ratelimiting\n",
1045 urandom_warning.missed);
1046 urandom_warning.missed = 0;
1047 }
1048 }
1049}
1050
1051static void _extract_crng(struct crng_state *crng,
1052 __u8 out[CHACHA_BLOCK_SIZE])
1053{
1054 unsigned long v, flags;
1055
1056 if (crng_ready() &&
1057 (time_after(crng_global_init_time, crng->init_time) ||
1058 time_after(jiffies, crng->init_time + CRNG_RESEED_INTERVAL)))
1059 crng_reseed(crng, crng == &primary_crng ? &input_pool : NULL);
1060 spin_lock_irqsave(&crng->lock, flags);
1061 if (arch_get_random_long(&v))
1062 crng->state[14] ^= v;
1063 chacha20_block(&crng->state[0], out);
1064 if (crng->state[12] == 0)
1065 crng->state[13]++;
1066 spin_unlock_irqrestore(&crng->lock, flags);
1067}
1068
1069static void extract_crng(__u8 out[CHACHA_BLOCK_SIZE])
1070{
1071 struct crng_state *crng = NULL;
1072
1073#ifdef CONFIG_NUMA
1074 if (crng_node_pool)
1075 crng = crng_node_pool[numa_node_id()];
1076 if (crng == NULL)
1077#endif
1078 crng = &primary_crng;
1079 _extract_crng(crng, out);
1080}
1081
1082/*
1083 * Use the leftover bytes from the CRNG block output (if there is
1084 * enough) to mutate the CRNG key to provide backtracking protection.
1085 */
1086static void _crng_backtrack_protect(struct crng_state *crng,
1087 __u8 tmp[CHACHA_BLOCK_SIZE], int used)
1088{
1089 unsigned long flags;
1090 __u32 *s, *d;
1091 int i;
1092
1093 used = round_up(used, sizeof(__u32));
1094 if (used + CHACHA_KEY_SIZE > CHACHA_BLOCK_SIZE) {
1095 extract_crng(tmp);
1096 used = 0;
1097 }
1098 spin_lock_irqsave(&crng->lock, flags);
1099 s = (__u32 *) &tmp[used];
1100 d = &crng->state[4];
1101 for (i=0; i < 8; i++)
1102 *d++ ^= *s++;
1103 spin_unlock_irqrestore(&crng->lock, flags);
1104}
1105
1106static void crng_backtrack_protect(__u8 tmp[CHACHA_BLOCK_SIZE], int used)
1107{
1108 struct crng_state *crng = NULL;
1109
1110#ifdef CONFIG_NUMA
1111 if (crng_node_pool)
1112 crng = crng_node_pool[numa_node_id()];
1113 if (crng == NULL)
1114#endif
1115 crng = &primary_crng;
1116 _crng_backtrack_protect(crng, tmp, used);
1117}
1118
1119static ssize_t extract_crng_user(void __user *buf, size_t nbytes)
1120{
1121 ssize_t ret = 0, i = CHACHA_BLOCK_SIZE;
1122 __u8 tmp[CHACHA_BLOCK_SIZE] __aligned(4);
1123 int large_request = (nbytes > 256);
1124
1125 while (nbytes) {
1126 if (large_request && need_resched()) {
1127 if (signal_pending(current)) {
1128 if (ret == 0)
1129 ret = -ERESTARTSYS;
1130 break;
1131 }
1132 schedule();
1133 }
1134
1135 extract_crng(tmp);
1136 i = min_t(int, nbytes, CHACHA_BLOCK_SIZE);
1137 if (copy_to_user(buf, tmp, i)) {
1138 ret = -EFAULT;
1139 break;
1140 }
1141
1142 nbytes -= i;
1143 buf += i;
1144 ret += i;
1145 }
1146 crng_backtrack_protect(tmp, i);
1147
1148 /* Wipe data just written to memory */
1149 memzero_explicit(tmp, sizeof(tmp));
1150
1151 return ret;
1152}
1153
1154
1155/*********************************************************************
1156 *
1157 * Entropy input management
1158 *
1159 *********************************************************************/
1160
1161/* There is one of these per entropy source */
1162struct timer_rand_state {
1163 cycles_t last_time;
1164 long last_delta, last_delta2;
1165};
1166
1167#define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
1168
1169/*
1170 * Add device- or boot-specific data to the input pool to help
1171 * initialize it.
1172 *
1173 * None of this adds any entropy; it is meant to avoid the problem of
1174 * the entropy pool having similar initial state across largely
1175 * identical devices.
1176 */
1177void add_device_randomness(const void *buf, unsigned int size)
1178{
1179 unsigned long time = random_get_entropy() ^ jiffies;
1180 unsigned long flags;
1181
1182 if (!crng_ready() && size)
1183 crng_slow_load(buf, size);
1184
1185 trace_add_device_randomness(size, _RET_IP_);
1186 spin_lock_irqsave(&input_pool.lock, flags);
1187 _mix_pool_bytes(&input_pool, buf, size);
1188 _mix_pool_bytes(&input_pool, &time, sizeof(time));
1189 spin_unlock_irqrestore(&input_pool.lock, flags);
1190}
1191EXPORT_SYMBOL(add_device_randomness);
1192
1193static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
1194
1195/*
1196 * This function adds entropy to the entropy "pool" by using timing
1197 * delays. It uses the timer_rand_state structure to make an estimate
1198 * of how many bits of entropy this call has added to the pool.
1199 *
1200 * The number "num" is also added to the pool - it should somehow describe
1201 * the type of event which just happened. This is currently 0-255 for
1202 * keyboard scan codes, and 256 upwards for interrupts.
1203 *
1204 */
1205static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
1206{
1207 struct entropy_store *r;
1208 struct {
1209 long jiffies;
1210 unsigned cycles;
1211 unsigned num;
1212 } sample;
1213 long delta, delta2, delta3;
1214
1215 sample.jiffies = jiffies;
1216 sample.cycles = random_get_entropy();
1217 sample.num = num;
1218 r = &input_pool;
1219 mix_pool_bytes(r, &sample, sizeof(sample));
1220
1221 /*
1222 * Calculate number of bits of randomness we probably added.
1223 * We take into account the first, second and third-order deltas
1224 * in order to make our estimate.
1225 */
1226 delta = sample.jiffies - state->last_time;
1227 state->last_time = sample.jiffies;
1228
1229 delta2 = delta - state->last_delta;
1230 state->last_delta = delta;
1231
1232 delta3 = delta2 - state->last_delta2;
1233 state->last_delta2 = delta2;
1234
1235 if (delta < 0)
1236 delta = -delta;
1237 if (delta2 < 0)
1238 delta2 = -delta2;
1239 if (delta3 < 0)
1240 delta3 = -delta3;
1241 if (delta > delta2)
1242 delta = delta2;
1243 if (delta > delta3)
1244 delta = delta3;
1245
1246 /*
1247 * delta is now minimum absolute delta.
1248 * Round down by 1 bit on general principles,
1249 * and limit entropy entimate to 12 bits.
1250 */
1251 credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
1252}
1253
1254void add_input_randomness(unsigned int type, unsigned int code,
1255 unsigned int value)
1256{
1257 static unsigned char last_value;
1258
1259 /* ignore autorepeat and the like */
1260 if (value == last_value)
1261 return;
1262
1263 last_value = value;
1264 add_timer_randomness(&input_timer_state,
1265 (type << 4) ^ code ^ (code >> 4) ^ value);
1266 trace_add_input_randomness(ENTROPY_BITS(&input_pool));
1267}
1268EXPORT_SYMBOL_GPL(add_input_randomness);
1269
1270static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
1271
1272#ifdef ADD_INTERRUPT_BENCH
1273static unsigned long avg_cycles, avg_deviation;
1274
1275#define AVG_SHIFT 8 /* Exponential average factor k=1/256 */
1276#define FIXED_1_2 (1 << (AVG_SHIFT-1))
1277
1278static void add_interrupt_bench(cycles_t start)
1279{
1280 long delta = random_get_entropy() - start;
1281
1282 /* Use a weighted moving average */
1283 delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
1284 avg_cycles += delta;
1285 /* And average deviation */
1286 delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
1287 avg_deviation += delta;
1288}
1289#else
1290#define add_interrupt_bench(x)
1291#endif
1292
1293static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
1294{
1295 __u32 *ptr = (__u32 *) regs;
1296 unsigned int idx;
1297
1298 if (regs == NULL)
1299 return 0;
1300 idx = READ_ONCE(f->reg_idx);
1301 if (idx >= sizeof(struct pt_regs) / sizeof(__u32))
1302 idx = 0;
1303 ptr += idx++;
1304 WRITE_ONCE(f->reg_idx, idx);
1305 return *ptr;
1306}
1307
1308void add_interrupt_randomness(int irq, int irq_flags)
1309{
1310 struct entropy_store *r;
1311 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1312 struct pt_regs *regs = get_irq_regs();
1313 unsigned long now = jiffies;
1314 cycles_t cycles = random_get_entropy();
1315 __u32 c_high, j_high;
1316 __u64 ip;
1317 unsigned long seed;
1318 int credit = 0;
1319
1320 if (cycles == 0)
1321 cycles = get_reg(fast_pool, regs);
1322 c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
1323 j_high = (sizeof(now) > 4) ? now >> 32 : 0;
1324 fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
1325 fast_pool->pool[1] ^= now ^ c_high;
1326 ip = regs ? instruction_pointer(regs) : _RET_IP_;
1327 fast_pool->pool[2] ^= ip;
1328 fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
1329 get_reg(fast_pool, regs);
1330
1331 fast_mix(fast_pool);
1332 add_interrupt_bench(cycles);
1333
1334 if (unlikely(crng_init == 0)) {
1335 if ((fast_pool->count >= 64) &&
1336 crng_fast_load((char *) fast_pool->pool,
1337 sizeof(fast_pool->pool))) {
1338 fast_pool->count = 0;
1339 fast_pool->last = now;
1340 }
1341 return;
1342 }
1343
1344 if ((fast_pool->count < 64) &&
1345 !time_after(now, fast_pool->last + HZ))
1346 return;
1347
1348 r = &input_pool;
1349 if (!spin_trylock(&r->lock))
1350 return;
1351
1352 fast_pool->last = now;
1353 __mix_pool_bytes(r, &fast_pool->pool, sizeof(fast_pool->pool));
1354
1355 /*
1356 * If we have architectural seed generator, produce a seed and
1357 * add it to the pool. For the sake of paranoia don't let the
1358 * architectural seed generator dominate the input from the
1359 * interrupt noise.
1360 */
1361 if (arch_get_random_seed_long(&seed)) {
1362 __mix_pool_bytes(r, &seed, sizeof(seed));
1363 credit = 1;
1364 }
1365 spin_unlock(&r->lock);
1366
1367 fast_pool->count = 0;
1368
1369 /* award one bit for the contents of the fast pool */
1370 credit_entropy_bits(r, credit + 1);
1371}
1372EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1373
1374#ifdef CONFIG_BLOCK
1375void add_disk_randomness(struct gendisk *disk)
1376{
1377 if (!disk || !disk->random)
1378 return;
1379 /* first major is 1, so we get >= 0x200 here */
1380 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1381 trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
1382}
1383EXPORT_SYMBOL_GPL(add_disk_randomness);
1384#endif
1385
1386/*********************************************************************
1387 *
1388 * Entropy extraction routines
1389 *
1390 *********************************************************************/
1391
1392/*
1393 * This utility inline function is responsible for transferring entropy
1394 * from the primary pool to the secondary extraction pool. We make
1395 * sure we pull enough for a 'catastrophic reseed'.
1396 */
1397static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes);
1398static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1399{
1400 if (!r->pull ||
1401 r->entropy_count >= (nbytes << (ENTROPY_SHIFT + 3)) ||
1402 r->entropy_count > r->poolinfo->poolfracbits)
1403 return;
1404
1405 _xfer_secondary_pool(r, nbytes);
1406}
1407
1408static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1409{
1410 __u32 tmp[OUTPUT_POOL_WORDS];
1411
1412 int bytes = nbytes;
1413
1414 /* pull at least as much as a wakeup */
1415 bytes = max_t(int, bytes, random_read_wakeup_bits / 8);
1416 /* but never more than the buffer size */
1417 bytes = min_t(int, bytes, sizeof(tmp));
1418
1419 trace_xfer_secondary_pool(r->name, bytes * 8, nbytes * 8,
1420 ENTROPY_BITS(r), ENTROPY_BITS(r->pull));
1421 bytes = extract_entropy(r->pull, tmp, bytes,
1422 random_read_wakeup_bits / 8, 0);
1423 mix_pool_bytes(r, tmp, bytes);
1424 credit_entropy_bits(r, bytes*8);
1425}
1426
1427/*
1428 * Used as a workqueue function so that when the input pool is getting
1429 * full, we can "spill over" some entropy to the output pools. That
1430 * way the output pools can store some of the excess entropy instead
1431 * of letting it go to waste.
1432 */
1433static void push_to_pool(struct work_struct *work)
1434{
1435 struct entropy_store *r = container_of(work, struct entropy_store,
1436 push_work);
1437 BUG_ON(!r);
1438 _xfer_secondary_pool(r, random_read_wakeup_bits/8);
1439 trace_push_to_pool(r->name, r->entropy_count >> ENTROPY_SHIFT,
1440 r->pull->entropy_count >> ENTROPY_SHIFT);
1441}
1442
1443/*
1444 * This function decides how many bytes to actually take from the
1445 * given pool, and also debits the entropy count accordingly.
1446 */
1447static size_t account(struct entropy_store *r, size_t nbytes, int min,
1448 int reserved)
1449{
1450 int entropy_count, orig, have_bytes;
1451 size_t ibytes, nfrac;
1452
1453 BUG_ON(r->entropy_count > r->poolinfo->poolfracbits);
1454
1455 /* Can we pull enough? */
1456retry:
1457 entropy_count = orig = READ_ONCE(r->entropy_count);
1458 ibytes = nbytes;
1459 /* never pull more than available */
1460 have_bytes = entropy_count >> (ENTROPY_SHIFT + 3);
1461
1462 if ((have_bytes -= reserved) < 0)
1463 have_bytes = 0;
1464 ibytes = min_t(size_t, ibytes, have_bytes);
1465 if (ibytes < min)
1466 ibytes = 0;
1467
1468 if (unlikely(entropy_count < 0)) {
1469 pr_warn("random: negative entropy count: pool %s count %d\n",
1470 r->name, entropy_count);
1471 WARN_ON(1);
1472 entropy_count = 0;
1473 }
1474 nfrac = ibytes << (ENTROPY_SHIFT + 3);
1475 if ((size_t) entropy_count > nfrac)
1476 entropy_count -= nfrac;
1477 else
1478 entropy_count = 0;
1479
1480 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
1481 goto retry;
1482
1483 trace_debit_entropy(r->name, 8 * ibytes);
1484 if (ibytes &&
1485 (r->entropy_count >> ENTROPY_SHIFT) < random_write_wakeup_bits) {
1486 wake_up_interruptible(&random_write_wait);
1487 kill_fasync(&fasync, SIGIO, POLL_OUT);
1488 }
1489
1490 return ibytes;
1491}
1492
1493/*
1494 * This function does the actual extraction for extract_entropy and
1495 * extract_entropy_user.
1496 *
1497 * Note: we assume that .poolwords is a multiple of 16 words.
1498 */
1499static void extract_buf(struct entropy_store *r, __u8 *out)
1500{
1501 int i;
1502 union {
1503 __u32 w[5];
1504 unsigned long l[LONGS(20)];
1505 } hash;
1506 __u32 workspace[SHA_WORKSPACE_WORDS];
1507 unsigned long flags;
1508
1509 /*
1510 * If we have an architectural hardware random number
1511 * generator, use it for SHA's initial vector
1512 */
1513 sha_init(hash.w);
1514 for (i = 0; i < LONGS(20); i++) {
1515 unsigned long v;
1516 if (!arch_get_random_long(&v))
1517 break;
1518 hash.l[i] = v;
1519 }
1520
1521 /* Generate a hash across the pool, 16 words (512 bits) at a time */
1522 spin_lock_irqsave(&r->lock, flags);
1523 for (i = 0; i < r->poolinfo->poolwords; i += 16)
1524 sha_transform(hash.w, (__u8 *)(r->pool + i), workspace);
1525
1526 /*
1527 * We mix the hash back into the pool to prevent backtracking
1528 * attacks (where the attacker knows the state of the pool
1529 * plus the current outputs, and attempts to find previous
1530 * ouputs), unless the hash function can be inverted. By
1531 * mixing at least a SHA1 worth of hash data back, we make
1532 * brute-forcing the feedback as hard as brute-forcing the
1533 * hash.
1534 */
1535 __mix_pool_bytes(r, hash.w, sizeof(hash.w));
1536 spin_unlock_irqrestore(&r->lock, flags);
1537
1538 memzero_explicit(workspace, sizeof(workspace));
1539
1540 /*
1541 * In case the hash function has some recognizable output
1542 * pattern, we fold it in half. Thus, we always feed back
1543 * twice as much data as we output.
1544 */
1545 hash.w[0] ^= hash.w[3];
1546 hash.w[1] ^= hash.w[4];
1547 hash.w[2] ^= rol32(hash.w[2], 16);
1548
1549 memcpy(out, &hash, EXTRACT_SIZE);
1550 memzero_explicit(&hash, sizeof(hash));
1551}
1552
1553static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
1554 size_t nbytes, int fips)
1555{
1556 ssize_t ret = 0, i;
1557 __u8 tmp[EXTRACT_SIZE];
1558 unsigned long flags;
1559
1560 while (nbytes) {
1561 extract_buf(r, tmp);
1562
1563 if (fips) {
1564 spin_lock_irqsave(&r->lock, flags);
1565 if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
1566 panic("Hardware RNG duplicated output!\n");
1567 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1568 spin_unlock_irqrestore(&r->lock, flags);
1569 }
1570 i = min_t(int, nbytes, EXTRACT_SIZE);
1571 memcpy(buf, tmp, i);
1572 nbytes -= i;
1573 buf += i;
1574 ret += i;
1575 }
1576
1577 /* Wipe data just returned from memory */
1578 memzero_explicit(tmp, sizeof(tmp));
1579
1580 return ret;
1581}
1582
1583/*
1584 * This function extracts randomness from the "entropy pool", and
1585 * returns it in a buffer.
1586 *
1587 * The min parameter specifies the minimum amount we can pull before
1588 * failing to avoid races that defeat catastrophic reseeding while the
1589 * reserved parameter indicates how much entropy we must leave in the
1590 * pool after each pull to avoid starving other readers.
1591 */
1592static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1593 size_t nbytes, int min, int reserved)
1594{
1595 __u8 tmp[EXTRACT_SIZE];
1596 unsigned long flags;
1597
1598 /* if last_data isn't primed, we need EXTRACT_SIZE extra bytes */
1599 if (fips_enabled) {
1600 spin_lock_irqsave(&r->lock, flags);
1601 if (!r->last_data_init) {
1602 r->last_data_init = 1;
1603 spin_unlock_irqrestore(&r->lock, flags);
1604 trace_extract_entropy(r->name, EXTRACT_SIZE,
1605 ENTROPY_BITS(r), _RET_IP_);
1606 xfer_secondary_pool(r, EXTRACT_SIZE);
1607 extract_buf(r, tmp);
1608 spin_lock_irqsave(&r->lock, flags);
1609 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1610 }
1611 spin_unlock_irqrestore(&r->lock, flags);
1612 }
1613
1614 trace_extract_entropy(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1615 xfer_secondary_pool(r, nbytes);
1616 nbytes = account(r, nbytes, min, reserved);
1617
1618 return _extract_entropy(r, buf, nbytes, fips_enabled);
1619}
1620
1621/*
1622 * This function extracts randomness from the "entropy pool", and
1623 * returns it in a userspace buffer.
1624 */
1625static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
1626 size_t nbytes)
1627{
1628 ssize_t ret = 0, i;
1629 __u8 tmp[EXTRACT_SIZE];
1630 int large_request = (nbytes > 256);
1631
1632 trace_extract_entropy_user(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1633 if (!r->initialized && r->pull) {
1634 xfer_secondary_pool(r, ENTROPY_BITS(r->pull)/8);
1635 if (!r->initialized)
1636 return 0;
1637 }
1638 xfer_secondary_pool(r, nbytes);
1639 nbytes = account(r, nbytes, 0, 0);
1640
1641 while (nbytes) {
1642 if (large_request && need_resched()) {
1643 if (signal_pending(current)) {
1644 if (ret == 0)
1645 ret = -ERESTARTSYS;
1646 break;
1647 }
1648 schedule();
1649 }
1650
1651 extract_buf(r, tmp);
1652 i = min_t(int, nbytes, EXTRACT_SIZE);
1653 if (copy_to_user(buf, tmp, i)) {
1654 ret = -EFAULT;
1655 break;
1656 }
1657
1658 nbytes -= i;
1659 buf += i;
1660 ret += i;
1661 }
1662
1663 /* Wipe data just returned from memory */
1664 memzero_explicit(tmp, sizeof(tmp));
1665
1666 return ret;
1667}
1668
1669#define warn_unseeded_randomness(previous) \
1670 _warn_unseeded_randomness(__func__, (void *) _RET_IP_, (previous))
1671
1672static void _warn_unseeded_randomness(const char *func_name, void *caller,
1673 void **previous)
1674{
1675#ifdef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1676 const bool print_once = false;
1677#else
1678 static bool print_once __read_mostly;
1679#endif
1680
1681 if (print_once ||
1682 crng_ready() ||
1683 (previous && (caller == READ_ONCE(*previous))))
1684 return;
1685 WRITE_ONCE(*previous, caller);
1686#ifndef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1687 print_once = true;
1688#endif
1689 if (__ratelimit(&unseeded_warning))
1690 pr_notice("random: %s called from %pS with crng_init=%d\n",
1691 func_name, caller, crng_init);
1692}
1693
1694/*
1695 * This function is the exported kernel interface. It returns some
1696 * number of good random numbers, suitable for key generation, seeding
1697 * TCP sequence numbers, etc. It does not rely on the hardware random
1698 * number generator. For random bytes direct from the hardware RNG
1699 * (when available), use get_random_bytes_arch(). In order to ensure
1700 * that the randomness provided by this function is okay, the function
1701 * wait_for_random_bytes() should be called and return 0 at least once
1702 * at any point prior.
1703 */
1704static void _get_random_bytes(void *buf, int nbytes)
1705{
1706 __u8 tmp[CHACHA_BLOCK_SIZE] __aligned(4);
1707
1708 trace_get_random_bytes(nbytes, _RET_IP_);
1709
1710 while (nbytes >= CHACHA_BLOCK_SIZE) {
1711 extract_crng(buf);
1712 buf += CHACHA_BLOCK_SIZE;
1713 nbytes -= CHACHA_BLOCK_SIZE;
1714 }
1715
1716 if (nbytes > 0) {
1717 extract_crng(tmp);
1718 memcpy(buf, tmp, nbytes);
1719 crng_backtrack_protect(tmp, nbytes);
1720 } else
1721 crng_backtrack_protect(tmp, CHACHA_BLOCK_SIZE);
1722 memzero_explicit(tmp, sizeof(tmp));
1723}
1724
1725void get_random_bytes(void *buf, int nbytes)
1726{
1727 static void *previous;
1728
1729 warn_unseeded_randomness(&previous);
1730 _get_random_bytes(buf, nbytes);
1731}
1732EXPORT_SYMBOL(get_random_bytes);
1733
1734
1735/*
1736 * Each time the timer fires, we expect that we got an unpredictable
1737 * jump in the cycle counter. Even if the timer is running on another
1738 * CPU, the timer activity will be touching the stack of the CPU that is
1739 * generating entropy..
1740 *
1741 * Note that we don't re-arm the timer in the timer itself - we are
1742 * happy to be scheduled away, since that just makes the load more
1743 * complex, but we do not want the timer to keep ticking unless the
1744 * entropy loop is running.
1745 *
1746 * So the re-arming always happens in the entropy loop itself.
1747 */
1748static void entropy_timer(struct timer_list *t)
1749{
1750 credit_entropy_bits(&input_pool, 1);
1751}
1752
1753/*
1754 * If we have an actual cycle counter, see if we can
1755 * generate enough entropy with timing noise
1756 */
1757static void try_to_generate_entropy(void)
1758{
1759 struct {
1760 unsigned long now;
1761 struct timer_list timer;
1762 } stack;
1763
1764 stack.now = random_get_entropy();
1765
1766 /* Slow counter - or none. Don't even bother */
1767 if (stack.now == random_get_entropy())
1768 return;
1769
1770 timer_setup_on_stack(&stack.timer, entropy_timer, 0);
1771 while (!crng_ready()) {
1772 if (!timer_pending(&stack.timer))
1773 mod_timer(&stack.timer, jiffies+1);
1774 mix_pool_bytes(&input_pool, &stack.now, sizeof(stack.now));
1775 schedule();
1776 stack.now = random_get_entropy();
1777 }
1778
1779 del_timer_sync(&stack.timer);
1780 destroy_timer_on_stack(&stack.timer);
1781 mix_pool_bytes(&input_pool, &stack.now, sizeof(stack.now));
1782}
1783
1784/*
1785 * Wait for the urandom pool to be seeded and thus guaranteed to supply
1786 * cryptographically secure random numbers. This applies to: the /dev/urandom
1787 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
1788 * family of functions. Using any of these functions without first calling
1789 * this function forfeits the guarantee of security.
1790 *
1791 * Returns: 0 if the urandom pool has been seeded.
1792 * -ERESTARTSYS if the function was interrupted by a signal.
1793 */
1794int wait_for_random_bytes(void)
1795{
1796 if (likely(crng_ready()))
1797 return 0;
1798
1799 do {
1800 int ret;
1801 ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ);
1802 if (ret)
1803 return ret > 0 ? 0 : ret;
1804
1805 try_to_generate_entropy();
1806 } while (!crng_ready());
1807
1808 return 0;
1809}
1810EXPORT_SYMBOL(wait_for_random_bytes);
1811
1812/*
1813 * Returns whether or not the urandom pool has been seeded and thus guaranteed
1814 * to supply cryptographically secure random numbers. This applies to: the
1815 * /dev/urandom device, the get_random_bytes function, and the get_random_{u32,
1816 * ,u64,int,long} family of functions.
1817 *
1818 * Returns: true if the urandom pool has been seeded.
1819 * false if the urandom pool has not been seeded.
1820 */
1821bool rng_is_initialized(void)
1822{
1823 return crng_ready();
1824}
1825EXPORT_SYMBOL(rng_is_initialized);
1826
1827/*
1828 * Add a callback function that will be invoked when the nonblocking
1829 * pool is initialised.
1830 *
1831 * returns: 0 if callback is successfully added
1832 * -EALREADY if pool is already initialised (callback not called)
1833 * -ENOENT if module for callback is not alive
1834 */
1835int add_random_ready_callback(struct random_ready_callback *rdy)
1836{
1837 struct module *owner;
1838 unsigned long flags;
1839 int err = -EALREADY;
1840
1841 if (crng_ready())
1842 return err;
1843
1844 owner = rdy->owner;
1845 if (!try_module_get(owner))
1846 return -ENOENT;
1847
1848 spin_lock_irqsave(&random_ready_list_lock, flags);
1849 if (crng_ready())
1850 goto out;
1851
1852 owner = NULL;
1853
1854 list_add(&rdy->list, &random_ready_list);
1855 err = 0;
1856
1857out:
1858 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1859
1860 module_put(owner);
1861
1862 return err;
1863}
1864EXPORT_SYMBOL(add_random_ready_callback);
1865
1866/*
1867 * Delete a previously registered readiness callback function.
1868 */
1869void del_random_ready_callback(struct random_ready_callback *rdy)
1870{
1871 unsigned long flags;
1872 struct module *owner = NULL;
1873
1874 spin_lock_irqsave(&random_ready_list_lock, flags);
1875 if (!list_empty(&rdy->list)) {
1876 list_del_init(&rdy->list);
1877 owner = rdy->owner;
1878 }
1879 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1880
1881 module_put(owner);
1882}
1883EXPORT_SYMBOL(del_random_ready_callback);
1884
1885/*
1886 * This function will use the architecture-specific hardware random
1887 * number generator if it is available. The arch-specific hw RNG will
1888 * almost certainly be faster than what we can do in software, but it
1889 * is impossible to verify that it is implemented securely (as
1890 * opposed, to, say, the AES encryption of a sequence number using a
1891 * key known by the NSA). So it's useful if we need the speed, but
1892 * only if we're willing to trust the hardware manufacturer not to
1893 * have put in a back door.
1894 *
1895 * Return number of bytes filled in.
1896 */
1897int __must_check get_random_bytes_arch(void *buf, int nbytes)
1898{
1899 int left = nbytes;
1900 char *p = buf;
1901
1902 trace_get_random_bytes_arch(left, _RET_IP_);
1903 while (left) {
1904 unsigned long v;
1905 int chunk = min_t(int, left, sizeof(unsigned long));
1906
1907 if (!arch_get_random_long(&v))
1908 break;
1909
1910 memcpy(p, &v, chunk);
1911 p += chunk;
1912 left -= chunk;
1913 }
1914
1915 return nbytes - left;
1916}
1917EXPORT_SYMBOL(get_random_bytes_arch);
1918
1919/*
1920 * init_std_data - initialize pool with system data
1921 *
1922 * @r: pool to initialize
1923 *
1924 * This function clears the pool's entropy count and mixes some system
1925 * data into the pool to prepare it for use. The pool is not cleared
1926 * as that can only decrease the entropy in the pool.
1927 */
1928static void __init init_std_data(struct entropy_store *r)
1929{
1930 int i;
1931 ktime_t now = ktime_get_real();
1932 unsigned long rv;
1933
1934 r->last_pulled = jiffies;
1935 mix_pool_bytes(r, &now, sizeof(now));
1936 for (i = r->poolinfo->poolbytes; i > 0; i -= sizeof(rv)) {
1937 if (!arch_get_random_seed_long(&rv) &&
1938 !arch_get_random_long(&rv))
1939 rv = random_get_entropy();
1940 mix_pool_bytes(r, &rv, sizeof(rv));
1941 }
1942 mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1943}
1944
1945/*
1946 * Note that setup_arch() may call add_device_randomness()
1947 * long before we get here. This allows seeding of the pools
1948 * with some platform dependent data very early in the boot
1949 * process. But it limits our options here. We must use
1950 * statically allocated structures that already have all
1951 * initializations complete at compile time. We should also
1952 * take care not to overwrite the precious per platform data
1953 * we were given.
1954 */
1955int __init rand_initialize(void)
1956{
1957 init_std_data(&input_pool);
1958 init_std_data(&blocking_pool);
1959 crng_initialize(&primary_crng);
1960 crng_global_init_time = jiffies;
1961 if (ratelimit_disable) {
1962 urandom_warning.interval = 0;
1963 unseeded_warning.interval = 0;
1964 }
1965 return 0;
1966}
1967
1968#ifdef CONFIG_BLOCK
1969void rand_initialize_disk(struct gendisk *disk)
1970{
1971 struct timer_rand_state *state;
1972
1973 /*
1974 * If kzalloc returns null, we just won't use that entropy
1975 * source.
1976 */
1977 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1978 if (state) {
1979 state->last_time = INITIAL_JIFFIES;
1980 disk->random = state;
1981 }
1982}
1983#endif
1984
1985static ssize_t
1986_random_read(int nonblock, char __user *buf, size_t nbytes)
1987{
1988 ssize_t n;
1989
1990 if (nbytes == 0)
1991 return 0;
1992
1993 nbytes = min_t(size_t, nbytes, SEC_XFER_SIZE);
1994 while (1) {
1995 n = extract_entropy_user(&blocking_pool, buf, nbytes);
1996 if (n < 0)
1997 return n;
1998 trace_random_read(n*8, (nbytes-n)*8,
1999 ENTROPY_BITS(&blocking_pool),
2000 ENTROPY_BITS(&input_pool));
2001 if (n > 0)
2002 return n;
2003
2004 /* Pool is (near) empty. Maybe wait and retry. */
2005 if (nonblock)
2006 return -EAGAIN;
2007
2008 wait_event_interruptible(random_read_wait,
2009 blocking_pool.initialized &&
2010 (ENTROPY_BITS(&input_pool) >= random_read_wakeup_bits));
2011 if (signal_pending(current))
2012 return -ERESTARTSYS;
2013 }
2014}
2015
2016static ssize_t
2017random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
2018{
2019 return _random_read(file->f_flags & O_NONBLOCK, buf, nbytes);
2020}
2021
2022static ssize_t
2023urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
2024{
2025 unsigned long flags;
2026 static int maxwarn = 10;
2027 int ret;
2028
2029 if (!crng_ready() && maxwarn > 0) {
2030 maxwarn--;
2031 if (__ratelimit(&urandom_warning))
2032 printk(KERN_NOTICE "random: %s: uninitialized "
2033 "urandom read (%zd bytes read)\n",
2034 current->comm, nbytes);
2035 spin_lock_irqsave(&primary_crng.lock, flags);
2036 crng_init_cnt = 0;
2037 spin_unlock_irqrestore(&primary_crng.lock, flags);
2038 }
2039 nbytes = min_t(size_t, nbytes, INT_MAX >> (ENTROPY_SHIFT + 3));
2040 ret = extract_crng_user(buf, nbytes);
2041 trace_urandom_read(8 * nbytes, 0, ENTROPY_BITS(&input_pool));
2042 return ret;
2043}
2044
2045static __poll_t
2046random_poll(struct file *file, poll_table * wait)
2047{
2048 __poll_t mask;
2049
2050 poll_wait(file, &random_read_wait, wait);
2051 poll_wait(file, &random_write_wait, wait);
2052 mask = 0;
2053 if (ENTROPY_BITS(&input_pool) >= random_read_wakeup_bits)
2054 mask |= EPOLLIN | EPOLLRDNORM;
2055 if (ENTROPY_BITS(&input_pool) < random_write_wakeup_bits)
2056 mask |= EPOLLOUT | EPOLLWRNORM;
2057 return mask;
2058}
2059
2060static int
2061write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
2062{
2063 size_t bytes;
2064 __u32 t, buf[16];
2065 const char __user *p = buffer;
2066
2067 while (count > 0) {
2068 int b, i = 0;
2069
2070 bytes = min(count, sizeof(buf));
2071 if (copy_from_user(&buf, p, bytes))
2072 return -EFAULT;
2073
2074 for (b = bytes ; b > 0 ; b -= sizeof(__u32), i++) {
2075 if (!arch_get_random_int(&t))
2076 break;
2077 buf[i] ^= t;
2078 }
2079
2080 count -= bytes;
2081 p += bytes;
2082
2083 mix_pool_bytes(r, buf, bytes);
2084 cond_resched();
2085 }
2086
2087 return 0;
2088}
2089
2090static ssize_t random_write(struct file *file, const char __user *buffer,
2091 size_t count, loff_t *ppos)
2092{
2093 size_t ret;
2094
2095 ret = write_pool(&input_pool, buffer, count);
2096 if (ret)
2097 return ret;
2098
2099 return (ssize_t)count;
2100}
2101
2102static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
2103{
2104 int size, ent_count;
2105 int __user *p = (int __user *)arg;
2106 int retval;
2107
2108 switch (cmd) {
2109 case RNDGETENTCNT:
2110 /* inherently racy, no point locking */
2111 ent_count = ENTROPY_BITS(&input_pool);
2112 if (put_user(ent_count, p))
2113 return -EFAULT;
2114 return 0;
2115 case RNDADDTOENTCNT:
2116 if (!capable(CAP_SYS_ADMIN))
2117 return -EPERM;
2118 if (get_user(ent_count, p))
2119 return -EFAULT;
2120 return credit_entropy_bits_safe(&input_pool, ent_count);
2121 case RNDADDENTROPY:
2122 if (!capable(CAP_SYS_ADMIN))
2123 return -EPERM;
2124 if (get_user(ent_count, p++))
2125 return -EFAULT;
2126 if (ent_count < 0)
2127 return -EINVAL;
2128 if (get_user(size, p++))
2129 return -EFAULT;
2130 retval = write_pool(&input_pool, (const char __user *)p,
2131 size);
2132 if (retval < 0)
2133 return retval;
2134 return credit_entropy_bits_safe(&input_pool, ent_count);
2135 case RNDZAPENTCNT:
2136 case RNDCLEARPOOL:
2137 /*
2138 * Clear the entropy pool counters. We no longer clear
2139 * the entropy pool, as that's silly.
2140 */
2141 if (!capable(CAP_SYS_ADMIN))
2142 return -EPERM;
2143 input_pool.entropy_count = 0;
2144 blocking_pool.entropy_count = 0;
2145 return 0;
2146 case RNDRESEEDCRNG:
2147 if (!capable(CAP_SYS_ADMIN))
2148 return -EPERM;
2149 if (crng_init < 2)
2150 return -ENODATA;
2151 crng_reseed(&primary_crng, NULL);
2152 crng_global_init_time = jiffies - 1;
2153 return 0;
2154 default:
2155 return -EINVAL;
2156 }
2157}
2158
2159static int random_fasync(int fd, struct file *filp, int on)
2160{
2161 return fasync_helper(fd, filp, on, &fasync);
2162}
2163
2164const struct file_operations random_fops = {
2165 .read = random_read,
2166 .write = random_write,
2167 .poll = random_poll,
2168 .unlocked_ioctl = random_ioctl,
2169 .fasync = random_fasync,
2170 .llseek = noop_llseek,
2171};
2172
2173const struct file_operations urandom_fops = {
2174 .read = urandom_read,
2175 .write = random_write,
2176 .unlocked_ioctl = random_ioctl,
2177 .fasync = random_fasync,
2178 .llseek = noop_llseek,
2179};
2180
2181SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count,
2182 unsigned int, flags)
2183{
2184 int ret;
2185
2186 if (flags & ~(GRND_NONBLOCK|GRND_RANDOM))
2187 return -EINVAL;
2188
2189 if (count > INT_MAX)
2190 count = INT_MAX;
2191
2192 if (flags & GRND_RANDOM)
2193 return _random_read(flags & GRND_NONBLOCK, buf, count);
2194
2195 if (!crng_ready()) {
2196 if (flags & GRND_NONBLOCK)
2197 return -EAGAIN;
2198 ret = wait_for_random_bytes();
2199 if (unlikely(ret))
2200 return ret;
2201 }
2202 return urandom_read(NULL, buf, count, NULL);
2203}
2204
2205/********************************************************************
2206 *
2207 * Sysctl interface
2208 *
2209 ********************************************************************/
2210
2211#ifdef CONFIG_SYSCTL
2212
2213#include <linux/sysctl.h>
2214
2215static int min_read_thresh = 8, min_write_thresh;
2216static int max_read_thresh = OUTPUT_POOL_WORDS * 32;
2217static int max_write_thresh = INPUT_POOL_WORDS * 32;
2218static int random_min_urandom_seed = 60;
2219static char sysctl_bootid[16];
2220
2221/*
2222 * This function is used to return both the bootid UUID, and random
2223 * UUID. The difference is in whether table->data is NULL; if it is,
2224 * then a new UUID is generated and returned to the user.
2225 *
2226 * If the user accesses this via the proc interface, the UUID will be
2227 * returned as an ASCII string in the standard UUID format; if via the
2228 * sysctl system call, as 16 bytes of binary data.
2229 */
2230static int proc_do_uuid(struct ctl_table *table, int write,
2231 void __user *buffer, size_t *lenp, loff_t *ppos)
2232{
2233 struct ctl_table fake_table;
2234 unsigned char buf[64], tmp_uuid[16], *uuid;
2235
2236 uuid = table->data;
2237 if (!uuid) {
2238 uuid = tmp_uuid;
2239 generate_random_uuid(uuid);
2240 } else {
2241 static DEFINE_SPINLOCK(bootid_spinlock);
2242
2243 spin_lock(&bootid_spinlock);
2244 if (!uuid[8])
2245 generate_random_uuid(uuid);
2246 spin_unlock(&bootid_spinlock);
2247 }
2248
2249 sprintf(buf, "%pU", uuid);
2250
2251 fake_table.data = buf;
2252 fake_table.maxlen = sizeof(buf);
2253
2254 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
2255}
2256
2257/*
2258 * Return entropy available scaled to integral bits
2259 */
2260static int proc_do_entropy(struct ctl_table *table, int write,
2261 void __user *buffer, size_t *lenp, loff_t *ppos)
2262{
2263 struct ctl_table fake_table;
2264 int entropy_count;
2265
2266 entropy_count = *(int *)table->data >> ENTROPY_SHIFT;
2267
2268 fake_table.data = &entropy_count;
2269 fake_table.maxlen = sizeof(entropy_count);
2270
2271 return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
2272}
2273
2274static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
2275extern struct ctl_table random_table[];
2276struct ctl_table random_table[] = {
2277 {
2278 .procname = "poolsize",
2279 .data = &sysctl_poolsize,
2280 .maxlen = sizeof(int),
2281 .mode = 0444,
2282 .proc_handler = proc_dointvec,
2283 },
2284 {
2285 .procname = "entropy_avail",
2286 .maxlen = sizeof(int),
2287 .mode = 0444,
2288 .proc_handler = proc_do_entropy,
2289 .data = &input_pool.entropy_count,
2290 },
2291 {
2292 .procname = "read_wakeup_threshold",
2293 .data = &random_read_wakeup_bits,
2294 .maxlen = sizeof(int),
2295 .mode = 0644,
2296 .proc_handler = proc_dointvec_minmax,
2297 .extra1 = &min_read_thresh,
2298 .extra2 = &max_read_thresh,
2299 },
2300 {
2301 .procname = "write_wakeup_threshold",
2302 .data = &random_write_wakeup_bits,
2303 .maxlen = sizeof(int),
2304 .mode = 0644,
2305 .proc_handler = proc_dointvec_minmax,
2306 .extra1 = &min_write_thresh,
2307 .extra2 = &max_write_thresh,
2308 },
2309 {
2310 .procname = "urandom_min_reseed_secs",
2311 .data = &random_min_urandom_seed,
2312 .maxlen = sizeof(int),
2313 .mode = 0644,
2314 .proc_handler = proc_dointvec,
2315 },
2316 {
2317 .procname = "boot_id",
2318 .data = &sysctl_bootid,
2319 .maxlen = 16,
2320 .mode = 0444,
2321 .proc_handler = proc_do_uuid,
2322 },
2323 {
2324 .procname = "uuid",
2325 .maxlen = 16,
2326 .mode = 0444,
2327 .proc_handler = proc_do_uuid,
2328 },
2329#ifdef ADD_INTERRUPT_BENCH
2330 {
2331 .procname = "add_interrupt_avg_cycles",
2332 .data = &avg_cycles,
2333 .maxlen = sizeof(avg_cycles),
2334 .mode = 0444,
2335 .proc_handler = proc_doulongvec_minmax,
2336 },
2337 {
2338 .procname = "add_interrupt_avg_deviation",
2339 .data = &avg_deviation,
2340 .maxlen = sizeof(avg_deviation),
2341 .mode = 0444,
2342 .proc_handler = proc_doulongvec_minmax,
2343 },
2344#endif
2345 { }
2346};
2347#endif /* CONFIG_SYSCTL */
2348
2349struct batched_entropy {
2350 union {
2351 u64 entropy_u64[CHACHA_BLOCK_SIZE / sizeof(u64)];
2352 u32 entropy_u32[CHACHA_BLOCK_SIZE / sizeof(u32)];
2353 };
2354 unsigned int position;
2355 spinlock_t batch_lock;
2356};
2357
2358/*
2359 * Get a random word for internal kernel use only. The quality of the random
2360 * number is either as good as RDRAND or as good as /dev/urandom, with the
2361 * goal of being quite fast and not depleting entropy. In order to ensure
2362 * that the randomness provided by this function is okay, the function
2363 * wait_for_random_bytes() should be called and return 0 at least once
2364 * at any point prior.
2365 */
2366static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64) = {
2367 .batch_lock = __SPIN_LOCK_UNLOCKED(batched_entropy_u64.lock),
2368};
2369
2370u64 get_random_u64(void)
2371{
2372 u64 ret;
2373 unsigned long flags;
2374 struct batched_entropy *batch;
2375 static void *previous;
2376
2377#if BITS_PER_LONG == 64
2378 if (arch_get_random_long((unsigned long *)&ret))
2379 return ret;
2380#else
2381 if (arch_get_random_long((unsigned long *)&ret) &&
2382 arch_get_random_long((unsigned long *)&ret + 1))
2383 return ret;
2384#endif
2385
2386 warn_unseeded_randomness(&previous);
2387
2388 batch = raw_cpu_ptr(&batched_entropy_u64);
2389 spin_lock_irqsave(&batch->batch_lock, flags);
2390 if (batch->position % ARRAY_SIZE(batch->entropy_u64) == 0) {
2391 extract_crng((u8 *)batch->entropy_u64);
2392 batch->position = 0;
2393 }
2394 ret = batch->entropy_u64[batch->position++];
2395 spin_unlock_irqrestore(&batch->batch_lock, flags);
2396 return ret;
2397}
2398EXPORT_SYMBOL(get_random_u64);
2399
2400static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32) = {
2401 .batch_lock = __SPIN_LOCK_UNLOCKED(batched_entropy_u32.lock),
2402};
2403u32 get_random_u32(void)
2404{
2405 u32 ret;
2406 unsigned long flags;
2407 struct batched_entropy *batch;
2408 static void *previous;
2409
2410 if (arch_get_random_int(&ret))
2411 return ret;
2412
2413 warn_unseeded_randomness(&previous);
2414
2415 batch = raw_cpu_ptr(&batched_entropy_u32);
2416 spin_lock_irqsave(&batch->batch_lock, flags);
2417 if (batch->position % ARRAY_SIZE(batch->entropy_u32) == 0) {
2418 extract_crng((u8 *)batch->entropy_u32);
2419 batch->position = 0;
2420 }
2421 ret = batch->entropy_u32[batch->position++];
2422 spin_unlock_irqrestore(&batch->batch_lock, flags);
2423 return ret;
2424}
2425EXPORT_SYMBOL(get_random_u32);
2426
2427/* It's important to invalidate all potential batched entropy that might
2428 * be stored before the crng is initialized, which we can do lazily by
2429 * simply resetting the counter to zero so that it's re-extracted on the
2430 * next usage. */
2431static void invalidate_batched_entropy(void)
2432{
2433 int cpu;
2434 unsigned long flags;
2435
2436 for_each_possible_cpu (cpu) {
2437 struct batched_entropy *batched_entropy;
2438
2439 batched_entropy = per_cpu_ptr(&batched_entropy_u32, cpu);
2440 spin_lock_irqsave(&batched_entropy->batch_lock, flags);
2441 batched_entropy->position = 0;
2442 spin_unlock(&batched_entropy->batch_lock);
2443
2444 batched_entropy = per_cpu_ptr(&batched_entropy_u64, cpu);
2445 spin_lock(&batched_entropy->batch_lock);
2446 batched_entropy->position = 0;
2447 spin_unlock_irqrestore(&batched_entropy->batch_lock, flags);
2448 }
2449}
2450
2451/**
2452 * randomize_page - Generate a random, page aligned address
2453 * @start: The smallest acceptable address the caller will take.
2454 * @range: The size of the area, starting at @start, within which the
2455 * random address must fall.
2456 *
2457 * If @start + @range would overflow, @range is capped.
2458 *
2459 * NOTE: Historical use of randomize_range, which this replaces, presumed that
2460 * @start was already page aligned. We now align it regardless.
2461 *
2462 * Return: A page aligned address within [start, start + range). On error,
2463 * @start is returned.
2464 */
2465unsigned long
2466randomize_page(unsigned long start, unsigned long range)
2467{
2468 if (!PAGE_ALIGNED(start)) {
2469 range -= PAGE_ALIGN(start) - start;
2470 start = PAGE_ALIGN(start);
2471 }
2472
2473 if (start > ULONG_MAX - range)
2474 range = ULONG_MAX - start;
2475
2476 range >>= PAGE_SHIFT;
2477
2478 if (range == 0)
2479 return start;
2480
2481 return start + (get_random_long() % range << PAGE_SHIFT);
2482}
2483
2484/* Interface for in-kernel drivers of true hardware RNGs.
2485 * Those devices may produce endless random bits and will be throttled
2486 * when our pool is full.
2487 */
2488void add_hwgenerator_randomness(const char *buffer, size_t count,
2489 size_t entropy)
2490{
2491 struct entropy_store *poolp = &input_pool;
2492
2493 if (unlikely(crng_init == 0)) {
2494 crng_fast_load(buffer, count);
2495 return;
2496 }
2497
2498 /* Suspend writing if we're above the trickle threshold.
2499 * We'll be woken up again once below random_write_wakeup_thresh,
2500 * or when the calling thread is about to terminate.
2501 */
2502 wait_event_interruptible(random_write_wait, kthread_should_stop() ||
2503 ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
2504 mix_pool_bytes(poolp, buffer, count);
2505 credit_entropy_bits(poolp, entropy);
2506}
2507EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
2508
2509/* Handle random seed passed by bootloader.
2510 * If the seed is trustworthy, it would be regarded as hardware RNGs. Otherwise
2511 * it would be regarded as device data.
2512 * The decision is controlled by CONFIG_RANDOM_TRUST_BOOTLOADER.
2513 */
2514void add_bootloader_randomness(const void *buf, unsigned int size)
2515{
2516 if (IS_ENABLED(CONFIG_RANDOM_TRUST_BOOTLOADER))
2517 add_hwgenerator_randomness(buf, size, size * 8);
2518 else
2519 add_device_randomness(buf, size);
2520}
2521EXPORT_SYMBOL_GPL(add_bootloader_randomness);