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