random.c

   1// SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause)
   2/*
   3 * Copyright (C) 2017-2024 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#ifdef CONFIG_VDSO_GETRANDOM
  60#include <vdso/getrandom.h>
  61#include <vdso/datapage.h>
  62#include <vdso/vsyscall.h>
  63#endif
  64#include <asm/archrandom.h>
  65#include <asm/processor.h>
  66#include <asm/irq.h>
  67#include <asm/irq_regs.h>
  68#include <asm/io.h>
  69
  70/*********************************************************************
  71 *
  72 * Initialization and readiness waiting.
  73 *
  74 * Much of the RNG infrastructure is devoted to various dependencies
  75 * being able to wait until the RNG has collected enough entropy and
  76 * is ready for safe consumption.
  77 *
  78 *********************************************************************/
  79
  80/*
  81 * crng_init is protected by base_crng->lock, and only increases
  82 * its value (from empty->early->ready).
  83 */
  84static enum {
  85	CRNG_EMPTY = 0, /* Little to no entropy collected */
  86	CRNG_EARLY = 1, /* At least POOL_EARLY_BITS collected */
  87	CRNG_READY = 2  /* Fully initialized with POOL_READY_BITS collected */
  88} crng_init __read_mostly = CRNG_EMPTY;
  89static DEFINE_STATIC_KEY_FALSE(crng_is_ready);
  90#define crng_ready() (static_branch_likely(&crng_is_ready) || crng_init >= CRNG_READY)
  91/* Various types of waiters for crng_init->CRNG_READY transition. */
  92static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
  93static struct fasync_struct *fasync;
  94static ATOMIC_NOTIFIER_HEAD(random_ready_notifier);
  95
  96/* Control how we warn userspace. */
  97static struct ratelimit_state urandom_warning =
  98	RATELIMIT_STATE_INIT_FLAGS("urandom_warning", HZ, 3, RATELIMIT_MSG_ON_RELEASE);
  99static int ratelimit_disable __read_mostly = 0;
 100module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
 101MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
 102
 103/*
 104 * Returns whether or not the input pool has been seeded and thus guaranteed
 105 * to supply cryptographically secure random numbers. This applies to: the
 106 * /dev/urandom device, the get_random_bytes function, and the get_random_{u8,
 107 * u16,u32,u64,long} family of functions.
 108 *
 109 * Returns: true if the input pool has been seeded.
 110 *          false if the input pool has not been seeded.
 111 */
 112bool rng_is_initialized(void)
 113{
 114	return crng_ready();
 115}
 116EXPORT_SYMBOL(rng_is_initialized);
 117
 118static void __cold crng_set_ready(struct work_struct *work)
 119{
 120	static_branch_enable(&crng_is_ready);
 121}
 122
 123/* Used by wait_for_random_bytes(), and considered an entropy collector, below. */
 124static void try_to_generate_entropy(void);
 125
 126/*
 127 * Wait for the input pool to be seeded and thus guaranteed to supply
 128 * cryptographically secure random numbers. This applies to: the /dev/urandom
 129 * device, the get_random_bytes function, and the get_random_{u8,u16,u32,u64,
 130 * long} family of functions. Using any of these functions without first
 131 * calling this function forfeits the guarantee of security.
 132 *
 133 * Returns: 0 if the input pool has been seeded.
 134 *          -ERESTARTSYS if the function was interrupted by a signal.
 135 */
 136int wait_for_random_bytes(void)
 137{
 138	while (!crng_ready()) {
 139		int ret;
 140
 141		try_to_generate_entropy();
 142		ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ);
 143		if (ret)
 144			return ret > 0 ? 0 : ret;
 145	}
 146	return 0;
 147}
 148EXPORT_SYMBOL(wait_for_random_bytes);
 149
 150/*
 151 * Add a callback function that will be invoked when the crng is initialised,
 152 * or immediately if it already has been. Only use this is you are absolutely
 153 * sure it is required. Most users should instead be able to test
 154 * `rng_is_initialized()` on demand, or make use of `get_random_bytes_wait()`.
 155 */
 156int __cold execute_with_initialized_rng(struct notifier_block *nb)
 157{
 158	unsigned long flags;
 159	int ret = 0;
 160
 161	spin_lock_irqsave(&random_ready_notifier.lock, flags);
 162	if (crng_ready())
 163		nb->notifier_call(nb, 0, NULL);
 164	else
 165		ret = raw_notifier_chain_register((struct raw_notifier_head *)&random_ready_notifier.head, nb);
 166	spin_unlock_irqrestore(&random_ready_notifier.lock, flags);
 167	return ret;
 168}
 169
 170/*********************************************************************
 171 *
 172 * Fast key erasure RNG, the "crng".
 173 *
 174 * These functions expand entropy from the entropy extractor into
 175 * long streams for external consumption using the "fast key erasure"
 176 * RNG described at <https://blog.cr.yp.to/20170723-random.html>.
 177 *
 178 * There are a few exported interfaces for use by other drivers:
 179 *
 180 *	void get_random_bytes(void *buf, size_t len)
 181 *	u8 get_random_u8()
 182 *	u16 get_random_u16()
 183 *	u32 get_random_u32()
 184 *	u32 get_random_u32_below(u32 ceil)
 185 *	u32 get_random_u32_above(u32 floor)
 186 *	u32 get_random_u32_inclusive(u32 floor, u32 ceil)
 187 *	u64 get_random_u64()
 188 *	unsigned long get_random_long()
 189 *
 190 * These interfaces will return the requested number of random bytes
 191 * into the given buffer or as a return value. This is equivalent to
 192 * a read from /dev/urandom. The u8, u16, u32, u64, long family of
 193 * functions may be higher performance for one-off random integers,
 194 * because they do a bit of buffering and do not invoke reseeding
 195 * until the buffer is emptied.
 196 *
 197 *********************************************************************/
 198
 199enum {
 200	CRNG_RESEED_START_INTERVAL = HZ,
 201	CRNG_RESEED_INTERVAL = 60 * HZ
 202};
 203
 204static struct {
 205	u8 key[CHACHA_KEY_SIZE] __aligned(__alignof__(long));
 206	unsigned long generation;
 207	spinlock_t lock;
 208} base_crng = {
 209	.lock = __SPIN_LOCK_UNLOCKED(base_crng.lock)
 210};
 211
 212struct crng {
 213	u8 key[CHACHA_KEY_SIZE];
 214	unsigned long generation;
 215	local_lock_t lock;
 216};
 217
 218static DEFINE_PER_CPU(struct crng, crngs) = {
 219	.generation = ULONG_MAX,
 220	.lock = INIT_LOCAL_LOCK(crngs.lock),
 221};
 222
 223/*
 224 * Return the interval until the next reseeding, which is normally
 225 * CRNG_RESEED_INTERVAL, but during early boot, it is at an interval
 226 * proportional to the uptime.
 227 */
 228static unsigned int crng_reseed_interval(void)
 229{
 230	static bool early_boot = true;
 231
 232	if (unlikely(READ_ONCE(early_boot))) {
 233		time64_t uptime = ktime_get_seconds();
 234		if (uptime >= CRNG_RESEED_INTERVAL / HZ * 2)
 235			WRITE_ONCE(early_boot, false);
 236		else
 237			return max_t(unsigned int, CRNG_RESEED_START_INTERVAL,
 238				     (unsigned int)uptime / 2 * HZ);
 239	}
 240	return CRNG_RESEED_INTERVAL;
 241}
 242
 243/* Used by crng_reseed() and crng_make_state() to extract a new seed from the input pool. */
 244static void extract_entropy(void *buf, size_t len);
 245
 246/* This extracts a new crng key from the input pool. */
 247static void crng_reseed(struct work_struct *work)
 248{
 249	static DECLARE_DELAYED_WORK(next_reseed, crng_reseed);
 250	unsigned long flags;
 251	unsigned long next_gen;
 252	u8 key[CHACHA_KEY_SIZE];
 253
 254	/* Immediately schedule the next reseeding, so that it fires sooner rather than later. */
 255	if (likely(system_dfl_wq))
 256		queue_delayed_work(system_dfl_wq, &next_reseed, crng_reseed_interval());
 257
 258	extract_entropy(key, sizeof(key));
 259
 260	/*
 261	 * We copy the new key into the base_crng, overwriting the old one,
 262	 * and update the generation counter. We avoid hitting ULONG_MAX,
 263	 * because the per-cpu crngs are initialized to ULONG_MAX, so this
 264	 * forces new CPUs that come online to always initialize.
 265	 */
 266	spin_lock_irqsave(&base_crng.lock, flags);
 267	memcpy(base_crng.key, key, sizeof(base_crng.key));
 268	next_gen = base_crng.generation + 1;
 269	if (next_gen == ULONG_MAX)
 270		++next_gen;
 271	WRITE_ONCE(base_crng.generation, next_gen);
 272#ifdef CONFIG_VDSO_GETRANDOM
 273	/* base_crng.generation's invalid value is ULONG_MAX, while
 274	 * vdso_k_rng_data->generation's invalid value is 0, so add one to the
 275	 * former to arrive at the latter. Use smp_store_release so that this
 276	 * is ordered with the write above to base_crng.generation. Pairs with
 277	 * the smp_rmb() before the syscall in the vDSO code.
 278	 *
 279	 * Cast to unsigned long for 32-bit architectures, since atomic 64-bit
 280	 * operations are not supported on those architectures. This is safe
 281	 * because base_crng.generation is a 32-bit value. On big-endian
 282	 * architectures it will be stored in the upper 32 bits, but that's okay
 283	 * because the vDSO side only checks whether the value changed, without
 284	 * actually using or interpreting the value.
 285	 */
 286	smp_store_release((unsigned long *)&vdso_k_rng_data->generation, next_gen + 1);
 287#endif
 288	if (!static_branch_likely(&crng_is_ready))
 289		crng_init = CRNG_READY;
 290	spin_unlock_irqrestore(&base_crng.lock, flags);
 291	memzero_explicit(key, sizeof(key));
 292}
 293
 294/*
 295 * This generates a ChaCha block using the provided key, and then
 296 * immediately overwrites that key with half the block. It returns
 297 * the resultant ChaCha state to the user, along with the second
 298 * half of the block containing 32 bytes of random data that may
 299 * be used; random_data_len may not be greater than 32.
 300 *
 301 * The returned ChaCha state contains within it a copy of the old
 302 * key value, at index 4, so the state should always be zeroed out
 303 * immediately after using in order to maintain forward secrecy.
 304 * If the state cannot be erased in a timely manner, then it is
 305 * safer to set the random_data parameter to &chacha_state->x[4]
 306 * so that this function overwrites it before returning.
 307 */
 308static void crng_fast_key_erasure(u8 key[CHACHA_KEY_SIZE],
 309				  struct chacha_state *chacha_state,
 310				  u8 *random_data, size_t random_data_len)
 311{
 312	u8 first_block[CHACHA_BLOCK_SIZE];
 313
 314	BUG_ON(random_data_len > 32);
 315
 316	chacha_init_consts(chacha_state);
 317	memcpy(&chacha_state->x[4], key, CHACHA_KEY_SIZE);
 318	memset(&chacha_state->x[12], 0, sizeof(u32) * 4);
 319	chacha20_block(chacha_state, first_block);
 320
 321	memcpy(key, first_block, CHACHA_KEY_SIZE);
 322	memcpy(random_data, first_block + CHACHA_KEY_SIZE, random_data_len);
 323	memzero_explicit(first_block, sizeof(first_block));
 324}
 325
 326/*
 327 * This function returns a ChaCha state that you may use for generating
 328 * random data. It also returns up to 32 bytes on its own of random data
 329 * that may be used; random_data_len may not be greater than 32.
 330 */
 331static void crng_make_state(struct chacha_state *chacha_state,
 332			    u8 *random_data, size_t random_data_len)
 333{
 334	unsigned long flags;
 335	struct crng *crng;
 336
 337	BUG_ON(random_data_len > 32);
 338
 339	/*
 340	 * For the fast path, we check whether we're ready, unlocked first, and
 341	 * then re-check once locked later. In the case where we're really not
 342	 * ready, we do fast key erasure with the base_crng directly, extracting
 343	 * when crng_init is CRNG_EMPTY.
 344	 */
 345	if (!crng_ready()) {
 346		bool ready;
 347
 348		spin_lock_irqsave(&base_crng.lock, flags);
 349		ready = crng_ready();
 350		if (!ready) {
 351			if (crng_init == CRNG_EMPTY)
 352				extract_entropy(base_crng.key, sizeof(base_crng.key));
 353			crng_fast_key_erasure(base_crng.key, chacha_state,
 354					      random_data, random_data_len);
 355		}
 356		spin_unlock_irqrestore(&base_crng.lock, flags);
 357		if (!ready)
 358			return;
 359	}
 360
 361	local_lock_irqsave(&crngs.lock, flags);
 362	crng = raw_cpu_ptr(&crngs);
 363
 364	/*
 365	 * If our per-cpu crng is older than the base_crng, then it means
 366	 * somebody reseeded the base_crng. In that case, we do fast key
 367	 * erasure on the base_crng, and use its output as the new key
 368	 * for our per-cpu crng. This brings us up to date with base_crng.
 369	 */
 370	if (unlikely(crng->generation != READ_ONCE(base_crng.generation))) {
 371		spin_lock(&base_crng.lock);
 372		crng_fast_key_erasure(base_crng.key, chacha_state,
 373				      crng->key, sizeof(crng->key));
 374		crng->generation = base_crng.generation;
 375		spin_unlock(&base_crng.lock);
 376	}
 377
 378	/*
 379	 * Finally, when we've made it this far, our per-cpu crng has an up
 380	 * to date key, and we can do fast key erasure with it to produce
 381	 * some random data and a ChaCha state for the caller. All other
 382	 * branches of this function are "unlikely", so most of the time we
 383	 * should wind up here immediately.
 384	 */
 385	crng_fast_key_erasure(crng->key, chacha_state, random_data, random_data_len);
 386	local_unlock_irqrestore(&crngs.lock, flags);
 387}
 388
 389static void _get_random_bytes(void *buf, size_t len)
 390{
 391	struct chacha_state chacha_state;
 392	u8 tmp[CHACHA_BLOCK_SIZE];
 393	size_t first_block_len;
 394
 395	if (!len)
 396		return;
 397
 398	first_block_len = min_t(size_t, 32, len);
 399	crng_make_state(&chacha_state, buf, first_block_len);
 400	len -= first_block_len;
 401	buf += first_block_len;
 402
 403	while (len) {
 404		if (len < CHACHA_BLOCK_SIZE) {
 405			chacha20_block(&chacha_state, tmp);
 406			memcpy(buf, tmp, len);
 407			memzero_explicit(tmp, sizeof(tmp));
 408			break;
 409		}
 410
 411		chacha20_block(&chacha_state, buf);
 412		if (unlikely(chacha_state.x[12] == 0))
 413			++chacha_state.x[13];
 414		len -= CHACHA_BLOCK_SIZE;
 415		buf += CHACHA_BLOCK_SIZE;
 416	}
 417
 418	chacha_zeroize_state(&chacha_state);
 419}
 420
 421/*
 422 * This returns random bytes in arbitrary quantities. The quality of the
 423 * random bytes is as good as /dev/urandom. In order to ensure that the
 424 * randomness provided by this function is okay, the function
 425 * wait_for_random_bytes() should be called and return 0 at least once
 426 * at any point prior.
 427 */
 428void get_random_bytes(void *buf, size_t len)
 429{
 430	_get_random_bytes(buf, len);
 431}
 432EXPORT_SYMBOL(get_random_bytes);
 433
 434static ssize_t get_random_bytes_user(struct iov_iter *iter)
 435{
 436	struct chacha_state chacha_state;
 437	u8 block[CHACHA_BLOCK_SIZE];
 438	size_t ret = 0, copied;
 439
 440	if (unlikely(!iov_iter_count(iter)))
 441		return 0;
 442
 443	/*
 444	 * Immediately overwrite the ChaCha key at index 4 with random
 445	 * bytes, in case userspace causes copy_to_iter() below to sleep
 446	 * forever, so that we still retain forward secrecy in that case.
 447	 */
 448	crng_make_state(&chacha_state, (u8 *)&chacha_state.x[4],
 449			CHACHA_KEY_SIZE);
 450	/*
 451	 * However, if we're doing a read of len <= 32, we don't need to
 452	 * use chacha_state after, so we can simply return those bytes to
 453	 * the user directly.
 454	 */
 455	if (iov_iter_count(iter) <= CHACHA_KEY_SIZE) {
 456		ret = copy_to_iter(&chacha_state.x[4], CHACHA_KEY_SIZE, iter);
 457		goto out_zero_chacha;
 458	}
 459
 460	for (;;) {
 461		chacha20_block(&chacha_state, block);
 462		if (unlikely(chacha_state.x[12] == 0))
 463			++chacha_state.x[13];
 464
 465		copied = copy_to_iter(block, sizeof(block), iter);
 466		ret += copied;
 467		if (!iov_iter_count(iter) || copied != sizeof(block))
 468			break;
 469
 470		BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0);
 471		if (ret % PAGE_SIZE == 0) {
 472			if (signal_pending(current))
 473				break;
 474			cond_resched();
 475		}
 476	}
 477
 478	memzero_explicit(block, sizeof(block));
 479out_zero_chacha:
 480	chacha_zeroize_state(&chacha_state);
 481	return ret ? ret : -EFAULT;
 482}
 483
 484/*
 485 * Batched entropy returns random integers. The quality of the random
 486 * number is as good as /dev/urandom. In order to ensure that the randomness
 487 * provided by this function is okay, the function wait_for_random_bytes()
 488 * should be called and return 0 at least once at any point prior.
 489 */
 490
 491#define DEFINE_BATCHED_ENTROPY(type)						\
 492struct batch_ ##type {								\
 493	/*									\
 494	 * We make this 1.5x a ChaCha block, so that we get the			\
 495	 * remaining 32 bytes from fast key erasure, plus one full		\
 496	 * block from the detached ChaCha state. We can increase		\
 497	 * the size of this later if needed so long as we keep the		\
 498	 * formula of (integer_blocks + 0.5) * CHACHA_BLOCK_SIZE.		\
 499	 */									\
 500	type entropy[CHACHA_BLOCK_SIZE * 3 / (2 * sizeof(type))];		\
 501	local_lock_t lock;							\
 502	unsigned long generation;						\
 503	unsigned int position;							\
 504};										\
 505										\
 506static DEFINE_PER_CPU(struct batch_ ##type, batched_entropy_ ##type) = {	\
 507	.lock = INIT_LOCAL_LOCK(batched_entropy_ ##type.lock),			\
 508	.position = UINT_MAX							\
 509};										\
 510										\
 511type get_random_ ##type(void)							\
 512{										\
 513	type ret;								\
 514	unsigned long flags;							\
 515	struct batch_ ##type *batch;						\
 516	unsigned long next_gen;							\
 517										\
 518	if  (!crng_ready()) {							\
 519		_get_random_bytes(&ret, sizeof(ret));				\
 520		return ret;							\
 521	}									\
 522										\
 523	local_lock_irqsave(&batched_entropy_ ##type.lock, flags);		\
 524	batch = raw_cpu_ptr(&batched_entropy_##type);				\
 525										\
 526	next_gen = READ_ONCE(base_crng.generation);				\
 527	if (batch->position >= ARRAY_SIZE(batch->entropy) ||			\
 528	    next_gen != batch->generation) {					\
 529		_get_random_bytes(batch->entropy, sizeof(batch->entropy));	\
 530		batch->position = 0;						\
 531		batch->generation = next_gen;					\
 532	}									\
 533										\
 534	ret = batch->entropy[batch->position];					\
 535	batch->entropy[batch->position] = 0;					\
 536	++batch->position;							\
 537	local_unlock_irqrestore(&batched_entropy_ ##type.lock, flags);		\
 538	return ret;								\
 539}										\
 540EXPORT_SYMBOL(get_random_ ##type);
 541
 542DEFINE_BATCHED_ENTROPY(u8)
 543DEFINE_BATCHED_ENTROPY(u16)
 544DEFINE_BATCHED_ENTROPY(u32)
 545DEFINE_BATCHED_ENTROPY(u64)
 546
 547u32 __get_random_u32_below(u32 ceil)
 548{
 549	/*
 550	 * This is the slow path for variable ceil. It is still fast, most of
 551	 * the time, by doing traditional reciprocal multiplication and
 552	 * opportunistically comparing the lower half to ceil itself, before
 553	 * falling back to computing a larger bound, and then rejecting samples
 554	 * whose lower half would indicate a range indivisible by ceil. The use
 555	 * of `-ceil % ceil` is analogous to `2^32 % ceil`, but is computable
 556	 * in 32-bits.
 557	 */
 558	u32 rand = get_random_u32();
 559	u64 mult;
 560
 561	/*
 562	 * This function is technically undefined for ceil == 0, and in fact
 563	 * for the non-underscored constant version in the header, we build bug
 564	 * on that. But for the non-constant case, it's convenient to have that
 565	 * evaluate to being a straight call to get_random_u32(), so that
 566	 * get_random_u32_inclusive() can work over its whole range without
 567	 * undefined behavior.
 568	 */
 569	if (unlikely(!ceil))
 570		return rand;
 571
 572	mult = (u64)ceil * rand;
 573	if (unlikely((u32)mult < ceil)) {
 574		u32 bound = -ceil % ceil;
 575		while (unlikely((u32)mult < bound))
 576			mult = (u64)ceil * get_random_u32();
 577	}
 578	return mult >> 32;
 579}
 580EXPORT_SYMBOL(__get_random_u32_below);
 581
 582#ifdef CONFIG_SMP
 583/*
 584 * This function is called when the CPU is coming up, with entry
 585 * CPUHP_RANDOM_PREPARE, which comes before CPUHP_WORKQUEUE_PREP.
 586 */
 587int __cold random_prepare_cpu(unsigned int cpu)
 588{
 589	/*
 590	 * When the cpu comes back online, immediately invalidate both
 591	 * the per-cpu crng and all batches, so that we serve fresh
 592	 * randomness.
 593	 */
 594	per_cpu_ptr(&crngs, cpu)->generation = ULONG_MAX;
 595	per_cpu_ptr(&batched_entropy_u8, cpu)->position = UINT_MAX;
 596	per_cpu_ptr(&batched_entropy_u16, cpu)->position = UINT_MAX;
 597	per_cpu_ptr(&batched_entropy_u32, cpu)->position = UINT_MAX;
 598	per_cpu_ptr(&batched_entropy_u64, cpu)->position = UINT_MAX;
 599	return 0;
 600}
 601#endif
 602
 603
 604/**********************************************************************
 605 *
 606 * Entropy accumulation and extraction routines.
 607 *
 608 * Callers may add entropy via:
 609 *
 610 *     static void mix_pool_bytes(const void *buf, size_t len)
 611 *
 612 * After which, if added entropy should be credited:
 613 *
 614 *     static void credit_init_bits(size_t bits)
 615 *
 616 * Finally, extract entropy via:
 617 *
 618 *     static void extract_entropy(void *buf, size_t len)
 619 *
 620 **********************************************************************/
 621
 622enum {
 623	POOL_BITS = BLAKE2S_HASH_SIZE * 8,
 624	POOL_READY_BITS = POOL_BITS, /* When crng_init->CRNG_READY */
 625	POOL_EARLY_BITS = POOL_READY_BITS / 2 /* When crng_init->CRNG_EARLY */
 626};
 627
 628static struct {
 629	struct blake2s_ctx hash;
 630	spinlock_t lock;
 631	unsigned int init_bits;
 632} input_pool = {
 633	.hash.h = { BLAKE2S_IV0 ^ (0x01010000 | BLAKE2S_HASH_SIZE),
 634		    BLAKE2S_IV1, BLAKE2S_IV2, BLAKE2S_IV3, BLAKE2S_IV4,
 635		    BLAKE2S_IV5, BLAKE2S_IV6, BLAKE2S_IV7 },
 636	.hash.outlen = BLAKE2S_HASH_SIZE,
 637	.lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
 638};
 639
 640static void _mix_pool_bytes(const void *buf, size_t len)
 641{
 642	blake2s_update(&input_pool.hash, buf, len);
 643}
 644
 645/*
 646 * This function adds bytes into the input pool. It does not
 647 * update the initialization bit counter; the caller should call
 648 * credit_init_bits if this is appropriate.
 649 */
 650static void mix_pool_bytes(const void *buf, size_t len)
 651{
 652	unsigned long flags;
 653
 654	spin_lock_irqsave(&input_pool.lock, flags);
 655	_mix_pool_bytes(buf, len);
 656	spin_unlock_irqrestore(&input_pool.lock, flags);
 657}
 658
 659/*
 660 * This is an HKDF-like construction for using the hashed collected entropy
 661 * as a PRF key, that's then expanded block-by-block.
 662 */
 663static void extract_entropy(void *buf, size_t len)
 664{
 665	unsigned long flags;
 666	u8 seed[BLAKE2S_HASH_SIZE], next_key[BLAKE2S_HASH_SIZE];
 667	struct {
 668		unsigned long rdseed[32 / sizeof(long)];
 669		size_t counter;
 670	} block;
 671	size_t i, longs;
 672
 673	for (i = 0; i < ARRAY_SIZE(block.rdseed);) {
 674		longs = arch_get_random_seed_longs(&block.rdseed[i], ARRAY_SIZE(block.rdseed) - i);
 675		if (longs) {
 676			i += longs;
 677			continue;
 678		}
 679		longs = arch_get_random_longs(&block.rdseed[i], ARRAY_SIZE(block.rdseed) - i);
 680		if (longs) {
 681			i += longs;
 682			continue;
 683		}
 684		block.rdseed[i++] = random_get_entropy();
 685	}
 686
 687	spin_lock_irqsave(&input_pool.lock, flags);
 688
 689	/* seed = HASHPRF(last_key, entropy_input) */
 690	blake2s_final(&input_pool.hash, seed);
 691
 692	/* next_key = HASHPRF(seed, RDSEED || 0) */
 693	block.counter = 0;
 694	blake2s(seed, sizeof(seed), (const u8 *)&block, sizeof(block), next_key, sizeof(next_key));
 695	blake2s_init_key(&input_pool.hash, BLAKE2S_HASH_SIZE, next_key, sizeof(next_key));
 696
 697	spin_unlock_irqrestore(&input_pool.lock, flags);
 698	memzero_explicit(next_key, sizeof(next_key));
 699
 700	while (len) {
 701		i = min_t(size_t, len, BLAKE2S_HASH_SIZE);
 702		/* output = HASHPRF(seed, RDSEED || ++counter) */
 703		++block.counter;
 704		blake2s(seed, sizeof(seed), (const u8 *)&block, sizeof(block), buf, i);
 705		len -= i;
 706		buf += i;
 707	}
 708
 709	memzero_explicit(seed, sizeof(seed));
 710	memzero_explicit(&block, sizeof(block));
 711}
 712
 713#define credit_init_bits(bits) if (!crng_ready()) _credit_init_bits(bits)
 714
 715static void __cold _credit_init_bits(size_t bits)
 716{
 717	static DECLARE_WORK(set_ready, crng_set_ready);
 718	unsigned int new, orig, add;
 719	unsigned long flags;
 720	int m;
 721
 722	if (!bits)
 723		return;
 724
 725	add = min_t(size_t, bits, POOL_BITS);
 726
 727	orig = READ_ONCE(input_pool.init_bits);
 728	do {
 729		new = min_t(unsigned int, POOL_BITS, orig + add);
 730	} while (!try_cmpxchg(&input_pool.init_bits, &orig, new));
 731
 732	if (orig < POOL_READY_BITS && new >= POOL_READY_BITS) {
 733		crng_reseed(NULL); /* Sets crng_init to CRNG_READY under base_crng.lock. */
 734		if (system_dfl_wq)
 735			queue_work(system_dfl_wq, &set_ready);
 736		atomic_notifier_call_chain(&random_ready_notifier, 0, NULL);
 737#ifdef CONFIG_VDSO_GETRANDOM
 738		WRITE_ONCE(vdso_k_rng_data->is_ready, true);
 739#endif
 740		wake_up_interruptible(&crng_init_wait);
 741		kill_fasync(&fasync, SIGIO, POLL_IN);
 742		pr_notice("crng init done\n");
 743		m = ratelimit_state_get_miss(&urandom_warning);
 744		if (m)
 745			pr_notice("%d urandom warning(s) missed due to ratelimiting\n", m);
 746	} else if (orig < POOL_EARLY_BITS && new >= POOL_EARLY_BITS) {
 747		spin_lock_irqsave(&base_crng.lock, flags);
 748		/* Check if crng_init is CRNG_EMPTY, to avoid race with crng_reseed(). */
 749		if (crng_init == CRNG_EMPTY) {
 750			extract_entropy(base_crng.key, sizeof(base_crng.key));
 751			crng_init = CRNG_EARLY;
 752		}
 753		spin_unlock_irqrestore(&base_crng.lock, flags);
 754	}
 755}
 756
 757
 758/**********************************************************************
 759 *
 760 * Entropy collection routines.
 761 *
 762 * The following exported functions are used for pushing entropy into
 763 * the above entropy accumulation routines:
 764 *
 765 *	void add_device_randomness(const void *buf, size_t len);
 766 *	void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy, bool sleep_after);
 767 *	void add_bootloader_randomness(const void *buf, size_t len);
 768 *	void add_vmfork_randomness(const void *unique_vm_id, size_t len);
 769 *	void add_interrupt_randomness(int irq);
 770 *	void add_input_randomness(unsigned int type, unsigned int code, unsigned int value);
 771 *	void add_disk_randomness(struct gendisk *disk);
 772 *
 773 * add_device_randomness() adds data to the input pool that
 774 * is likely to differ between two devices (or possibly even per boot).
 775 * This would be things like MAC addresses or serial numbers, or the
 776 * read-out of the RTC. This does *not* credit any actual entropy to
 777 * the pool, but it initializes the pool to different values for devices
 778 * that might otherwise be identical and have very little entropy
 779 * available to them (particularly common in the embedded world).
 780 *
 781 * add_hwgenerator_randomness() is for true hardware RNGs, and will credit
 782 * entropy as specified by the caller. If the entropy pool is full it will
 783 * block until more entropy is needed.
 784 *
 785 * add_bootloader_randomness() is called by bootloader drivers, such as EFI
 786 * and device tree, and credits its input depending on whether or not the
 787 * command line option 'random.trust_bootloader' is set.
 788 *
 789 * add_vmfork_randomness() adds a unique (but not necessarily secret) ID
 790 * representing the current instance of a VM to the pool, without crediting,
 791 * and then force-reseeds the crng so that it takes effect immediately.
 792 *
 793 * add_interrupt_randomness() uses the interrupt timing as random
 794 * inputs to the entropy pool. Using the cycle counters and the irq source
 795 * as inputs, it feeds the input pool roughly once a second or after 64
 796 * interrupts, crediting 1 bit of entropy for whichever comes first.
 797 *
 798 * add_input_randomness() uses the input layer interrupt timing, as well
 799 * as the event type information from the hardware.
 800 *
 801 * add_disk_randomness() uses what amounts to the seek time of block
 802 * layer request events, on a per-disk_devt basis, as input to the
 803 * entropy pool. Note that high-speed solid state drives with very low
 804 * seek times do not make for good sources of entropy, as their seek
 805 * times are usually fairly consistent.
 806 *
 807 * The last two routines try to estimate how many bits of entropy
 808 * to credit. They do this by keeping track of the first and second
 809 * order deltas of the event timings.
 810 *
 811 **********************************************************************/
 812
 813static bool trust_cpu __initdata = true;
 814static bool trust_bootloader __initdata = true;
 815static int __init parse_trust_cpu(char *arg)
 816{
 817	return kstrtobool(arg, &trust_cpu);
 818}
 819static int __init parse_trust_bootloader(char *arg)
 820{
 821	return kstrtobool(arg, &trust_bootloader);
 822}
 823early_param("random.trust_cpu", parse_trust_cpu);
 824early_param("random.trust_bootloader", parse_trust_bootloader);
 825
 826static int random_pm_notification(struct notifier_block *nb, unsigned long action, void *data)
 827{
 828	unsigned long flags, entropy = random_get_entropy();
 829
 830	/*
 831	 * Encode a representation of how long the system has been suspended,
 832	 * in a way that is distinct from prior system suspends.
 833	 */
 834	ktime_t stamps[] = { ktime_get(), ktime_get_boottime(), ktime_get_real() };
 835
 836	spin_lock_irqsave(&input_pool.lock, flags);
 837	_mix_pool_bytes(&action, sizeof(action));
 838	_mix_pool_bytes(stamps, sizeof(stamps));
 839	_mix_pool_bytes(&entropy, sizeof(entropy));
 840	spin_unlock_irqrestore(&input_pool.lock, flags);
 841
 842	if (crng_ready() && (action == PM_RESTORE_PREPARE ||
 843	    (action == PM_POST_SUSPEND && !IS_ENABLED(CONFIG_PM_AUTOSLEEP) &&
 844	     !IS_ENABLED(CONFIG_PM_USERSPACE_AUTOSLEEP)))) {
 845		crng_reseed(NULL);
 846		pr_notice("crng reseeded on system resumption\n");
 847	}
 848	return 0;
 849}
 850
 851static struct notifier_block pm_notifier = { .notifier_call = random_pm_notification };
 852
 853/*
 854 * This is called extremely early, before time keeping functionality is
 855 * available, but arch randomness is. Interrupts are not yet enabled.
 856 */
 857void __init random_init_early(const char *command_line)
 858{
 859	unsigned long entropy[BLAKE2S_BLOCK_SIZE / sizeof(long)];
 860	size_t i, longs, arch_bits;
 861
 862#if defined(LATENT_ENTROPY_PLUGIN)
 863	static const u8 compiletime_seed[BLAKE2S_BLOCK_SIZE] __initconst __latent_entropy;
 864	_mix_pool_bytes(compiletime_seed, sizeof(compiletime_seed));
 865#endif
 866
 867	for (i = 0, arch_bits = sizeof(entropy) * 8; i < ARRAY_SIZE(entropy);) {
 868		longs = arch_get_random_seed_longs(entropy, ARRAY_SIZE(entropy) - i);
 869		if (longs) {
 870			_mix_pool_bytes(entropy, sizeof(*entropy) * longs);
 871			i += longs;
 872			continue;
 873		}
 874		longs = arch_get_random_longs(entropy, ARRAY_SIZE(entropy) - i);
 875		if (longs) {
 876			_mix_pool_bytes(entropy, sizeof(*entropy) * longs);
 877			i += longs;
 878			continue;
 879		}
 880		arch_bits -= sizeof(*entropy) * 8;
 881		++i;
 882	}
 883
 884	_mix_pool_bytes(init_utsname(), sizeof(*(init_utsname())));
 885	_mix_pool_bytes(command_line, strlen(command_line));
 886
 887	/* Reseed if already seeded by earlier phases. */
 888	if (crng_ready())
 889		crng_reseed(NULL);
 890	else if (trust_cpu)
 891		_credit_init_bits(arch_bits);
 892}
 893
 894/*
 895 * This is called a little bit after the prior function, and now there is
 896 * access to timestamps counters. Interrupts are not yet enabled.
 897 */
 898void __init random_init(void)
 899{
 900	unsigned long entropy = random_get_entropy();
 901	ktime_t now = ktime_get_real();
 902
 903	_mix_pool_bytes(&now, sizeof(now));
 904	_mix_pool_bytes(&entropy, sizeof(entropy));
 905	add_latent_entropy();
 906
 907	/*
 908	 * If we were initialized by the cpu or bootloader before workqueues
 909	 * are initialized, then we should enable the static branch here.
 910	 */
 911	if (!static_branch_likely(&crng_is_ready) && crng_init >= CRNG_READY)
 912		crng_set_ready(NULL);
 913
 914	/* Reseed if already seeded by earlier phases. */
 915	if (crng_ready())
 916		crng_reseed(NULL);
 917
 918	WARN_ON(register_pm_notifier(&pm_notifier));
 919
 920	WARN(!entropy, "Missing cycle counter and fallback timer; RNG "
 921		       "entropy collection will consequently suffer.");
 922}
 923
 924/*
 925 * Add device- or boot-specific data to the input pool to help
 926 * initialize it.
 927 *
 928 * None of this adds any entropy; it is meant to avoid the problem of
 929 * the entropy pool having similar initial state across largely
 930 * identical devices.
 931 */
 932void add_device_randomness(const void *buf, size_t len)
 933{
 934	unsigned long entropy = random_get_entropy();
 935	unsigned long flags;
 936
 937	spin_lock_irqsave(&input_pool.lock, flags);
 938	_mix_pool_bytes(&entropy, sizeof(entropy));
 939	_mix_pool_bytes(buf, len);
 940	spin_unlock_irqrestore(&input_pool.lock, flags);
 941}
 942EXPORT_SYMBOL(add_device_randomness);
 943
 944/*
 945 * Interface for in-kernel drivers of true hardware RNGs. Those devices
 946 * may produce endless random bits, so this function will sleep for
 947 * some amount of time after, if the sleep_after parameter is true.
 948 */
 949void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy, bool sleep_after)
 950{
 951	mix_pool_bytes(buf, len);
 952	credit_init_bits(entropy);
 953
 954	/*
 955	 * Throttle writing to once every reseed interval, unless we're not yet
 956	 * initialized or no entropy is credited.
 957	 */
 958	if (sleep_after && !kthread_should_stop() && (crng_ready() || !entropy))
 959		schedule_timeout_interruptible(crng_reseed_interval());
 960}
 961EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
 962
 963/*
 964 * Handle random seed passed by bootloader, and credit it depending
 965 * on the command line option 'random.trust_bootloader'.
 966 */
 967void __init add_bootloader_randomness(const void *buf, size_t len)
 968{
 969	mix_pool_bytes(buf, len);
 970	if (trust_bootloader)
 971		credit_init_bits(len * 8);
 972}
 973
 974#if IS_ENABLED(CONFIG_VMGENID)
 975static BLOCKING_NOTIFIER_HEAD(vmfork_chain);
 976
 977/*
 978 * Handle a new unique VM ID, which is unique, not secret, so we
 979 * don't credit it, but we do immediately force a reseed after so
 980 * that it's used by the crng posthaste.
 981 */
 982void __cold add_vmfork_randomness(const void *unique_vm_id, size_t len)
 983{
 984	add_device_randomness(unique_vm_id, len);
 985	if (crng_ready()) {
 986		crng_reseed(NULL);
 987		pr_notice("crng reseeded due to virtual machine fork\n");
 988	}
 989	blocking_notifier_call_chain(&vmfork_chain, 0, NULL);
 990}
 991#if IS_MODULE(CONFIG_VMGENID)
 992EXPORT_SYMBOL_GPL(add_vmfork_randomness);
 993#endif
 994
 995int __cold register_random_vmfork_notifier(struct notifier_block *nb)
 996{
 997	return blocking_notifier_chain_register(&vmfork_chain, nb);
 998}
 999EXPORT_SYMBOL_GPL(register_random_vmfork_notifier);
1000
1001int __cold unregister_random_vmfork_notifier(struct notifier_block *nb)
1002{
1003	return blocking_notifier_chain_unregister(&vmfork_chain, nb);
1004}
1005EXPORT_SYMBOL_GPL(unregister_random_vmfork_notifier);
1006#endif
1007
1008struct fast_pool {
1009	unsigned long pool[4];
1010	unsigned long last;
1011	unsigned int count;
1012	struct timer_list mix;
1013};
1014
1015static void mix_interrupt_randomness(struct timer_list *work);
1016
1017static DEFINE_PER_CPU(struct fast_pool, irq_randomness) = {
1018#ifdef CONFIG_64BIT
1019#define FASTMIX_PERM SIPHASH_PERMUTATION
1020	.pool = { SIPHASH_CONST_0, SIPHASH_CONST_1, SIPHASH_CONST_2, SIPHASH_CONST_3 },
1021#else
1022#define FASTMIX_PERM HSIPHASH_PERMUTATION
1023	.pool = { HSIPHASH_CONST_0, HSIPHASH_CONST_1, HSIPHASH_CONST_2, HSIPHASH_CONST_3 },
1024#endif
1025	.mix = __TIMER_INITIALIZER(mix_interrupt_randomness, 0)
1026};
1027
1028/*
1029 * This is [Half]SipHash-1-x, starting from an empty key. Because
1030 * the key is fixed, it assumes that its inputs are non-malicious,
1031 * and therefore this has no security on its own. s represents the
1032 * four-word SipHash state, while v represents a two-word input.
1033 */
1034static void fast_mix(unsigned long s[4], unsigned long v1, unsigned long v2)
1035{
1036	s[3] ^= v1;
1037	FASTMIX_PERM(s[0], s[1], s[2], s[3]);
1038	s[0] ^= v1;
1039	s[3] ^= v2;
1040	FASTMIX_PERM(s[0], s[1], s[2], s[3]);
1041	s[0] ^= v2;
1042}
1043
1044#ifdef CONFIG_SMP
1045/*
1046 * This function is called when the CPU has just come online, with
1047 * entry CPUHP_AP_RANDOM_ONLINE, just after CPUHP_AP_WORKQUEUE_ONLINE.
1048 */
1049int __cold random_online_cpu(unsigned int cpu)
1050{
1051	/*
1052	 * During CPU shutdown and before CPU onlining, add_interrupt_
1053	 * randomness() may schedule mix_interrupt_randomness(), and
1054	 * set the MIX_INFLIGHT flag. However, because the worker can
1055	 * be scheduled on a different CPU during this period, that
1056	 * flag will never be cleared. For that reason, we zero out
1057	 * the flag here, which runs just after workqueues are onlined
1058	 * for the CPU again. This also has the effect of setting the
1059	 * irq randomness count to zero so that new accumulated irqs
1060	 * are fresh.
1061	 */
1062	per_cpu_ptr(&irq_randomness, cpu)->count = 0;
1063	return 0;
1064}
1065#endif
1066
1067static void mix_interrupt_randomness(struct timer_list *work)
1068{
1069	struct fast_pool *fast_pool = container_of(work, struct fast_pool, mix);
1070	/*
1071	 * The size of the copied stack pool is explicitly 2 longs so that we
1072	 * only ever ingest half of the siphash output each time, retaining
1073	 * the other half as the next "key" that carries over. The entropy is
1074	 * supposed to be sufficiently dispersed between bits so on average
1075	 * we don't wind up "losing" some.
1076	 */
1077	unsigned long pool[2];
1078	unsigned int count;
1079
1080	/* Check to see if we're running on the wrong CPU due to hotplug. */
1081	local_irq_disable();
1082	if (fast_pool != this_cpu_ptr(&irq_randomness)) {
1083		local_irq_enable();
1084		return;
1085	}
1086
1087	/*
1088	 * Copy the pool to the stack so that the mixer always has a
1089	 * consistent view, before we reenable irqs again.
1090	 */
1091	memcpy(pool, fast_pool->pool, sizeof(pool));
1092	count = fast_pool->count;
1093	fast_pool->count = 0;
1094	fast_pool->last = jiffies;
1095	local_irq_enable();
1096
1097	mix_pool_bytes(pool, sizeof(pool));
1098	credit_init_bits(clamp_t(unsigned int, (count & U16_MAX) / 64, 1, sizeof(pool) * 8));
1099
1100	memzero_explicit(pool, sizeof(pool));
1101}
1102
1103void add_interrupt_randomness(int irq)
1104{
1105	enum { MIX_INFLIGHT = 1U << 31 };
1106	unsigned long entropy = random_get_entropy();
1107	struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1108	struct pt_regs *regs = get_irq_regs();
1109	unsigned int new_count;
1110
1111	fast_mix(fast_pool->pool, entropy,
1112		 (regs ? instruction_pointer(regs) : _RET_IP_) ^ swab(irq));
1113	new_count = ++fast_pool->count;
1114
1115	if (new_count & MIX_INFLIGHT)
1116		return;
1117
1118	if (new_count < 1024 && !time_is_before_jiffies(fast_pool->last + HZ))
1119		return;
1120
1121	fast_pool->count |= MIX_INFLIGHT;
1122	if (!timer_pending(&fast_pool->mix)) {
1123		fast_pool->mix.expires = jiffies;
1124		add_timer_on(&fast_pool->mix, raw_smp_processor_id());
1125	}
1126}
1127EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1128
1129/* There is one of these per entropy source */
1130struct timer_rand_state {
1131	unsigned long last_time;
1132	long last_delta, last_delta2;
1133};
1134
1135/*
1136 * This function adds entropy to the entropy "pool" by using timing
1137 * delays. It uses the timer_rand_state structure to make an estimate
1138 * of how many bits of entropy this call has added to the pool. The
1139 * value "num" is also added to the pool; it should somehow describe
1140 * the type of event that just happened.
1141 */
1142static void add_timer_randomness(struct timer_rand_state *state, unsigned int num)
1143{
1144	unsigned long entropy = random_get_entropy(), now = jiffies, flags;
1145	long delta, delta2, delta3;
1146	unsigned int bits;
1147
1148	/*
1149	 * If we're in a hard IRQ, add_interrupt_randomness() will be called
1150	 * sometime after, so mix into the fast pool.
1151	 */
1152	if (in_hardirq()) {
1153		fast_mix(this_cpu_ptr(&irq_randomness)->pool, entropy, num);
1154	} else {
1155		spin_lock_irqsave(&input_pool.lock, flags);
1156		_mix_pool_bytes(&entropy, sizeof(entropy));
1157		_mix_pool_bytes(&num, sizeof(num));
1158		spin_unlock_irqrestore(&input_pool.lock, flags);
1159	}
1160
1161	if (crng_ready())
1162		return;
1163
1164	/*
1165	 * Calculate number of bits of randomness we probably added.
1166	 * We take into account the first, second and third-order deltas
1167	 * in order to make our estimate.
1168	 */
1169	delta = now - READ_ONCE(state->last_time);
1170	WRITE_ONCE(state->last_time, now);
1171
1172	delta2 = delta - READ_ONCE(state->last_delta);
1173	WRITE_ONCE(state->last_delta, delta);
1174
1175	delta3 = delta2 - READ_ONCE(state->last_delta2);
1176	WRITE_ONCE(state->last_delta2, delta2);
1177
1178	if (delta < 0)
1179		delta = -delta;
1180	if (delta2 < 0)
1181		delta2 = -delta2;
1182	if (delta3 < 0)
1183		delta3 = -delta3;
1184	if (delta > delta2)
1185		delta = delta2;
1186	if (delta > delta3)
1187		delta = delta3;
1188
1189	/*
1190	 * delta is now minimum absolute delta. Round down by 1 bit
1191	 * on general principles, and limit entropy estimate to 11 bits.
1192	 */
1193	bits = min(fls(delta >> 1), 11);
1194
1195	/*
1196	 * As mentioned above, if we're in a hard IRQ, add_interrupt_randomness()
1197	 * will run after this, which uses a different crediting scheme of 1 bit
1198	 * per every 64 interrupts. In order to let that function do accounting
1199	 * close to the one in this function, we credit a full 64/64 bit per bit,
1200	 * and then subtract one to account for the extra one added.
1201	 */
1202	if (in_hardirq())
1203		this_cpu_ptr(&irq_randomness)->count += max(1u, bits * 64) - 1;
1204	else
1205		_credit_init_bits(bits);
1206}
1207
1208void add_input_randomness(unsigned int type, unsigned int code, unsigned int value)
1209{
1210	static unsigned char last_value;
1211	static struct timer_rand_state input_timer_state = { INITIAL_JIFFIES };
1212
1213	/* Ignore autorepeat and the like. */
1214	if (value == last_value)
1215		return;
1216
1217	last_value = value;
1218	add_timer_randomness(&input_timer_state,
1219			     (type << 4) ^ code ^ (code >> 4) ^ value);
1220}
1221EXPORT_SYMBOL_GPL(add_input_randomness);
1222
1223#ifdef CONFIG_BLOCK
1224void add_disk_randomness(struct gendisk *disk)
1225{
1226	if (!disk || !disk->random)
1227		return;
1228	/* First major is 1, so we get >= 0x200 here. */
1229	add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1230}
1231EXPORT_SYMBOL_GPL(add_disk_randomness);
1232
1233void __cold rand_initialize_disk(struct gendisk *disk)
1234{
1235	struct timer_rand_state *state;
1236
1237	/*
1238	 * If kzalloc returns null, we just won't use that entropy
1239	 * source.
1240	 */
1241	state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1242	if (state) {
1243		state->last_time = INITIAL_JIFFIES;
1244		disk->random = state;
1245	}
1246}
1247#endif
1248
1249struct entropy_timer_state {
1250	unsigned long entropy;
1251	struct timer_list timer;
1252	atomic_t samples;
1253	unsigned int samples_per_bit;
1254};
1255
1256/*
1257 * Each time the timer fires, we expect that we got an unpredictable jump in
1258 * the cycle counter. Even if the timer is running on another CPU, the timer
1259 * activity will be touching the stack of the CPU that is generating entropy.
1260 *
1261 * Note that we don't re-arm the timer in the timer itself - we are happy to be
1262 * scheduled away, since that just makes the load more complex, but we do not
1263 * want the timer to keep ticking unless the entropy loop is running.
1264 *
1265 * So the re-arming always happens in the entropy loop itself.
1266 */
1267static void __cold entropy_timer(struct timer_list *timer)
1268{
1269	struct entropy_timer_state *state = container_of(timer, struct entropy_timer_state, timer);
1270	unsigned long entropy = random_get_entropy();
1271
1272	mix_pool_bytes(&entropy, sizeof(entropy));
1273	if (atomic_inc_return(&state->samples) % state->samples_per_bit == 0)
1274		credit_init_bits(1);
1275}
1276
1277/*
1278 * If we have an actual cycle counter, see if we can generate enough entropy
1279 * with timing noise.
1280 */
1281static void __cold try_to_generate_entropy(void)
1282{
1283	enum { NUM_TRIAL_SAMPLES = 8192, MAX_SAMPLES_PER_BIT = HZ / 15 };
1284	u8 stack_bytes[sizeof(struct entropy_timer_state) + SMP_CACHE_BYTES - 1];
1285	struct entropy_timer_state *stack = PTR_ALIGN((void *)stack_bytes, SMP_CACHE_BYTES);
1286	unsigned int i, num_different = 0;
1287	unsigned long last = random_get_entropy();
1288	cpumask_var_t timer_cpus;
1289	int cpu = -1;
1290
1291	for (i = 0; i < NUM_TRIAL_SAMPLES - 1; ++i) {
1292		stack->entropy = random_get_entropy();
1293		if (stack->entropy != last)
1294			++num_different;
1295		last = stack->entropy;
1296	}
1297	stack->samples_per_bit = DIV_ROUND_UP(NUM_TRIAL_SAMPLES, num_different + 1);
1298	if (stack->samples_per_bit > MAX_SAMPLES_PER_BIT)
1299		return;
1300
1301	atomic_set(&stack->samples, 0);
1302	timer_setup_on_stack(&stack->timer, entropy_timer, 0);
1303	if (!alloc_cpumask_var(&timer_cpus, GFP_KERNEL))
1304		goto out;
1305
1306	while (!crng_ready() && !signal_pending(current)) {
1307		/*
1308		 * Check !timer_pending() and then ensure that any previous callback has finished
1309		 * executing by checking timer_delete_sync_try(), before queueing the next one.
1310		 */
1311		if (!timer_pending(&stack->timer) && timer_delete_sync_try(&stack->timer) >= 0) {
1312			unsigned int num_cpus;
1313
1314			/*
1315			 * Preemption must be disabled here, both to read the current CPU number
1316			 * and to avoid scheduling a timer on a dead CPU.
1317			 */
1318			preempt_disable();
1319
1320			/* Only schedule callbacks on timer CPUs that are online. */
1321			cpumask_and(timer_cpus, housekeeping_cpumask(HK_TYPE_TIMER), cpu_online_mask);
1322			num_cpus = cpumask_weight(timer_cpus);
1323			/* In very bizarre case of misconfiguration, fallback to all online. */
1324			if (unlikely(num_cpus == 0)) {
1325				*timer_cpus = *cpu_online_mask;
1326				num_cpus = cpumask_weight(timer_cpus);
1327			}
1328
1329			/* Basic CPU round-robin, which avoids the current CPU. */
1330			do {
1331				cpu = cpumask_next(cpu, timer_cpus);
1332				if (cpu >= nr_cpu_ids)
1333					cpu = cpumask_first(timer_cpus);
1334			} while (cpu == smp_processor_id() && num_cpus > 1);
1335
1336			/* Expiring the timer at `jiffies` means it's the next tick. */
1337			stack->timer.expires = jiffies;
1338
1339			add_timer_on(&stack->timer, cpu);
1340
1341			preempt_enable();
1342		}
1343		mix_pool_bytes(&stack->entropy, sizeof(stack->entropy));
1344		schedule();
1345		stack->entropy = random_get_entropy();
1346	}
1347	mix_pool_bytes(&stack->entropy, sizeof(stack->entropy));
1348
1349	free_cpumask_var(timer_cpus);
1350out:
1351	timer_delete_sync(&stack->timer);
1352	timer_destroy_on_stack(&stack->timer);
1353}
1354
1355
1356/**********************************************************************
1357 *
1358 * Userspace reader/writer interfaces.
1359 *
1360 * getrandom(2) is the primary modern interface into the RNG and should
1361 * be used in preference to anything else.
1362 *
1363 * Reading from /dev/random has the same functionality as calling
1364 * getrandom(2) with flags=0. In earlier versions, however, it had
1365 * vastly different semantics and should therefore be avoided, to
1366 * prevent backwards compatibility issues.
1367 *
1368 * Reading from /dev/urandom has the same functionality as calling
1369 * getrandom(2) with flags=GRND_INSECURE. Because it does not block
1370 * waiting for the RNG to be ready, it should not be used.
1371 *
1372 * Writing to either /dev/random or /dev/urandom adds entropy to
1373 * the input pool but does not credit it.
1374 *
1375 * Polling on /dev/random indicates when the RNG is initialized, on
1376 * the read side, and when it wants new entropy, on the write side.
1377 *
1378 * Both /dev/random and /dev/urandom have the same set of ioctls for
1379 * adding entropy, getting the entropy count, zeroing the count, and
1380 * reseeding the crng.
1381 *
1382 **********************************************************************/
1383
1384SYSCALL_DEFINE3(getrandom, char __user *, ubuf, size_t, len, unsigned int, flags)
1385{
1386	struct iov_iter iter;
1387	int ret;
1388
1389	if (flags & ~(GRND_NONBLOCK | GRND_RANDOM | GRND_INSECURE))
1390		return -EINVAL;
1391
1392	/*
1393	 * Requesting insecure and blocking randomness at the same time makes
1394	 * no sense.
1395	 */
1396	if ((flags & (GRND_INSECURE | GRND_RANDOM)) == (GRND_INSECURE | GRND_RANDOM))
1397		return -EINVAL;
1398
1399	if (!crng_ready() && !(flags & GRND_INSECURE)) {
1400		if (flags & GRND_NONBLOCK)
1401			return -EAGAIN;
1402		ret = wait_for_random_bytes();
1403		if (unlikely(ret))
1404			return ret;
1405	}
1406
1407	ret = import_ubuf(ITER_DEST, ubuf, len, &iter);
1408	if (unlikely(ret))
1409		return ret;
1410	return get_random_bytes_user(&iter);
1411}
1412
1413static __poll_t random_poll(struct file *file, poll_table *wait)
1414{
1415	poll_wait(file, &crng_init_wait, wait);
1416	return crng_ready() ? EPOLLIN | EPOLLRDNORM : EPOLLOUT | EPOLLWRNORM;
1417}
1418
1419static ssize_t write_pool_user(struct iov_iter *iter)
1420{
1421	u8 block[BLAKE2S_BLOCK_SIZE];
1422	ssize_t ret = 0;
1423	size_t copied;
1424
1425	if (unlikely(!iov_iter_count(iter)))
1426		return 0;
1427
1428	for (;;) {
1429		copied = copy_from_iter(block, sizeof(block), iter);
1430		ret += copied;
1431		mix_pool_bytes(block, copied);
1432		if (!iov_iter_count(iter) || copied != sizeof(block))
1433			break;
1434
1435		BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0);
1436		if (ret % PAGE_SIZE == 0) {
1437			if (signal_pending(current))
1438				break;
1439			cond_resched();
1440		}
1441	}
1442
1443	memzero_explicit(block, sizeof(block));
1444	return ret ? ret : -EFAULT;
1445}
1446
1447static ssize_t random_write_iter(struct kiocb *kiocb, struct iov_iter *iter)
1448{
1449	return write_pool_user(iter);
1450}
1451
1452static ssize_t urandom_read_iter(struct kiocb *kiocb, struct iov_iter *iter)
1453{
1454	static int maxwarn = 10;
1455
1456	/*
1457	 * Opportunistically attempt to initialize the RNG on platforms that
1458	 * have fast cycle counters, but don't (for now) require it to succeed.
1459	 */
1460	if (!crng_ready())
1461		try_to_generate_entropy();
1462
1463	if (!crng_ready()) {
1464		if (!ratelimit_disable && maxwarn <= 0)
1465			ratelimit_state_inc_miss(&urandom_warning);
1466		else if (ratelimit_disable || __ratelimit(&urandom_warning)) {
1467			--maxwarn;
1468			pr_notice("%s: uninitialized urandom read (%zu bytes read)\n",
1469				  current->comm, iov_iter_count(iter));
1470		}
1471	}
1472
1473	return get_random_bytes_user(iter);
1474}
1475
1476static ssize_t random_read_iter(struct kiocb *kiocb, struct iov_iter *iter)
1477{
1478	int ret;
1479
1480	if (!crng_ready() &&
1481	    ((kiocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO)) ||
1482	     (kiocb->ki_filp->f_flags & O_NONBLOCK)))
1483		return -EAGAIN;
1484
1485	ret = wait_for_random_bytes();
1486	if (ret != 0)
1487		return ret;
1488	return get_random_bytes_user(iter);
1489}
1490
1491static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1492{
1493	int __user *p = (int __user *)arg;
1494	int ent_count;
1495
1496	switch (cmd) {
1497	case RNDGETENTCNT:
1498		/* Inherently racy, no point locking. */
1499		if (put_user(input_pool.init_bits, p))
1500			return -EFAULT;
1501		return 0;
1502	case RNDADDTOENTCNT:
1503		if (!capable(CAP_SYS_ADMIN))
1504			return -EPERM;
1505		if (get_user(ent_count, p))
1506			return -EFAULT;
1507		if (ent_count < 0)
1508			return -EINVAL;
1509		credit_init_bits(ent_count);
1510		return 0;
1511	case RNDADDENTROPY: {
1512		struct iov_iter iter;
1513		ssize_t ret;
1514		int len;
1515
1516		if (!capable(CAP_SYS_ADMIN))
1517			return -EPERM;
1518		if (get_user(ent_count, p++))
1519			return -EFAULT;
1520		if (ent_count < 0)
1521			return -EINVAL;
1522		if (get_user(len, p++))
1523			return -EFAULT;
1524		ret = import_ubuf(ITER_SOURCE, p, len, &iter);
1525		if (unlikely(ret))
1526			return ret;
1527		ret = write_pool_user(&iter);
1528		if (unlikely(ret < 0))
1529			return ret;
1530		/* Since we're crediting, enforce that it was all written into the pool. */
1531		if (unlikely(ret != len))
1532			return -EFAULT;
1533		credit_init_bits(ent_count);
1534		return 0;
1535	}
1536	case RNDZAPENTCNT:
1537	case RNDCLEARPOOL:
1538		/* No longer has any effect. */
1539		if (!capable(CAP_SYS_ADMIN))
1540			return -EPERM;
1541		return 0;
1542	case RNDRESEEDCRNG:
1543		if (!capable(CAP_SYS_ADMIN))
1544			return -EPERM;
1545		if (!crng_ready())
1546			return -ENODATA;
1547		crng_reseed(NULL);
1548		return 0;
1549	default:
1550		return -EINVAL;
1551	}
1552}
1553
1554static int random_fasync(int fd, struct file *filp, int on)
1555{
1556	return fasync_helper(fd, filp, on, &fasync);
1557}
1558
1559const struct file_operations random_fops = {
1560	.read_iter = random_read_iter,
1561	.write_iter = random_write_iter,
1562	.poll = random_poll,
1563	.unlocked_ioctl = random_ioctl,
1564	.compat_ioctl = compat_ptr_ioctl,
1565	.fasync = random_fasync,
1566	.llseek = noop_llseek,
1567	.splice_read = copy_splice_read,
1568	.splice_write = iter_file_splice_write,
1569};
1570
1571const struct file_operations urandom_fops = {
1572	.read_iter = urandom_read_iter,
1573	.write_iter = random_write_iter,
1574	.unlocked_ioctl = random_ioctl,
1575	.compat_ioctl = compat_ptr_ioctl,
1576	.fasync = random_fasync,
1577	.llseek = noop_llseek,
1578	.splice_read = copy_splice_read,
1579	.splice_write = iter_file_splice_write,
1580};
1581
1582
1583/********************************************************************
1584 *
1585 * Sysctl interface.
1586 *
1587 * These are partly unused legacy knobs with dummy values to not break
1588 * userspace and partly still useful things. They are usually accessible
1589 * in /proc/sys/kernel/random/ and are as follows:
1590 *
1591 * - boot_id - a UUID representing the current boot.
1592 *
1593 * - uuid - a random UUID, different each time the file is read.
1594 *
1595 * - poolsize - the number of bits of entropy that the input pool can
1596 *   hold, tied to the POOL_BITS constant.
1597 *
1598 * - entropy_avail - the number of bits of entropy currently in the
1599 *   input pool. Always <= poolsize.
1600 *
1601 * - write_wakeup_threshold - the amount of entropy in the input pool
1602 *   below which write polls to /dev/random will unblock, requesting
1603 *   more entropy, tied to the POOL_READY_BITS constant. It is writable
1604 *   to avoid breaking old userspaces, but writing to it does not
1605 *   change any behavior of the RNG.
1606 *
1607 * - urandom_min_reseed_secs - fixed to the value CRNG_RESEED_INTERVAL.
1608 *   It is writable to avoid breaking old userspaces, but writing
1609 *   to it does not change any behavior of the RNG.
1610 *
1611 ********************************************************************/
1612
1613#ifdef CONFIG_SYSCTL
1614
1615#include <linux/sysctl.h>
1616
1617static int sysctl_random_min_urandom_seed = CRNG_RESEED_INTERVAL / HZ;
1618static int sysctl_random_write_wakeup_bits = POOL_READY_BITS;
1619static int sysctl_poolsize = POOL_BITS;
1620static u8 sysctl_bootid[UUID_SIZE];
1621
1622/*
1623 * This function is used to return both the bootid UUID, and random
1624 * UUID. The difference is in whether table->data is NULL; if it is,
1625 * then a new UUID is generated and returned to the user.
1626 */
1627static int proc_do_uuid(const struct ctl_table *table, int write, void *buf,
1628			size_t *lenp, loff_t *ppos)
1629{
1630	u8 tmp_uuid[UUID_SIZE], *uuid;
1631	char uuid_string[UUID_STRING_LEN + 1];
1632	struct ctl_table fake_table = {
1633		.data = uuid_string,
1634		.maxlen = UUID_STRING_LEN
1635	};
1636
1637	if (write)
1638		return -EPERM;
1639
1640	uuid = table->data;
1641	if (!uuid) {
1642		uuid = tmp_uuid;
1643		generate_random_uuid(uuid);
1644	} else {
1645		static DEFINE_SPINLOCK(bootid_spinlock);
1646
1647		spin_lock(&bootid_spinlock);
1648		if (!uuid[8])
1649			generate_random_uuid(uuid);
1650		spin_unlock(&bootid_spinlock);
1651	}
1652
1653	snprintf(uuid_string, sizeof(uuid_string), "%pU", uuid);
1654	return proc_dostring(&fake_table, 0, buf, lenp, ppos);
1655}
1656
1657/* The same as proc_dointvec, but writes don't change anything. */
1658static int proc_do_rointvec(const struct ctl_table *table, int write, void *buf,
1659			    size_t *lenp, loff_t *ppos)
1660{
1661	return write ? 0 : proc_dointvec(table, 0, buf, lenp, ppos);
1662}
1663
1664static const struct ctl_table random_table[] = {
1665	{
1666		.procname	= "poolsize",
1667		.data		= &sysctl_poolsize,
1668		.maxlen		= sizeof(int),
1669		.mode		= 0444,
1670		.proc_handler	= proc_dointvec,
1671	},
1672	{
1673		.procname	= "entropy_avail",
1674		.data		= &input_pool.init_bits,
1675		.maxlen		= sizeof(int),
1676		.mode		= 0444,
1677		.proc_handler	= proc_dointvec,
1678	},
1679	{
1680		.procname	= "write_wakeup_threshold",
1681		.data		= &sysctl_random_write_wakeup_bits,
1682		.maxlen		= sizeof(int),
1683		.mode		= 0644,
1684		.proc_handler	= proc_do_rointvec,
1685	},
1686	{
1687		.procname	= "urandom_min_reseed_secs",
1688		.data		= &sysctl_random_min_urandom_seed,
1689		.maxlen		= sizeof(int),
1690		.mode		= 0644,
1691		.proc_handler	= proc_do_rointvec,
1692	},
1693	{
1694		.procname	= "boot_id",
1695		.data		= &sysctl_bootid,
1696		.mode		= 0444,
1697		.proc_handler	= proc_do_uuid,
1698	},
1699	{
1700		.procname	= "uuid",
1701		.mode		= 0444,
1702		.proc_handler	= proc_do_uuid,
1703	},
1704};
1705
1706/*
1707 * random_init() is called before sysctl_init(),
1708 * so we cannot call register_sysctl_init() in random_init()
1709 */
1710static int __init random_sysctls_init(void)
1711{
1712	register_sysctl_init("kernel/random", random_table);
1713	return 0;
1714}
1715device_initcall(random_sysctls_init);
1716#endif