Linux Audio

Check our new training course

Loading...
v6.8
   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
v4.10.11
 
   1/*
   2 * random.c -- A strong random number generator
   3 *
   4 * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005
 
   5 *
   6 * Copyright Theodore Ts'o, 1994, 1995, 1996, 1997, 1998, 1999.  All
   7 * rights reserved.
   8 *
   9 * Redistribution and use in source and binary forms, with or without
  10 * modification, are permitted provided that the following conditions
  11 * are met:
  12 * 1. Redistributions of source code must retain the above copyright
  13 *    notice, and the entire permission notice in its entirety,
  14 *    including the disclaimer of warranties.
  15 * 2. Redistributions in binary form must reproduce the above copyright
  16 *    notice, this list of conditions and the following disclaimer in the
  17 *    documentation and/or other materials provided with the distribution.
  18 * 3. The name of the author may not be used to endorse or promote
  19 *    products derived from this software without specific prior
  20 *    written permission.
  21 *
  22 * ALTERNATIVELY, this product may be distributed under the terms of
  23 * the GNU General Public License, in which case the provisions of the GPL are
  24 * required INSTEAD OF the above restrictions.  (This clause is
  25 * necessary due to a potential bad interaction between the GPL and
  26 * the restrictions contained in a BSD-style copyright.)
  27 *
  28 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  29 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  30 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ALL OF
  31 * WHICH ARE HEREBY DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE
  32 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  33 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT
  34 * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  35 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  36 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  37 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
  38 * USE OF THIS SOFTWARE, EVEN IF NOT ADVISED OF THE POSSIBILITY OF SUCH
  39 * DAMAGE.
  40 */
  41
  42/*
  43 * (now, with legal B.S. out of the way.....)
  44 *
  45 * This routine gathers environmental noise from device drivers, etc.,
  46 * and returns good random numbers, suitable for cryptographic use.
  47 * Besides the obvious cryptographic uses, these numbers are also good
  48 * for seeding TCP sequence numbers, and other places where it is
  49 * desirable to have numbers which are not only random, but hard to
  50 * predict by an attacker.
  51 *
  52 * Theory of operation
  53 * ===================
  54 *
  55 * Computers are very predictable devices.  Hence it is extremely hard
  56 * to produce truly random numbers on a computer --- as opposed to
  57 * pseudo-random numbers, which can easily generated by using a
  58 * algorithm.  Unfortunately, it is very easy for attackers to guess
  59 * the sequence of pseudo-random number generators, and for some
  60 * applications this is not acceptable.  So instead, we must try to
  61 * gather "environmental noise" from the computer's environment, which
  62 * must be hard for outside attackers to observe, and use that to
  63 * generate random numbers.  In a Unix environment, this is best done
  64 * from inside the kernel.
  65 *
  66 * Sources of randomness from the environment include inter-keyboard
  67 * timings, inter-interrupt timings from some interrupts, and other
  68 * events which are both (a) non-deterministic and (b) hard for an
  69 * outside observer to measure.  Randomness from these sources are
  70 * added to an "entropy pool", which is mixed using a CRC-like function.
  71 * This is not cryptographically strong, but it is adequate assuming
  72 * the randomness is not chosen maliciously, and it is fast enough that
  73 * the overhead of doing it on every interrupt is very reasonable.
  74 * As random bytes are mixed into the entropy pool, the routines keep
  75 * an *estimate* of how many bits of randomness have been stored into
  76 * the random number generator's internal state.
  77 *
  78 * When random bytes are desired, they are obtained by taking the SHA
  79 * hash of the contents of the "entropy pool".  The SHA hash avoids
  80 * exposing the internal state of the entropy pool.  It is believed to
  81 * be computationally infeasible to derive any useful information
  82 * about the input of SHA from its output.  Even if it is possible to
  83 * analyze SHA in some clever way, as long as the amount of data
  84 * returned from the generator is less than the inherent entropy in
  85 * the pool, the output data is totally unpredictable.  For this
  86 * reason, the routine decreases its internal estimate of how many
  87 * bits of "true randomness" are contained in the entropy pool as it
  88 * outputs random numbers.
  89 *
  90 * If this estimate goes to zero, the routine can still generate
  91 * random numbers; however, an attacker may (at least in theory) be
  92 * able to infer the future output of the generator from prior
  93 * outputs.  This requires successful cryptanalysis of SHA, which is
  94 * not believed to be feasible, but there is a remote possibility.
  95 * Nonetheless, these numbers should be useful for the vast majority
  96 * of purposes.
  97 *
  98 * Exported interfaces ---- output
  99 * ===============================
 100 *
 101 * There are three exported interfaces; the first is one designed to
 102 * be used from within the kernel:
 103 *
 104 * 	void get_random_bytes(void *buf, int nbytes);
 105 *
 106 * This interface will return the requested number of random bytes,
 107 * and place it in the requested buffer.
 108 *
 109 * The two other interfaces are two character devices /dev/random and
 110 * /dev/urandom.  /dev/random is suitable for use when very high
 111 * quality randomness is desired (for example, for key generation or
 112 * one-time pads), as it will only return a maximum of the number of
 113 * bits of randomness (as estimated by the random number generator)
 114 * contained in the entropy pool.
 115 *
 116 * The /dev/urandom device does not have this limit, and will return
 117 * as many bytes as are requested.  As more and more random bytes are
 118 * requested without giving time for the entropy pool to recharge,
 119 * this will result in random numbers that are merely cryptographically
 120 * strong.  For many applications, however, this is acceptable.
 121 *
 122 * Exported interfaces ---- input
 123 * ==============================
 124 *
 125 * The current exported interfaces for gathering environmental noise
 126 * from the devices are:
 127 *
 128 *	void add_device_randomness(const void *buf, unsigned int size);
 129 * 	void add_input_randomness(unsigned int type, unsigned int code,
 130 *                                unsigned int value);
 131 *	void add_interrupt_randomness(int irq, int irq_flags);
 132 * 	void add_disk_randomness(struct gendisk *disk);
 133 *
 134 * add_device_randomness() is for adding data to the random pool that
 135 * is likely to differ between two devices (or possibly even per boot).
 136 * This would be things like MAC addresses or serial numbers, or the
 137 * read-out of the RTC. This does *not* add any actual entropy to the
 138 * pool, but it initializes the pool to different values for devices
 139 * that might otherwise be identical and have very little entropy
 140 * available to them (particularly common in the embedded world).
 141 *
 142 * add_input_randomness() uses the input layer interrupt timing, as well as
 143 * the event type information from the hardware.
 144 *
 145 * add_interrupt_randomness() uses the interrupt timing as random
 146 * inputs to the entropy pool. Using the cycle counters and the irq source
 147 * as inputs, it feeds the randomness roughly once a second.
 148 *
 149 * add_disk_randomness() uses what amounts to the seek time of block
 150 * layer request events, on a per-disk_devt basis, as input to the
 151 * entropy pool. Note that high-speed solid state drives with very low
 152 * seek times do not make for good sources of entropy, as their seek
 153 * times are usually fairly consistent.
 154 *
 155 * All of these routines try to estimate how many bits of randomness a
 156 * particular randomness source.  They do this by keeping track of the
 157 * first and second order deltas of the event timings.
 158 *
 159 * Ensuring unpredictability at system startup
 160 * ============================================
 161 *
 162 * When any operating system starts up, it will go through a sequence
 163 * of actions that are fairly predictable by an adversary, especially
 164 * if the start-up does not involve interaction with a human operator.
 165 * This reduces the actual number of bits of unpredictability in the
 166 * entropy pool below the value in entropy_count.  In order to
 167 * counteract this effect, it helps to carry information in the
 168 * entropy pool across shut-downs and start-ups.  To do this, put the
 169 * following lines an appropriate script which is run during the boot
 170 * sequence:
 171 *
 172 *	echo "Initializing random number generator..."
 173 *	random_seed=/var/run/random-seed
 174 *	# Carry a random seed from start-up to start-up
 175 *	# Load and then save the whole entropy pool
 176 *	if [ -f $random_seed ]; then
 177 *		cat $random_seed >/dev/urandom
 178 *	else
 179 *		touch $random_seed
 180 *	fi
 181 *	chmod 600 $random_seed
 182 *	dd if=/dev/urandom of=$random_seed count=1 bs=512
 183 *
 184 * and the following lines in an appropriate script which is run as
 185 * the system is shutdown:
 186 *
 187 *	# Carry a random seed from shut-down to start-up
 188 *	# Save the whole entropy pool
 189 *	echo "Saving random seed..."
 190 *	random_seed=/var/run/random-seed
 191 *	touch $random_seed
 192 *	chmod 600 $random_seed
 193 *	dd if=/dev/urandom of=$random_seed count=1 bs=512
 194 *
 195 * For example, on most modern systems using the System V init
 196 * scripts, such code fragments would be found in
 197 * /etc/rc.d/init.d/random.  On older Linux systems, the correct script
 198 * location might be in /etc/rcb.d/rc.local or /etc/rc.d/rc.0.
 199 *
 200 * Effectively, these commands cause the contents of the entropy pool
 201 * to be saved at shut-down time and reloaded into the entropy pool at
 202 * start-up.  (The 'dd' in the addition to the bootup script is to
 203 * make sure that /etc/random-seed is different for every start-up,
 204 * even if the system crashes without executing rc.0.)  Even with
 205 * complete knowledge of the start-up activities, predicting the state
 206 * of the entropy pool requires knowledge of the previous history of
 207 * the system.
 208 *
 209 * Configuring the /dev/random driver under Linux
 210 * ==============================================
 211 *
 212 * The /dev/random driver under Linux uses minor numbers 8 and 9 of
 213 * the /dev/mem major number (#1).  So if your system does not have
 214 * /dev/random and /dev/urandom created already, they can be created
 215 * by using the commands:
 216 *
 217 * 	mknod /dev/random c 1 8
 218 * 	mknod /dev/urandom c 1 9
 219 *
 220 * Acknowledgements:
 221 * =================
 222 *
 223 * Ideas for constructing this random number generator were derived
 224 * from Pretty Good Privacy's random number generator, and from private
 225 * discussions with Phil Karn.  Colin Plumb provided a faster random
 226 * number generator, which speed up the mixing function of the entropy
 227 * pool, taken from PGPfone.  Dale Worley has also contributed many
 228 * useful ideas and suggestions to improve this driver.
 229 *
 230 * Any flaws in the design are solely my responsibility, and should
 231 * not be attributed to the Phil, Colin, or any of authors of PGP.
 232 *
 233 * Further background information on this topic may be obtained from
 234 * RFC 1750, "Randomness Recommendations for Security", by Donald
 235 * Eastlake, Steve Crocker, and Jeff Schiller.
 236 */
 237
 
 
 238#include <linux/utsname.h>
 239#include <linux/module.h>
 240#include <linux/kernel.h>
 241#include <linux/major.h>
 242#include <linux/string.h>
 243#include <linux/fcntl.h>
 244#include <linux/slab.h>
 245#include <linux/random.h>
 246#include <linux/poll.h>
 247#include <linux/init.h>
 248#include <linux/fs.h>
 249#include <linux/genhd.h>
 250#include <linux/interrupt.h>
 251#include <linux/mm.h>
 252#include <linux/nodemask.h>
 253#include <linux/spinlock.h>
 254#include <linux/kthread.h>
 255#include <linux/percpu.h>
 256#include <linux/cryptohash.h>
 257#include <linux/fips.h>
 258#include <linux/ptrace.h>
 259#include <linux/kmemcheck.h>
 260#include <linux/workqueue.h>
 261#include <linux/irq.h>
 
 262#include <linux/syscalls.h>
 263#include <linux/completion.h>
 264#include <linux/uuid.h>
 265#include <crypto/chacha20.h>
 266
 
 
 
 
 
 267#include <asm/processor.h>
 268#include <linux/uaccess.h>
 269#include <asm/irq.h>
 270#include <asm/irq_regs.h>
 271#include <asm/io.h>
 272
 273#define CREATE_TRACE_POINTS
 274#include <trace/events/random.h>
 275
 276/* #define ADD_INTERRUPT_BENCH */
 
 
 
 
 
 277
 278/*
 279 * Configuration information
 
 280 */
 281#define INPUT_POOL_SHIFT	12
 282#define INPUT_POOL_WORDS	(1 << (INPUT_POOL_SHIFT-5))
 283#define OUTPUT_POOL_SHIFT	10
 284#define OUTPUT_POOL_WORDS	(1 << (OUTPUT_POOL_SHIFT-5))
 285#define SEC_XFER_SIZE		512
 286#define EXTRACT_SIZE		10
 
 
 
 
 
 287
 288#define DEBUG_RANDOM_BOOT 0
 289
 290#define LONGS(x) (((x) + sizeof(unsigned long) - 1)/sizeof(unsigned long))
 
 
 
 
 291
 292/*
 293 * To allow fractional bits to be tracked, the entropy_count field is
 294 * denominated in units of 1/8th bits.
 
 
 295 *
 296 * 2*(ENTROPY_SHIFT + log2(poolbits)) must <= 31, or the multiply in
 297 * credit_entropy_bits() needs to be 64 bits wide.
 298 */
 299#define ENTROPY_SHIFT 3
 300#define ENTROPY_BITS(r) ((r)->entropy_count >> ENTROPY_SHIFT)
 
 
 
 301
 302/*
 303 * The minimum number of bits of entropy before we wake up a read on
 304 * /dev/random.  Should be enough to do a significant reseed.
 305 */
 306static int random_read_wakeup_bits = 64;
 307
 308/*
 309 * If the entropy count falls under this number of bits, then we
 310 * should wake up processes which are selecting or polling on write
 311 * access to /dev/random.
 312 */
 313static int random_write_wakeup_bits = 28 * OUTPUT_POOL_WORDS;
 314
 315/*
 316 * The minimum number of seconds between urandom pool reseeding.  We
 317 * do this to limit the amount of entropy that can be drained from the
 318 * input pool even if there are heavy demands on /dev/urandom.
 319 */
 320static int random_min_urandom_seed = 60;
 321
 322/*
 323 * Originally, we used a primitive polynomial of degree .poolwords
 324 * over GF(2).  The taps for various sizes are defined below.  They
 325 * were chosen to be evenly spaced except for the last tap, which is 1
 326 * to get the twisting happening as fast as possible.
 
 327 *
 328 * For the purposes of better mixing, we use the CRC-32 polynomial as
 329 * well to make a (modified) twisted Generalized Feedback Shift
 330 * Register.  (See M. Matsumoto & Y. Kurita, 1992.  Twisted GFSR
 331 * generators.  ACM Transactions on Modeling and Computer Simulation
 332 * 2(3):179-194.  Also see M. Matsumoto & Y. Kurita, 1994.  Twisted
 333 * GFSR generators II.  ACM Transactions on Modeling and Computer
 334 * Simulation 4:254-266)
 335 *
 336 * Thanks to Colin Plumb for suggesting this.
 337 *
 338 * The mixing operation is much less sensitive than the output hash,
 339 * where we use SHA-1.  All that we want of mixing operation is that
 340 * it be a good non-cryptographic hash; i.e. it not produce collisions
 341 * when fed "random" data of the sort we expect to see.  As long as
 342 * the pool state differs for different inputs, we have preserved the
 343 * input entropy and done a good job.  The fact that an intelligent
 344 * attacker can construct inputs that will produce controlled
 345 * alterations to the pool's state is not important because we don't
 346 * consider such inputs to contribute any randomness.  The only
 347 * property we need with respect to them is that the attacker can't
 348 * increase his/her knowledge of the pool's state.  Since all
 349 * additions are reversible (knowing the final state and the input,
 350 * you can reconstruct the initial state), if an attacker has any
 351 * uncertainty about the initial state, he/she can only shuffle that
 352 * uncertainty about, but never cause any collisions (which would
 353 * decrease the uncertainty).
 354 *
 355 * Our mixing functions were analyzed by Lacharme, Roeck, Strubel, and
 356 * Videau in their paper, "The Linux Pseudorandom Number Generator
 357 * Revisited" (see: http://eprint.iacr.org/2012/251.pdf).  In their
 358 * paper, they point out that we are not using a true Twisted GFSR,
 359 * since Matsumoto & Kurita used a trinomial feedback polynomial (that
 360 * is, with only three taps, instead of the six that we are using).
 361 * As a result, the resulting polynomial is neither primitive nor
 362 * irreducible, and hence does not have a maximal period over
 363 * GF(2**32).  They suggest a slight change to the generator
 364 * polynomial which improves the resulting TGFSR polynomial to be
 365 * irreducible, which we have made here.
 366 */
 367static struct poolinfo {
 368	int poolbitshift, poolwords, poolbytes, poolbits, poolfracbits;
 369#define S(x) ilog2(x)+5, (x), (x)*4, (x)*32, (x) << (ENTROPY_SHIFT+5)
 370	int tap1, tap2, tap3, tap4, tap5;
 371} poolinfo_table[] = {
 372	/* was: x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 */
 373	/* x^128 + x^104 + x^76 + x^51 +x^25 + x + 1 */
 374	{ S(128),	104,	76,	51,	25,	1 },
 375	/* was: x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 */
 376	/* x^32 + x^26 + x^19 + x^14 + x^7 + x + 1 */
 377	{ S(32),	26,	19,	14,	7,	1 },
 378#if 0
 379	/* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1  -- 115 */
 380	{ S(2048),	1638,	1231,	819,	411,	1 },
 381
 382	/* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
 383	{ S(1024),	817,	615,	412,	204,	1 },
 384
 385	/* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
 386	{ S(1024),	819,	616,	410,	207,	2 },
 387
 388	/* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
 389	{ S(512),	411,	308,	208,	104,	1 },
 390
 391	/* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
 392	{ S(512),	409,	307,	206,	102,	2 },
 393	/* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
 394	{ S(512),	409,	309,	205,	103,	2 },
 395
 396	/* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
 397	{ S(256),	205,	155,	101,	52,	1 },
 398
 399	/* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
 400	{ S(128),	103,	78,	51,	27,	2 },
 401
 402	/* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
 403	{ S(64),	52,	39,	26,	14,	1 },
 404#endif
 405};
 406
 407/*
 408 * Static global variables
 
 
 
 409 */
 410static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
 411static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
 412static DECLARE_WAIT_QUEUE_HEAD(urandom_init_wait);
 413static struct fasync_struct *fasync;
 414
 415static DEFINE_SPINLOCK(random_ready_list_lock);
 416static LIST_HEAD(random_ready_list);
 
 
 
 
 
 
 417
 418struct crng_state {
 419	__u32		state[16];
 420	unsigned long	init_time;
 421	spinlock_t	lock;
 422};
 423
 424struct crng_state primary_crng = {
 425	.lock = __SPIN_LOCK_UNLOCKED(primary_crng.lock),
 426};
 427
 428/*
 429 * crng_init =  0 --> Uninitialized
 430 *		1 --> Initialized
 431 *		2 --> Initialized from input_pool
 432 *
 433 * crng_init is protected by primary_crng->lock, and only increases
 434 * its value (from 0->1->2).
 435 */
 436static int crng_init = 0;
 437#define crng_ready() (likely(crng_init > 0))
 438static int crng_init_cnt = 0;
 439#define CRNG_INIT_CNT_THRESH (2*CHACHA20_KEY_SIZE)
 440static void _extract_crng(struct crng_state *crng,
 441			  __u8 out[CHACHA20_BLOCK_SIZE]);
 442static void _crng_backtrack_protect(struct crng_state *crng,
 443				    __u8 tmp[CHACHA20_BLOCK_SIZE], int used);
 444static void process_random_ready_list(void);
 445
 446/**********************************************************************
 447 *
 448 * OS independent entropy store.   Here are the functions which handle
 449 * storing entropy in an entropy pool.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 450 *
 451 **********************************************************************/
 452
 453struct entropy_store;
 454struct entropy_store {
 455	/* read-only data: */
 456	const struct poolinfo *poolinfo;
 457	__u32 *pool;
 458	const char *name;
 459	struct entropy_store *pull;
 460	struct work_struct push_work;
 461
 462	/* read-write data: */
 463	unsigned long last_pulled;
 
 464	spinlock_t lock;
 465	unsigned short add_ptr;
 466	unsigned short input_rotate;
 467	int entropy_count;
 468	int entropy_total;
 469	unsigned int initialized:1;
 470	unsigned int limit:1;
 471	unsigned int last_data_init:1;
 472	__u8 last_data[EXTRACT_SIZE];
 473};
 474
 475static ssize_t extract_entropy(struct entropy_store *r, void *buf,
 476			       size_t nbytes, int min, int rsvd);
 477static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
 478				size_t nbytes, int fips);
 479
 480static void crng_reseed(struct crng_state *crng, struct entropy_store *r);
 481static void push_to_pool(struct work_struct *work);
 482static __u32 input_pool_data[INPUT_POOL_WORDS] __latent_entropy;
 483static __u32 blocking_pool_data[OUTPUT_POOL_WORDS] __latent_entropy;
 484
 485static struct entropy_store input_pool = {
 486	.poolinfo = &poolinfo_table[0],
 487	.name = "input",
 488	.limit = 1,
 489	.lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
 490	.pool = input_pool_data
 491};
 492
 493static struct entropy_store blocking_pool = {
 494	.poolinfo = &poolinfo_table[1],
 495	.name = "blocking",
 496	.limit = 1,
 497	.pull = &input_pool,
 498	.lock = __SPIN_LOCK_UNLOCKED(blocking_pool.lock),
 499	.pool = blocking_pool_data,
 500	.push_work = __WORK_INITIALIZER(blocking_pool.push_work,
 501					push_to_pool),
 502};
 503
 504static __u32 const twist_table[8] = {
 505	0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
 506	0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
 507
 508/*
 509 * This function adds bytes into the entropy "pool".  It does not
 510 * update the entropy estimate.  The caller should call
 511 * credit_entropy_bits if this is appropriate.
 512 *
 513 * The pool is stirred with a primitive polynomial of the appropriate
 514 * degree, and then twisted.  We twist by three bits at a time because
 515 * it's cheap to do so and helps slightly in the expected case where
 516 * the entropy is concentrated in the low-order bits.
 517 */
 518static void _mix_pool_bytes(struct entropy_store *r, const void *in,
 519			    int nbytes)
 520{
 521	unsigned long i, tap1, tap2, tap3, tap4, tap5;
 522	int input_rotate;
 523	int wordmask = r->poolinfo->poolwords - 1;
 524	const char *bytes = in;
 525	__u32 w;
 526
 527	tap1 = r->poolinfo->tap1;
 528	tap2 = r->poolinfo->tap2;
 529	tap3 = r->poolinfo->tap3;
 530	tap4 = r->poolinfo->tap4;
 531	tap5 = r->poolinfo->tap5;
 532
 533	input_rotate = r->input_rotate;
 534	i = r->add_ptr;
 535
 536	/* mix one byte at a time to simplify size handling and churn faster */
 537	while (nbytes--) {
 538		w = rol32(*bytes++, input_rotate);
 539		i = (i - 1) & wordmask;
 540
 541		/* XOR in the various taps */
 542		w ^= r->pool[i];
 543		w ^= r->pool[(i + tap1) & wordmask];
 544		w ^= r->pool[(i + tap2) & wordmask];
 545		w ^= r->pool[(i + tap3) & wordmask];
 546		w ^= r->pool[(i + tap4) & wordmask];
 547		w ^= r->pool[(i + tap5) & wordmask];
 548
 549		/* Mix the result back in with a twist */
 550		r->pool[i] = (w >> 3) ^ twist_table[w & 7];
 551
 552		/*
 553		 * Normally, we add 7 bits of rotation to the pool.
 554		 * At the beginning of the pool, add an extra 7 bits
 555		 * rotation, so that successive passes spread the
 556		 * input bits across the pool evenly.
 557		 */
 558		input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
 559	}
 
 
 560
 561	r->input_rotate = input_rotate;
 562	r->add_ptr = i;
 563}
 564
 565static void __mix_pool_bytes(struct entropy_store *r, const void *in,
 566			     int nbytes)
 567{
 568	trace_mix_pool_bytes_nolock(r->name, nbytes, _RET_IP_);
 569	_mix_pool_bytes(r, in, nbytes);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 570}
 571
 572static void mix_pool_bytes(struct entropy_store *r, const void *in,
 573			   int nbytes)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 574{
 575	unsigned long flags;
 
 
 
 
 
 
 
 576
 577	trace_mix_pool_bytes(r->name, nbytes, _RET_IP_);
 578	spin_lock_irqsave(&r->lock, flags);
 579	_mix_pool_bytes(r, in, nbytes);
 580	spin_unlock_irqrestore(&r->lock, flags);
 581}
 582
 583struct fast_pool {
 584	__u32		pool[4];
 585	unsigned long	last;
 586	unsigned short	reg_idx;
 587	unsigned char	count;
 588};
 589
 590/*
 591 * This is a fast mixing routine used by the interrupt randomness
 592 * collector.  It's hardcoded for an 128 bit pool and assumes that any
 593 * locks that might be needed are taken by the caller.
 594 */
 595static void fast_mix(struct fast_pool *f)
 
 596{
 597	__u32 a = f->pool[0],	b = f->pool[1];
 598	__u32 c = f->pool[2],	d = f->pool[3];
 599
 600	a += b;			c += d;
 601	b = rol32(b, 6);	d = rol32(d, 27);
 602	d ^= a;			b ^= c;
 603
 604	a += b;			c += d;
 605	b = rol32(b, 16);	d = rol32(d, 14);
 606	d ^= a;			b ^= c;
 
 
 
 
 
 607
 608	a += b;			c += d;
 609	b = rol32(b, 6);	d = rol32(d, 27);
 610	d ^= a;			b ^= c;
 
 
 
 
 
 
 
 
 
 
 
 
 611
 612	a += b;			c += d;
 613	b = rol32(b, 16);	d = rol32(d, 14);
 614	d ^= a;			b ^= c;
 
 
 
 
 
 
 
 
 
 
 615
 616	f->pool[0] = a;  f->pool[1] = b;
 617	f->pool[2] = c;  f->pool[3] = d;
 618	f->count++;
 
 
 
 
 
 
 619}
 620
 621static void process_random_ready_list(void)
 622{
 623	unsigned long flags;
 624	struct random_ready_callback *rdy, *tmp;
 
 
 
 
 625
 626	spin_lock_irqsave(&random_ready_list_lock, flags);
 627	list_for_each_entry_safe(rdy, tmp, &random_ready_list, list) {
 628		struct module *owner = rdy->owner;
 
 
 
 
 
 
 
 
 
 629
 630		list_del_init(&rdy->list);
 631		rdy->func(rdy);
 632		module_put(owner);
 
 
 633	}
 634	spin_unlock_irqrestore(&random_ready_list_lock, flags);
 
 635}
 636
 637/*
 638 * Credit (or debit) the entropy store with n bits of entropy.
 639 * Use credit_entropy_bits_safe() if the value comes from userspace
 640 * or otherwise should be checked for extreme values.
 
 
 641 */
 642static void credit_entropy_bits(struct entropy_store *r, int nbits)
 
 
 
 
 
 
 
 643{
 644	int entropy_count, orig;
 645	const int pool_size = r->poolinfo->poolfracbits;
 646	int nfrac = nbits << ENTROPY_SHIFT;
 647
 648	if (!nbits)
 649		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 650
 651retry:
 652	entropy_count = orig = ACCESS_ONCE(r->entropy_count);
 653	if (nfrac < 0) {
 654		/* Debit */
 655		entropy_count += nfrac;
 656	} else {
 657		/*
 658		 * Credit: we have to account for the possibility of
 659		 * overwriting already present entropy.	 Even in the
 660		 * ideal case of pure Shannon entropy, new contributions
 661		 * approach the full value asymptotically:
 662		 *
 663		 * entropy <- entropy + (pool_size - entropy) *
 664		 *	(1 - exp(-add_entropy/pool_size))
 665		 *
 666		 * For add_entropy <= pool_size/2 then
 667		 * (1 - exp(-add_entropy/pool_size)) >=
 668		 *    (add_entropy/pool_size)*0.7869...
 669		 * so we can approximate the exponential with
 670		 * 3/4*add_entropy/pool_size and still be on the
 671		 * safe side by adding at most pool_size/2 at a time.
 672		 *
 673		 * The use of pool_size-2 in the while statement is to
 674		 * prevent rounding artifacts from making the loop
 675		 * arbitrarily long; this limits the loop to log2(pool_size)*2
 676		 * turns no matter how large nbits is.
 677		 */
 678		int pnfrac = nfrac;
 679		const int s = r->poolinfo->poolbitshift + ENTROPY_SHIFT + 2;
 680		/* The +2 corresponds to the /4 in the denominator */
 681
 682		do {
 683			unsigned int anfrac = min(pnfrac, pool_size/2);
 684			unsigned int add =
 685				((pool_size - entropy_count)*anfrac*3) >> s;
 686
 687			entropy_count += add;
 688			pnfrac -= anfrac;
 689		} while (unlikely(entropy_count < pool_size-2 && pnfrac));
 690	}
 691
 692	if (unlikely(entropy_count < 0)) {
 693		pr_warn("random: negative entropy/overflow: pool %s count %d\n",
 694			r->name, entropy_count);
 695		WARN_ON(1);
 696		entropy_count = 0;
 697	} else if (entropy_count > pool_size)
 698		entropy_count = pool_size;
 699	if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
 700		goto retry;
 701
 702	r->entropy_total += nbits;
 703	if (!r->initialized && r->entropy_total > 128) {
 704		r->initialized = 1;
 705		r->entropy_total = 0;
 706	}
 707
 708	trace_credit_entropy_bits(r->name, nbits,
 709				  entropy_count >> ENTROPY_SHIFT,
 710				  r->entropy_total, _RET_IP_);
 711
 712	if (r == &input_pool) {
 713		int entropy_bits = entropy_count >> ENTROPY_SHIFT;
 714
 715		if (crng_init < 2 && entropy_bits >= 128) {
 716			crng_reseed(&primary_crng, r);
 717			entropy_bits = r->entropy_count >> ENTROPY_SHIFT;
 718		}
 719
 720		/* should we wake readers? */
 721		if (entropy_bits >= random_read_wakeup_bits) {
 722			wake_up_interruptible(&random_read_wait);
 723			kill_fasync(&fasync, SIGIO, POLL_IN);
 724		}
 725		/* If the input pool is getting full, send some
 726		 * entropy to the blocking pool until it is 75% full.
 727		 */
 728		if (entropy_bits > random_write_wakeup_bits &&
 729		    r->initialized &&
 730		    r->entropy_total >= 2*random_read_wakeup_bits) {
 731			struct entropy_store *other = &blocking_pool;
 732
 733			if (other->entropy_count <=
 734			    3 * other->poolinfo->poolfracbits / 4) {
 735				schedule_work(&other->push_work);
 736				r->entropy_total = 0;
 737			}
 738		}
 739	}
 740}
 741
 742static int credit_entropy_bits_safe(struct entropy_store *r, int nbits)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 743{
 744	const int nbits_max = (int)(~0U >> (ENTROPY_SHIFT + 1));
 
 
 
 
 
 
 
 
 
 
 745
 746	if (nbits < 0)
 747		return -EINVAL;
 
 
 
 
 
 
 
 
 748
 749	/* Cap the value to avoid overflows */
 750	nbits = min(nbits,  nbits_max);
 
 
 
 
 
 
 
 751
 752	credit_entropy_bits(r, nbits);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 753	return 0;
 754}
 
 
 755
 756/*********************************************************************
 
 
 
 
 
 
 
 
 
 
 
 
 757 *
 758 * CRNG using CHACHA20
 759 *
 760 *********************************************************************/
 
 
 
 
 
 
 761
 762#define CRNG_RESEED_INTERVAL (300*HZ)
 
 
 
 
 
 
 
 
 
 
 763
 764static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
 
 
 
 765
 766#ifdef CONFIG_NUMA
 767/*
 768 * Hack to deal with crazy userspace progams when they are all trying
 769 * to access /dev/urandom in parallel.  The programs are almost
 770 * certainly doing something terribly wrong, but we'll work around
 771 * their brain damage.
 772 */
 773static struct crng_state **crng_node_pool __read_mostly;
 774#endif
 775
 776static void crng_initialize(struct crng_state *crng)
 777{
 778	int		i;
 779	unsigned long	rv;
 780
 781	memcpy(&crng->state[0], "expand 32-byte k", 16);
 782	if (crng == &primary_crng)
 783		_extract_entropy(&input_pool, &crng->state[4],
 784				 sizeof(__u32) * 12, 0);
 785	else
 786		get_random_bytes(&crng->state[4], sizeof(__u32) * 12);
 787	for (i = 4; i < 16; i++) {
 788		if (!arch_get_random_seed_long(&rv) &&
 789		    !arch_get_random_long(&rv))
 790			rv = random_get_entropy();
 791		crng->state[i] ^= rv;
 792	}
 793	crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
 794}
 795
 796static int crng_fast_load(const char *cp, size_t len)
 
 
 
 
 797{
 798	unsigned long flags;
 799	char *p;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 800
 801	if (!spin_trylock_irqsave(&primary_crng.lock, flags))
 802		return 0;
 803	if (crng_ready()) {
 804		spin_unlock_irqrestore(&primary_crng.lock, flags);
 805		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 806	}
 807	p = (unsigned char *) &primary_crng.state[4];
 808	while (len > 0 && crng_init_cnt < CRNG_INIT_CNT_THRESH) {
 809		p[crng_init_cnt % CHACHA20_KEY_SIZE] ^= *cp;
 810		cp++; crng_init_cnt++; len--;
 811	}
 812	if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
 813		crng_init = 1;
 814		wake_up_interruptible(&crng_init_wait);
 815		pr_notice("random: fast init done\n");
 816	}
 817	spin_unlock_irqrestore(&primary_crng.lock, flags);
 818	return 1;
 819}
 820
 821static void crng_reseed(struct crng_state *crng, struct entropy_store *r)
 
 
 822{
 823	unsigned long	flags;
 824	int		i, num;
 825	union {
 826		__u8	block[CHACHA20_BLOCK_SIZE];
 827		__u32	key[8];
 828	} buf;
 829
 830	if (r) {
 831		num = extract_entropy(r, &buf, 32, 16, 0);
 832		if (num == 0)
 833			return;
 834	} else {
 835		_extract_crng(&primary_crng, buf.block);
 836		_crng_backtrack_protect(&primary_crng, buf.block,
 837					CHACHA20_KEY_SIZE);
 838	}
 839	spin_lock_irqsave(&primary_crng.lock, flags);
 840	for (i = 0; i < 8; i++) {
 841		unsigned long	rv;
 842		if (!arch_get_random_seed_long(&rv) &&
 843		    !arch_get_random_long(&rv))
 844			rv = random_get_entropy();
 845		crng->state[i+4] ^= buf.key[i] ^ rv;
 846	}
 847	memzero_explicit(&buf, sizeof(buf));
 848	crng->init_time = jiffies;
 849	if (crng == &primary_crng && crng_init < 2) {
 850		crng_init = 2;
 851		process_random_ready_list();
 852		wake_up_interruptible(&crng_init_wait);
 853		pr_notice("random: crng init done\n");
 
 
 
 
 
 
 
 
 
 
 
 
 854	}
 855	spin_unlock_irqrestore(&primary_crng.lock, flags);
 856}
 857
 858static inline void maybe_reseed_primary_crng(void)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 859{
 860	if (crng_init > 2 &&
 861	    time_after(jiffies, primary_crng.init_time + CRNG_RESEED_INTERVAL))
 862		crng_reseed(&primary_crng, &input_pool);
 863}
 864
 865static inline void crng_wait_ready(void)
 866{
 867	wait_event_interruptible(crng_init_wait, crng_ready());
 868}
 
 
 869
 870static void _extract_crng(struct crng_state *crng,
 871			  __u8 out[CHACHA20_BLOCK_SIZE])
 872{
 873	unsigned long v, flags;
 874
 875	if (crng_init > 1 &&
 876	    time_after(jiffies, crng->init_time + CRNG_RESEED_INTERVAL))
 877		crng_reseed(crng, crng == &primary_crng ? &input_pool : NULL);
 878	spin_lock_irqsave(&crng->lock, flags);
 879	if (arch_get_random_long(&v))
 880		crng->state[14] ^= v;
 881	chacha20_block(&crng->state[0], out);
 882	if (crng->state[12] == 0)
 883		crng->state[13]++;
 884	spin_unlock_irqrestore(&crng->lock, flags);
 885}
 886
 887static void extract_crng(__u8 out[CHACHA20_BLOCK_SIZE])
 888{
 889	struct crng_state *crng = NULL;
 
 
 890
 891#ifdef CONFIG_NUMA
 892	if (crng_node_pool)
 893		crng = crng_node_pool[numa_node_id()];
 894	if (crng == NULL)
 895#endif
 896		crng = &primary_crng;
 897	_extract_crng(crng, out);
 898}
 899
 
 
 900/*
 901 * Use the leftover bytes from the CRNG block output (if there is
 902 * enough) to mutate the CRNG key to provide backtracking protection.
 903 */
 904static void _crng_backtrack_protect(struct crng_state *crng,
 905				    __u8 tmp[CHACHA20_BLOCK_SIZE], int used)
 906{
 907	unsigned long	flags;
 908	__u32		*s, *d;
 909	int		i;
 
 
 
 
 910
 911	used = round_up(used, sizeof(__u32));
 912	if (used + CHACHA20_KEY_SIZE > CHACHA20_BLOCK_SIZE) {
 913		extract_crng(tmp);
 914		used = 0;
 
 
 
 
 
 
 
 
 
 
 
 915	}
 916	spin_lock_irqsave(&crng->lock, flags);
 917	s = (__u32 *) &tmp[used];
 918	d = &crng->state[4];
 919	for (i=0; i < 8; i++)
 920		*d++ ^= *s++;
 921	spin_unlock_irqrestore(&crng->lock, flags);
 922}
 923
 924static void crng_backtrack_protect(__u8 tmp[CHACHA20_BLOCK_SIZE], int used)
 925{
 926	struct crng_state *crng = NULL;
 927
 928#ifdef CONFIG_NUMA
 929	if (crng_node_pool)
 930		crng = crng_node_pool[numa_node_id()];
 931	if (crng == NULL)
 932#endif
 933		crng = &primary_crng;
 934	_crng_backtrack_protect(crng, tmp, used);
 935}
 936
 937static ssize_t extract_crng_user(void __user *buf, size_t nbytes)
 
 
 
 
 938{
 939	ssize_t ret = 0, i = CHACHA20_BLOCK_SIZE;
 940	__u8 tmp[CHACHA20_BLOCK_SIZE];
 941	int large_request = (nbytes > 256);
 942
 943	while (nbytes) {
 944		if (large_request && need_resched()) {
 945			if (signal_pending(current)) {
 946				if (ret == 0)
 947					ret = -ERESTARTSYS;
 948				break;
 949			}
 950			schedule();
 951		}
 952
 953		extract_crng(tmp);
 954		i = min_t(int, nbytes, CHACHA20_BLOCK_SIZE);
 955		if (copy_to_user(buf, tmp, i)) {
 956			ret = -EFAULT;
 957			break;
 958		}
 
 959
 960		nbytes -= i;
 961		buf += i;
 962		ret += i;
 963	}
 964	crng_backtrack_protect(tmp, i);
 965
 966	/* Wipe data just written to memory */
 967	memzero_explicit(tmp, sizeof(tmp));
 968
 969	return ret;
 
 970}
 971
 972
 973/*********************************************************************
 974 *
 975 * Entropy input management
 976 *
 977 *********************************************************************/
 978
 979/* There is one of these per entropy source */
 980struct timer_rand_state {
 981	cycles_t last_time;
 982	long last_delta, last_delta2;
 983	unsigned dont_count_entropy:1;
 984};
 985
 986#define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
 987
 988/*
 989 * Add device- or boot-specific data to the input pool to help
 990 * initialize it.
 991 *
 992 * None of this adds any entropy; it is meant to avoid the problem of
 993 * the entropy pool having similar initial state across largely
 994 * identical devices.
 995 */
 996void add_device_randomness(const void *buf, unsigned int size)
 997{
 998	unsigned long time = random_get_entropy() ^ jiffies;
 999	unsigned long flags;
1000
1001	trace_add_device_randomness(size, _RET_IP_);
1002	spin_lock_irqsave(&input_pool.lock, flags);
1003	_mix_pool_bytes(&input_pool, buf, size);
1004	_mix_pool_bytes(&input_pool, &time, sizeof(time));
1005	spin_unlock_irqrestore(&input_pool.lock, flags);
1006}
1007EXPORT_SYMBOL(add_device_randomness);
1008
1009static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
1010
1011/*
1012 * This function adds entropy to the entropy "pool" by using timing
1013 * delays.  It uses the timer_rand_state structure to make an estimate
1014 * of how many bits of entropy this call has added to the pool.
1015 *
1016 * The number "num" is also added to the pool - it should somehow describe
1017 * the type of event which just happened.  This is currently 0-255 for
1018 * keyboard scan codes, and 256 upwards for interrupts.
1019 *
1020 */
1021static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
1022{
1023	struct entropy_store	*r;
1024	struct {
1025		long jiffies;
1026		unsigned cycles;
1027		unsigned num;
1028	} sample;
1029	long delta, delta2, delta3;
1030
1031	preempt_disable();
1032
1033	sample.jiffies = jiffies;
1034	sample.cycles = random_get_entropy();
1035	sample.num = num;
1036	r = &input_pool;
1037	mix_pool_bytes(r, &sample, sizeof(sample));
1038
1039	/*
1040	 * Calculate number of bits of randomness we probably added.
1041	 * We take into account the first, second and third-order deltas
1042	 * in order to make our estimate.
1043	 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1044
1045	if (!state->dont_count_entropy) {
1046		delta = sample.jiffies - state->last_time;
1047		state->last_time = sample.jiffies;
1048
1049		delta2 = delta - state->last_delta;
1050		state->last_delta = delta;
1051
1052		delta3 = delta2 - state->last_delta2;
1053		state->last_delta2 = delta2;
1054
1055		if (delta < 0)
1056			delta = -delta;
1057		if (delta2 < 0)
1058			delta2 = -delta2;
1059		if (delta3 < 0)
1060			delta3 = -delta3;
1061		if (delta > delta2)
1062			delta = delta2;
1063		if (delta > delta3)
1064			delta = delta3;
1065
1066		/*
1067		 * delta is now minimum absolute delta.
1068		 * Round down by 1 bit on general principles,
1069		 * and limit entropy entimate to 12 bits.
1070		 */
1071		credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
 
 
 
 
 
1072	}
1073	preempt_enable();
1074}
 
 
 
1075
1076void add_input_randomness(unsigned int type, unsigned int code,
1077				 unsigned int value)
1078{
1079	static unsigned char last_value;
 
 
1080
1081	/* ignore autorepeat and the like */
1082	if (value == last_value)
1083		return;
1084
1085	last_value = value;
1086	add_timer_randomness(&input_timer_state,
1087			     (type << 4) ^ code ^ (code >> 4) ^ value);
1088	trace_add_input_randomness(ENTROPY_BITS(&input_pool));
1089}
1090EXPORT_SYMBOL_GPL(add_input_randomness);
 
1091
1092static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
 
 
 
 
 
1093
1094#ifdef ADD_INTERRUPT_BENCH
1095static unsigned long avg_cycles, avg_deviation;
1096
1097#define AVG_SHIFT 8     /* Exponential average factor k=1/256 */
1098#define FIXED_1_2 (1 << (AVG_SHIFT-1))
 
 
 
 
 
 
 
 
1099
1100static void add_interrupt_bench(cycles_t start)
 
 
 
 
 
 
1101{
1102        long delta = random_get_entropy() - start;
 
 
 
 
 
 
1103
1104        /* Use a weighted moving average */
1105        delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
1106        avg_cycles += delta;
1107        /* And average deviation */
1108        delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
1109        avg_deviation += delta;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1110}
1111#else
1112#define add_interrupt_bench(x)
1113#endif
1114
1115static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
1116{
1117	__u32 *ptr = (__u32 *) regs;
1118
1119	if (regs == NULL)
1120		return 0;
1121	if (f->reg_idx >= sizeof(struct pt_regs) / sizeof(__u32))
1122		f->reg_idx = 0;
1123	return *(ptr + f->reg_idx++);
1124}
1125
1126void add_interrupt_randomness(int irq, int irq_flags)
1127{
1128	struct entropy_store	*r;
1129	struct fast_pool	*fast_pool = this_cpu_ptr(&irq_randomness);
1130	struct pt_regs		*regs = get_irq_regs();
1131	unsigned long		now = jiffies;
1132	cycles_t		cycles = random_get_entropy();
1133	__u32			c_high, j_high;
1134	__u64			ip;
1135	unsigned long		seed;
1136	int			credit = 0;
1137
1138	if (cycles == 0)
1139		cycles = get_reg(fast_pool, regs);
1140	c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
1141	j_high = (sizeof(now) > 4) ? now >> 32 : 0;
1142	fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
1143	fast_pool->pool[1] ^= now ^ c_high;
1144	ip = regs ? instruction_pointer(regs) : _RET_IP_;
1145	fast_pool->pool[2] ^= ip;
1146	fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
1147		get_reg(fast_pool, regs);
1148
1149	fast_mix(fast_pool);
1150	add_interrupt_bench(cycles);
1151
1152	if (!crng_ready()) {
1153		if ((fast_pool->count >= 64) &&
1154		    crng_fast_load((char *) fast_pool->pool,
1155				   sizeof(fast_pool->pool))) {
1156			fast_pool->count = 0;
1157			fast_pool->last = now;
1158		}
1159		return;
1160	}
1161
1162	if ((fast_pool->count < 64) &&
1163	    !time_after(now, fast_pool->last + HZ))
1164		return;
1165
1166	r = &input_pool;
1167	if (!spin_trylock(&r->lock))
1168		return;
1169
1170	fast_pool->last = now;
1171	__mix_pool_bytes(r, &fast_pool->pool, sizeof(fast_pool->pool));
1172
1173	/*
1174	 * If we have architectural seed generator, produce a seed and
1175	 * add it to the pool.  For the sake of paranoia don't let the
1176	 * architectural seed generator dominate the input from the
1177	 * interrupt noise.
1178	 */
1179	if (arch_get_random_seed_long(&seed)) {
1180		__mix_pool_bytes(r, &seed, sizeof(seed));
1181		credit = 1;
1182	}
1183	spin_unlock(&r->lock);
1184
1185	fast_pool->count = 0;
 
1186
1187	/* award one bit for the contents of the fast pool */
1188	credit_entropy_bits(r, credit + 1);
1189}
1190EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1191
1192#ifdef CONFIG_BLOCK
1193void add_disk_randomness(struct gendisk *disk)
1194{
1195	if (!disk || !disk->random)
1196		return;
1197	/* first major is 1, so we get >= 0x200 here */
1198	add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1199	trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
1200}
1201EXPORT_SYMBOL_GPL(add_disk_randomness);
1202#endif
1203
1204/*********************************************************************
1205 *
1206 * Entropy extraction routines
1207 *
1208 *********************************************************************/
1209
1210/*
1211 * This utility inline function is responsible for transferring entropy
1212 * from the primary pool to the secondary extraction pool. We make
1213 * sure we pull enough for a 'catastrophic reseed'.
1214 */
1215static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes);
1216static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1217{
1218	if (!r->pull ||
1219	    r->entropy_count >= (nbytes << (ENTROPY_SHIFT + 3)) ||
1220	    r->entropy_count > r->poolinfo->poolfracbits)
1221		return;
1222
1223	if (r->limit == 0 && random_min_urandom_seed) {
1224		unsigned long now = jiffies;
1225
1226		if (time_before(now,
1227				r->last_pulled + random_min_urandom_seed * HZ))
1228			return;
1229		r->last_pulled = now;
1230	}
1231
1232	_xfer_secondary_pool(r, nbytes);
1233}
 
1234
1235static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1236{
1237	__u32	tmp[OUTPUT_POOL_WORDS];
1238
1239	/* For /dev/random's pool, always leave two wakeups' worth */
1240	int rsvd_bytes = r->limit ? 0 : random_read_wakeup_bits / 4;
1241	int bytes = nbytes;
1242
1243	/* pull at least as much as a wakeup */
1244	bytes = max_t(int, bytes, random_read_wakeup_bits / 8);
1245	/* but never more than the buffer size */
1246	bytes = min_t(int, bytes, sizeof(tmp));
1247
1248	trace_xfer_secondary_pool(r->name, bytes * 8, nbytes * 8,
1249				  ENTROPY_BITS(r), ENTROPY_BITS(r->pull));
1250	bytes = extract_entropy(r->pull, tmp, bytes,
1251				random_read_wakeup_bits / 8, rsvd_bytes);
1252	mix_pool_bytes(r, tmp, bytes);
1253	credit_entropy_bits(r, bytes*8);
1254}
1255
1256/*
1257 * Used as a workqueue function so that when the input pool is getting
1258 * full, we can "spill over" some entropy to the output pools.  That
1259 * way the output pools can store some of the excess entropy instead
1260 * of letting it go to waste.
 
1261 */
1262static void push_to_pool(struct work_struct *work)
1263{
1264	struct entropy_store *r = container_of(work, struct entropy_store,
1265					      push_work);
1266	BUG_ON(!r);
1267	_xfer_secondary_pool(r, random_read_wakeup_bits/8);
1268	trace_push_to_pool(r->name, r->entropy_count >> ENTROPY_SHIFT,
1269			   r->pull->entropy_count >> ENTROPY_SHIFT);
1270}
1271
1272/*
1273 * This function decides how many bytes to actually take from the
1274 * given pool, and also debits the entropy count accordingly.
1275 */
1276static size_t account(struct entropy_store *r, size_t nbytes, int min,
1277		      int reserved)
1278{
1279	int entropy_count, orig;
1280	size_t ibytes, nfrac;
1281
1282	BUG_ON(r->entropy_count > r->poolinfo->poolfracbits);
1283
1284	/* Can we pull enough? */
1285retry:
1286	entropy_count = orig = ACCESS_ONCE(r->entropy_count);
1287	ibytes = nbytes;
1288	/* If limited, never pull more than available */
1289	if (r->limit) {
1290		int have_bytes = entropy_count >> (ENTROPY_SHIFT + 3);
1291
1292		if ((have_bytes -= reserved) < 0)
1293			have_bytes = 0;
1294		ibytes = min_t(size_t, ibytes, have_bytes);
1295	}
1296	if (ibytes < min)
1297		ibytes = 0;
1298
1299	if (unlikely(entropy_count < 0)) {
1300		pr_warn("random: negative entropy count: pool %s count %d\n",
1301			r->name, entropy_count);
1302		WARN_ON(1);
1303		entropy_count = 0;
1304	}
1305	nfrac = ibytes << (ENTROPY_SHIFT + 3);
1306	if ((size_t) entropy_count > nfrac)
1307		entropy_count -= nfrac;
1308	else
1309		entropy_count = 0;
1310
1311	if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
1312		goto retry;
1313
1314	trace_debit_entropy(r->name, 8 * ibytes);
1315	if (ibytes &&
1316	    (r->entropy_count >> ENTROPY_SHIFT) < random_write_wakeup_bits) {
1317		wake_up_interruptible(&random_write_wait);
1318		kill_fasync(&fasync, SIGIO, POLL_OUT);
1319	}
1320
1321	return ibytes;
1322}
1323
1324/*
1325 * This function does the actual extraction for extract_entropy and
1326 * extract_entropy_user.
1327 *
1328 * Note: we assume that .poolwords is a multiple of 16 words.
1329 */
1330static void extract_buf(struct entropy_store *r, __u8 *out)
1331{
1332	int i;
1333	union {
1334		__u32 w[5];
1335		unsigned long l[LONGS(20)];
1336	} hash;
1337	__u32 workspace[SHA_WORKSPACE_WORDS];
1338	unsigned long flags;
1339
1340	/*
1341	 * If we have an architectural hardware random number
1342	 * generator, use it for SHA's initial vector
1343	 */
1344	sha_init(hash.w);
1345	for (i = 0; i < LONGS(20); i++) {
1346		unsigned long v;
1347		if (!arch_get_random_long(&v))
1348			break;
1349		hash.l[i] = v;
1350	}
1351
1352	/* Generate a hash across the pool, 16 words (512 bits) at a time */
1353	spin_lock_irqsave(&r->lock, flags);
1354	for (i = 0; i < r->poolinfo->poolwords; i += 16)
1355		sha_transform(hash.w, (__u8 *)(r->pool + i), workspace);
 
 
 
 
 
 
1356
1357	/*
1358	 * We mix the hash back into the pool to prevent backtracking
1359	 * attacks (where the attacker knows the state of the pool
1360	 * plus the current outputs, and attempts to find previous
1361	 * ouputs), unless the hash function can be inverted. By
1362	 * mixing at least a SHA1 worth of hash data back, we make
1363	 * brute-forcing the feedback as hard as brute-forcing the
1364	 * hash.
1365	 */
1366	__mix_pool_bytes(r, hash.w, sizeof(hash.w));
1367	spin_unlock_irqrestore(&r->lock, flags);
1368
1369	memzero_explicit(workspace, sizeof(workspace));
1370
1371	/*
1372	 * In case the hash function has some recognizable output
1373	 * pattern, we fold it in half. Thus, we always feed back
1374	 * twice as much data as we output.
 
 
1375	 */
1376	hash.w[0] ^= hash.w[3];
1377	hash.w[1] ^= hash.w[4];
1378	hash.w[2] ^= rol32(hash.w[2], 16);
1379
1380	memcpy(out, &hash, EXTRACT_SIZE);
1381	memzero_explicit(&hash, sizeof(hash));
1382}
1383
1384static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
1385				size_t nbytes, int fips)
1386{
1387	ssize_t ret = 0, i;
1388	__u8 tmp[EXTRACT_SIZE];
1389	unsigned long flags;
1390
1391	while (nbytes) {
1392		extract_buf(r, tmp);
 
1393
1394		if (fips) {
1395			spin_lock_irqsave(&r->lock, flags);
1396			if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
1397				panic("Hardware RNG duplicated output!\n");
1398			memcpy(r->last_data, tmp, EXTRACT_SIZE);
1399			spin_unlock_irqrestore(&r->lock, flags);
1400		}
1401		i = min_t(int, nbytes, EXTRACT_SIZE);
1402		memcpy(buf, tmp, i);
1403		nbytes -= i;
1404		buf += i;
1405		ret += i;
1406	}
1407
1408	/* Wipe data just returned from memory */
1409	memzero_explicit(tmp, sizeof(tmp));
1410
1411	return ret;
1412}
 
1413
1414/*
1415 * This function extracts randomness from the "entropy pool", and
1416 * returns it in a buffer.
1417 *
1418 * The min parameter specifies the minimum amount we can pull before
1419 * failing to avoid races that defeat catastrophic reseeding while the
1420 * reserved parameter indicates how much entropy we must leave in the
1421 * pool after each pull to avoid starving other readers.
1422 */
1423static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1424				 size_t nbytes, int min, int reserved)
1425{
1426	__u8 tmp[EXTRACT_SIZE];
1427	unsigned long flags;
1428
1429	/* if last_data isn't primed, we need EXTRACT_SIZE extra bytes */
1430	if (fips_enabled) {
1431		spin_lock_irqsave(&r->lock, flags);
1432		if (!r->last_data_init) {
1433			r->last_data_init = 1;
1434			spin_unlock_irqrestore(&r->lock, flags);
1435			trace_extract_entropy(r->name, EXTRACT_SIZE,
1436					      ENTROPY_BITS(r), _RET_IP_);
1437			xfer_secondary_pool(r, EXTRACT_SIZE);
1438			extract_buf(r, tmp);
1439			spin_lock_irqsave(&r->lock, flags);
1440			memcpy(r->last_data, tmp, EXTRACT_SIZE);
1441		}
1442		spin_unlock_irqrestore(&r->lock, flags);
1443	}
1444
1445	trace_extract_entropy(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1446	xfer_secondary_pool(r, nbytes);
1447	nbytes = account(r, nbytes, min, reserved);
1448
1449	return _extract_entropy(r, buf, nbytes, fips_enabled);
1450}
 
1451
1452/*
1453 * This function extracts randomness from the "entropy pool", and
1454 * returns it in a userspace buffer.
1455 */
1456static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
1457				    size_t nbytes)
1458{
1459	ssize_t ret = 0, i;
1460	__u8 tmp[EXTRACT_SIZE];
1461	int large_request = (nbytes > 256);
1462
1463	trace_extract_entropy_user(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1464	xfer_secondary_pool(r, nbytes);
1465	nbytes = account(r, nbytes, 0, 0);
1466
1467	while (nbytes) {
1468		if (large_request && need_resched()) {
1469			if (signal_pending(current)) {
1470				if (ret == 0)
1471					ret = -ERESTARTSYS;
1472				break;
1473			}
1474			schedule();
1475		}
1476
1477		extract_buf(r, tmp);
1478		i = min_t(int, nbytes, EXTRACT_SIZE);
1479		if (copy_to_user(buf, tmp, i)) {
1480			ret = -EFAULT;
1481			break;
1482		}
1483
1484		nbytes -= i;
1485		buf += i;
1486		ret += i;
1487	}
1488
1489	/* Wipe data just returned from memory */
1490	memzero_explicit(tmp, sizeof(tmp));
1491
1492	return ret;
1493}
1494
1495/*
1496 * This function is the exported kernel interface.  It returns some
1497 * number of good random numbers, suitable for key generation, seeding
1498 * TCP sequence numbers, etc.  It does not rely on the hardware random
1499 * number generator.  For random bytes direct from the hardware RNG
1500 * (when available), use get_random_bytes_arch().
1501 */
1502void get_random_bytes(void *buf, int nbytes)
1503{
1504	__u8 tmp[CHACHA20_BLOCK_SIZE];
1505
1506#if DEBUG_RANDOM_BOOT > 0
1507	if (!crng_ready())
1508		printk(KERN_NOTICE "random: %pF get_random_bytes called "
1509		       "with crng_init = %d\n", (void *) _RET_IP_, crng_init);
1510#endif
1511	trace_get_random_bytes(nbytes, _RET_IP_);
1512
1513	while (nbytes >= CHACHA20_BLOCK_SIZE) {
1514		extract_crng(buf);
1515		buf += CHACHA20_BLOCK_SIZE;
1516		nbytes -= CHACHA20_BLOCK_SIZE;
1517	}
1518
1519	if (nbytes > 0) {
1520		extract_crng(tmp);
1521		memcpy(buf, tmp, nbytes);
1522		crng_backtrack_protect(tmp, nbytes);
1523	} else
1524		crng_backtrack_protect(tmp, CHACHA20_BLOCK_SIZE);
1525	memzero_explicit(tmp, sizeof(tmp));
1526}
1527EXPORT_SYMBOL(get_random_bytes);
1528
1529/*
1530 * Add a callback function that will be invoked when the nonblocking
1531 * pool is initialised.
 
 
 
 
 
1532 *
1533 * returns: 0 if callback is successfully added
1534 *	    -EALREADY if pool is already initialised (callback not called)
1535 *	    -ENOENT if module for callback is not alive
1536 */
1537int add_random_ready_callback(struct random_ready_callback *rdy)
1538{
1539	struct module *owner;
1540	unsigned long flags;
1541	int err = -EALREADY;
1542
1543	if (crng_ready())
1544		return err;
1545
1546	owner = rdy->owner;
1547	if (!try_module_get(owner))
1548		return -ENOENT;
1549
1550	spin_lock_irqsave(&random_ready_list_lock, flags);
1551	if (crng_ready())
1552		goto out;
1553
1554	owner = NULL;
1555
1556	list_add(&rdy->list, &random_ready_list);
1557	err = 0;
1558
1559out:
1560	spin_unlock_irqrestore(&random_ready_list_lock, flags);
1561
1562	module_put(owner);
1563
1564	return err;
1565}
1566EXPORT_SYMBOL(add_random_ready_callback);
1567
1568/*
1569 * Delete a previously registered readiness callback function.
 
1570 */
1571void del_random_ready_callback(struct random_ready_callback *rdy)
1572{
1573	unsigned long flags;
1574	struct module *owner = NULL;
 
 
 
 
1575
1576	spin_lock_irqsave(&random_ready_list_lock, flags);
1577	if (!list_empty(&rdy->list)) {
1578		list_del_init(&rdy->list);
1579		owner = rdy->owner;
 
1580	}
1581	spin_unlock_irqrestore(&random_ready_list_lock, flags);
 
 
1582
1583	module_put(owner);
1584}
1585EXPORT_SYMBOL(del_random_ready_callback);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1586
1587/*
1588 * This function will use the architecture-specific hardware random
1589 * number generator if it is available.  The arch-specific hw RNG will
1590 * almost certainly be faster than what we can do in software, but it
1591 * is impossible to verify that it is implemented securely (as
1592 * opposed, to, say, the AES encryption of a sequence number using a
1593 * key known by the NSA).  So it's useful if we need the speed, but
1594 * only if we're willing to trust the hardware manufacturer not to
1595 * have put in a back door.
1596 */
1597void get_random_bytes_arch(void *buf, int nbytes)
1598{
1599	char *p = buf;
1600
1601	trace_get_random_bytes_arch(nbytes, _RET_IP_);
1602	while (nbytes) {
1603		unsigned long v;
1604		int chunk = min(nbytes, (int)sizeof(unsigned long));
1605
1606		if (!arch_get_random_long(&v))
1607			break;
1608		
1609		memcpy(p, &v, chunk);
1610		p += chunk;
1611		nbytes -= chunk;
1612	}
 
1613
1614	if (nbytes)
1615		get_random_bytes(p, nbytes);
1616}
1617EXPORT_SYMBOL(get_random_bytes_arch);
1618
1619
1620/*
1621 * init_std_data - initialize pool with system data
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1622 *
1623 * @r: pool to initialize
 
 
1624 *
1625 * This function clears the pool's entropy count and mixes some system
1626 * data into the pool to prepare it for use. The pool is not cleared
1627 * as that can only decrease the entropy in the pool.
1628 */
1629static void init_std_data(struct entropy_store *r)
1630{
1631	int i;
1632	ktime_t now = ktime_get_real();
1633	unsigned long rv;
1634
1635	r->last_pulled = jiffies;
1636	mix_pool_bytes(r, &now, sizeof(now));
1637	for (i = r->poolinfo->poolbytes; i > 0; i -= sizeof(rv)) {
1638		if (!arch_get_random_seed_long(&rv) &&
1639		    !arch_get_random_long(&rv))
1640			rv = random_get_entropy();
1641		mix_pool_bytes(r, &rv, sizeof(rv));
1642	}
1643	mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1644}
1645
1646/*
1647 * Note that setup_arch() may call add_device_randomness()
1648 * long before we get here. This allows seeding of the pools
1649 * with some platform dependent data very early in the boot
1650 * process. But it limits our options here. We must use
1651 * statically allocated structures that already have all
1652 * initializations complete at compile time. We should also
1653 * take care not to overwrite the precious per platform data
1654 * we were given.
1655 */
1656static int rand_initialize(void)
1657{
1658#ifdef CONFIG_NUMA
1659	int i;
1660	struct crng_state *crng;
1661	struct crng_state **pool;
1662#endif
1663
1664	init_std_data(&input_pool);
1665	init_std_data(&blocking_pool);
1666	crng_initialize(&primary_crng);
1667
1668#ifdef CONFIG_NUMA
1669	pool = kcalloc(nr_node_ids, sizeof(*pool), GFP_KERNEL|__GFP_NOFAIL);
1670	for_each_online_node(i) {
1671		crng = kmalloc_node(sizeof(struct crng_state),
1672				    GFP_KERNEL | __GFP_NOFAIL, i);
1673		spin_lock_init(&crng->lock);
1674		crng_initialize(crng);
1675		pool[i] = crng;
1676	}
1677	mb();
1678	crng_node_pool = pool;
1679#endif
1680	return 0;
 
1681}
1682early_initcall(rand_initialize);
1683
1684#ifdef CONFIG_BLOCK
1685void rand_initialize_disk(struct gendisk *disk)
1686{
1687	struct timer_rand_state *state;
1688
1689	/*
1690	 * If kzalloc returns null, we just won't use that entropy
1691	 * source.
1692	 */
1693	state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1694	if (state) {
1695		state->last_time = INITIAL_JIFFIES;
1696		disk->random = state;
1697	}
1698}
1699#endif
1700
1701static ssize_t
1702_random_read(int nonblock, char __user *buf, size_t nbytes)
1703{
1704	ssize_t n;
 
 
1705
1706	if (nbytes == 0)
1707		return 0;
1708
1709	nbytes = min_t(size_t, nbytes, SEC_XFER_SIZE);
1710	while (1) {
1711		n = extract_entropy_user(&blocking_pool, buf, nbytes);
1712		if (n < 0)
1713			return n;
1714		trace_random_read(n*8, (nbytes-n)*8,
1715				  ENTROPY_BITS(&blocking_pool),
1716				  ENTROPY_BITS(&input_pool));
1717		if (n > 0)
1718			return n;
1719
1720		/* Pool is (near) empty.  Maybe wait and retry. */
1721		if (nonblock)
1722			return -EAGAIN;
 
 
 
 
1723
1724		wait_event_interruptible(random_read_wait,
1725			ENTROPY_BITS(&input_pool) >=
1726			random_read_wakeup_bits);
1727		if (signal_pending(current))
1728			return -ERESTARTSYS;
1729	}
1730}
1731
1732static ssize_t
1733random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1734{
1735	return _random_read(file->f_flags & O_NONBLOCK, buf, nbytes);
1736}
1737
1738static ssize_t
1739urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1740{
1741	unsigned long flags;
1742	static int maxwarn = 10;
1743	int ret;
1744
1745	if (!crng_ready() && maxwarn > 0) {
1746		maxwarn--;
1747		printk(KERN_NOTICE "random: %s: uninitialized urandom read "
1748		       "(%zd bytes read)\n",
1749		       current->comm, nbytes);
1750		spin_lock_irqsave(&primary_crng.lock, flags);
1751		crng_init_cnt = 0;
1752		spin_unlock_irqrestore(&primary_crng.lock, flags);
1753	}
1754	nbytes = min_t(size_t, nbytes, INT_MAX >> (ENTROPY_SHIFT + 3));
1755	ret = extract_crng_user(buf, nbytes);
1756	trace_urandom_read(8 * nbytes, 0, ENTROPY_BITS(&input_pool));
1757	return ret;
1758}
1759
1760static unsigned int
1761random_poll(struct file *file, poll_table * wait)
1762{
1763	unsigned int mask;
1764
1765	poll_wait(file, &random_read_wait, wait);
1766	poll_wait(file, &random_write_wait, wait);
1767	mask = 0;
1768	if (ENTROPY_BITS(&input_pool) >= random_read_wakeup_bits)
1769		mask |= POLLIN | POLLRDNORM;
1770	if (ENTROPY_BITS(&input_pool) < random_write_wakeup_bits)
1771		mask |= POLLOUT | POLLWRNORM;
1772	return mask;
1773}
1774
1775static int
1776write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1777{
1778	size_t bytes;
1779	__u32 buf[16];
1780	const char __user *p = buffer;
1781
1782	while (count > 0) {
1783		bytes = min(count, sizeof(buf));
1784		if (copy_from_user(&buf, p, bytes))
1785			return -EFAULT;
1786
1787		count -= bytes;
1788		p += bytes;
1789
1790		mix_pool_bytes(r, buf, bytes);
1791		cond_resched();
1792	}
1793
1794	return 0;
1795}
1796
1797static ssize_t random_write(struct file *file, const char __user *buffer,
1798			    size_t count, loff_t *ppos)
1799{
1800	size_t ret;
1801
1802	ret = write_pool(&input_pool, buffer, count);
1803	if (ret)
 
 
 
 
 
1804		return ret;
1805
1806	return (ssize_t)count;
1807}
1808
1809static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1810{
1811	int size, ent_count;
1812	int __user *p = (int __user *)arg;
1813	int retval;
1814
1815	switch (cmd) {
1816	case RNDGETENTCNT:
1817		/* inherently racy, no point locking */
1818		ent_count = ENTROPY_BITS(&input_pool);
1819		if (put_user(ent_count, p))
1820			return -EFAULT;
1821		return 0;
1822	case RNDADDTOENTCNT:
1823		if (!capable(CAP_SYS_ADMIN))
1824			return -EPERM;
1825		if (get_user(ent_count, p))
1826			return -EFAULT;
1827		return credit_entropy_bits_safe(&input_pool, ent_count);
1828	case RNDADDENTROPY:
 
 
 
 
 
 
 
1829		if (!capable(CAP_SYS_ADMIN))
1830			return -EPERM;
1831		if (get_user(ent_count, p++))
1832			return -EFAULT;
1833		if (ent_count < 0)
1834			return -EINVAL;
1835		if (get_user(size, p++))
 
 
 
 
 
 
 
 
 
1836			return -EFAULT;
1837		retval = write_pool(&input_pool, (const char __user *)p,
1838				    size);
1839		if (retval < 0)
1840			return retval;
1841		return credit_entropy_bits_safe(&input_pool, ent_count);
1842	case RNDZAPENTCNT:
1843	case RNDCLEARPOOL:
1844		/*
1845		 * Clear the entropy pool counters. We no longer clear
1846		 * the entropy pool, as that's silly.
1847		 */
 
1848		if (!capable(CAP_SYS_ADMIN))
1849			return -EPERM;
1850		input_pool.entropy_count = 0;
1851		blocking_pool.entropy_count = 0;
 
1852		return 0;
1853	default:
1854		return -EINVAL;
1855	}
1856}
1857
1858static int random_fasync(int fd, struct file *filp, int on)
1859{
1860	return fasync_helper(fd, filp, on, &fasync);
1861}
1862
1863const struct file_operations random_fops = {
1864	.read  = random_read,
1865	.write = random_write,
1866	.poll  = random_poll,
1867	.unlocked_ioctl = random_ioctl,
 
1868	.fasync = random_fasync,
1869	.llseek = noop_llseek,
 
 
1870};
1871
1872const struct file_operations urandom_fops = {
1873	.read  = urandom_read,
1874	.write = random_write,
1875	.unlocked_ioctl = random_ioctl,
 
1876	.fasync = random_fasync,
1877	.llseek = noop_llseek,
 
 
1878};
1879
1880SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count,
1881		unsigned int, flags)
1882{
1883	if (flags & ~(GRND_NONBLOCK|GRND_RANDOM))
1884		return -EINVAL;
1885
1886	if (count > INT_MAX)
1887		count = INT_MAX;
1888
1889	if (flags & GRND_RANDOM)
1890		return _random_read(flags & GRND_NONBLOCK, buf, count);
1891
1892	if (!crng_ready()) {
1893		if (flags & GRND_NONBLOCK)
1894			return -EAGAIN;
1895		crng_wait_ready();
1896		if (signal_pending(current))
1897			return -ERESTARTSYS;
1898	}
1899	return urandom_read(NULL, buf, count, NULL);
1900}
1901
1902/********************************************************************
1903 *
1904 * Sysctl interface
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1905 *
1906 ********************************************************************/
1907
1908#ifdef CONFIG_SYSCTL
1909
1910#include <linux/sysctl.h>
1911
1912static int min_read_thresh = 8, min_write_thresh;
1913static int max_read_thresh = OUTPUT_POOL_WORDS * 32;
1914static int max_write_thresh = INPUT_POOL_WORDS * 32;
1915static char sysctl_bootid[16];
1916
1917/*
1918 * This function is used to return both the bootid UUID, and random
1919 * UUID.  The difference is in whether table->data is NULL; if it is,
1920 * then a new UUID is generated and returned to the user.
1921 *
1922 * If the user accesses this via the proc interface, the UUID will be
1923 * returned as an ASCII string in the standard UUID format; if via the
1924 * sysctl system call, as 16 bytes of binary data.
1925 */
1926static int proc_do_uuid(struct ctl_table *table, int write,
1927			void __user *buffer, size_t *lenp, loff_t *ppos)
1928{
1929	struct ctl_table fake_table;
1930	unsigned char buf[64], tmp_uuid[16], *uuid;
 
 
 
 
 
 
 
1931
1932	uuid = table->data;
1933	if (!uuid) {
1934		uuid = tmp_uuid;
1935		generate_random_uuid(uuid);
1936	} else {
1937		static DEFINE_SPINLOCK(bootid_spinlock);
1938
1939		spin_lock(&bootid_spinlock);
1940		if (!uuid[8])
1941			generate_random_uuid(uuid);
1942		spin_unlock(&bootid_spinlock);
1943	}
1944
1945	sprintf(buf, "%pU", uuid);
1946
1947	fake_table.data = buf;
1948	fake_table.maxlen = sizeof(buf);
1949
1950	return proc_dostring(&fake_table, write, buffer, lenp, ppos);
1951}
1952
1953/*
1954 * Return entropy available scaled to integral bits
1955 */
1956static int proc_do_entropy(struct ctl_table *table, int write,
1957			   void __user *buffer, size_t *lenp, loff_t *ppos)
1958{
1959	struct ctl_table fake_table;
1960	int entropy_count;
1961
1962	entropy_count = *(int *)table->data >> ENTROPY_SHIFT;
1963
1964	fake_table.data = &entropy_count;
1965	fake_table.maxlen = sizeof(entropy_count);
1966
1967	return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
1968}
1969
1970static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
1971extern struct ctl_table random_table[];
1972struct ctl_table random_table[] = {
1973	{
1974		.procname	= "poolsize",
1975		.data		= &sysctl_poolsize,
1976		.maxlen		= sizeof(int),
1977		.mode		= 0444,
1978		.proc_handler	= proc_dointvec,
1979	},
1980	{
1981		.procname	= "entropy_avail",
 
1982		.maxlen		= sizeof(int),
1983		.mode		= 0444,
1984		.proc_handler	= proc_do_entropy,
1985		.data		= &input_pool.entropy_count,
1986	},
1987	{
1988		.procname	= "read_wakeup_threshold",
1989		.data		= &random_read_wakeup_bits,
1990		.maxlen		= sizeof(int),
1991		.mode		= 0644,
1992		.proc_handler	= proc_dointvec_minmax,
1993		.extra1		= &min_read_thresh,
1994		.extra2		= &max_read_thresh,
1995	},
1996	{
1997		.procname	= "write_wakeup_threshold",
1998		.data		= &random_write_wakeup_bits,
1999		.maxlen		= sizeof(int),
2000		.mode		= 0644,
2001		.proc_handler	= proc_dointvec_minmax,
2002		.extra1		= &min_write_thresh,
2003		.extra2		= &max_write_thresh,
2004	},
2005	{
2006		.procname	= "urandom_min_reseed_secs",
2007		.data		= &random_min_urandom_seed,
2008		.maxlen		= sizeof(int),
2009		.mode		= 0644,
2010		.proc_handler	= proc_dointvec,
2011	},
2012	{
2013		.procname	= "boot_id",
2014		.data		= &sysctl_bootid,
2015		.maxlen		= 16,
2016		.mode		= 0444,
2017		.proc_handler	= proc_do_uuid,
2018	},
2019	{
2020		.procname	= "uuid",
2021		.maxlen		= 16,
2022		.mode		= 0444,
2023		.proc_handler	= proc_do_uuid,
2024	},
2025#ifdef ADD_INTERRUPT_BENCH
2026	{
2027		.procname	= "add_interrupt_avg_cycles",
2028		.data		= &avg_cycles,
2029		.maxlen		= sizeof(avg_cycles),
2030		.mode		= 0444,
2031		.proc_handler	= proc_doulongvec_minmax,
2032	},
2033	{
2034		.procname	= "add_interrupt_avg_deviation",
2035		.data		= &avg_deviation,
2036		.maxlen		= sizeof(avg_deviation),
2037		.mode		= 0444,
2038		.proc_handler	= proc_doulongvec_minmax,
2039	},
2040#endif
2041	{ }
2042};
2043#endif 	/* CONFIG_SYSCTL */
2044
2045struct batched_entropy {
2046	union {
2047		unsigned long entropy_long[CHACHA20_BLOCK_SIZE / sizeof(unsigned long)];
2048		unsigned int entropy_int[CHACHA20_BLOCK_SIZE / sizeof(unsigned int)];
2049	};
2050	unsigned int position;
2051};
2052
2053/*
2054 * Get a random word for internal kernel use only. The quality of the random
2055 * number is either as good as RDRAND or as good as /dev/urandom, with the
2056 * goal of being quite fast and not depleting entropy.
2057 */
2058static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_long);
2059unsigned long get_random_long(void)
2060{
2061	unsigned long ret;
2062	struct batched_entropy *batch;
2063
2064	if (arch_get_random_long(&ret))
2065		return ret;
2066
2067	batch = &get_cpu_var(batched_entropy_long);
2068	if (batch->position % ARRAY_SIZE(batch->entropy_long) == 0) {
2069		extract_crng((u8 *)batch->entropy_long);
2070		batch->position = 0;
2071	}
2072	ret = batch->entropy_long[batch->position++];
2073	put_cpu_var(batched_entropy_long);
2074	return ret;
2075}
2076EXPORT_SYMBOL(get_random_long);
2077
2078#if BITS_PER_LONG == 32
2079unsigned int get_random_int(void)
2080{
2081	return get_random_long();
2082}
2083#else
2084static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_int);
2085unsigned int get_random_int(void)
2086{
2087	unsigned int ret;
2088	struct batched_entropy *batch;
2089
2090	if (arch_get_random_int(&ret))
2091		return ret;
2092
2093	batch = &get_cpu_var(batched_entropy_int);
2094	if (batch->position % ARRAY_SIZE(batch->entropy_int) == 0) {
2095		extract_crng((u8 *)batch->entropy_int);
2096		batch->position = 0;
2097	}
2098	ret = batch->entropy_int[batch->position++];
2099	put_cpu_var(batched_entropy_int);
2100	return ret;
2101}
 
2102#endif
2103EXPORT_SYMBOL(get_random_int);
2104
2105/**
2106 * randomize_page - Generate a random, page aligned address
2107 * @start:	The smallest acceptable address the caller will take.
2108 * @range:	The size of the area, starting at @start, within which the
2109 *		random address must fall.
2110 *
2111 * If @start + @range would overflow, @range is capped.
2112 *
2113 * NOTE: Historical use of randomize_range, which this replaces, presumed that
2114 * @start was already page aligned.  We now align it regardless.
2115 *
2116 * Return: A page aligned address within [start, start + range).  On error,
2117 * @start is returned.
2118 */
2119unsigned long
2120randomize_page(unsigned long start, unsigned long range)
2121{
2122	if (!PAGE_ALIGNED(start)) {
2123		range -= PAGE_ALIGN(start) - start;
2124		start = PAGE_ALIGN(start);
2125	}
2126
2127	if (start > ULONG_MAX - range)
2128		range = ULONG_MAX - start;
2129
2130	range >>= PAGE_SHIFT;
2131
2132	if (range == 0)
2133		return start;
2134
2135	return start + (get_random_long() % range << PAGE_SHIFT);
2136}
2137
2138/* Interface for in-kernel drivers of true hardware RNGs.
2139 * Those devices may produce endless random bits and will be throttled
2140 * when our pool is full.
2141 */
2142void add_hwgenerator_randomness(const char *buffer, size_t count,
2143				size_t entropy)
2144{
2145	struct entropy_store *poolp = &input_pool;
2146
2147	if (!crng_ready()) {
2148		crng_fast_load(buffer, count);
2149		return;
2150	}
2151
2152	/* Suspend writing if we're above the trickle threshold.
2153	 * We'll be woken up again once below random_write_wakeup_thresh,
2154	 * or when the calling thread is about to terminate.
2155	 */
2156	wait_event_interruptible(random_write_wait, kthread_should_stop() ||
2157			ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
2158	mix_pool_bytes(poolp, buffer, count);
2159	credit_entropy_bits(poolp, entropy);
2160}
2161EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);