Back to home page

OSCL-LXR

 
 

    


0001 // SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause)
0002 /*
0003  * Copyright (C) 2017-2022 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
0004  * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005
0005  * Copyright Theodore Ts'o, 1994, 1995, 1996, 1997, 1998, 1999. All rights reserved.
0006  *
0007  * This driver produces cryptographically secure pseudorandom data. It is divided
0008  * into roughly six sections, each with a section header:
0009  *
0010  *   - Initialization and readiness waiting.
0011  *   - Fast key erasure RNG, the "crng".
0012  *   - Entropy accumulation and extraction routines.
0013  *   - Entropy collection routines.
0014  *   - Userspace reader/writer interfaces.
0015  *   - Sysctl interface.
0016  *
0017  * The high level overview is that there is one input pool, into which
0018  * various pieces of data are hashed. Prior to initialization, some of that
0019  * data is then "credited" as having a certain number of bits of entropy.
0020  * When enough bits of entropy are available, the hash is finalized and
0021  * handed as a key to a stream cipher that expands it indefinitely for
0022  * various consumers. This key is periodically refreshed as the various
0023  * entropy collectors, described below, add data to the input pool.
0024  */
0025 
0026 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
0027 
0028 #include <linux/utsname.h>
0029 #include <linux/module.h>
0030 #include <linux/kernel.h>
0031 #include <linux/major.h>
0032 #include <linux/string.h>
0033 #include <linux/fcntl.h>
0034 #include <linux/slab.h>
0035 #include <linux/random.h>
0036 #include <linux/poll.h>
0037 #include <linux/init.h>
0038 #include <linux/fs.h>
0039 #include <linux/blkdev.h>
0040 #include <linux/interrupt.h>
0041 #include <linux/mm.h>
0042 #include <linux/nodemask.h>
0043 #include <linux/spinlock.h>
0044 #include <linux/kthread.h>
0045 #include <linux/percpu.h>
0046 #include <linux/ptrace.h>
0047 #include <linux/workqueue.h>
0048 #include <linux/irq.h>
0049 #include <linux/ratelimit.h>
0050 #include <linux/syscalls.h>
0051 #include <linux/completion.h>
0052 #include <linux/uuid.h>
0053 #include <linux/uaccess.h>
0054 #include <linux/suspend.h>
0055 #include <linux/siphash.h>
0056 #include <crypto/chacha.h>
0057 #include <crypto/blake2s.h>
0058 #include <asm/processor.h>
0059 #include <asm/irq.h>
0060 #include <asm/irq_regs.h>
0061 #include <asm/io.h>
0062 
0063 /*********************************************************************
0064  *
0065  * Initialization and readiness waiting.
0066  *
0067  * Much of the RNG infrastructure is devoted to various dependencies
0068  * being able to wait until the RNG has collected enough entropy and
0069  * is ready for safe consumption.
0070  *
0071  *********************************************************************/
0072 
0073 /*
0074  * crng_init is protected by base_crng->lock, and only increases
0075  * its value (from empty->early->ready).
0076  */
0077 static enum {
0078     CRNG_EMPTY = 0, /* Little to no entropy collected */
0079     CRNG_EARLY = 1, /* At least POOL_EARLY_BITS collected */
0080     CRNG_READY = 2  /* Fully initialized with POOL_READY_BITS collected */
0081 } crng_init __read_mostly = CRNG_EMPTY;
0082 static DEFINE_STATIC_KEY_FALSE(crng_is_ready);
0083 #define crng_ready() (static_branch_likely(&crng_is_ready) || crng_init >= CRNG_READY)
0084 /* Various types of waiters for crng_init->CRNG_READY transition. */
0085 static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
0086 static struct fasync_struct *fasync;
0087 
0088 /* Control how we warn userspace. */
0089 static struct ratelimit_state urandom_warning =
0090     RATELIMIT_STATE_INIT_FLAGS("urandom_warning", HZ, 3, RATELIMIT_MSG_ON_RELEASE);
0091 static int ratelimit_disable __read_mostly =
0092     IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM);
0093 module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
0094 MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
0095 
0096 /*
0097  * Returns whether or not the input pool has been seeded and thus guaranteed
0098  * to supply cryptographically secure random numbers. This applies to: the
0099  * /dev/urandom device, the get_random_bytes function, and the get_random_{u32,
0100  * ,u64,int,long} family of functions.
0101  *
0102  * Returns: true if the input pool has been seeded.
0103  *          false if the input pool has not been seeded.
0104  */
0105 bool rng_is_initialized(void)
0106 {
0107     return crng_ready();
0108 }
0109 EXPORT_SYMBOL(rng_is_initialized);
0110 
0111 static void __cold crng_set_ready(struct work_struct *work)
0112 {
0113     static_branch_enable(&crng_is_ready);
0114 }
0115 
0116 /* Used by wait_for_random_bytes(), and considered an entropy collector, below. */
0117 static void try_to_generate_entropy(void);
0118 
0119 /*
0120  * Wait for the input pool to be seeded and thus guaranteed to supply
0121  * cryptographically secure random numbers. This applies to: the /dev/urandom
0122  * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
0123  * family of functions. Using any of these functions without first calling
0124  * this function forfeits the guarantee of security.
0125  *
0126  * Returns: 0 if the input pool has been seeded.
0127  *          -ERESTARTSYS if the function was interrupted by a signal.
0128  */
0129 int wait_for_random_bytes(void)
0130 {
0131     while (!crng_ready()) {
0132         int ret;
0133 
0134         try_to_generate_entropy();
0135         ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ);
0136         if (ret)
0137             return ret > 0 ? 0 : ret;
0138     }
0139     return 0;
0140 }
0141 EXPORT_SYMBOL(wait_for_random_bytes);
0142 
0143 #define warn_unseeded_randomness() \
0144     if (IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM) && !crng_ready()) \
0145         printk_deferred(KERN_NOTICE "random: %s called from %pS with crng_init=%d\n", \
0146                 __func__, (void *)_RET_IP_, crng_init)
0147 
0148 
0149 /*********************************************************************
0150  *
0151  * Fast key erasure RNG, the "crng".
0152  *
0153  * These functions expand entropy from the entropy extractor into
0154  * long streams for external consumption using the "fast key erasure"
0155  * RNG described at <https://blog.cr.yp.to/20170723-random.html>.
0156  *
0157  * There are a few exported interfaces for use by other drivers:
0158  *
0159  *  void get_random_bytes(void *buf, size_t len)
0160  *  u32 get_random_u32()
0161  *  u64 get_random_u64()
0162  *  unsigned int get_random_int()
0163  *  unsigned long get_random_long()
0164  *
0165  * These interfaces will return the requested number of random bytes
0166  * into the given buffer or as a return value. This is equivalent to
0167  * a read from /dev/urandom. The u32, u64, int, and long family of
0168  * functions may be higher performance for one-off random integers,
0169  * because they do a bit of buffering and do not invoke reseeding
0170  * until the buffer is emptied.
0171  *
0172  *********************************************************************/
0173 
0174 enum {
0175     CRNG_RESEED_START_INTERVAL = HZ,
0176     CRNG_RESEED_INTERVAL = 60 * HZ
0177 };
0178 
0179 static struct {
0180     u8 key[CHACHA_KEY_SIZE] __aligned(__alignof__(long));
0181     unsigned long birth;
0182     unsigned long generation;
0183     spinlock_t lock;
0184 } base_crng = {
0185     .lock = __SPIN_LOCK_UNLOCKED(base_crng.lock)
0186 };
0187 
0188 struct crng {
0189     u8 key[CHACHA_KEY_SIZE];
0190     unsigned long generation;
0191     local_lock_t lock;
0192 };
0193 
0194 static DEFINE_PER_CPU(struct crng, crngs) = {
0195     .generation = ULONG_MAX,
0196     .lock = INIT_LOCAL_LOCK(crngs.lock),
0197 };
0198 
0199 /* Used by crng_reseed() and crng_make_state() to extract a new seed from the input pool. */
0200 static void extract_entropy(void *buf, size_t len);
0201 
0202 /* This extracts a new crng key from the input pool. */
0203 static void crng_reseed(void)
0204 {
0205     unsigned long flags;
0206     unsigned long next_gen;
0207     u8 key[CHACHA_KEY_SIZE];
0208 
0209     extract_entropy(key, sizeof(key));
0210 
0211     /*
0212      * We copy the new key into the base_crng, overwriting the old one,
0213      * and update the generation counter. We avoid hitting ULONG_MAX,
0214      * because the per-cpu crngs are initialized to ULONG_MAX, so this
0215      * forces new CPUs that come online to always initialize.
0216      */
0217     spin_lock_irqsave(&base_crng.lock, flags);
0218     memcpy(base_crng.key, key, sizeof(base_crng.key));
0219     next_gen = base_crng.generation + 1;
0220     if (next_gen == ULONG_MAX)
0221         ++next_gen;
0222     WRITE_ONCE(base_crng.generation, next_gen);
0223     WRITE_ONCE(base_crng.birth, jiffies);
0224     if (!static_branch_likely(&crng_is_ready))
0225         crng_init = CRNG_READY;
0226     spin_unlock_irqrestore(&base_crng.lock, flags);
0227     memzero_explicit(key, sizeof(key));
0228 }
0229 
0230 /*
0231  * This generates a ChaCha block using the provided key, and then
0232  * immediately overwrites that key with half the block. It returns
0233  * the resultant ChaCha state to the user, along with the second
0234  * half of the block containing 32 bytes of random data that may
0235  * be used; random_data_len may not be greater than 32.
0236  *
0237  * The returned ChaCha state contains within it a copy of the old
0238  * key value, at index 4, so the state should always be zeroed out
0239  * immediately after using in order to maintain forward secrecy.
0240  * If the state cannot be erased in a timely manner, then it is
0241  * safer to set the random_data parameter to &chacha_state[4] so
0242  * that this function overwrites it before returning.
0243  */
0244 static void crng_fast_key_erasure(u8 key[CHACHA_KEY_SIZE],
0245                   u32 chacha_state[CHACHA_STATE_WORDS],
0246                   u8 *random_data, size_t random_data_len)
0247 {
0248     u8 first_block[CHACHA_BLOCK_SIZE];
0249 
0250     BUG_ON(random_data_len > 32);
0251 
0252     chacha_init_consts(chacha_state);
0253     memcpy(&chacha_state[4], key, CHACHA_KEY_SIZE);
0254     memset(&chacha_state[12], 0, sizeof(u32) * 4);
0255     chacha20_block(chacha_state, first_block);
0256 
0257     memcpy(key, first_block, CHACHA_KEY_SIZE);
0258     memcpy(random_data, first_block + CHACHA_KEY_SIZE, random_data_len);
0259     memzero_explicit(first_block, sizeof(first_block));
0260 }
0261 
0262 /*
0263  * Return whether the crng seed is considered to be sufficiently old
0264  * that a reseeding is needed. This happens if the last reseeding
0265  * was CRNG_RESEED_INTERVAL ago, or during early boot, at an interval
0266  * proportional to the uptime.
0267  */
0268 static bool crng_has_old_seed(void)
0269 {
0270     static bool early_boot = true;
0271     unsigned long interval = CRNG_RESEED_INTERVAL;
0272 
0273     if (unlikely(READ_ONCE(early_boot))) {
0274         time64_t uptime = ktime_get_seconds();
0275         if (uptime >= CRNG_RESEED_INTERVAL / HZ * 2)
0276             WRITE_ONCE(early_boot, false);
0277         else
0278             interval = max_t(unsigned int, CRNG_RESEED_START_INTERVAL,
0279                      (unsigned int)uptime / 2 * HZ);
0280     }
0281     return time_is_before_jiffies(READ_ONCE(base_crng.birth) + interval);
0282 }
0283 
0284 /*
0285  * This function returns a ChaCha state that you may use for generating
0286  * random data. It also returns up to 32 bytes on its own of random data
0287  * that may be used; random_data_len may not be greater than 32.
0288  */
0289 static void crng_make_state(u32 chacha_state[CHACHA_STATE_WORDS],
0290                 u8 *random_data, size_t random_data_len)
0291 {
0292     unsigned long flags;
0293     struct crng *crng;
0294 
0295     BUG_ON(random_data_len > 32);
0296 
0297     /*
0298      * For the fast path, we check whether we're ready, unlocked first, and
0299      * then re-check once locked later. In the case where we're really not
0300      * ready, we do fast key erasure with the base_crng directly, extracting
0301      * when crng_init is CRNG_EMPTY.
0302      */
0303     if (!crng_ready()) {
0304         bool ready;
0305 
0306         spin_lock_irqsave(&base_crng.lock, flags);
0307         ready = crng_ready();
0308         if (!ready) {
0309             if (crng_init == CRNG_EMPTY)
0310                 extract_entropy(base_crng.key, sizeof(base_crng.key));
0311             crng_fast_key_erasure(base_crng.key, chacha_state,
0312                           random_data, random_data_len);
0313         }
0314         spin_unlock_irqrestore(&base_crng.lock, flags);
0315         if (!ready)
0316             return;
0317     }
0318 
0319     /*
0320      * If the base_crng is old enough, we reseed, which in turn bumps the
0321      * generation counter that we check below.
0322      */
0323     if (unlikely(crng_has_old_seed()))
0324         crng_reseed();
0325 
0326     local_lock_irqsave(&crngs.lock, flags);
0327     crng = raw_cpu_ptr(&crngs);
0328 
0329     /*
0330      * If our per-cpu crng is older than the base_crng, then it means
0331      * somebody reseeded the base_crng. In that case, we do fast key
0332      * erasure on the base_crng, and use its output as the new key
0333      * for our per-cpu crng. This brings us up to date with base_crng.
0334      */
0335     if (unlikely(crng->generation != READ_ONCE(base_crng.generation))) {
0336         spin_lock(&base_crng.lock);
0337         crng_fast_key_erasure(base_crng.key, chacha_state,
0338                       crng->key, sizeof(crng->key));
0339         crng->generation = base_crng.generation;
0340         spin_unlock(&base_crng.lock);
0341     }
0342 
0343     /*
0344      * Finally, when we've made it this far, our per-cpu crng has an up
0345      * to date key, and we can do fast key erasure with it to produce
0346      * some random data and a ChaCha state for the caller. All other
0347      * branches of this function are "unlikely", so most of the time we
0348      * should wind up here immediately.
0349      */
0350     crng_fast_key_erasure(crng->key, chacha_state, random_data, random_data_len);
0351     local_unlock_irqrestore(&crngs.lock, flags);
0352 }
0353 
0354 static void _get_random_bytes(void *buf, size_t len)
0355 {
0356     u32 chacha_state[CHACHA_STATE_WORDS];
0357     u8 tmp[CHACHA_BLOCK_SIZE];
0358     size_t first_block_len;
0359 
0360     if (!len)
0361         return;
0362 
0363     first_block_len = min_t(size_t, 32, len);
0364     crng_make_state(chacha_state, buf, first_block_len);
0365     len -= first_block_len;
0366     buf += first_block_len;
0367 
0368     while (len) {
0369         if (len < CHACHA_BLOCK_SIZE) {
0370             chacha20_block(chacha_state, tmp);
0371             memcpy(buf, tmp, len);
0372             memzero_explicit(tmp, sizeof(tmp));
0373             break;
0374         }
0375 
0376         chacha20_block(chacha_state, buf);
0377         if (unlikely(chacha_state[12] == 0))
0378             ++chacha_state[13];
0379         len -= CHACHA_BLOCK_SIZE;
0380         buf += CHACHA_BLOCK_SIZE;
0381     }
0382 
0383     memzero_explicit(chacha_state, sizeof(chacha_state));
0384 }
0385 
0386 /*
0387  * This function is the exported kernel interface.  It returns some
0388  * number of good random numbers, suitable for key generation, seeding
0389  * TCP sequence numbers, etc. In order to ensure that the randomness
0390  * by this function is okay, the function wait_for_random_bytes()
0391  * should be called and return 0 at least once at any point prior.
0392  */
0393 void get_random_bytes(void *buf, size_t len)
0394 {
0395     warn_unseeded_randomness();
0396     _get_random_bytes(buf, len);
0397 }
0398 EXPORT_SYMBOL(get_random_bytes);
0399 
0400 static ssize_t get_random_bytes_user(struct iov_iter *iter)
0401 {
0402     u32 chacha_state[CHACHA_STATE_WORDS];
0403     u8 block[CHACHA_BLOCK_SIZE];
0404     size_t ret = 0, copied;
0405 
0406     if (unlikely(!iov_iter_count(iter)))
0407         return 0;
0408 
0409     /*
0410      * Immediately overwrite the ChaCha key at index 4 with random
0411      * bytes, in case userspace causes copy_to_iter() below to sleep
0412      * forever, so that we still retain forward secrecy in that case.
0413      */
0414     crng_make_state(chacha_state, (u8 *)&chacha_state[4], CHACHA_KEY_SIZE);
0415     /*
0416      * However, if we're doing a read of len <= 32, we don't need to
0417      * use chacha_state after, so we can simply return those bytes to
0418      * the user directly.
0419      */
0420     if (iov_iter_count(iter) <= CHACHA_KEY_SIZE) {
0421         ret = copy_to_iter(&chacha_state[4], CHACHA_KEY_SIZE, iter);
0422         goto out_zero_chacha;
0423     }
0424 
0425     for (;;) {
0426         chacha20_block(chacha_state, block);
0427         if (unlikely(chacha_state[12] == 0))
0428             ++chacha_state[13];
0429 
0430         copied = copy_to_iter(block, sizeof(block), iter);
0431         ret += copied;
0432         if (!iov_iter_count(iter) || copied != sizeof(block))
0433             break;
0434 
0435         BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0);
0436         if (ret % PAGE_SIZE == 0) {
0437             if (signal_pending(current))
0438                 break;
0439             cond_resched();
0440         }
0441     }
0442 
0443     memzero_explicit(block, sizeof(block));
0444 out_zero_chacha:
0445     memzero_explicit(chacha_state, sizeof(chacha_state));
0446     return ret ? ret : -EFAULT;
0447 }
0448 
0449 /*
0450  * Batched entropy returns random integers. The quality of the random
0451  * number is good as /dev/urandom. In order to ensure that the randomness
0452  * provided by this function is okay, the function wait_for_random_bytes()
0453  * should be called and return 0 at least once at any point prior.
0454  */
0455 
0456 #define DEFINE_BATCHED_ENTROPY(type)                        \
0457 struct batch_ ##type {                              \
0458     /*                                  \
0459      * We make this 1.5x a ChaCha block, so that we get the         \
0460      * remaining 32 bytes from fast key erasure, plus one full      \
0461      * block from the detached ChaCha state. We can increase        \
0462      * the size of this later if needed so long as we keep the      \
0463      * formula of (integer_blocks + 0.5) * CHACHA_BLOCK_SIZE.       \
0464      */                                 \
0465     type entropy[CHACHA_BLOCK_SIZE * 3 / (2 * sizeof(type))];       \
0466     local_lock_t lock;                          \
0467     unsigned long generation;                       \
0468     unsigned int position;                          \
0469 };                                      \
0470                                         \
0471 static DEFINE_PER_CPU(struct batch_ ##type, batched_entropy_ ##type) = {    \
0472     .lock = INIT_LOCAL_LOCK(batched_entropy_ ##type.lock),          \
0473     .position = UINT_MAX                            \
0474 };                                      \
0475                                         \
0476 type get_random_ ##type(void)                           \
0477 {                                       \
0478     type ret;                               \
0479     unsigned long flags;                            \
0480     struct batch_ ##type *batch;                        \
0481     unsigned long next_gen;                         \
0482                                         \
0483     warn_unseeded_randomness();                     \
0484                                         \
0485     if  (!crng_ready()) {                           \
0486         _get_random_bytes(&ret, sizeof(ret));               \
0487         return ret;                         \
0488     }                                   \
0489                                         \
0490     local_lock_irqsave(&batched_entropy_ ##type.lock, flags);       \
0491     batch = raw_cpu_ptr(&batched_entropy_##type);               \
0492                                         \
0493     next_gen = READ_ONCE(base_crng.generation);             \
0494     if (batch->position >= ARRAY_SIZE(batch->entropy) ||            \
0495         next_gen != batch->generation) {                    \
0496         _get_random_bytes(batch->entropy, sizeof(batch->entropy));  \
0497         batch->position = 0;                        \
0498         batch->generation = next_gen;                   \
0499     }                                   \
0500                                         \
0501     ret = batch->entropy[batch->position];                  \
0502     batch->entropy[batch->position] = 0;                    \
0503     ++batch->position;                          \
0504     local_unlock_irqrestore(&batched_entropy_ ##type.lock, flags);      \
0505     return ret;                             \
0506 }                                       \
0507 EXPORT_SYMBOL(get_random_ ##type);
0508 
0509 DEFINE_BATCHED_ENTROPY(u64)
0510 DEFINE_BATCHED_ENTROPY(u32)
0511 
0512 #ifdef CONFIG_SMP
0513 /*
0514  * This function is called when the CPU is coming up, with entry
0515  * CPUHP_RANDOM_PREPARE, which comes before CPUHP_WORKQUEUE_PREP.
0516  */
0517 int __cold random_prepare_cpu(unsigned int cpu)
0518 {
0519     /*
0520      * When the cpu comes back online, immediately invalidate both
0521      * the per-cpu crng and all batches, so that we serve fresh
0522      * randomness.
0523      */
0524     per_cpu_ptr(&crngs, cpu)->generation = ULONG_MAX;
0525     per_cpu_ptr(&batched_entropy_u32, cpu)->position = UINT_MAX;
0526     per_cpu_ptr(&batched_entropy_u64, cpu)->position = UINT_MAX;
0527     return 0;
0528 }
0529 #endif
0530 
0531 
0532 /**********************************************************************
0533  *
0534  * Entropy accumulation and extraction routines.
0535  *
0536  * Callers may add entropy via:
0537  *
0538  *     static void mix_pool_bytes(const void *buf, size_t len)
0539  *
0540  * After which, if added entropy should be credited:
0541  *
0542  *     static void credit_init_bits(size_t bits)
0543  *
0544  * Finally, extract entropy via:
0545  *
0546  *     static void extract_entropy(void *buf, size_t len)
0547  *
0548  **********************************************************************/
0549 
0550 enum {
0551     POOL_BITS = BLAKE2S_HASH_SIZE * 8,
0552     POOL_READY_BITS = POOL_BITS, /* When crng_init->CRNG_READY */
0553     POOL_EARLY_BITS = POOL_READY_BITS / 2 /* When crng_init->CRNG_EARLY */
0554 };
0555 
0556 static struct {
0557     struct blake2s_state hash;
0558     spinlock_t lock;
0559     unsigned int init_bits;
0560 } input_pool = {
0561     .hash.h = { BLAKE2S_IV0 ^ (0x01010000 | BLAKE2S_HASH_SIZE),
0562             BLAKE2S_IV1, BLAKE2S_IV2, BLAKE2S_IV3, BLAKE2S_IV4,
0563             BLAKE2S_IV5, BLAKE2S_IV6, BLAKE2S_IV7 },
0564     .hash.outlen = BLAKE2S_HASH_SIZE,
0565     .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
0566 };
0567 
0568 static void _mix_pool_bytes(const void *buf, size_t len)
0569 {
0570     blake2s_update(&input_pool.hash, buf, len);
0571 }
0572 
0573 /*
0574  * This function adds bytes into the input pool. It does not
0575  * update the initialization bit counter; the caller should call
0576  * credit_init_bits if this is appropriate.
0577  */
0578 static void mix_pool_bytes(const void *buf, size_t len)
0579 {
0580     unsigned long flags;
0581 
0582     spin_lock_irqsave(&input_pool.lock, flags);
0583     _mix_pool_bytes(buf, len);
0584     spin_unlock_irqrestore(&input_pool.lock, flags);
0585 }
0586 
0587 /*
0588  * This is an HKDF-like construction for using the hashed collected entropy
0589  * as a PRF key, that's then expanded block-by-block.
0590  */
0591 static void extract_entropy(void *buf, size_t len)
0592 {
0593     unsigned long flags;
0594     u8 seed[BLAKE2S_HASH_SIZE], next_key[BLAKE2S_HASH_SIZE];
0595     struct {
0596         unsigned long rdseed[32 / sizeof(long)];
0597         size_t counter;
0598     } block;
0599     size_t i, longs;
0600 
0601     for (i = 0; i < ARRAY_SIZE(block.rdseed);) {
0602         longs = arch_get_random_seed_longs(&block.rdseed[i], ARRAY_SIZE(block.rdseed) - i);
0603         if (longs) {
0604             i += longs;
0605             continue;
0606         }
0607         longs = arch_get_random_longs(&block.rdseed[i], ARRAY_SIZE(block.rdseed) - i);
0608         if (longs) {
0609             i += longs;
0610             continue;
0611         }
0612         block.rdseed[i++] = random_get_entropy();
0613     }
0614 
0615     spin_lock_irqsave(&input_pool.lock, flags);
0616 
0617     /* seed = HASHPRF(last_key, entropy_input) */
0618     blake2s_final(&input_pool.hash, seed);
0619 
0620     /* next_key = HASHPRF(seed, RDSEED || 0) */
0621     block.counter = 0;
0622     blake2s(next_key, (u8 *)&block, seed, sizeof(next_key), sizeof(block), sizeof(seed));
0623     blake2s_init_key(&input_pool.hash, BLAKE2S_HASH_SIZE, next_key, sizeof(next_key));
0624 
0625     spin_unlock_irqrestore(&input_pool.lock, flags);
0626     memzero_explicit(next_key, sizeof(next_key));
0627 
0628     while (len) {
0629         i = min_t(size_t, len, BLAKE2S_HASH_SIZE);
0630         /* output = HASHPRF(seed, RDSEED || ++counter) */
0631         ++block.counter;
0632         blake2s(buf, (u8 *)&block, seed, i, sizeof(block), sizeof(seed));
0633         len -= i;
0634         buf += i;
0635     }
0636 
0637     memzero_explicit(seed, sizeof(seed));
0638     memzero_explicit(&block, sizeof(block));
0639 }
0640 
0641 #define credit_init_bits(bits) if (!crng_ready()) _credit_init_bits(bits)
0642 
0643 static void __cold _credit_init_bits(size_t bits)
0644 {
0645     static struct execute_work set_ready;
0646     unsigned int new, orig, add;
0647     unsigned long flags;
0648 
0649     if (!bits)
0650         return;
0651 
0652     add = min_t(size_t, bits, POOL_BITS);
0653 
0654     orig = READ_ONCE(input_pool.init_bits);
0655     do {
0656         new = min_t(unsigned int, POOL_BITS, orig + add);
0657     } while (!try_cmpxchg(&input_pool.init_bits, &orig, new));
0658 
0659     if (orig < POOL_READY_BITS && new >= POOL_READY_BITS) {
0660         crng_reseed(); /* Sets crng_init to CRNG_READY under base_crng.lock. */
0661         if (static_key_initialized)
0662             execute_in_process_context(crng_set_ready, &set_ready);
0663         wake_up_interruptible(&crng_init_wait);
0664         kill_fasync(&fasync, SIGIO, POLL_IN);
0665         pr_notice("crng init done\n");
0666         if (urandom_warning.missed)
0667             pr_notice("%d urandom warning(s) missed due to ratelimiting\n",
0668                   urandom_warning.missed);
0669     } else if (orig < POOL_EARLY_BITS && new >= POOL_EARLY_BITS) {
0670         spin_lock_irqsave(&base_crng.lock, flags);
0671         /* Check if crng_init is CRNG_EMPTY, to avoid race with crng_reseed(). */
0672         if (crng_init == CRNG_EMPTY) {
0673             extract_entropy(base_crng.key, sizeof(base_crng.key));
0674             crng_init = CRNG_EARLY;
0675         }
0676         spin_unlock_irqrestore(&base_crng.lock, flags);
0677     }
0678 }
0679 
0680 
0681 /**********************************************************************
0682  *
0683  * Entropy collection routines.
0684  *
0685  * The following exported functions are used for pushing entropy into
0686  * the above entropy accumulation routines:
0687  *
0688  *  void add_device_randomness(const void *buf, size_t len);
0689  *  void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy);
0690  *  void add_bootloader_randomness(const void *buf, size_t len);
0691  *  void add_vmfork_randomness(const void *unique_vm_id, size_t len);
0692  *  void add_interrupt_randomness(int irq);
0693  *  void add_input_randomness(unsigned int type, unsigned int code, unsigned int value);
0694  *  void add_disk_randomness(struct gendisk *disk);
0695  *
0696  * add_device_randomness() adds data to the input pool that
0697  * is likely to differ between two devices (or possibly even per boot).
0698  * This would be things like MAC addresses or serial numbers, or the
0699  * read-out of the RTC. This does *not* credit any actual entropy to
0700  * the pool, but it initializes the pool to different values for devices
0701  * that might otherwise be identical and have very little entropy
0702  * available to them (particularly common in the embedded world).
0703  *
0704  * add_hwgenerator_randomness() is for true hardware RNGs, and will credit
0705  * entropy as specified by the caller. If the entropy pool is full it will
0706  * block until more entropy is needed.
0707  *
0708  * add_bootloader_randomness() is called by bootloader drivers, such as EFI
0709  * and device tree, and credits its input depending on whether or not the
0710  * configuration option CONFIG_RANDOM_TRUST_BOOTLOADER is set.
0711  *
0712  * add_vmfork_randomness() adds a unique (but not necessarily secret) ID
0713  * representing the current instance of a VM to the pool, without crediting,
0714  * and then force-reseeds the crng so that it takes effect immediately.
0715  *
0716  * add_interrupt_randomness() uses the interrupt timing as random
0717  * inputs to the entropy pool. Using the cycle counters and the irq source
0718  * as inputs, it feeds the input pool roughly once a second or after 64
0719  * interrupts, crediting 1 bit of entropy for whichever comes first.
0720  *
0721  * add_input_randomness() uses the input layer interrupt timing, as well
0722  * as the event type information from the hardware.
0723  *
0724  * add_disk_randomness() uses what amounts to the seek time of block
0725  * layer request events, on a per-disk_devt basis, as input to the
0726  * entropy pool. Note that high-speed solid state drives with very low
0727  * seek times do not make for good sources of entropy, as their seek
0728  * times are usually fairly consistent.
0729  *
0730  * The last two routines try to estimate how many bits of entropy
0731  * to credit. They do this by keeping track of the first and second
0732  * order deltas of the event timings.
0733  *
0734  **********************************************************************/
0735 
0736 static bool trust_cpu __initdata = IS_ENABLED(CONFIG_RANDOM_TRUST_CPU);
0737 static bool trust_bootloader __initdata = IS_ENABLED(CONFIG_RANDOM_TRUST_BOOTLOADER);
0738 static int __init parse_trust_cpu(char *arg)
0739 {
0740     return kstrtobool(arg, &trust_cpu);
0741 }
0742 static int __init parse_trust_bootloader(char *arg)
0743 {
0744     return kstrtobool(arg, &trust_bootloader);
0745 }
0746 early_param("random.trust_cpu", parse_trust_cpu);
0747 early_param("random.trust_bootloader", parse_trust_bootloader);
0748 
0749 static int random_pm_notification(struct notifier_block *nb, unsigned long action, void *data)
0750 {
0751     unsigned long flags, entropy = random_get_entropy();
0752 
0753     /*
0754      * Encode a representation of how long the system has been suspended,
0755      * in a way that is distinct from prior system suspends.
0756      */
0757     ktime_t stamps[] = { ktime_get(), ktime_get_boottime(), ktime_get_real() };
0758 
0759     spin_lock_irqsave(&input_pool.lock, flags);
0760     _mix_pool_bytes(&action, sizeof(action));
0761     _mix_pool_bytes(stamps, sizeof(stamps));
0762     _mix_pool_bytes(&entropy, sizeof(entropy));
0763     spin_unlock_irqrestore(&input_pool.lock, flags);
0764 
0765     if (crng_ready() && (action == PM_RESTORE_PREPARE ||
0766         (action == PM_POST_SUSPEND && !IS_ENABLED(CONFIG_PM_AUTOSLEEP) &&
0767          !IS_ENABLED(CONFIG_PM_USERSPACE_AUTOSLEEP)))) {
0768         crng_reseed();
0769         pr_notice("crng reseeded on system resumption\n");
0770     }
0771     return 0;
0772 }
0773 
0774 static struct notifier_block pm_notifier = { .notifier_call = random_pm_notification };
0775 
0776 /*
0777  * The first collection of entropy occurs at system boot while interrupts
0778  * are still turned off. Here we push in latent entropy, RDSEED, a timestamp,
0779  * utsname(), and the command line. Depending on the above configuration knob,
0780  * RDSEED may be considered sufficient for initialization. Note that much
0781  * earlier setup may already have pushed entropy into the input pool by the
0782  * time we get here.
0783  */
0784 int __init random_init(const char *command_line)
0785 {
0786     ktime_t now = ktime_get_real();
0787     size_t i, longs, arch_bits;
0788     unsigned long entropy[BLAKE2S_BLOCK_SIZE / sizeof(long)];
0789 
0790 #if defined(LATENT_ENTROPY_PLUGIN)
0791     static const u8 compiletime_seed[BLAKE2S_BLOCK_SIZE] __initconst __latent_entropy;
0792     _mix_pool_bytes(compiletime_seed, sizeof(compiletime_seed));
0793 #endif
0794 
0795     for (i = 0, arch_bits = sizeof(entropy) * 8; i < ARRAY_SIZE(entropy);) {
0796         longs = arch_get_random_seed_longs(entropy, ARRAY_SIZE(entropy) - i);
0797         if (longs) {
0798             _mix_pool_bytes(entropy, sizeof(*entropy) * longs);
0799             i += longs;
0800             continue;
0801         }
0802         longs = arch_get_random_longs(entropy, ARRAY_SIZE(entropy) - i);
0803         if (longs) {
0804             _mix_pool_bytes(entropy, sizeof(*entropy) * longs);
0805             i += longs;
0806             continue;
0807         }
0808         entropy[0] = random_get_entropy();
0809         _mix_pool_bytes(entropy, sizeof(*entropy));
0810         arch_bits -= sizeof(*entropy) * 8;
0811         ++i;
0812     }
0813     _mix_pool_bytes(&now, sizeof(now));
0814     _mix_pool_bytes(utsname(), sizeof(*(utsname())));
0815     _mix_pool_bytes(command_line, strlen(command_line));
0816     add_latent_entropy();
0817 
0818     /*
0819      * If we were initialized by the bootloader before jump labels are
0820      * initialized, then we should enable the static branch here, where
0821      * it's guaranteed that jump labels have been initialized.
0822      */
0823     if (!static_branch_likely(&crng_is_ready) && crng_init >= CRNG_READY)
0824         crng_set_ready(NULL);
0825 
0826     if (crng_ready())
0827         crng_reseed();
0828     else if (trust_cpu)
0829         _credit_init_bits(arch_bits);
0830 
0831     WARN_ON(register_pm_notifier(&pm_notifier));
0832 
0833     WARN(!random_get_entropy(), "Missing cycle counter and fallback timer; RNG "
0834                     "entropy collection will consequently suffer.");
0835     return 0;
0836 }
0837 
0838 /*
0839  * Add device- or boot-specific data to the input pool to help
0840  * initialize it.
0841  *
0842  * None of this adds any entropy; it is meant to avoid the problem of
0843  * the entropy pool having similar initial state across largely
0844  * identical devices.
0845  */
0846 void add_device_randomness(const void *buf, size_t len)
0847 {
0848     unsigned long entropy = random_get_entropy();
0849     unsigned long flags;
0850 
0851     spin_lock_irqsave(&input_pool.lock, flags);
0852     _mix_pool_bytes(&entropy, sizeof(entropy));
0853     _mix_pool_bytes(buf, len);
0854     spin_unlock_irqrestore(&input_pool.lock, flags);
0855 }
0856 EXPORT_SYMBOL(add_device_randomness);
0857 
0858 /*
0859  * Interface for in-kernel drivers of true hardware RNGs.
0860  * Those devices may produce endless random bits and will be throttled
0861  * when our pool is full.
0862  */
0863 void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy)
0864 {
0865     mix_pool_bytes(buf, len);
0866     credit_init_bits(entropy);
0867 
0868     /*
0869      * Throttle writing to once every CRNG_RESEED_INTERVAL, unless
0870      * we're not yet initialized.
0871      */
0872     if (!kthread_should_stop() && crng_ready())
0873         schedule_timeout_interruptible(CRNG_RESEED_INTERVAL);
0874 }
0875 EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
0876 
0877 /*
0878  * Handle random seed passed by bootloader, and credit it if
0879  * CONFIG_RANDOM_TRUST_BOOTLOADER is set.
0880  */
0881 void __init add_bootloader_randomness(const void *buf, size_t len)
0882 {
0883     mix_pool_bytes(buf, len);
0884     if (trust_bootloader)
0885         credit_init_bits(len * 8);
0886 }
0887 
0888 #if IS_ENABLED(CONFIG_VMGENID)
0889 static BLOCKING_NOTIFIER_HEAD(vmfork_chain);
0890 
0891 /*
0892  * Handle a new unique VM ID, which is unique, not secret, so we
0893  * don't credit it, but we do immediately force a reseed after so
0894  * that it's used by the crng posthaste.
0895  */
0896 void __cold add_vmfork_randomness(const void *unique_vm_id, size_t len)
0897 {
0898     add_device_randomness(unique_vm_id, len);
0899     if (crng_ready()) {
0900         crng_reseed();
0901         pr_notice("crng reseeded due to virtual machine fork\n");
0902     }
0903     blocking_notifier_call_chain(&vmfork_chain, 0, NULL);
0904 }
0905 #if IS_MODULE(CONFIG_VMGENID)
0906 EXPORT_SYMBOL_GPL(add_vmfork_randomness);
0907 #endif
0908 
0909 int __cold register_random_vmfork_notifier(struct notifier_block *nb)
0910 {
0911     return blocking_notifier_chain_register(&vmfork_chain, nb);
0912 }
0913 EXPORT_SYMBOL_GPL(register_random_vmfork_notifier);
0914 
0915 int __cold unregister_random_vmfork_notifier(struct notifier_block *nb)
0916 {
0917     return blocking_notifier_chain_unregister(&vmfork_chain, nb);
0918 }
0919 EXPORT_SYMBOL_GPL(unregister_random_vmfork_notifier);
0920 #endif
0921 
0922 struct fast_pool {
0923     struct work_struct mix;
0924     unsigned long pool[4];
0925     unsigned long last;
0926     unsigned int count;
0927 };
0928 
0929 static DEFINE_PER_CPU(struct fast_pool, irq_randomness) = {
0930 #ifdef CONFIG_64BIT
0931 #define FASTMIX_PERM SIPHASH_PERMUTATION
0932     .pool = { SIPHASH_CONST_0, SIPHASH_CONST_1, SIPHASH_CONST_2, SIPHASH_CONST_3 }
0933 #else
0934 #define FASTMIX_PERM HSIPHASH_PERMUTATION
0935     .pool = { HSIPHASH_CONST_0, HSIPHASH_CONST_1, HSIPHASH_CONST_2, HSIPHASH_CONST_3 }
0936 #endif
0937 };
0938 
0939 /*
0940  * This is [Half]SipHash-1-x, starting from an empty key. Because
0941  * the key is fixed, it assumes that its inputs are non-malicious,
0942  * and therefore this has no security on its own. s represents the
0943  * four-word SipHash state, while v represents a two-word input.
0944  */
0945 static void fast_mix(unsigned long s[4], unsigned long v1, unsigned long v2)
0946 {
0947     s[3] ^= v1;
0948     FASTMIX_PERM(s[0], s[1], s[2], s[3]);
0949     s[0] ^= v1;
0950     s[3] ^= v2;
0951     FASTMIX_PERM(s[0], s[1], s[2], s[3]);
0952     s[0] ^= v2;
0953 }
0954 
0955 #ifdef CONFIG_SMP
0956 /*
0957  * This function is called when the CPU has just come online, with
0958  * entry CPUHP_AP_RANDOM_ONLINE, just after CPUHP_AP_WORKQUEUE_ONLINE.
0959  */
0960 int __cold random_online_cpu(unsigned int cpu)
0961 {
0962     /*
0963      * During CPU shutdown and before CPU onlining, add_interrupt_
0964      * randomness() may schedule mix_interrupt_randomness(), and
0965      * set the MIX_INFLIGHT flag. However, because the worker can
0966      * be scheduled on a different CPU during this period, that
0967      * flag will never be cleared. For that reason, we zero out
0968      * the flag here, which runs just after workqueues are onlined
0969      * for the CPU again. This also has the effect of setting the
0970      * irq randomness count to zero so that new accumulated irqs
0971      * are fresh.
0972      */
0973     per_cpu_ptr(&irq_randomness, cpu)->count = 0;
0974     return 0;
0975 }
0976 #endif
0977 
0978 static void mix_interrupt_randomness(struct work_struct *work)
0979 {
0980     struct fast_pool *fast_pool = container_of(work, struct fast_pool, mix);
0981     /*
0982      * The size of the copied stack pool is explicitly 2 longs so that we
0983      * only ever ingest half of the siphash output each time, retaining
0984      * the other half as the next "key" that carries over. The entropy is
0985      * supposed to be sufficiently dispersed between bits so on average
0986      * we don't wind up "losing" some.
0987      */
0988     unsigned long pool[2];
0989     unsigned int count;
0990 
0991     /* Check to see if we're running on the wrong CPU due to hotplug. */
0992     local_irq_disable();
0993     if (fast_pool != this_cpu_ptr(&irq_randomness)) {
0994         local_irq_enable();
0995         return;
0996     }
0997 
0998     /*
0999      * Copy the pool to the stack so that the mixer always has a
1000      * consistent view, before we reenable irqs again.
1001      */
1002     memcpy(pool, fast_pool->pool, sizeof(pool));
1003     count = fast_pool->count;
1004     fast_pool->count = 0;
1005     fast_pool->last = jiffies;
1006     local_irq_enable();
1007 
1008     mix_pool_bytes(pool, sizeof(pool));
1009     credit_init_bits(max(1u, (count & U16_MAX) / 64));
1010 
1011     memzero_explicit(pool, sizeof(pool));
1012 }
1013 
1014 void add_interrupt_randomness(int irq)
1015 {
1016     enum { MIX_INFLIGHT = 1U << 31 };
1017     unsigned long entropy = random_get_entropy();
1018     struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1019     struct pt_regs *regs = get_irq_regs();
1020     unsigned int new_count;
1021 
1022     fast_mix(fast_pool->pool, entropy,
1023          (regs ? instruction_pointer(regs) : _RET_IP_) ^ swab(irq));
1024     new_count = ++fast_pool->count;
1025 
1026     if (new_count & MIX_INFLIGHT)
1027         return;
1028 
1029     if (new_count < 1024 && !time_is_before_jiffies(fast_pool->last + HZ))
1030         return;
1031 
1032     if (unlikely(!fast_pool->mix.func))
1033         INIT_WORK(&fast_pool->mix, mix_interrupt_randomness);
1034     fast_pool->count |= MIX_INFLIGHT;
1035     queue_work_on(raw_smp_processor_id(), system_highpri_wq, &fast_pool->mix);
1036 }
1037 EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1038 
1039 /* There is one of these per entropy source */
1040 struct timer_rand_state {
1041     unsigned long last_time;
1042     long last_delta, last_delta2;
1043 };
1044 
1045 /*
1046  * This function adds entropy to the entropy "pool" by using timing
1047  * delays. It uses the timer_rand_state structure to make an estimate
1048  * of how many bits of entropy this call has added to the pool. The
1049  * value "num" is also added to the pool; it should somehow describe
1050  * the type of event that just happened.
1051  */
1052 static void add_timer_randomness(struct timer_rand_state *state, unsigned int num)
1053 {
1054     unsigned long entropy = random_get_entropy(), now = jiffies, flags;
1055     long delta, delta2, delta3;
1056     unsigned int bits;
1057 
1058     /*
1059      * If we're in a hard IRQ, add_interrupt_randomness() will be called
1060      * sometime after, so mix into the fast pool.
1061      */
1062     if (in_hardirq()) {
1063         fast_mix(this_cpu_ptr(&irq_randomness)->pool, entropy, num);
1064     } else {
1065         spin_lock_irqsave(&input_pool.lock, flags);
1066         _mix_pool_bytes(&entropy, sizeof(entropy));
1067         _mix_pool_bytes(&num, sizeof(num));
1068         spin_unlock_irqrestore(&input_pool.lock, flags);
1069     }
1070 
1071     if (crng_ready())
1072         return;
1073 
1074     /*
1075      * Calculate number of bits of randomness we probably added.
1076      * We take into account the first, second and third-order deltas
1077      * in order to make our estimate.
1078      */
1079     delta = now - READ_ONCE(state->last_time);
1080     WRITE_ONCE(state->last_time, now);
1081 
1082     delta2 = delta - READ_ONCE(state->last_delta);
1083     WRITE_ONCE(state->last_delta, delta);
1084 
1085     delta3 = delta2 - READ_ONCE(state->last_delta2);
1086     WRITE_ONCE(state->last_delta2, delta2);
1087 
1088     if (delta < 0)
1089         delta = -delta;
1090     if (delta2 < 0)
1091         delta2 = -delta2;
1092     if (delta3 < 0)
1093         delta3 = -delta3;
1094     if (delta > delta2)
1095         delta = delta2;
1096     if (delta > delta3)
1097         delta = delta3;
1098 
1099     /*
1100      * delta is now minimum absolute delta. Round down by 1 bit
1101      * on general principles, and limit entropy estimate to 11 bits.
1102      */
1103     bits = min(fls(delta >> 1), 11);
1104 
1105     /*
1106      * As mentioned above, if we're in a hard IRQ, add_interrupt_randomness()
1107      * will run after this, which uses a different crediting scheme of 1 bit
1108      * per every 64 interrupts. In order to let that function do accounting
1109      * close to the one in this function, we credit a full 64/64 bit per bit,
1110      * and then subtract one to account for the extra one added.
1111      */
1112     if (in_hardirq())
1113         this_cpu_ptr(&irq_randomness)->count += max(1u, bits * 64) - 1;
1114     else
1115         _credit_init_bits(bits);
1116 }
1117 
1118 void add_input_randomness(unsigned int type, unsigned int code, unsigned int value)
1119 {
1120     static unsigned char last_value;
1121     static struct timer_rand_state input_timer_state = { INITIAL_JIFFIES };
1122 
1123     /* Ignore autorepeat and the like. */
1124     if (value == last_value)
1125         return;
1126 
1127     last_value = value;
1128     add_timer_randomness(&input_timer_state,
1129                  (type << 4) ^ code ^ (code >> 4) ^ value);
1130 }
1131 EXPORT_SYMBOL_GPL(add_input_randomness);
1132 
1133 #ifdef CONFIG_BLOCK
1134 void add_disk_randomness(struct gendisk *disk)
1135 {
1136     if (!disk || !disk->random)
1137         return;
1138     /* First major is 1, so we get >= 0x200 here. */
1139     add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1140 }
1141 EXPORT_SYMBOL_GPL(add_disk_randomness);
1142 
1143 void __cold rand_initialize_disk(struct gendisk *disk)
1144 {
1145     struct timer_rand_state *state;
1146 
1147     /*
1148      * If kzalloc returns null, we just won't use that entropy
1149      * source.
1150      */
1151     state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1152     if (state) {
1153         state->last_time = INITIAL_JIFFIES;
1154         disk->random = state;
1155     }
1156 }
1157 #endif
1158 
1159 struct entropy_timer_state {
1160     unsigned long entropy;
1161     struct timer_list timer;
1162     unsigned int samples, samples_per_bit;
1163 };
1164 
1165 /*
1166  * Each time the timer fires, we expect that we got an unpredictable
1167  * jump in the cycle counter. Even if the timer is running on another
1168  * CPU, the timer activity will be touching the stack of the CPU that is
1169  * generating entropy..
1170  *
1171  * Note that we don't re-arm the timer in the timer itself - we are
1172  * happy to be scheduled away, since that just makes the load more
1173  * complex, but we do not want the timer to keep ticking unless the
1174  * entropy loop is running.
1175  *
1176  * So the re-arming always happens in the entropy loop itself.
1177  */
1178 static void __cold entropy_timer(struct timer_list *timer)
1179 {
1180     struct entropy_timer_state *state = container_of(timer, struct entropy_timer_state, timer);
1181 
1182     if (++state->samples == state->samples_per_bit) {
1183         credit_init_bits(1);
1184         state->samples = 0;
1185     }
1186 }
1187 
1188 /*
1189  * If we have an actual cycle counter, see if we can
1190  * generate enough entropy with timing noise
1191  */
1192 static void __cold try_to_generate_entropy(void)
1193 {
1194     enum { NUM_TRIAL_SAMPLES = 8192, MAX_SAMPLES_PER_BIT = HZ / 30 };
1195     struct entropy_timer_state stack;
1196     unsigned int i, num_different = 0;
1197     unsigned long last = random_get_entropy();
1198 
1199     for (i = 0; i < NUM_TRIAL_SAMPLES - 1; ++i) {
1200         stack.entropy = random_get_entropy();
1201         if (stack.entropy != last)
1202             ++num_different;
1203         last = stack.entropy;
1204     }
1205     stack.samples_per_bit = DIV_ROUND_UP(NUM_TRIAL_SAMPLES, num_different + 1);
1206     if (stack.samples_per_bit > MAX_SAMPLES_PER_BIT)
1207         return;
1208 
1209     stack.samples = 0;
1210     timer_setup_on_stack(&stack.timer, entropy_timer, 0);
1211     while (!crng_ready() && !signal_pending(current)) {
1212         if (!timer_pending(&stack.timer))
1213             mod_timer(&stack.timer, jiffies + 1);
1214         mix_pool_bytes(&stack.entropy, sizeof(stack.entropy));
1215         schedule();
1216         stack.entropy = random_get_entropy();
1217     }
1218 
1219     del_timer_sync(&stack.timer);
1220     destroy_timer_on_stack(&stack.timer);
1221     mix_pool_bytes(&stack.entropy, sizeof(stack.entropy));
1222 }
1223 
1224 
1225 /**********************************************************************
1226  *
1227  * Userspace reader/writer interfaces.
1228  *
1229  * getrandom(2) is the primary modern interface into the RNG and should
1230  * be used in preference to anything else.
1231  *
1232  * Reading from /dev/random has the same functionality as calling
1233  * getrandom(2) with flags=0. In earlier versions, however, it had
1234  * vastly different semantics and should therefore be avoided, to
1235  * prevent backwards compatibility issues.
1236  *
1237  * Reading from /dev/urandom has the same functionality as calling
1238  * getrandom(2) with flags=GRND_INSECURE. Because it does not block
1239  * waiting for the RNG to be ready, it should not be used.
1240  *
1241  * Writing to either /dev/random or /dev/urandom adds entropy to
1242  * the input pool but does not credit it.
1243  *
1244  * Polling on /dev/random indicates when the RNG is initialized, on
1245  * the read side, and when it wants new entropy, on the write side.
1246  *
1247  * Both /dev/random and /dev/urandom have the same set of ioctls for
1248  * adding entropy, getting the entropy count, zeroing the count, and
1249  * reseeding the crng.
1250  *
1251  **********************************************************************/
1252 
1253 SYSCALL_DEFINE3(getrandom, char __user *, ubuf, size_t, len, unsigned int, flags)
1254 {
1255     struct iov_iter iter;
1256     struct iovec iov;
1257     int ret;
1258 
1259     if (flags & ~(GRND_NONBLOCK | GRND_RANDOM | GRND_INSECURE))
1260         return -EINVAL;
1261 
1262     /*
1263      * Requesting insecure and blocking randomness at the same time makes
1264      * no sense.
1265      */
1266     if ((flags & (GRND_INSECURE | GRND_RANDOM)) == (GRND_INSECURE | GRND_RANDOM))
1267         return -EINVAL;
1268 
1269     if (!crng_ready() && !(flags & GRND_INSECURE)) {
1270         if (flags & GRND_NONBLOCK)
1271             return -EAGAIN;
1272         ret = wait_for_random_bytes();
1273         if (unlikely(ret))
1274             return ret;
1275     }
1276 
1277     ret = import_single_range(READ, ubuf, len, &iov, &iter);
1278     if (unlikely(ret))
1279         return ret;
1280     return get_random_bytes_user(&iter);
1281 }
1282 
1283 static __poll_t random_poll(struct file *file, poll_table *wait)
1284 {
1285     poll_wait(file, &crng_init_wait, wait);
1286     return crng_ready() ? EPOLLIN | EPOLLRDNORM : EPOLLOUT | EPOLLWRNORM;
1287 }
1288 
1289 static ssize_t write_pool_user(struct iov_iter *iter)
1290 {
1291     u8 block[BLAKE2S_BLOCK_SIZE];
1292     ssize_t ret = 0;
1293     size_t copied;
1294 
1295     if (unlikely(!iov_iter_count(iter)))
1296         return 0;
1297 
1298     for (;;) {
1299         copied = copy_from_iter(block, sizeof(block), iter);
1300         ret += copied;
1301         mix_pool_bytes(block, copied);
1302         if (!iov_iter_count(iter) || copied != sizeof(block))
1303             break;
1304 
1305         BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0);
1306         if (ret % PAGE_SIZE == 0) {
1307             if (signal_pending(current))
1308                 break;
1309             cond_resched();
1310         }
1311     }
1312 
1313     memzero_explicit(block, sizeof(block));
1314     return ret ? ret : -EFAULT;
1315 }
1316 
1317 static ssize_t random_write_iter(struct kiocb *kiocb, struct iov_iter *iter)
1318 {
1319     return write_pool_user(iter);
1320 }
1321 
1322 static ssize_t urandom_read_iter(struct kiocb *kiocb, struct iov_iter *iter)
1323 {
1324     static int maxwarn = 10;
1325 
1326     /*
1327      * Opportunistically attempt to initialize the RNG on platforms that
1328      * have fast cycle counters, but don't (for now) require it to succeed.
1329      */
1330     if (!crng_ready())
1331         try_to_generate_entropy();
1332 
1333     if (!crng_ready()) {
1334         if (!ratelimit_disable && maxwarn <= 0)
1335             ++urandom_warning.missed;
1336         else if (ratelimit_disable || __ratelimit(&urandom_warning)) {
1337             --maxwarn;
1338             pr_notice("%s: uninitialized urandom read (%zu bytes read)\n",
1339                   current->comm, iov_iter_count(iter));
1340         }
1341     }
1342 
1343     return get_random_bytes_user(iter);
1344 }
1345 
1346 static ssize_t random_read_iter(struct kiocb *kiocb, struct iov_iter *iter)
1347 {
1348     int ret;
1349 
1350     ret = wait_for_random_bytes();
1351     if (ret != 0)
1352         return ret;
1353     return get_random_bytes_user(iter);
1354 }
1355 
1356 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1357 {
1358     int __user *p = (int __user *)arg;
1359     int ent_count;
1360 
1361     switch (cmd) {
1362     case RNDGETENTCNT:
1363         /* Inherently racy, no point locking. */
1364         if (put_user(input_pool.init_bits, p))
1365             return -EFAULT;
1366         return 0;
1367     case RNDADDTOENTCNT:
1368         if (!capable(CAP_SYS_ADMIN))
1369             return -EPERM;
1370         if (get_user(ent_count, p))
1371             return -EFAULT;
1372         if (ent_count < 0)
1373             return -EINVAL;
1374         credit_init_bits(ent_count);
1375         return 0;
1376     case RNDADDENTROPY: {
1377         struct iov_iter iter;
1378         struct iovec iov;
1379         ssize_t ret;
1380         int len;
1381 
1382         if (!capable(CAP_SYS_ADMIN))
1383             return -EPERM;
1384         if (get_user(ent_count, p++))
1385             return -EFAULT;
1386         if (ent_count < 0)
1387             return -EINVAL;
1388         if (get_user(len, p++))
1389             return -EFAULT;
1390         ret = import_single_range(WRITE, p, len, &iov, &iter);
1391         if (unlikely(ret))
1392             return ret;
1393         ret = write_pool_user(&iter);
1394         if (unlikely(ret < 0))
1395             return ret;
1396         /* Since we're crediting, enforce that it was all written into the pool. */
1397         if (unlikely(ret != len))
1398             return -EFAULT;
1399         credit_init_bits(ent_count);
1400         return 0;
1401     }
1402     case RNDZAPENTCNT:
1403     case RNDCLEARPOOL:
1404         /* No longer has any effect. */
1405         if (!capable(CAP_SYS_ADMIN))
1406             return -EPERM;
1407         return 0;
1408     case RNDRESEEDCRNG:
1409         if (!capable(CAP_SYS_ADMIN))
1410             return -EPERM;
1411         if (!crng_ready())
1412             return -ENODATA;
1413         crng_reseed();
1414         return 0;
1415     default:
1416         return -EINVAL;
1417     }
1418 }
1419 
1420 static int random_fasync(int fd, struct file *filp, int on)
1421 {
1422     return fasync_helper(fd, filp, on, &fasync);
1423 }
1424 
1425 const struct file_operations random_fops = {
1426     .read_iter = random_read_iter,
1427     .write_iter = random_write_iter,
1428     .poll = random_poll,
1429     .unlocked_ioctl = random_ioctl,
1430     .compat_ioctl = compat_ptr_ioctl,
1431     .fasync = random_fasync,
1432     .llseek = noop_llseek,
1433     .splice_read = generic_file_splice_read,
1434     .splice_write = iter_file_splice_write,
1435 };
1436 
1437 const struct file_operations urandom_fops = {
1438     .read_iter = urandom_read_iter,
1439     .write_iter = random_write_iter,
1440     .unlocked_ioctl = random_ioctl,
1441     .compat_ioctl = compat_ptr_ioctl,
1442     .fasync = random_fasync,
1443     .llseek = noop_llseek,
1444     .splice_read = generic_file_splice_read,
1445     .splice_write = iter_file_splice_write,
1446 };
1447 
1448 
1449 /********************************************************************
1450  *
1451  * Sysctl interface.
1452  *
1453  * These are partly unused legacy knobs with dummy values to not break
1454  * userspace and partly still useful things. They are usually accessible
1455  * in /proc/sys/kernel/random/ and are as follows:
1456  *
1457  * - boot_id - a UUID representing the current boot.
1458  *
1459  * - uuid - a random UUID, different each time the file is read.
1460  *
1461  * - poolsize - the number of bits of entropy that the input pool can
1462  *   hold, tied to the POOL_BITS constant.
1463  *
1464  * - entropy_avail - the number of bits of entropy currently in the
1465  *   input pool. Always <= poolsize.
1466  *
1467  * - write_wakeup_threshold - the amount of entropy in the input pool
1468  *   below which write polls to /dev/random will unblock, requesting
1469  *   more entropy, tied to the POOL_READY_BITS constant. It is writable
1470  *   to avoid breaking old userspaces, but writing to it does not
1471  *   change any behavior of the RNG.
1472  *
1473  * - urandom_min_reseed_secs - fixed to the value CRNG_RESEED_INTERVAL.
1474  *   It is writable to avoid breaking old userspaces, but writing
1475  *   to it does not change any behavior of the RNG.
1476  *
1477  ********************************************************************/
1478 
1479 #ifdef CONFIG_SYSCTL
1480 
1481 #include <linux/sysctl.h>
1482 
1483 static int sysctl_random_min_urandom_seed = CRNG_RESEED_INTERVAL / HZ;
1484 static int sysctl_random_write_wakeup_bits = POOL_READY_BITS;
1485 static int sysctl_poolsize = POOL_BITS;
1486 static u8 sysctl_bootid[UUID_SIZE];
1487 
1488 /*
1489  * This function is used to return both the bootid UUID, and random
1490  * UUID. The difference is in whether table->data is NULL; if it is,
1491  * then a new UUID is generated and returned to the user.
1492  */
1493 static int proc_do_uuid(struct ctl_table *table, int write, void *buf,
1494             size_t *lenp, loff_t *ppos)
1495 {
1496     u8 tmp_uuid[UUID_SIZE], *uuid;
1497     char uuid_string[UUID_STRING_LEN + 1];
1498     struct ctl_table fake_table = {
1499         .data = uuid_string,
1500         .maxlen = UUID_STRING_LEN
1501     };
1502 
1503     if (write)
1504         return -EPERM;
1505 
1506     uuid = table->data;
1507     if (!uuid) {
1508         uuid = tmp_uuid;
1509         generate_random_uuid(uuid);
1510     } else {
1511         static DEFINE_SPINLOCK(bootid_spinlock);
1512 
1513         spin_lock(&bootid_spinlock);
1514         if (!uuid[8])
1515             generate_random_uuid(uuid);
1516         spin_unlock(&bootid_spinlock);
1517     }
1518 
1519     snprintf(uuid_string, sizeof(uuid_string), "%pU", uuid);
1520     return proc_dostring(&fake_table, 0, buf, lenp, ppos);
1521 }
1522 
1523 /* The same as proc_dointvec, but writes don't change anything. */
1524 static int proc_do_rointvec(struct ctl_table *table, int write, void *buf,
1525                 size_t *lenp, loff_t *ppos)
1526 {
1527     return write ? 0 : proc_dointvec(table, 0, buf, lenp, ppos);
1528 }
1529 
1530 static struct ctl_table random_table[] = {
1531     {
1532         .procname   = "poolsize",
1533         .data       = &sysctl_poolsize,
1534         .maxlen     = sizeof(int),
1535         .mode       = 0444,
1536         .proc_handler   = proc_dointvec,
1537     },
1538     {
1539         .procname   = "entropy_avail",
1540         .data       = &input_pool.init_bits,
1541         .maxlen     = sizeof(int),
1542         .mode       = 0444,
1543         .proc_handler   = proc_dointvec,
1544     },
1545     {
1546         .procname   = "write_wakeup_threshold",
1547         .data       = &sysctl_random_write_wakeup_bits,
1548         .maxlen     = sizeof(int),
1549         .mode       = 0644,
1550         .proc_handler   = proc_do_rointvec,
1551     },
1552     {
1553         .procname   = "urandom_min_reseed_secs",
1554         .data       = &sysctl_random_min_urandom_seed,
1555         .maxlen     = sizeof(int),
1556         .mode       = 0644,
1557         .proc_handler   = proc_do_rointvec,
1558     },
1559     {
1560         .procname   = "boot_id",
1561         .data       = &sysctl_bootid,
1562         .mode       = 0444,
1563         .proc_handler   = proc_do_uuid,
1564     },
1565     {
1566         .procname   = "uuid",
1567         .mode       = 0444,
1568         .proc_handler   = proc_do_uuid,
1569     },
1570     { }
1571 };
1572 
1573 /*
1574  * random_init() is called before sysctl_init(),
1575  * so we cannot call register_sysctl_init() in random_init()
1576  */
1577 static int __init random_sysctls_init(void)
1578 {
1579     register_sysctl_init("kernel/random", random_table);
1580     return 0;
1581 }
1582 device_initcall(random_sysctls_init);
1583 #endif