Commit | Line | Data |
---|---|---|
a07fdae3 | 1 | // SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause) |
1da177e4 | 2 | /* |
9f9eff85 | 3 | * Copyright (C) 2017-2022 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved. |
9e95ce27 | 4 | * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005 |
5f75d9f3 JD |
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 | |
e85c0fc1 JD |
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. | |
1da177e4 LT |
24 | */ |
25 | ||
12cd53af YL |
26 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
27 | ||
1da177e4 | 28 | #include <linux/utsname.h> |
1da177e4 LT |
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> | |
322cbb50 | 39 | #include <linux/blkdev.h> |
1da177e4 | 40 | #include <linux/interrupt.h> |
27ac792c | 41 | #include <linux/mm.h> |
dd0f0cf5 | 42 | #include <linux/nodemask.h> |
1da177e4 | 43 | #include <linux/spinlock.h> |
c84dbf61 | 44 | #include <linux/kthread.h> |
1da177e4 | 45 | #include <linux/percpu.h> |
775f4b29 | 46 | #include <linux/ptrace.h> |
6265e169 | 47 | #include <linux/workqueue.h> |
0244ad00 | 48 | #include <linux/irq.h> |
4e00b339 | 49 | #include <linux/ratelimit.h> |
c6e9d6f3 TT |
50 | #include <linux/syscalls.h> |
51 | #include <linux/completion.h> | |
8da4b8c4 | 52 | #include <linux/uuid.h> |
87e7d5ab | 53 | #include <linux/uaccess.h> |
b7b67d13 | 54 | #include <linux/suspend.h> |
e73aaae2 | 55 | #include <linux/siphash.h> |
1c21fe00 | 56 | #include <linux/sched/isolation.h> |
1ca1b917 | 57 | #include <crypto/chacha.h> |
9f9eff85 | 58 | #include <crypto/blake2s.h> |
6bb20c15 | 59 | #include <asm/archrandom.h> |
1da177e4 | 60 | #include <asm/processor.h> |
1da177e4 | 61 | #include <asm/irq.h> |
775f4b29 | 62 | #include <asm/irq_regs.h> |
1da177e4 LT |
63 | #include <asm/io.h> |
64 | ||
5f1bb112 JD |
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 | *********************************************************************/ | |
205a525c | 74 | |
e192be9d | 75 | /* |
5f1bb112 | 76 | * crng_init is protected by base_crng->lock, and only increases |
e3d2c5e7 | 77 | * its value (from empty->early->ready). |
e192be9d | 78 | */ |
e3d2c5e7 JD |
79 | static 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 */ | |
f5bda35f JD |
83 | } crng_init __read_mostly = CRNG_EMPTY; |
84 | static DEFINE_STATIC_KEY_FALSE(crng_is_ready); | |
85 | #define crng_ready() (static_branch_likely(&crng_is_ready) || crng_init >= CRNG_READY) | |
e3d2c5e7 | 86 | /* Various types of waiters for crng_init->CRNG_READY transition. */ |
5f1bb112 JD |
87 | static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait); |
88 | static struct fasync_struct *fasync; | |
bbc7e1be | 89 | static ATOMIC_NOTIFIER_HEAD(random_ready_notifier); |
e192be9d | 90 | |
5f1bb112 | 91 | /* Control how we warn userspace. */ |
0313bc27 | 92 | static struct ratelimit_state urandom_warning = |
c01d4d0a | 93 | RATELIMIT_STATE_INIT_FLAGS("urandom_warning", HZ, 3, RATELIMIT_MSG_ON_RELEASE); |
cc1e127b JD |
94 | static int ratelimit_disable __read_mostly = |
95 | IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM); | |
4e00b339 TT |
96 | module_param_named(ratelimit_disable, ratelimit_disable, int, 0644); |
97 | MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression"); | |
98 | ||
5f1bb112 JD |
99 | /* |
100 | * Returns whether or not the input pool has been seeded and thus guaranteed | |
0313bc27 | 101 | * to supply cryptographically secure random numbers. This applies to: the |
a890d1c6 | 102 | * /dev/urandom device, the get_random_bytes function, and the get_random_{u8, |
de492c83 | 103 | * u16,u32,u64,long} family of functions. |
5f1bb112 JD |
104 | * |
105 | * Returns: true if the input pool has been seeded. | |
106 | * false if the input pool has not been seeded. | |
107 | */ | |
108 | bool rng_is_initialized(void) | |
109 | { | |
110 | return crng_ready(); | |
111 | } | |
112 | EXPORT_SYMBOL(rng_is_initialized); | |
113 | ||
560181c2 | 114 | static void __cold crng_set_ready(struct work_struct *work) |
f5bda35f JD |
115 | { |
116 | static_branch_enable(&crng_is_ready); | |
117 | } | |
118 | ||
5f1bb112 JD |
119 | /* Used by wait_for_random_bytes(), and considered an entropy collector, below. */ |
120 | static void try_to_generate_entropy(void); | |
121 | ||
122 | /* | |
123 | * Wait for the input pool to be seeded and thus guaranteed to supply | |
0313bc27 | 124 | * cryptographically secure random numbers. This applies to: the /dev/urandom |
a890d1c6 | 125 | * device, the get_random_bytes function, and the get_random_{u8,u16,u32,u64, |
b240bab5 | 126 | * long} family of functions. Using any of these functions without first |
a890d1c6 | 127 | * calling this function forfeits the guarantee of security. |
5f1bb112 JD |
128 | * |
129 | * Returns: 0 if the input pool has been seeded. | |
130 | * -ERESTARTSYS if the function was interrupted by a signal. | |
131 | */ | |
132 | int wait_for_random_bytes(void) | |
133 | { | |
a96cfe2d | 134 | while (!crng_ready()) { |
5f1bb112 | 135 | int ret; |
3e504d20 JD |
136 | |
137 | try_to_generate_entropy(); | |
5f1bb112 JD |
138 | ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ); |
139 | if (ret) | |
140 | return ret > 0 ? 0 : ret; | |
a96cfe2d | 141 | } |
5f1bb112 JD |
142 | return 0; |
143 | } | |
144 | EXPORT_SYMBOL(wait_for_random_bytes); | |
145 | ||
bbc7e1be JD |
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 | */ | |
152 | int __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 | ||
cc1e127b | 166 | #define warn_unseeded_randomness() \ |
560181c2 JD |
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) | |
5f1bb112 JD |
170 | |
171 | ||
3655adc7 | 172 | /********************************************************************* |
1da177e4 | 173 | * |
3655adc7 | 174 | * Fast key erasure RNG, the "crng". |
1da177e4 | 175 | * |
3655adc7 JD |
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>. | |
e192be9d | 179 | * |
3655adc7 JD |
180 | * There are a few exported interfaces for use by other drivers: |
181 | * | |
a1940263 | 182 | * void get_random_bytes(void *buf, size_t len) |
a890d1c6 JD |
183 | * u8 get_random_u8() |
184 | * u16 get_random_u16() | |
3655adc7 | 185 | * u32 get_random_u32() |
e9a688bc | 186 | * u32 get_random_u32_below(u32 ceil) |
7f576b25 JD |
187 | * u32 get_random_u32_above(u32 floor) |
188 | * u32 get_random_u32_inclusive(u32 floor, u32 ceil) | |
3655adc7 | 189 | * u64 get_random_u64() |
3655adc7 JD |
190 | * unsigned long get_random_long() |
191 | * | |
192 | * These interfaces will return the requested number of random bytes | |
0313bc27 | 193 | * into the given buffer or as a return value. This is equivalent to |
de492c83 JD |
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. | |
e192be9d TT |
198 | * |
199 | *********************************************************************/ | |
200 | ||
e85c0fc1 JD |
201 | enum { |
202 | CRNG_RESEED_START_INTERVAL = HZ, | |
203 | CRNG_RESEED_INTERVAL = 60 * HZ | |
204 | }; | |
186873c5 JD |
205 | |
206 | static struct { | |
207 | u8 key[CHACHA_KEY_SIZE] __aligned(__alignof__(long)); | |
186873c5 JD |
208 | unsigned long generation; |
209 | spinlock_t lock; | |
210 | } base_crng = { | |
211 | .lock = __SPIN_LOCK_UNLOCKED(base_crng.lock) | |
212 | }; | |
213 | ||
214 | struct crng { | |
215 | u8 key[CHACHA_KEY_SIZE]; | |
216 | unsigned long generation; | |
217 | local_lock_t lock; | |
218 | }; | |
219 | ||
220 | static DEFINE_PER_CPU(struct crng, crngs) = { | |
221 | .generation = ULONG_MAX, | |
222 | .lock = INIT_LOCAL_LOCK(crngs.lock), | |
223 | }; | |
e192be9d | 224 | |
9148de31 JD |
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 | */ | |
230 | static 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 | ||
e85c0fc1 | 245 | /* Used by crng_reseed() and crng_make_state() to extract a new seed from the input pool. */ |
a1940263 | 246 | static void extract_entropy(void *buf, size_t len); |
e192be9d | 247 | |
e85c0fc1 | 248 | /* This extracts a new crng key from the input pool. */ |
9148de31 | 249 | static void crng_reseed(struct work_struct *work) |
e192be9d | 250 | { |
9148de31 | 251 | static DECLARE_DELAYED_WORK(next_reseed, crng_reseed); |
248045b8 | 252 | unsigned long flags; |
186873c5 JD |
253 | unsigned long next_gen; |
254 | u8 key[CHACHA_KEY_SIZE]; | |
e192be9d | 255 | |
9148de31 JD |
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 | ||
e85c0fc1 | 260 | extract_entropy(key, sizeof(key)); |
a9412d51 | 261 | |
186873c5 JD |
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); | |
f5bda35f | 274 | if (!static_branch_likely(&crng_is_ready)) |
e3d2c5e7 | 275 | crng_init = CRNG_READY; |
7191c628 DB |
276 | spin_unlock_irqrestore(&base_crng.lock, flags); |
277 | memzero_explicit(key, sizeof(key)); | |
e192be9d TT |
278 | } |
279 | ||
186873c5 | 280 | /* |
3655adc7 | 281 | * This generates a ChaCha block using the provided key, and then |
7f637be4 | 282 | * immediately overwrites that key with half the block. It returns |
3655adc7 JD |
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. | |
8717627d JD |
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. | |
186873c5 JD |
293 | */ |
294 | static 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) | |
e192be9d | 297 | { |
186873c5 | 298 | u8 first_block[CHACHA_BLOCK_SIZE]; |
009ba856 | 299 | |
186873c5 JD |
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); | |
8717627d | 308 | memcpy(random_data, first_block + CHACHA_KEY_SIZE, random_data_len); |
186873c5 | 309 | memzero_explicit(first_block, sizeof(first_block)); |
1e7f583a TT |
310 | } |
311 | ||
c92e040d | 312 | /* |
186873c5 JD |
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. | |
c92e040d | 316 | */ |
186873c5 JD |
317 | static void crng_make_state(u32 chacha_state[CHACHA_STATE_WORDS], |
318 | u8 *random_data, size_t random_data_len) | |
c92e040d | 319 | { |
248045b8 | 320 | unsigned long flags; |
186873c5 | 321 | struct crng *crng; |
c92e040d | 322 | |
186873c5 JD |
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 | |
5c3b747e | 328 | * ready, we do fast key erasure with the base_crng directly, extracting |
e3d2c5e7 | 329 | * when crng_init is CRNG_EMPTY. |
186873c5 | 330 | */ |
a96cfe2d | 331 | if (!crng_ready()) { |
186873c5 JD |
332 | bool ready; |
333 | ||
334 | spin_lock_irqsave(&base_crng.lock, flags); | |
335 | ready = crng_ready(); | |
5c3b747e | 336 | if (!ready) { |
e3d2c5e7 | 337 | if (crng_init == CRNG_EMPTY) |
5c3b747e | 338 | extract_entropy(base_crng.key, sizeof(base_crng.key)); |
186873c5 JD |
339 | crng_fast_key_erasure(base_crng.key, chacha_state, |
340 | random_data, random_data_len); | |
5c3b747e | 341 | } |
186873c5 JD |
342 | spin_unlock_irqrestore(&base_crng.lock, flags); |
343 | if (!ready) | |
344 | return; | |
c92e040d | 345 | } |
186873c5 | 346 | |
186873c5 JD |
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); | |
c92e040d TT |
373 | } |
374 | ||
a1940263 | 375 | static void _get_random_bytes(void *buf, size_t len) |
e192be9d | 376 | { |
186873c5 | 377 | u32 chacha_state[CHACHA_STATE_WORDS]; |
3655adc7 | 378 | u8 tmp[CHACHA_BLOCK_SIZE]; |
a1940263 | 379 | size_t first_block_len; |
3655adc7 | 380 | |
a1940263 | 381 | if (!len) |
3655adc7 JD |
382 | return; |
383 | ||
a1940263 JD |
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; | |
3655adc7 | 388 | |
a1940263 JD |
389 | while (len) { |
390 | if (len < CHACHA_BLOCK_SIZE) { | |
3655adc7 | 391 | chacha20_block(chacha_state, tmp); |
a1940263 | 392 | memcpy(buf, tmp, len); |
3655adc7 JD |
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]; | |
a1940263 | 400 | len -= CHACHA_BLOCK_SIZE; |
3655adc7 JD |
401 | buf += CHACHA_BLOCK_SIZE; |
402 | } | |
403 | ||
404 | memzero_explicit(chacha_state, sizeof(chacha_state)); | |
405 | } | |
406 | ||
407 | /* | |
19258d05 JD |
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. | |
3655adc7 | 413 | */ |
a1940263 | 414 | void get_random_bytes(void *buf, size_t len) |
3655adc7 | 415 | { |
cc1e127b | 416 | warn_unseeded_randomness(); |
a1940263 | 417 | _get_random_bytes(buf, len); |
3655adc7 JD |
418 | } |
419 | EXPORT_SYMBOL(get_random_bytes); | |
420 | ||
1b388e77 | 421 | static ssize_t get_random_bytes_user(struct iov_iter *iter) |
3655adc7 | 422 | { |
3655adc7 | 423 | u32 chacha_state[CHACHA_STATE_WORDS]; |
1b388e77 JA |
424 | u8 block[CHACHA_BLOCK_SIZE]; |
425 | size_t ret = 0, copied; | |
3655adc7 | 426 | |
1b388e77 | 427 | if (unlikely(!iov_iter_count(iter))) |
3655adc7 JD |
428 | return 0; |
429 | ||
aba120cc JD |
430 | /* |
431 | * Immediately overwrite the ChaCha key at index 4 with random | |
63b8ea5e | 432 | * bytes, in case userspace causes copy_to_iter() below to sleep |
aba120cc JD |
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 | */ | |
1b388e77 JA |
441 | if (iov_iter_count(iter) <= CHACHA_KEY_SIZE) { |
442 | ret = copy_to_iter(&chacha_state[4], CHACHA_KEY_SIZE, iter); | |
aba120cc JD |
443 | goto out_zero_chacha; |
444 | } | |
3655adc7 | 445 | |
5209aed5 | 446 | for (;;) { |
1b388e77 | 447 | chacha20_block(chacha_state, block); |
3655adc7 JD |
448 | if (unlikely(chacha_state[12] == 0)) |
449 | ++chacha_state[13]; | |
450 | ||
1b388e77 JA |
451 | copied = copy_to_iter(block, sizeof(block), iter); |
452 | ret += copied; | |
453 | if (!iov_iter_count(iter) || copied != sizeof(block)) | |
5209aed5 | 454 | break; |
e3c1c4fd | 455 | |
1b388e77 | 456 | BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0); |
5209aed5 | 457 | if (ret % PAGE_SIZE == 0) { |
e3c1c4fd JD |
458 | if (signal_pending(current)) |
459 | break; | |
460 | cond_resched(); | |
461 | } | |
5209aed5 | 462 | } |
3655adc7 | 463 | |
1b388e77 | 464 | memzero_explicit(block, sizeof(block)); |
aba120cc JD |
465 | out_zero_chacha: |
466 | memzero_explicit(chacha_state, sizeof(chacha_state)); | |
5209aed5 | 467 | return ret ? ret : -EFAULT; |
3655adc7 JD |
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 | */ | |
3655adc7 | 476 | |
3092adce JD |
477 | #define DEFINE_BATCHED_ENTROPY(type) \ |
478 | struct 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 | \ | |
492 | static DEFINE_PER_CPU(struct batch_ ##type, batched_entropy_ ##type) = { \ | |
493 | .lock = INIT_LOCAL_LOCK(batched_entropy_ ##type.lock), \ | |
494 | .position = UINT_MAX \ | |
495 | }; \ | |
496 | \ | |
497 | type 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 | } \ | |
528 | EXPORT_SYMBOL(get_random_ ##type); | |
529 | ||
585cd5fe | 530 | DEFINE_BATCHED_ENTROPY(u8) |
a890d1c6 JD |
531 | DEFINE_BATCHED_ENTROPY(u16) |
532 | DEFINE_BATCHED_ENTROPY(u32) | |
533 | DEFINE_BATCHED_ENTROPY(u64) | |
3655adc7 | 534 | |
e9a688bc JD |
535 | u32 __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 | */ | |
7f576b25 JD |
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; | |
e9a688bc JD |
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 | } | |
568 | EXPORT_SYMBOL(__get_random_u32_below); | |
569 | ||
3191dd5a JD |
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 | */ | |
560181c2 | 575 | int __cold random_prepare_cpu(unsigned int cpu) |
3191dd5a JD |
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; | |
a890d1c6 JD |
583 | per_cpu_ptr(&batched_entropy_u8, cpu)->position = UINT_MAX; |
584 | per_cpu_ptr(&batched_entropy_u16, cpu)->position = UINT_MAX; | |
3191dd5a JD |
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 | ||
a5ed7cb1 JD |
591 | |
592 | /********************************************************************** | |
593 | * | |
594 | * Entropy accumulation and extraction routines. | |
595 | * | |
596 | * Callers may add entropy via: | |
597 | * | |
a1940263 | 598 | * static void mix_pool_bytes(const void *buf, size_t len) |
a5ed7cb1 JD |
599 | * |
600 | * After which, if added entropy should be credited: | |
601 | * | |
a1940263 | 602 | * static void credit_init_bits(size_t bits) |
a5ed7cb1 | 603 | * |
e85c0fc1 | 604 | * Finally, extract entropy via: |
a5ed7cb1 | 605 | * |
a1940263 | 606 | * static void extract_entropy(void *buf, size_t len) |
a5ed7cb1 JD |
607 | * |
608 | **********************************************************************/ | |
609 | ||
3655adc7 JD |
610 | enum { |
611 | POOL_BITS = BLAKE2S_HASH_SIZE * 8, | |
e3d2c5e7 JD |
612 | POOL_READY_BITS = POOL_BITS, /* When crng_init->CRNG_READY */ |
613 | POOL_EARLY_BITS = POOL_READY_BITS / 2 /* When crng_init->CRNG_EARLY */ | |
3655adc7 JD |
614 | }; |
615 | ||
3655adc7 JD |
616 | static struct { |
617 | struct blake2s_state hash; | |
618 | spinlock_t lock; | |
e85c0fc1 | 619 | unsigned int init_bits; |
3655adc7 JD |
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 | ||
a1940263 | 628 | static void _mix_pool_bytes(const void *buf, size_t len) |
a5ed7cb1 | 629 | { |
a1940263 | 630 | blake2s_update(&input_pool.hash, buf, len); |
a5ed7cb1 | 631 | } |
3655adc7 JD |
632 | |
633 | /* | |
e85c0fc1 JD |
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. | |
3655adc7 | 637 | */ |
a1940263 | 638 | static void mix_pool_bytes(const void *buf, size_t len) |
3655adc7 | 639 | { |
a5ed7cb1 JD |
640 | unsigned long flags; |
641 | ||
642 | spin_lock_irqsave(&input_pool.lock, flags); | |
a1940263 | 643 | _mix_pool_bytes(buf, len); |
a5ed7cb1 | 644 | spin_unlock_irqrestore(&input_pool.lock, flags); |
3655adc7 JD |
645 | } |
646 | ||
a5ed7cb1 JD |
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 | */ | |
a1940263 | 651 | static void extract_entropy(void *buf, size_t len) |
3655adc7 JD |
652 | { |
653 | unsigned long flags; | |
a5ed7cb1 JD |
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; | |
d349ab99 | 659 | size_t i, longs; |
a5ed7cb1 | 660 | |
d349ab99 JD |
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(); | |
a5ed7cb1 | 673 | } |
3655adc7 JD |
674 | |
675 | spin_lock_irqsave(&input_pool.lock, flags); | |
a5ed7cb1 JD |
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 | ||
3655adc7 | 685 | spin_unlock_irqrestore(&input_pool.lock, flags); |
a5ed7cb1 JD |
686 | memzero_explicit(next_key, sizeof(next_key)); |
687 | ||
a1940263 JD |
688 | while (len) { |
689 | i = min_t(size_t, len, BLAKE2S_HASH_SIZE); | |
a5ed7cb1 JD |
690 | /* output = HASHPRF(seed, RDSEED || ++counter) */ |
691 | ++block.counter; | |
692 | blake2s(buf, (u8 *)&block, seed, i, sizeof(block), sizeof(seed)); | |
a1940263 | 693 | len -= i; |
a5ed7cb1 JD |
694 | buf += i; |
695 | } | |
696 | ||
697 | memzero_explicit(seed, sizeof(seed)); | |
698 | memzero_explicit(&block, sizeof(block)); | |
699 | } | |
700 | ||
560181c2 JD |
701 | #define credit_init_bits(bits) if (!crng_ready()) _credit_init_bits(bits) |
702 | ||
703 | static void __cold _credit_init_bits(size_t bits) | |
5c3b747e | 704 | { |
f5bda35f | 705 | static struct execute_work set_ready; |
fed7ef06 | 706 | unsigned int new, orig, add; |
5c3b747e JD |
707 | unsigned long flags; |
708 | ||
560181c2 | 709 | if (!bits) |
5c3b747e JD |
710 | return; |
711 | ||
a1940263 | 712 | add = min_t(size_t, bits, POOL_BITS); |
5c3b747e | 713 | |
b7a68f67 | 714 | orig = READ_ONCE(input_pool.init_bits); |
5c3b747e | 715 | do { |
fed7ef06 | 716 | new = min_t(unsigned int, POOL_BITS, orig + add); |
b7a68f67 | 717 | } while (!try_cmpxchg(&input_pool.init_bits, &orig, new)); |
5c3b747e | 718 | |
68c9c8b1 | 719 | if (orig < POOL_READY_BITS && new >= POOL_READY_BITS) { |
9148de31 | 720 | crng_reseed(NULL); /* Sets crng_init to CRNG_READY under base_crng.lock. */ |
60e5b288 JD |
721 | if (static_key_initialized) |
722 | execute_in_process_context(crng_set_ready, &set_ready); | |
bbc7e1be | 723 | atomic_notifier_call_chain(&random_ready_notifier, 0, NULL); |
68c9c8b1 JD |
724 | wake_up_interruptible(&crng_init_wait); |
725 | kill_fasync(&fasync, SIGIO, POLL_IN); | |
726 | pr_notice("crng init done\n"); | |
cc1e127b | 727 | if (urandom_warning.missed) |
68c9c8b1 JD |
728 | pr_notice("%d urandom warning(s) missed due to ratelimiting\n", |
729 | urandom_warning.missed); | |
68c9c8b1 | 730 | } else if (orig < POOL_EARLY_BITS && new >= POOL_EARLY_BITS) { |
5c3b747e | 731 | spin_lock_irqsave(&base_crng.lock, flags); |
68c9c8b1 | 732 | /* Check if crng_init is CRNG_EMPTY, to avoid race with crng_reseed(). */ |
e3d2c5e7 | 733 | if (crng_init == CRNG_EMPTY) { |
5c3b747e | 734 | extract_entropy(base_crng.key, sizeof(base_crng.key)); |
e3d2c5e7 | 735 | crng_init = CRNG_EARLY; |
5c3b747e JD |
736 | } |
737 | spin_unlock_irqrestore(&base_crng.lock, flags); | |
738 | } | |
739 | } | |
740 | ||
92c653cf JD |
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 | * | |
a1940263 | 749 | * void add_device_randomness(const void *buf, size_t len); |
db516da9 | 750 | * void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy, bool sleep_after); |
a1940263 JD |
751 | * void add_bootloader_randomness(const void *buf, size_t len); |
752 | * void add_vmfork_randomness(const void *unique_vm_id, size_t len); | |
92c653cf | 753 | * void add_interrupt_randomness(int irq); |
a1940263 | 754 | * void add_input_randomness(unsigned int type, unsigned int code, unsigned int value); |
a4b5c26b | 755 | * void add_disk_randomness(struct gendisk *disk); |
92c653cf JD |
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 | * | |
92c653cf JD |
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 | * | |
5c3b747e JD |
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 | |
b9b01a56 | 771 | * command line option 'random.trust_bootloader'. |
92c653cf | 772 | * |
ae099e8e JD |
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 | * | |
92c653cf JD |
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 | * | |
a4b5c26b JD |
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 | * | |
92c653cf JD |
795 | **********************************************************************/ |
796 | ||
b9b01a56 JD |
797 | static bool trust_cpu __initdata = true; |
798 | static bool trust_bootloader __initdata = true; | |
92c653cf JD |
799 | static int __init parse_trust_cpu(char *arg) |
800 | { | |
801 | return kstrtobool(arg, &trust_cpu); | |
802 | } | |
d97c68d1 JD |
803 | static int __init parse_trust_bootloader(char *arg) |
804 | { | |
805 | return kstrtobool(arg, &trust_bootloader); | |
806 | } | |
92c653cf | 807 | early_param("random.trust_cpu", parse_trust_cpu); |
d97c68d1 | 808 | early_param("random.trust_bootloader", parse_trust_bootloader); |
3655adc7 | 809 | |
b7b67d13 JD |
810 | static 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 || | |
261e224d KS |
827 | (action == PM_POST_SUSPEND && !IS_ENABLED(CONFIG_PM_AUTOSLEEP) && |
828 | !IS_ENABLED(CONFIG_PM_USERSPACE_AUTOSLEEP)))) { | |
9148de31 | 829 | crng_reseed(NULL); |
b7b67d13 JD |
830 | pr_notice("crng reseeded on system resumption\n"); |
831 | } | |
832 | return 0; | |
833 | } | |
834 | ||
835 | static struct notifier_block pm_notifier = { .notifier_call = random_pm_notification }; | |
836 | ||
3655adc7 | 837 | /* |
f6238499 JD |
838 | * This is called extremely early, before time keeping functionality is |
839 | * available, but arch randomness is. Interrupts are not yet enabled. | |
3655adc7 | 840 | */ |
f6238499 | 841 | void __init random_init_early(const char *command_line) |
3655adc7 | 842 | { |
d349ab99 | 843 | unsigned long entropy[BLAKE2S_BLOCK_SIZE / sizeof(long)]; |
f6238499 | 844 | size_t i, longs, arch_bits; |
186873c5 | 845 | |
1754abb3 JD |
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 | ||
d349ab99 | 851 | for (i = 0, arch_bits = sizeof(entropy) * 8; i < ARRAY_SIZE(entropy);) { |
2c03e16f | 852 | longs = arch_get_random_seed_longs(entropy, ARRAY_SIZE(entropy) - i); |
d349ab99 JD |
853 | if (longs) { |
854 | _mix_pool_bytes(entropy, sizeof(*entropy) * longs); | |
855 | i += longs; | |
856 | continue; | |
857 | } | |
2c03e16f | 858 | longs = arch_get_random_longs(entropy, ARRAY_SIZE(entropy) - i); |
d349ab99 JD |
859 | if (longs) { |
860 | _mix_pool_bytes(entropy, sizeof(*entropy) * longs); | |
861 | i += longs; | |
862 | continue; | |
92c653cf | 863 | } |
d349ab99 JD |
864 | arch_bits -= sizeof(*entropy) * 8; |
865 | ++i; | |
92c653cf | 866 | } |
f6238499 | 867 | |
dd54fd7d | 868 | _mix_pool_bytes(init_utsname(), sizeof(*(init_utsname()))); |
2f14062b | 869 | _mix_pool_bytes(command_line, strlen(command_line)); |
f6238499 JD |
870 | |
871 | /* Reseed if already seeded by earlier phases. */ | |
872 | if (crng_ready()) | |
9148de31 | 873 | crng_reseed(NULL); |
f6238499 JD |
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 | */ | |
882 | void __init random_init(void) | |
883 | { | |
884 | unsigned long entropy = random_get_entropy(); | |
885 | ktime_t now = ktime_get_real(); | |
886 | ||
f6238499 JD |
887 | _mix_pool_bytes(&now, sizeof(now)); |
888 | _mix_pool_bytes(&entropy, sizeof(entropy)); | |
2f14062b | 889 | add_latent_entropy(); |
186873c5 | 890 | |
60e5b288 | 891 | /* |
f6238499 JD |
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 | |
60e5b288 JD |
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 | ||
f6238499 | 899 | /* Reseed if already seeded by earlier phases. */ |
e85c0fc1 | 900 | if (crng_ready()) |
9148de31 | 901 | crng_reseed(NULL); |
e192be9d | 902 | |
b7b67d13 JD |
903 | WARN_ON(register_pm_notifier(&pm_notifier)); |
904 | ||
f6238499 JD |
905 | WARN(!entropy, "Missing cycle counter and fallback timer; RNG " |
906 | "entropy collection will consequently suffer."); | |
3655adc7 | 907 | } |
e192be9d | 908 | |
a2080a67 | 909 | /* |
e192be9d TT |
910 | * Add device- or boot-specific data to the input pool to help |
911 | * initialize it. | |
a2080a67 | 912 | * |
e192be9d TT |
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. | |
a2080a67 | 916 | */ |
a1940263 | 917 | void add_device_randomness(const void *buf, size_t len) |
a2080a67 | 918 | { |
4b758eda JD |
919 | unsigned long entropy = random_get_entropy(); |
920 | unsigned long flags; | |
a2080a67 | 921 | |
3ef4cb2d | 922 | spin_lock_irqsave(&input_pool.lock, flags); |
4b758eda | 923 | _mix_pool_bytes(&entropy, sizeof(entropy)); |
a1940263 | 924 | _mix_pool_bytes(buf, len); |
3ef4cb2d | 925 | spin_unlock_irqrestore(&input_pool.lock, flags); |
a2080a67 LT |
926 | } |
927 | EXPORT_SYMBOL(add_device_randomness); | |
928 | ||
92c653cf | 929 | /* |
db516da9 JD |
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. | |
92c653cf | 933 | */ |
db516da9 | 934 | void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy, bool sleep_after) |
92c653cf | 935 | { |
a1940263 | 936 | mix_pool_bytes(buf, len); |
e85c0fc1 JD |
937 | credit_init_bits(entropy); |
938 | ||
92c653cf | 939 | /* |
745558f9 | 940 | * Throttle writing to once every reseed interval, unless we're not yet |
d775335e | 941 | * initialized or no entropy is credited. |
92c653cf | 942 | */ |
db516da9 | 943 | if (sleep_after && !kthread_should_stop() && (crng_ready() || !entropy)) |
745558f9 | 944 | schedule_timeout_interruptible(crng_reseed_interval()); |
92c653cf JD |
945 | } |
946 | EXPORT_SYMBOL_GPL(add_hwgenerator_randomness); | |
947 | ||
948 | /* | |
b9b01a56 JD |
949 | * Handle random seed passed by bootloader, and credit it depending |
950 | * on the command line option 'random.trust_bootloader'. | |
92c653cf | 951 | */ |
39e0f991 | 952 | void __init add_bootloader_randomness(const void *buf, size_t len) |
92c653cf | 953 | { |
a1940263 | 954 | mix_pool_bytes(buf, len); |
d97c68d1 | 955 | if (trust_bootloader) |
a1940263 | 956 | credit_init_bits(len * 8); |
92c653cf | 957 | } |
92c653cf | 958 | |
a4107d34 | 959 | #if IS_ENABLED(CONFIG_VMGENID) |
f3c2682b JD |
960 | static BLOCKING_NOTIFIER_HEAD(vmfork_chain); |
961 | ||
ae099e8e JD |
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 | */ | |
560181c2 | 967 | void __cold add_vmfork_randomness(const void *unique_vm_id, size_t len) |
ae099e8e | 968 | { |
a1940263 | 969 | add_device_randomness(unique_vm_id, len); |
ae099e8e | 970 | if (crng_ready()) { |
9148de31 | 971 | crng_reseed(NULL); |
ae099e8e JD |
972 | pr_notice("crng reseeded due to virtual machine fork\n"); |
973 | } | |
f3c2682b | 974 | blocking_notifier_call_chain(&vmfork_chain, 0, NULL); |
ae099e8e | 975 | } |
a4107d34 | 976 | #if IS_MODULE(CONFIG_VMGENID) |
ae099e8e | 977 | EXPORT_SYMBOL_GPL(add_vmfork_randomness); |
a4107d34 | 978 | #endif |
f3c2682b | 979 | |
560181c2 | 980 | int __cold register_random_vmfork_notifier(struct notifier_block *nb) |
f3c2682b JD |
981 | { |
982 | return blocking_notifier_chain_register(&vmfork_chain, nb); | |
983 | } | |
984 | EXPORT_SYMBOL_GPL(register_random_vmfork_notifier); | |
985 | ||
560181c2 | 986 | int __cold unregister_random_vmfork_notifier(struct notifier_block *nb) |
f3c2682b JD |
987 | { |
988 | return blocking_notifier_chain_unregister(&vmfork_chain, nb); | |
989 | } | |
990 | EXPORT_SYMBOL_GPL(unregister_random_vmfork_notifier); | |
a4107d34 | 991 | #endif |
ae099e8e | 992 | |
92c653cf | 993 | struct fast_pool { |
f5eab0e2 | 994 | unsigned long pool[4]; |
92c653cf | 995 | unsigned long last; |
3191dd5a | 996 | unsigned int count; |
748bc4dd | 997 | struct timer_list mix; |
92c653cf JD |
998 | }; |
999 | ||
748bc4dd JD |
1000 | static void mix_interrupt_randomness(struct timer_list *work); |
1001 | ||
f5eab0e2 JD |
1002 | static DEFINE_PER_CPU(struct fast_pool, irq_randomness) = { |
1003 | #ifdef CONFIG_64BIT | |
e73aaae2 | 1004 | #define FASTMIX_PERM SIPHASH_PERMUTATION |
748bc4dd | 1005 | .pool = { SIPHASH_CONST_0, SIPHASH_CONST_1, SIPHASH_CONST_2, SIPHASH_CONST_3 }, |
f5eab0e2 | 1006 | #else |
e73aaae2 | 1007 | #define FASTMIX_PERM HSIPHASH_PERMUTATION |
748bc4dd | 1008 | .pool = { HSIPHASH_CONST_0, HSIPHASH_CONST_1, HSIPHASH_CONST_2, HSIPHASH_CONST_3 }, |
f5eab0e2 | 1009 | #endif |
748bc4dd | 1010 | .mix = __TIMER_INITIALIZER(mix_interrupt_randomness, 0) |
f5eab0e2 JD |
1011 | }; |
1012 | ||
92c653cf | 1013 | /* |
f5eab0e2 JD |
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 | |
4b758eda | 1017 | * four-word SipHash state, while v represents a two-word input. |
92c653cf | 1018 | */ |
791332b3 | 1019 | static void fast_mix(unsigned long s[4], unsigned long v1, unsigned long v2) |
92c653cf | 1020 | { |
791332b3 | 1021 | s[3] ^= v1; |
e73aaae2 | 1022 | FASTMIX_PERM(s[0], s[1], s[2], s[3]); |
791332b3 JD |
1023 | s[0] ^= v1; |
1024 | s[3] ^= v2; | |
e73aaae2 | 1025 | FASTMIX_PERM(s[0], s[1], s[2], s[3]); |
791332b3 | 1026 | s[0] ^= v2; |
92c653cf JD |
1027 | } |
1028 | ||
3191dd5a JD |
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 | */ | |
560181c2 | 1034 | int __cold random_online_cpu(unsigned int cpu) |
3191dd5a JD |
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 | ||
748bc4dd | 1052 | static void mix_interrupt_randomness(struct timer_list *work) |
58340f8e JD |
1053 | { |
1054 | struct fast_pool *fast_pool = container_of(work, struct fast_pool, mix); | |
f5eab0e2 | 1055 | /* |
4b758eda JD |
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. | |
f5eab0e2 | 1061 | */ |
4b758eda | 1062 | unsigned long pool[2]; |
e3e33fc2 | 1063 | unsigned int count; |
58340f8e JD |
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(); | |
58340f8e JD |
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 | */ | |
f5eab0e2 | 1076 | memcpy(pool, fast_pool->pool, sizeof(pool)); |
e3e33fc2 | 1077 | count = fast_pool->count; |
3191dd5a | 1078 | fast_pool->count = 0; |
58340f8e JD |
1079 | fast_pool->last = jiffies; |
1080 | local_irq_enable(); | |
1081 | ||
5c3b747e | 1082 | mix_pool_bytes(pool, sizeof(pool)); |
e78a802a | 1083 | credit_init_bits(clamp_t(unsigned int, (count & U16_MAX) / 64, 1, sizeof(pool) * 8)); |
c2a7de4f | 1084 | |
58340f8e JD |
1085 | memzero_explicit(pool, sizeof(pool)); |
1086 | } | |
1087 | ||
703f7066 | 1088 | void add_interrupt_randomness(int irq) |
1da177e4 | 1089 | { |
58340f8e | 1090 | enum { MIX_INFLIGHT = 1U << 31 }; |
4b758eda | 1091 | unsigned long entropy = random_get_entropy(); |
248045b8 JD |
1092 | struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness); |
1093 | struct pt_regs *regs = get_irq_regs(); | |
58340f8e | 1094 | unsigned int new_count; |
b2f408fe | 1095 | |
791332b3 JD |
1096 | fast_mix(fast_pool->pool, entropy, |
1097 | (regs ? instruction_pointer(regs) : _RET_IP_) ^ swab(irq)); | |
3191dd5a | 1098 | new_count = ++fast_pool->count; |
3060d6fe | 1099 | |
58340f8e | 1100 | if (new_count & MIX_INFLIGHT) |
1da177e4 LT |
1101 | return; |
1102 | ||
534d2eaf | 1103 | if (new_count < 1024 && !time_is_before_jiffies(fast_pool->last + HZ)) |
91fcb532 | 1104 | return; |
83664a69 | 1105 | |
3191dd5a | 1106 | fast_pool->count |= MIX_INFLIGHT; |
748bc4dd JD |
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 | } | |
1da177e4 | 1111 | } |
4b44f2d1 | 1112 | EXPORT_SYMBOL_GPL(add_interrupt_randomness); |
1da177e4 | 1113 | |
a4b5c26b JD |
1114 | /* There is one of these per entropy source */ |
1115 | struct 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 | |
e3e33fc2 JD |
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. | |
a4b5c26b JD |
1126 | */ |
1127 | static 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; | |
e3e33fc2 | 1131 | unsigned int bits; |
a4b5c26b | 1132 | |
e3e33fc2 JD |
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()) { | |
791332b3 | 1138 | fast_mix(this_cpu_ptr(&irq_randomness)->pool, entropy, num); |
e3e33fc2 JD |
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 | } | |
a4b5c26b JD |
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 | /* | |
e3e33fc2 JD |
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. | |
a4b5c26b | 1186 | */ |
e3e33fc2 JD |
1187 | if (in_hardirq()) |
1188 | this_cpu_ptr(&irq_randomness)->count += max(1u, bits * 64) - 1; | |
1189 | else | |
560181c2 | 1190 | _credit_init_bits(bits); |
a4b5c26b JD |
1191 | } |
1192 | ||
a1940263 | 1193 | void add_input_randomness(unsigned int type, unsigned int code, unsigned int value) |
a4b5c26b JD |
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 | } | |
1206 | EXPORT_SYMBOL_GPL(add_input_randomness); | |
1207 | ||
1208 | #ifdef CONFIG_BLOCK | |
1209 | void 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 | } | |
1216 | EXPORT_SYMBOL_GPL(add_disk_randomness); | |
1217 | ||
560181c2 | 1218 | void __cold rand_initialize_disk(struct gendisk *disk) |
a4b5c26b JD |
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 | ||
78c768e6 JD |
1234 | struct entropy_timer_state { |
1235 | unsigned long entropy; | |
1236 | struct timer_list timer; | |
1c21fe00 JD |
1237 | atomic_t samples; |
1238 | unsigned int samples_per_bit; | |
78c768e6 JD |
1239 | }; |
1240 | ||
50ee7529 | 1241 | /* |
0e42d14b JD |
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. | |
50ee7529 | 1245 | * |
0e42d14b JD |
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. | |
50ee7529 LT |
1249 | * |
1250 | * So the re-arming always happens in the entropy loop itself. | |
1251 | */ | |
560181c2 | 1252 | static void __cold entropy_timer(struct timer_list *timer) |
50ee7529 | 1253 | { |
78c768e6 | 1254 | struct entropy_timer_state *state = container_of(timer, struct entropy_timer_state, timer); |
b83e45fd | 1255 | unsigned long entropy = random_get_entropy(); |
78c768e6 | 1256 | |
b83e45fd | 1257 | mix_pool_bytes(&entropy, sizeof(entropy)); |
1c21fe00 | 1258 | if (atomic_inc_return(&state->samples) % state->samples_per_bit == 0) |
e85c0fc1 | 1259 | credit_init_bits(1); |
50ee7529 LT |
1260 | } |
1261 | ||
1262 | /* | |
0e42d14b JD |
1263 | * If we have an actual cycle counter, see if we can generate enough entropy |
1264 | * with timing noise. | |
50ee7529 | 1265 | */ |
560181c2 | 1266 | static void __cold try_to_generate_entropy(void) |
50ee7529 | 1267 | { |
12273347 | 1268 | enum { NUM_TRIAL_SAMPLES = 8192, MAX_SAMPLES_PER_BIT = HZ / 15 }; |
39ec9e6b JD |
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); | |
78c768e6 JD |
1271 | unsigned int i, num_different = 0; |
1272 | unsigned long last = random_get_entropy(); | |
1c21fe00 | 1273 | int cpu = -1; |
50ee7529 | 1274 | |
78c768e6 | 1275 | for (i = 0; i < NUM_TRIAL_SAMPLES - 1; ++i) { |
39ec9e6b JD |
1276 | stack->entropy = random_get_entropy(); |
1277 | if (stack->entropy != last) | |
78c768e6 | 1278 | ++num_different; |
39ec9e6b | 1279 | last = stack->entropy; |
78c768e6 | 1280 | } |
39ec9e6b JD |
1281 | stack->samples_per_bit = DIV_ROUND_UP(NUM_TRIAL_SAMPLES, num_different + 1); |
1282 | if (stack->samples_per_bit > MAX_SAMPLES_PER_BIT) | |
50ee7529 LT |
1283 | return; |
1284 | ||
39ec9e6b JD |
1285 | atomic_set(&stack->samples, 0); |
1286 | timer_setup_on_stack(&stack->timer, entropy_timer, 0); | |
3e504d20 | 1287 | while (!crng_ready() && !signal_pending(current)) { |
1c21fe00 JD |
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 | */ | |
39ec9e6b | 1292 | if (!timer_pending(&stack->timer) && try_to_del_timer_sync(&stack->timer) >= 0) { |
1c21fe00 JD |
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); | |
8ca09d5f | 1314 | if (cpu >= nr_cpu_ids) |
1c21fe00 JD |
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. */ | |
39ec9e6b | 1319 | stack->timer.expires = jiffies; |
1c21fe00 | 1320 | |
39ec9e6b | 1321 | add_timer_on(&stack->timer, cpu); |
1c21fe00 JD |
1322 | |
1323 | preempt_enable(); | |
1324 | } | |
39ec9e6b | 1325 | mix_pool_bytes(&stack->entropy, sizeof(stack->entropy)); |
50ee7529 | 1326 | schedule(); |
39ec9e6b | 1327 | stack->entropy = random_get_entropy(); |
50ee7529 | 1328 | } |
39ec9e6b | 1329 | mix_pool_bytes(&stack->entropy, sizeof(stack->entropy)); |
50ee7529 | 1330 | |
39ec9e6b JD |
1331 | del_timer_sync(&stack->timer); |
1332 | destroy_timer_on_stack(&stack->timer); | |
50ee7529 LT |
1333 | } |
1334 | ||
a6adf8e7 JD |
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 | * | |
0313bc27 LT |
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. | |
a6adf8e7 JD |
1351 | * |
1352 | * Writing to either /dev/random or /dev/urandom adds entropy to | |
1353 | * the input pool but does not credit it. | |
1354 | * | |
0313bc27 LT |
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. | |
a6adf8e7 JD |
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 | ||
a1940263 | 1364 | SYSCALL_DEFINE3(getrandom, char __user *, ubuf, size_t, len, unsigned int, flags) |
1da177e4 | 1365 | { |
1b388e77 | 1366 | struct iov_iter iter; |
1b388e77 JA |
1367 | int ret; |
1368 | ||
a6adf8e7 JD |
1369 | if (flags & ~(GRND_NONBLOCK | GRND_RANDOM | GRND_INSECURE)) |
1370 | return -EINVAL; | |
301f0595 | 1371 | |
a6adf8e7 JD |
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; | |
c6f1deb1 | 1378 | |
f5bda35f | 1379 | if (!crng_ready() && !(flags & GRND_INSECURE)) { |
a6adf8e7 JD |
1380 | if (flags & GRND_NONBLOCK) |
1381 | return -EAGAIN; | |
1382 | ret = wait_for_random_bytes(); | |
1383 | if (unlikely(ret)) | |
1384 | return ret; | |
1385 | } | |
1b388e77 | 1386 | |
9fd7874c | 1387 | ret = import_ubuf(ITER_DEST, ubuf, len, &iter); |
1b388e77 JA |
1388 | if (unlikely(ret)) |
1389 | return ret; | |
1390 | return get_random_bytes_user(&iter); | |
30c08efe AL |
1391 | } |
1392 | ||
248045b8 | 1393 | static __poll_t random_poll(struct file *file, poll_table *wait) |
1da177e4 | 1394 | { |
30c08efe | 1395 | poll_wait(file, &crng_init_wait, wait); |
e85c0fc1 | 1396 | return crng_ready() ? EPOLLIN | EPOLLRDNORM : EPOLLOUT | EPOLLWRNORM; |
1da177e4 LT |
1397 | } |
1398 | ||
1ce6c8d6 | 1399 | static ssize_t write_pool_user(struct iov_iter *iter) |
1da177e4 | 1400 | { |
04ec96b7 | 1401 | u8 block[BLAKE2S_BLOCK_SIZE]; |
22b0a222 JA |
1402 | ssize_t ret = 0; |
1403 | size_t copied; | |
1da177e4 | 1404 | |
22b0a222 JA |
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; | |
1ce6c8d6 JD |
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 | } | |
1da177e4 | 1421 | } |
7f397dcd | 1422 | |
7b5164fb | 1423 | memzero_explicit(block, sizeof(block)); |
22b0a222 | 1424 | return ret ? ret : -EFAULT; |
7f397dcd MM |
1425 | } |
1426 | ||
22b0a222 | 1427 | static ssize_t random_write_iter(struct kiocb *kiocb, struct iov_iter *iter) |
7f397dcd | 1428 | { |
1ce6c8d6 | 1429 | return write_pool_user(iter); |
1da177e4 LT |
1430 | } |
1431 | ||
1b388e77 | 1432 | static ssize_t urandom_read_iter(struct kiocb *kiocb, struct iov_iter *iter) |
0313bc27 LT |
1433 | { |
1434 | static int maxwarn = 10; | |
1435 | ||
48bff105 JD |
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 | ||
cc1e127b JD |
1443 | if (!crng_ready()) { |
1444 | if (!ratelimit_disable && maxwarn <= 0) | |
1445 | ++urandom_warning.missed; | |
1446 | else if (ratelimit_disable || __ratelimit(&urandom_warning)) { | |
1447 | --maxwarn; | |
1b388e77 JA |
1448 | pr_notice("%s: uninitialized urandom read (%zu bytes read)\n", |
1449 | current->comm, iov_iter_count(iter)); | |
cc1e127b | 1450 | } |
0313bc27 LT |
1451 | } |
1452 | ||
1b388e77 | 1453 | return get_random_bytes_user(iter); |
0313bc27 LT |
1454 | } |
1455 | ||
1b388e77 | 1456 | static ssize_t random_read_iter(struct kiocb *kiocb, struct iov_iter *iter) |
a6adf8e7 JD |
1457 | { |
1458 | int ret; | |
1459 | ||
cd4f24ae JD |
1460 | if (!crng_ready() && |
1461 | ((kiocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO)) || | |
1462 | (kiocb->ki_filp->f_flags & O_NONBLOCK))) | |
1463 | return -EAGAIN; | |
1464 | ||
a6adf8e7 JD |
1465 | ret = wait_for_random_bytes(); |
1466 | if (ret != 0) | |
1467 | return ret; | |
1b388e77 | 1468 | return get_random_bytes_user(iter); |
a6adf8e7 JD |
1469 | } |
1470 | ||
43ae4860 | 1471 | static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg) |
1da177e4 | 1472 | { |
1da177e4 | 1473 | int __user *p = (int __user *)arg; |
22b0a222 | 1474 | int ent_count; |
1da177e4 LT |
1475 | |
1476 | switch (cmd) { | |
1477 | case RNDGETENTCNT: | |
a6adf8e7 | 1478 | /* Inherently racy, no point locking. */ |
e85c0fc1 | 1479 | if (put_user(input_pool.init_bits, p)) |
1da177e4 LT |
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; | |
a49c010e JD |
1487 | if (ent_count < 0) |
1488 | return -EINVAL; | |
e85c0fc1 | 1489 | credit_init_bits(ent_count); |
a49c010e | 1490 | return 0; |
22b0a222 JA |
1491 | case RNDADDENTROPY: { |
1492 | struct iov_iter iter; | |
22b0a222 JA |
1493 | ssize_t ret; |
1494 | int len; | |
1495 | ||
1da177e4 LT |
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; | |
22b0a222 JA |
1502 | if (get_user(len, p++)) |
1503 | return -EFAULT; | |
9fd7874c | 1504 | ret = import_ubuf(ITER_SOURCE, p, len, &iter); |
22b0a222 JA |
1505 | if (unlikely(ret)) |
1506 | return ret; | |
1ce6c8d6 | 1507 | ret = write_pool_user(&iter); |
22b0a222 JA |
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)) | |
1da177e4 | 1512 | return -EFAULT; |
e85c0fc1 | 1513 | credit_init_bits(ent_count); |
a49c010e | 1514 | return 0; |
22b0a222 | 1515 | } |
1da177e4 LT |
1516 | case RNDZAPENTCNT: |
1517 | case RNDCLEARPOOL: | |
e85c0fc1 | 1518 | /* No longer has any effect. */ |
1da177e4 LT |
1519 | if (!capable(CAP_SYS_ADMIN)) |
1520 | return -EPERM; | |
1da177e4 | 1521 | return 0; |
d848e5f8 TT |
1522 | case RNDRESEEDCRNG: |
1523 | if (!capable(CAP_SYS_ADMIN)) | |
1524 | return -EPERM; | |
a96cfe2d | 1525 | if (!crng_ready()) |
d848e5f8 | 1526 | return -ENODATA; |
9148de31 | 1527 | crng_reseed(NULL); |
d848e5f8 | 1528 | return 0; |
1da177e4 LT |
1529 | default: |
1530 | return -EINVAL; | |
1531 | } | |
1532 | } | |
1533 | ||
9a6f70bb JD |
1534 | static int random_fasync(int fd, struct file *filp, int on) |
1535 | { | |
1536 | return fasync_helper(fd, filp, on, &fasync); | |
1537 | } | |
1538 | ||
2b8693c0 | 1539 | const struct file_operations random_fops = { |
1b388e77 | 1540 | .read_iter = random_read_iter, |
22b0a222 | 1541 | .write_iter = random_write_iter, |
248045b8 | 1542 | .poll = random_poll, |
43ae4860 | 1543 | .unlocked_ioctl = random_ioctl, |
507e4e2b | 1544 | .compat_ioctl = compat_ptr_ioctl, |
9a6f70bb | 1545 | .fasync = random_fasync, |
6038f373 | 1546 | .llseek = noop_llseek, |
b0072734 | 1547 | .splice_read = copy_splice_read, |
79025e72 | 1548 | .splice_write = iter_file_splice_write, |
1da177e4 LT |
1549 | }; |
1550 | ||
0313bc27 | 1551 | const struct file_operations urandom_fops = { |
1b388e77 | 1552 | .read_iter = urandom_read_iter, |
22b0a222 | 1553 | .write_iter = random_write_iter, |
0313bc27 LT |
1554 | .unlocked_ioctl = random_ioctl, |
1555 | .compat_ioctl = compat_ptr_ioctl, | |
1556 | .fasync = random_fasync, | |
1557 | .llseek = noop_llseek, | |
b0072734 | 1558 | .splice_read = copy_splice_read, |
79025e72 | 1559 | .splice_write = iter_file_splice_write, |
0313bc27 LT |
1560 | }; |
1561 | ||
0deff3c4 | 1562 | |
1da177e4 LT |
1563 | /******************************************************************** |
1564 | * | |
0deff3c4 JD |
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 | |
e3d2c5e7 | 1583 | * more entropy, tied to the POOL_READY_BITS constant. It is writable |
0deff3c4 JD |
1584 | * to avoid breaking old userspaces, but writing to it does not |
1585 | * change any behavior of the RNG. | |
1586 | * | |
d0efdf35 | 1587 | * - urandom_min_reseed_secs - fixed to the value CRNG_RESEED_INTERVAL. |
0deff3c4 JD |
1588 | * It is writable to avoid breaking old userspaces, but writing |
1589 | * to it does not change any behavior of the RNG. | |
1da177e4 LT |
1590 | * |
1591 | ********************************************************************/ | |
1592 | ||
1593 | #ifdef CONFIG_SYSCTL | |
1594 | ||
1595 | #include <linux/sysctl.h> | |
1596 | ||
d0efdf35 | 1597 | static int sysctl_random_min_urandom_seed = CRNG_RESEED_INTERVAL / HZ; |
e3d2c5e7 | 1598 | static int sysctl_random_write_wakeup_bits = POOL_READY_BITS; |
489c7fc4 | 1599 | static int sysctl_poolsize = POOL_BITS; |
64276a99 | 1600 | static u8 sysctl_bootid[UUID_SIZE]; |
1da177e4 LT |
1601 | |
1602 | /* | |
f22052b2 | 1603 | * This function is used to return both the bootid UUID, and random |
64276a99 | 1604 | * UUID. The difference is in whether table->data is NULL; if it is, |
1da177e4 | 1605 | * then a new UUID is generated and returned to the user. |
1da177e4 | 1606 | */ |
a1940263 | 1607 | static int proc_do_uuid(struct ctl_table *table, int write, void *buf, |
248045b8 | 1608 | size_t *lenp, loff_t *ppos) |
1da177e4 | 1609 | { |
64276a99 JD |
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; | |
1da177e4 LT |
1619 | |
1620 | uuid = table->data; | |
1621 | if (!uuid) { | |
1622 | uuid = tmp_uuid; | |
1da177e4 | 1623 | generate_random_uuid(uuid); |
44e4360f MD |
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 | } | |
1da177e4 | 1632 | |
64276a99 | 1633 | snprintf(uuid_string, sizeof(uuid_string), "%pU", uuid); |
a1940263 | 1634 | return proc_dostring(&fake_table, 0, buf, lenp, ppos); |
1da177e4 LT |
1635 | } |
1636 | ||
77553cf8 | 1637 | /* The same as proc_dointvec, but writes don't change anything. */ |
a1940263 | 1638 | static int proc_do_rointvec(struct ctl_table *table, int write, void *buf, |
77553cf8 JD |
1639 | size_t *lenp, loff_t *ppos) |
1640 | { | |
a1940263 | 1641 | return write ? 0 : proc_dointvec(table, 0, buf, lenp, ppos); |
77553cf8 JD |
1642 | } |
1643 | ||
5475e8f0 | 1644 | static struct ctl_table random_table[] = { |
1da177e4 | 1645 | { |
1da177e4 LT |
1646 | .procname = "poolsize", |
1647 | .data = &sysctl_poolsize, | |
1648 | .maxlen = sizeof(int), | |
1649 | .mode = 0444, | |
6d456111 | 1650 | .proc_handler = proc_dointvec, |
1da177e4 LT |
1651 | }, |
1652 | { | |
1da177e4 | 1653 | .procname = "entropy_avail", |
e85c0fc1 | 1654 | .data = &input_pool.init_bits, |
1da177e4 LT |
1655 | .maxlen = sizeof(int), |
1656 | .mode = 0444, | |
c5704490 | 1657 | .proc_handler = proc_dointvec, |
1da177e4 | 1658 | }, |
1da177e4 | 1659 | { |
1da177e4 | 1660 | .procname = "write_wakeup_threshold", |
0deff3c4 | 1661 | .data = &sysctl_random_write_wakeup_bits, |
1da177e4 LT |
1662 | .maxlen = sizeof(int), |
1663 | .mode = 0644, | |
77553cf8 | 1664 | .proc_handler = proc_do_rointvec, |
1da177e4 | 1665 | }, |
f5c2742c TT |
1666 | { |
1667 | .procname = "urandom_min_reseed_secs", | |
0deff3c4 | 1668 | .data = &sysctl_random_min_urandom_seed, |
f5c2742c TT |
1669 | .maxlen = sizeof(int), |
1670 | .mode = 0644, | |
77553cf8 | 1671 | .proc_handler = proc_do_rointvec, |
f5c2742c | 1672 | }, |
1da177e4 | 1673 | { |
1da177e4 LT |
1674 | .procname = "boot_id", |
1675 | .data = &sysctl_bootid, | |
1da177e4 | 1676 | .mode = 0444, |
6d456111 | 1677 | .proc_handler = proc_do_uuid, |
1da177e4 LT |
1678 | }, |
1679 | { | |
1da177e4 | 1680 | .procname = "uuid", |
1da177e4 | 1681 | .mode = 0444, |
6d456111 | 1682 | .proc_handler = proc_do_uuid, |
1da177e4 | 1683 | }, |
1da177e4 | 1684 | }; |
5475e8f0 XN |
1685 | |
1686 | /* | |
2f14062b JD |
1687 | * random_init() is called before sysctl_init(), |
1688 | * so we cannot call register_sysctl_init() in random_init() | |
5475e8f0 XN |
1689 | */ |
1690 | static int __init random_sysctls_init(void) | |
1691 | { | |
1692 | register_sysctl_init("kernel/random", random_table); | |
1693 | return 0; | |
1694 | } | |
1695 | device_initcall(random_sysctls_init); | |
0deff3c4 | 1696 | #endif |