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> |
1ca1b917 | 56 | #include <crypto/chacha.h> |
9f9eff85 | 57 | #include <crypto/blake2s.h> |
1da177e4 | 58 | #include <asm/processor.h> |
1da177e4 | 59 | #include <asm/irq.h> |
775f4b29 | 60 | #include <asm/irq_regs.h> |
1da177e4 LT |
61 | #include <asm/io.h> |
62 | ||
5f1bb112 JD |
63 | /********************************************************************* |
64 | * | |
65 | * Initialization and readiness waiting. | |
66 | * | |
67 | * Much of the RNG infrastructure is devoted to various dependencies | |
68 | * being able to wait until the RNG has collected enough entropy and | |
69 | * is ready for safe consumption. | |
70 | * | |
71 | *********************************************************************/ | |
205a525c | 72 | |
e192be9d | 73 | /* |
5f1bb112 | 74 | * crng_init is protected by base_crng->lock, and only increases |
e3d2c5e7 | 75 | * its value (from empty->early->ready). |
e192be9d | 76 | */ |
e3d2c5e7 JD |
77 | static enum { |
78 | CRNG_EMPTY = 0, /* Little to no entropy collected */ | |
79 | CRNG_EARLY = 1, /* At least POOL_EARLY_BITS collected */ | |
80 | CRNG_READY = 2 /* Fully initialized with POOL_READY_BITS collected */ | |
f5bda35f JD |
81 | } crng_init __read_mostly = CRNG_EMPTY; |
82 | static DEFINE_STATIC_KEY_FALSE(crng_is_ready); | |
83 | #define crng_ready() (static_branch_likely(&crng_is_ready) || crng_init >= CRNG_READY) | |
e3d2c5e7 | 84 | /* Various types of waiters for crng_init->CRNG_READY transition. */ |
5f1bb112 JD |
85 | static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait); |
86 | static struct fasync_struct *fasync; | |
e192be9d | 87 | |
5f1bb112 | 88 | /* Control how we warn userspace. */ |
0313bc27 | 89 | static struct ratelimit_state urandom_warning = |
c01d4d0a | 90 | RATELIMIT_STATE_INIT_FLAGS("urandom_warning", HZ, 3, RATELIMIT_MSG_ON_RELEASE); |
cc1e127b JD |
91 | static int ratelimit_disable __read_mostly = |
92 | IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM); | |
4e00b339 TT |
93 | module_param_named(ratelimit_disable, ratelimit_disable, int, 0644); |
94 | MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression"); | |
95 | ||
5f1bb112 JD |
96 | /* |
97 | * Returns whether or not the input pool has been seeded and thus guaranteed | |
0313bc27 | 98 | * to supply cryptographically secure random numbers. This applies to: the |
a890d1c6 | 99 | * /dev/urandom device, the get_random_bytes function, and the get_random_{u8, |
de492c83 | 100 | * u16,u32,u64,long} family of functions. |
5f1bb112 JD |
101 | * |
102 | * Returns: true if the input pool has been seeded. | |
103 | * false if the input pool has not been seeded. | |
104 | */ | |
105 | bool rng_is_initialized(void) | |
106 | { | |
107 | return crng_ready(); | |
108 | } | |
109 | EXPORT_SYMBOL(rng_is_initialized); | |
110 | ||
560181c2 | 111 | static void __cold crng_set_ready(struct work_struct *work) |
f5bda35f JD |
112 | { |
113 | static_branch_enable(&crng_is_ready); | |
114 | } | |
115 | ||
5f1bb112 JD |
116 | /* Used by wait_for_random_bytes(), and considered an entropy collector, below. */ |
117 | static void try_to_generate_entropy(void); | |
118 | ||
119 | /* | |
120 | * Wait for the input pool to be seeded and thus guaranteed to supply | |
0313bc27 | 121 | * cryptographically secure random numbers. This applies to: the /dev/urandom |
a890d1c6 JD |
122 | * device, the get_random_bytes function, and the get_random_{u8,u16,u32,u64, |
123 | * int,long} family of functions. Using any of these functions without first | |
124 | * calling this function forfeits the guarantee of security. | |
5f1bb112 JD |
125 | * |
126 | * Returns: 0 if the input pool has been seeded. | |
127 | * -ERESTARTSYS if the function was interrupted by a signal. | |
128 | */ | |
129 | int wait_for_random_bytes(void) | |
130 | { | |
a96cfe2d | 131 | while (!crng_ready()) { |
5f1bb112 | 132 | int ret; |
3e504d20 JD |
133 | |
134 | try_to_generate_entropy(); | |
5f1bb112 JD |
135 | ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ); |
136 | if (ret) | |
137 | return ret > 0 ? 0 : ret; | |
a96cfe2d | 138 | } |
5f1bb112 JD |
139 | return 0; |
140 | } | |
141 | EXPORT_SYMBOL(wait_for_random_bytes); | |
142 | ||
cc1e127b | 143 | #define warn_unseeded_randomness() \ |
560181c2 JD |
144 | if (IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM) && !crng_ready()) \ |
145 | printk_deferred(KERN_NOTICE "random: %s called from %pS with crng_init=%d\n", \ | |
146 | __func__, (void *)_RET_IP_, crng_init) | |
5f1bb112 JD |
147 | |
148 | ||
3655adc7 | 149 | /********************************************************************* |
1da177e4 | 150 | * |
3655adc7 | 151 | * Fast key erasure RNG, the "crng". |
1da177e4 | 152 | * |
3655adc7 JD |
153 | * These functions expand entropy from the entropy extractor into |
154 | * long streams for external consumption using the "fast key erasure" | |
155 | * RNG described at <https://blog.cr.yp.to/20170723-random.html>. | |
e192be9d | 156 | * |
3655adc7 JD |
157 | * There are a few exported interfaces for use by other drivers: |
158 | * | |
a1940263 | 159 | * void get_random_bytes(void *buf, size_t len) |
a890d1c6 JD |
160 | * u8 get_random_u8() |
161 | * u16 get_random_u16() | |
3655adc7 | 162 | * u32 get_random_u32() |
e9a688bc | 163 | * u32 get_random_u32_below(u32 ceil) |
7f576b25 JD |
164 | * u32 get_random_u32_above(u32 floor) |
165 | * u32 get_random_u32_inclusive(u32 floor, u32 ceil) | |
3655adc7 | 166 | * u64 get_random_u64() |
3655adc7 JD |
167 | * unsigned long get_random_long() |
168 | * | |
169 | * These interfaces will return the requested number of random bytes | |
0313bc27 | 170 | * into the given buffer or as a return value. This is equivalent to |
de492c83 JD |
171 | * a read from /dev/urandom. The u8, u16, u32, u64, long family of |
172 | * functions may be higher performance for one-off random integers, | |
173 | * because they do a bit of buffering and do not invoke reseeding | |
174 | * until the buffer is emptied. | |
e192be9d TT |
175 | * |
176 | *********************************************************************/ | |
177 | ||
e85c0fc1 JD |
178 | enum { |
179 | CRNG_RESEED_START_INTERVAL = HZ, | |
180 | CRNG_RESEED_INTERVAL = 60 * HZ | |
181 | }; | |
186873c5 JD |
182 | |
183 | static struct { | |
184 | u8 key[CHACHA_KEY_SIZE] __aligned(__alignof__(long)); | |
185 | unsigned long birth; | |
186 | unsigned long generation; | |
187 | spinlock_t lock; | |
188 | } base_crng = { | |
189 | .lock = __SPIN_LOCK_UNLOCKED(base_crng.lock) | |
190 | }; | |
191 | ||
192 | struct crng { | |
193 | u8 key[CHACHA_KEY_SIZE]; | |
194 | unsigned long generation; | |
195 | local_lock_t lock; | |
196 | }; | |
197 | ||
198 | static DEFINE_PER_CPU(struct crng, crngs) = { | |
199 | .generation = ULONG_MAX, | |
200 | .lock = INIT_LOCAL_LOCK(crngs.lock), | |
201 | }; | |
e192be9d | 202 | |
e85c0fc1 | 203 | /* Used by crng_reseed() and crng_make_state() to extract a new seed from the input pool. */ |
a1940263 | 204 | static void extract_entropy(void *buf, size_t len); |
e192be9d | 205 | |
e85c0fc1 JD |
206 | /* This extracts a new crng key from the input pool. */ |
207 | static void crng_reseed(void) | |
e192be9d | 208 | { |
248045b8 | 209 | unsigned long flags; |
186873c5 JD |
210 | unsigned long next_gen; |
211 | u8 key[CHACHA_KEY_SIZE]; | |
e192be9d | 212 | |
e85c0fc1 | 213 | extract_entropy(key, sizeof(key)); |
a9412d51 | 214 | |
186873c5 JD |
215 | /* |
216 | * We copy the new key into the base_crng, overwriting the old one, | |
217 | * and update the generation counter. We avoid hitting ULONG_MAX, | |
218 | * because the per-cpu crngs are initialized to ULONG_MAX, so this | |
219 | * forces new CPUs that come online to always initialize. | |
220 | */ | |
221 | spin_lock_irqsave(&base_crng.lock, flags); | |
222 | memcpy(base_crng.key, key, sizeof(base_crng.key)); | |
223 | next_gen = base_crng.generation + 1; | |
224 | if (next_gen == ULONG_MAX) | |
225 | ++next_gen; | |
226 | WRITE_ONCE(base_crng.generation, next_gen); | |
227 | WRITE_ONCE(base_crng.birth, jiffies); | |
f5bda35f | 228 | if (!static_branch_likely(&crng_is_ready)) |
e3d2c5e7 | 229 | crng_init = CRNG_READY; |
7191c628 DB |
230 | spin_unlock_irqrestore(&base_crng.lock, flags); |
231 | memzero_explicit(key, sizeof(key)); | |
e192be9d TT |
232 | } |
233 | ||
186873c5 | 234 | /* |
3655adc7 | 235 | * This generates a ChaCha block using the provided key, and then |
7f637be4 | 236 | * immediately overwrites that key with half the block. It returns |
3655adc7 JD |
237 | * the resultant ChaCha state to the user, along with the second |
238 | * half of the block containing 32 bytes of random data that may | |
239 | * be used; random_data_len may not be greater than 32. | |
8717627d JD |
240 | * |
241 | * The returned ChaCha state contains within it a copy of the old | |
242 | * key value, at index 4, so the state should always be zeroed out | |
243 | * immediately after using in order to maintain forward secrecy. | |
244 | * If the state cannot be erased in a timely manner, then it is | |
245 | * safer to set the random_data parameter to &chacha_state[4] so | |
246 | * that this function overwrites it before returning. | |
186873c5 JD |
247 | */ |
248 | static void crng_fast_key_erasure(u8 key[CHACHA_KEY_SIZE], | |
249 | u32 chacha_state[CHACHA_STATE_WORDS], | |
250 | u8 *random_data, size_t random_data_len) | |
e192be9d | 251 | { |
186873c5 | 252 | u8 first_block[CHACHA_BLOCK_SIZE]; |
009ba856 | 253 | |
186873c5 JD |
254 | BUG_ON(random_data_len > 32); |
255 | ||
256 | chacha_init_consts(chacha_state); | |
257 | memcpy(&chacha_state[4], key, CHACHA_KEY_SIZE); | |
258 | memset(&chacha_state[12], 0, sizeof(u32) * 4); | |
259 | chacha20_block(chacha_state, first_block); | |
260 | ||
261 | memcpy(key, first_block, CHACHA_KEY_SIZE); | |
8717627d | 262 | memcpy(random_data, first_block + CHACHA_KEY_SIZE, random_data_len); |
186873c5 | 263 | memzero_explicit(first_block, sizeof(first_block)); |
1e7f583a TT |
264 | } |
265 | ||
7a7ff644 | 266 | /* |
745558f9 DB |
267 | * Return the interval until the next reseeding, which is normally |
268 | * CRNG_RESEED_INTERVAL, but during early boot, it is at an interval | |
e85c0fc1 | 269 | * proportional to the uptime. |
7a7ff644 | 270 | */ |
745558f9 | 271 | static unsigned int crng_reseed_interval(void) |
7a7ff644 JD |
272 | { |
273 | static bool early_boot = true; | |
7a7ff644 JD |
274 | |
275 | if (unlikely(READ_ONCE(early_boot))) { | |
276 | time64_t uptime = ktime_get_seconds(); | |
277 | if (uptime >= CRNG_RESEED_INTERVAL / HZ * 2) | |
278 | WRITE_ONCE(early_boot, false); | |
279 | else | |
745558f9 DB |
280 | return max_t(unsigned int, CRNG_RESEED_START_INTERVAL, |
281 | (unsigned int)uptime / 2 * HZ); | |
7a7ff644 | 282 | } |
745558f9 | 283 | return CRNG_RESEED_INTERVAL; |
7a7ff644 JD |
284 | } |
285 | ||
c92e040d | 286 | /* |
186873c5 JD |
287 | * This function returns a ChaCha state that you may use for generating |
288 | * random data. It also returns up to 32 bytes on its own of random data | |
289 | * that may be used; random_data_len may not be greater than 32. | |
c92e040d | 290 | */ |
186873c5 JD |
291 | static void crng_make_state(u32 chacha_state[CHACHA_STATE_WORDS], |
292 | u8 *random_data, size_t random_data_len) | |
c92e040d | 293 | { |
248045b8 | 294 | unsigned long flags; |
186873c5 | 295 | struct crng *crng; |
c92e040d | 296 | |
186873c5 JD |
297 | BUG_ON(random_data_len > 32); |
298 | ||
299 | /* | |
300 | * For the fast path, we check whether we're ready, unlocked first, and | |
301 | * then re-check once locked later. In the case where we're really not | |
5c3b747e | 302 | * ready, we do fast key erasure with the base_crng directly, extracting |
e3d2c5e7 | 303 | * when crng_init is CRNG_EMPTY. |
186873c5 | 304 | */ |
a96cfe2d | 305 | if (!crng_ready()) { |
186873c5 JD |
306 | bool ready; |
307 | ||
308 | spin_lock_irqsave(&base_crng.lock, flags); | |
309 | ready = crng_ready(); | |
5c3b747e | 310 | if (!ready) { |
e3d2c5e7 | 311 | if (crng_init == CRNG_EMPTY) |
5c3b747e | 312 | extract_entropy(base_crng.key, sizeof(base_crng.key)); |
186873c5 JD |
313 | crng_fast_key_erasure(base_crng.key, chacha_state, |
314 | random_data, random_data_len); | |
5c3b747e | 315 | } |
186873c5 JD |
316 | spin_unlock_irqrestore(&base_crng.lock, flags); |
317 | if (!ready) | |
318 | return; | |
c92e040d | 319 | } |
186873c5 JD |
320 | |
321 | /* | |
e85c0fc1 JD |
322 | * If the base_crng is old enough, we reseed, which in turn bumps the |
323 | * generation counter that we check below. | |
186873c5 | 324 | */ |
745558f9 | 325 | if (unlikely(time_is_before_jiffies(READ_ONCE(base_crng.birth) + crng_reseed_interval()))) |
e85c0fc1 | 326 | crng_reseed(); |
186873c5 JD |
327 | |
328 | local_lock_irqsave(&crngs.lock, flags); | |
329 | crng = raw_cpu_ptr(&crngs); | |
330 | ||
331 | /* | |
332 | * If our per-cpu crng is older than the base_crng, then it means | |
333 | * somebody reseeded the base_crng. In that case, we do fast key | |
334 | * erasure on the base_crng, and use its output as the new key | |
335 | * for our per-cpu crng. This brings us up to date with base_crng. | |
336 | */ | |
337 | if (unlikely(crng->generation != READ_ONCE(base_crng.generation))) { | |
338 | spin_lock(&base_crng.lock); | |
339 | crng_fast_key_erasure(base_crng.key, chacha_state, | |
340 | crng->key, sizeof(crng->key)); | |
341 | crng->generation = base_crng.generation; | |
342 | spin_unlock(&base_crng.lock); | |
343 | } | |
344 | ||
345 | /* | |
346 | * Finally, when we've made it this far, our per-cpu crng has an up | |
347 | * to date key, and we can do fast key erasure with it to produce | |
348 | * some random data and a ChaCha state for the caller. All other | |
349 | * branches of this function are "unlikely", so most of the time we | |
350 | * should wind up here immediately. | |
351 | */ | |
352 | crng_fast_key_erasure(crng->key, chacha_state, random_data, random_data_len); | |
353 | local_unlock_irqrestore(&crngs.lock, flags); | |
c92e040d TT |
354 | } |
355 | ||
a1940263 | 356 | static void _get_random_bytes(void *buf, size_t len) |
e192be9d | 357 | { |
186873c5 | 358 | u32 chacha_state[CHACHA_STATE_WORDS]; |
3655adc7 | 359 | u8 tmp[CHACHA_BLOCK_SIZE]; |
a1940263 | 360 | size_t first_block_len; |
3655adc7 | 361 | |
a1940263 | 362 | if (!len) |
3655adc7 JD |
363 | return; |
364 | ||
a1940263 JD |
365 | first_block_len = min_t(size_t, 32, len); |
366 | crng_make_state(chacha_state, buf, first_block_len); | |
367 | len -= first_block_len; | |
368 | buf += first_block_len; | |
3655adc7 | 369 | |
a1940263 JD |
370 | while (len) { |
371 | if (len < CHACHA_BLOCK_SIZE) { | |
3655adc7 | 372 | chacha20_block(chacha_state, tmp); |
a1940263 | 373 | memcpy(buf, tmp, len); |
3655adc7 JD |
374 | memzero_explicit(tmp, sizeof(tmp)); |
375 | break; | |
376 | } | |
377 | ||
378 | chacha20_block(chacha_state, buf); | |
379 | if (unlikely(chacha_state[12] == 0)) | |
380 | ++chacha_state[13]; | |
a1940263 | 381 | len -= CHACHA_BLOCK_SIZE; |
3655adc7 JD |
382 | buf += CHACHA_BLOCK_SIZE; |
383 | } | |
384 | ||
385 | memzero_explicit(chacha_state, sizeof(chacha_state)); | |
386 | } | |
387 | ||
388 | /* | |
d687772e WZ |
389 | * This function is the exported kernel interface. It returns some number of |
390 | * good random numbers, suitable for key generation, seeding TCP sequence | |
391 | * numbers, etc. In order to ensure that the randomness returned by this | |
392 | * function is okay, the function wait_for_random_bytes() should be called and | |
393 | * return 0 at least once at any point prior. | |
3655adc7 | 394 | */ |
a1940263 | 395 | void get_random_bytes(void *buf, size_t len) |
3655adc7 | 396 | { |
cc1e127b | 397 | warn_unseeded_randomness(); |
a1940263 | 398 | _get_random_bytes(buf, len); |
3655adc7 JD |
399 | } |
400 | EXPORT_SYMBOL(get_random_bytes); | |
401 | ||
1b388e77 | 402 | static ssize_t get_random_bytes_user(struct iov_iter *iter) |
3655adc7 | 403 | { |
3655adc7 | 404 | u32 chacha_state[CHACHA_STATE_WORDS]; |
1b388e77 JA |
405 | u8 block[CHACHA_BLOCK_SIZE]; |
406 | size_t ret = 0, copied; | |
3655adc7 | 407 | |
1b388e77 | 408 | if (unlikely(!iov_iter_count(iter))) |
3655adc7 JD |
409 | return 0; |
410 | ||
aba120cc JD |
411 | /* |
412 | * Immediately overwrite the ChaCha key at index 4 with random | |
63b8ea5e | 413 | * bytes, in case userspace causes copy_to_iter() below to sleep |
aba120cc JD |
414 | * forever, so that we still retain forward secrecy in that case. |
415 | */ | |
416 | crng_make_state(chacha_state, (u8 *)&chacha_state[4], CHACHA_KEY_SIZE); | |
417 | /* | |
418 | * However, if we're doing a read of len <= 32, we don't need to | |
419 | * use chacha_state after, so we can simply return those bytes to | |
420 | * the user directly. | |
421 | */ | |
1b388e77 JA |
422 | if (iov_iter_count(iter) <= CHACHA_KEY_SIZE) { |
423 | ret = copy_to_iter(&chacha_state[4], CHACHA_KEY_SIZE, iter); | |
aba120cc JD |
424 | goto out_zero_chacha; |
425 | } | |
3655adc7 | 426 | |
5209aed5 | 427 | for (;;) { |
1b388e77 | 428 | chacha20_block(chacha_state, block); |
3655adc7 JD |
429 | if (unlikely(chacha_state[12] == 0)) |
430 | ++chacha_state[13]; | |
431 | ||
1b388e77 JA |
432 | copied = copy_to_iter(block, sizeof(block), iter); |
433 | ret += copied; | |
434 | if (!iov_iter_count(iter) || copied != sizeof(block)) | |
5209aed5 | 435 | break; |
e3c1c4fd | 436 | |
1b388e77 | 437 | BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0); |
5209aed5 | 438 | if (ret % PAGE_SIZE == 0) { |
e3c1c4fd JD |
439 | if (signal_pending(current)) |
440 | break; | |
441 | cond_resched(); | |
442 | } | |
5209aed5 | 443 | } |
3655adc7 | 444 | |
1b388e77 | 445 | memzero_explicit(block, sizeof(block)); |
aba120cc JD |
446 | out_zero_chacha: |
447 | memzero_explicit(chacha_state, sizeof(chacha_state)); | |
5209aed5 | 448 | return ret ? ret : -EFAULT; |
3655adc7 JD |
449 | } |
450 | ||
451 | /* | |
452 | * Batched entropy returns random integers. The quality of the random | |
453 | * number is good as /dev/urandom. In order to ensure that the randomness | |
454 | * provided by this function is okay, the function wait_for_random_bytes() | |
455 | * should be called and return 0 at least once at any point prior. | |
456 | */ | |
3655adc7 | 457 | |
3092adce JD |
458 | #define DEFINE_BATCHED_ENTROPY(type) \ |
459 | struct batch_ ##type { \ | |
460 | /* \ | |
461 | * We make this 1.5x a ChaCha block, so that we get the \ | |
462 | * remaining 32 bytes from fast key erasure, plus one full \ | |
463 | * block from the detached ChaCha state. We can increase \ | |
464 | * the size of this later if needed so long as we keep the \ | |
465 | * formula of (integer_blocks + 0.5) * CHACHA_BLOCK_SIZE. \ | |
466 | */ \ | |
467 | type entropy[CHACHA_BLOCK_SIZE * 3 / (2 * sizeof(type))]; \ | |
468 | local_lock_t lock; \ | |
469 | unsigned long generation; \ | |
470 | unsigned int position; \ | |
471 | }; \ | |
472 | \ | |
473 | static DEFINE_PER_CPU(struct batch_ ##type, batched_entropy_ ##type) = { \ | |
474 | .lock = INIT_LOCAL_LOCK(batched_entropy_ ##type.lock), \ | |
475 | .position = UINT_MAX \ | |
476 | }; \ | |
477 | \ | |
478 | type get_random_ ##type(void) \ | |
479 | { \ | |
480 | type ret; \ | |
481 | unsigned long flags; \ | |
482 | struct batch_ ##type *batch; \ | |
483 | unsigned long next_gen; \ | |
484 | \ | |
485 | warn_unseeded_randomness(); \ | |
486 | \ | |
487 | if (!crng_ready()) { \ | |
488 | _get_random_bytes(&ret, sizeof(ret)); \ | |
489 | return ret; \ | |
490 | } \ | |
491 | \ | |
492 | local_lock_irqsave(&batched_entropy_ ##type.lock, flags); \ | |
493 | batch = raw_cpu_ptr(&batched_entropy_##type); \ | |
494 | \ | |
495 | next_gen = READ_ONCE(base_crng.generation); \ | |
496 | if (batch->position >= ARRAY_SIZE(batch->entropy) || \ | |
497 | next_gen != batch->generation) { \ | |
498 | _get_random_bytes(batch->entropy, sizeof(batch->entropy)); \ | |
499 | batch->position = 0; \ | |
500 | batch->generation = next_gen; \ | |
501 | } \ | |
502 | \ | |
503 | ret = batch->entropy[batch->position]; \ | |
504 | batch->entropy[batch->position] = 0; \ | |
505 | ++batch->position; \ | |
506 | local_unlock_irqrestore(&batched_entropy_ ##type.lock, flags); \ | |
507 | return ret; \ | |
508 | } \ | |
509 | EXPORT_SYMBOL(get_random_ ##type); | |
510 | ||
585cd5fe | 511 | DEFINE_BATCHED_ENTROPY(u8) |
a890d1c6 JD |
512 | DEFINE_BATCHED_ENTROPY(u16) |
513 | DEFINE_BATCHED_ENTROPY(u32) | |
514 | DEFINE_BATCHED_ENTROPY(u64) | |
3655adc7 | 515 | |
e9a688bc JD |
516 | u32 __get_random_u32_below(u32 ceil) |
517 | { | |
518 | /* | |
519 | * This is the slow path for variable ceil. It is still fast, most of | |
520 | * the time, by doing traditional reciprocal multiplication and | |
521 | * opportunistically comparing the lower half to ceil itself, before | |
522 | * falling back to computing a larger bound, and then rejecting samples | |
523 | * whose lower half would indicate a range indivisible by ceil. The use | |
524 | * of `-ceil % ceil` is analogous to `2^32 % ceil`, but is computable | |
525 | * in 32-bits. | |
526 | */ | |
7f576b25 JD |
527 | u32 rand = get_random_u32(); |
528 | u64 mult; | |
529 | ||
530 | /* | |
531 | * This function is technically undefined for ceil == 0, and in fact | |
532 | * for the non-underscored constant version in the header, we build bug | |
533 | * on that. But for the non-constant case, it's convenient to have that | |
534 | * evaluate to being a straight call to get_random_u32(), so that | |
535 | * get_random_u32_inclusive() can work over its whole range without | |
536 | * undefined behavior. | |
537 | */ | |
538 | if (unlikely(!ceil)) | |
539 | return rand; | |
540 | ||
541 | mult = (u64)ceil * rand; | |
e9a688bc JD |
542 | if (unlikely((u32)mult < ceil)) { |
543 | u32 bound = -ceil % ceil; | |
544 | while (unlikely((u32)mult < bound)) | |
545 | mult = (u64)ceil * get_random_u32(); | |
546 | } | |
547 | return mult >> 32; | |
548 | } | |
549 | EXPORT_SYMBOL(__get_random_u32_below); | |
550 | ||
3191dd5a JD |
551 | #ifdef CONFIG_SMP |
552 | /* | |
553 | * This function is called when the CPU is coming up, with entry | |
554 | * CPUHP_RANDOM_PREPARE, which comes before CPUHP_WORKQUEUE_PREP. | |
555 | */ | |
560181c2 | 556 | int __cold random_prepare_cpu(unsigned int cpu) |
3191dd5a JD |
557 | { |
558 | /* | |
559 | * When the cpu comes back online, immediately invalidate both | |
560 | * the per-cpu crng and all batches, so that we serve fresh | |
561 | * randomness. | |
562 | */ | |
563 | per_cpu_ptr(&crngs, cpu)->generation = ULONG_MAX; | |
a890d1c6 JD |
564 | per_cpu_ptr(&batched_entropy_u8, cpu)->position = UINT_MAX; |
565 | per_cpu_ptr(&batched_entropy_u16, cpu)->position = UINT_MAX; | |
3191dd5a JD |
566 | per_cpu_ptr(&batched_entropy_u32, cpu)->position = UINT_MAX; |
567 | per_cpu_ptr(&batched_entropy_u64, cpu)->position = UINT_MAX; | |
568 | return 0; | |
569 | } | |
570 | #endif | |
571 | ||
a5ed7cb1 JD |
572 | |
573 | /********************************************************************** | |
574 | * | |
575 | * Entropy accumulation and extraction routines. | |
576 | * | |
577 | * Callers may add entropy via: | |
578 | * | |
a1940263 | 579 | * static void mix_pool_bytes(const void *buf, size_t len) |
a5ed7cb1 JD |
580 | * |
581 | * After which, if added entropy should be credited: | |
582 | * | |
a1940263 | 583 | * static void credit_init_bits(size_t bits) |
a5ed7cb1 | 584 | * |
e85c0fc1 | 585 | * Finally, extract entropy via: |
a5ed7cb1 | 586 | * |
a1940263 | 587 | * static void extract_entropy(void *buf, size_t len) |
a5ed7cb1 JD |
588 | * |
589 | **********************************************************************/ | |
590 | ||
3655adc7 JD |
591 | enum { |
592 | POOL_BITS = BLAKE2S_HASH_SIZE * 8, | |
e3d2c5e7 JD |
593 | POOL_READY_BITS = POOL_BITS, /* When crng_init->CRNG_READY */ |
594 | POOL_EARLY_BITS = POOL_READY_BITS / 2 /* When crng_init->CRNG_EARLY */ | |
3655adc7 JD |
595 | }; |
596 | ||
3655adc7 JD |
597 | static struct { |
598 | struct blake2s_state hash; | |
599 | spinlock_t lock; | |
e85c0fc1 | 600 | unsigned int init_bits; |
3655adc7 JD |
601 | } input_pool = { |
602 | .hash.h = { BLAKE2S_IV0 ^ (0x01010000 | BLAKE2S_HASH_SIZE), | |
603 | BLAKE2S_IV1, BLAKE2S_IV2, BLAKE2S_IV3, BLAKE2S_IV4, | |
604 | BLAKE2S_IV5, BLAKE2S_IV6, BLAKE2S_IV7 }, | |
605 | .hash.outlen = BLAKE2S_HASH_SIZE, | |
606 | .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock), | |
607 | }; | |
608 | ||
a1940263 | 609 | static void _mix_pool_bytes(const void *buf, size_t len) |
a5ed7cb1 | 610 | { |
a1940263 | 611 | blake2s_update(&input_pool.hash, buf, len); |
a5ed7cb1 | 612 | } |
3655adc7 JD |
613 | |
614 | /* | |
e85c0fc1 JD |
615 | * This function adds bytes into the input pool. It does not |
616 | * update the initialization bit counter; the caller should call | |
617 | * credit_init_bits if this is appropriate. | |
3655adc7 | 618 | */ |
a1940263 | 619 | static void mix_pool_bytes(const void *buf, size_t len) |
3655adc7 | 620 | { |
a5ed7cb1 JD |
621 | unsigned long flags; |
622 | ||
623 | spin_lock_irqsave(&input_pool.lock, flags); | |
a1940263 | 624 | _mix_pool_bytes(buf, len); |
a5ed7cb1 | 625 | spin_unlock_irqrestore(&input_pool.lock, flags); |
3655adc7 JD |
626 | } |
627 | ||
a5ed7cb1 JD |
628 | /* |
629 | * This is an HKDF-like construction for using the hashed collected entropy | |
630 | * as a PRF key, that's then expanded block-by-block. | |
631 | */ | |
a1940263 | 632 | static void extract_entropy(void *buf, size_t len) |
3655adc7 JD |
633 | { |
634 | unsigned long flags; | |
a5ed7cb1 JD |
635 | u8 seed[BLAKE2S_HASH_SIZE], next_key[BLAKE2S_HASH_SIZE]; |
636 | struct { | |
637 | unsigned long rdseed[32 / sizeof(long)]; | |
638 | size_t counter; | |
639 | } block; | |
d349ab99 | 640 | size_t i, longs; |
a5ed7cb1 | 641 | |
d349ab99 JD |
642 | for (i = 0; i < ARRAY_SIZE(block.rdseed);) { |
643 | longs = arch_get_random_seed_longs(&block.rdseed[i], ARRAY_SIZE(block.rdseed) - i); | |
644 | if (longs) { | |
645 | i += longs; | |
646 | continue; | |
647 | } | |
648 | longs = arch_get_random_longs(&block.rdseed[i], ARRAY_SIZE(block.rdseed) - i); | |
649 | if (longs) { | |
650 | i += longs; | |
651 | continue; | |
652 | } | |
653 | block.rdseed[i++] = random_get_entropy(); | |
a5ed7cb1 | 654 | } |
3655adc7 JD |
655 | |
656 | spin_lock_irqsave(&input_pool.lock, flags); | |
a5ed7cb1 JD |
657 | |
658 | /* seed = HASHPRF(last_key, entropy_input) */ | |
659 | blake2s_final(&input_pool.hash, seed); | |
660 | ||
661 | /* next_key = HASHPRF(seed, RDSEED || 0) */ | |
662 | block.counter = 0; | |
663 | blake2s(next_key, (u8 *)&block, seed, sizeof(next_key), sizeof(block), sizeof(seed)); | |
664 | blake2s_init_key(&input_pool.hash, BLAKE2S_HASH_SIZE, next_key, sizeof(next_key)); | |
665 | ||
3655adc7 | 666 | spin_unlock_irqrestore(&input_pool.lock, flags); |
a5ed7cb1 JD |
667 | memzero_explicit(next_key, sizeof(next_key)); |
668 | ||
a1940263 JD |
669 | while (len) { |
670 | i = min_t(size_t, len, BLAKE2S_HASH_SIZE); | |
a5ed7cb1 JD |
671 | /* output = HASHPRF(seed, RDSEED || ++counter) */ |
672 | ++block.counter; | |
673 | blake2s(buf, (u8 *)&block, seed, i, sizeof(block), sizeof(seed)); | |
a1940263 | 674 | len -= i; |
a5ed7cb1 JD |
675 | buf += i; |
676 | } | |
677 | ||
678 | memzero_explicit(seed, sizeof(seed)); | |
679 | memzero_explicit(&block, sizeof(block)); | |
680 | } | |
681 | ||
560181c2 JD |
682 | #define credit_init_bits(bits) if (!crng_ready()) _credit_init_bits(bits) |
683 | ||
684 | static void __cold _credit_init_bits(size_t bits) | |
5c3b747e | 685 | { |
f5bda35f | 686 | static struct execute_work set_ready; |
fed7ef06 | 687 | unsigned int new, orig, add; |
5c3b747e JD |
688 | unsigned long flags; |
689 | ||
560181c2 | 690 | if (!bits) |
5c3b747e JD |
691 | return; |
692 | ||
a1940263 | 693 | add = min_t(size_t, bits, POOL_BITS); |
5c3b747e | 694 | |
b7a68f67 | 695 | orig = READ_ONCE(input_pool.init_bits); |
5c3b747e | 696 | do { |
fed7ef06 | 697 | new = min_t(unsigned int, POOL_BITS, orig + add); |
b7a68f67 | 698 | } while (!try_cmpxchg(&input_pool.init_bits, &orig, new)); |
5c3b747e | 699 | |
68c9c8b1 JD |
700 | if (orig < POOL_READY_BITS && new >= POOL_READY_BITS) { |
701 | crng_reseed(); /* Sets crng_init to CRNG_READY under base_crng.lock. */ | |
60e5b288 JD |
702 | if (static_key_initialized) |
703 | execute_in_process_context(crng_set_ready, &set_ready); | |
68c9c8b1 JD |
704 | wake_up_interruptible(&crng_init_wait); |
705 | kill_fasync(&fasync, SIGIO, POLL_IN); | |
706 | pr_notice("crng init done\n"); | |
cc1e127b | 707 | if (urandom_warning.missed) |
68c9c8b1 JD |
708 | pr_notice("%d urandom warning(s) missed due to ratelimiting\n", |
709 | urandom_warning.missed); | |
68c9c8b1 | 710 | } else if (orig < POOL_EARLY_BITS && new >= POOL_EARLY_BITS) { |
5c3b747e | 711 | spin_lock_irqsave(&base_crng.lock, flags); |
68c9c8b1 | 712 | /* Check if crng_init is CRNG_EMPTY, to avoid race with crng_reseed(). */ |
e3d2c5e7 | 713 | if (crng_init == CRNG_EMPTY) { |
5c3b747e | 714 | extract_entropy(base_crng.key, sizeof(base_crng.key)); |
e3d2c5e7 | 715 | crng_init = CRNG_EARLY; |
5c3b747e JD |
716 | } |
717 | spin_unlock_irqrestore(&base_crng.lock, flags); | |
718 | } | |
719 | } | |
720 | ||
92c653cf JD |
721 | |
722 | /********************************************************************** | |
723 | * | |
724 | * Entropy collection routines. | |
725 | * | |
726 | * The following exported functions are used for pushing entropy into | |
727 | * the above entropy accumulation routines: | |
728 | * | |
a1940263 JD |
729 | * void add_device_randomness(const void *buf, size_t len); |
730 | * void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy); | |
731 | * void add_bootloader_randomness(const void *buf, size_t len); | |
732 | * void add_vmfork_randomness(const void *unique_vm_id, size_t len); | |
92c653cf | 733 | * void add_interrupt_randomness(int irq); |
a1940263 | 734 | * void add_input_randomness(unsigned int type, unsigned int code, unsigned int value); |
a4b5c26b | 735 | * void add_disk_randomness(struct gendisk *disk); |
92c653cf JD |
736 | * |
737 | * add_device_randomness() adds data to the input pool that | |
738 | * is likely to differ between two devices (or possibly even per boot). | |
739 | * This would be things like MAC addresses or serial numbers, or the | |
740 | * read-out of the RTC. This does *not* credit any actual entropy to | |
741 | * the pool, but it initializes the pool to different values for devices | |
742 | * that might otherwise be identical and have very little entropy | |
743 | * available to them (particularly common in the embedded world). | |
744 | * | |
92c653cf JD |
745 | * add_hwgenerator_randomness() is for true hardware RNGs, and will credit |
746 | * entropy as specified by the caller. If the entropy pool is full it will | |
747 | * block until more entropy is needed. | |
748 | * | |
5c3b747e JD |
749 | * add_bootloader_randomness() is called by bootloader drivers, such as EFI |
750 | * and device tree, and credits its input depending on whether or not the | |
751 | * configuration option CONFIG_RANDOM_TRUST_BOOTLOADER is set. | |
92c653cf | 752 | * |
ae099e8e JD |
753 | * add_vmfork_randomness() adds a unique (but not necessarily secret) ID |
754 | * representing the current instance of a VM to the pool, without crediting, | |
755 | * and then force-reseeds the crng so that it takes effect immediately. | |
756 | * | |
92c653cf JD |
757 | * add_interrupt_randomness() uses the interrupt timing as random |
758 | * inputs to the entropy pool. Using the cycle counters and the irq source | |
759 | * as inputs, it feeds the input pool roughly once a second or after 64 | |
760 | * interrupts, crediting 1 bit of entropy for whichever comes first. | |
761 | * | |
a4b5c26b JD |
762 | * add_input_randomness() uses the input layer interrupt timing, as well |
763 | * as the event type information from the hardware. | |
764 | * | |
765 | * add_disk_randomness() uses what amounts to the seek time of block | |
766 | * layer request events, on a per-disk_devt basis, as input to the | |
767 | * entropy pool. Note that high-speed solid state drives with very low | |
768 | * seek times do not make for good sources of entropy, as their seek | |
769 | * times are usually fairly consistent. | |
770 | * | |
771 | * The last two routines try to estimate how many bits of entropy | |
772 | * to credit. They do this by keeping track of the first and second | |
773 | * order deltas of the event timings. | |
774 | * | |
92c653cf JD |
775 | **********************************************************************/ |
776 | ||
39e0f991 JD |
777 | static bool trust_cpu __initdata = IS_ENABLED(CONFIG_RANDOM_TRUST_CPU); |
778 | static bool trust_bootloader __initdata = IS_ENABLED(CONFIG_RANDOM_TRUST_BOOTLOADER); | |
92c653cf JD |
779 | static int __init parse_trust_cpu(char *arg) |
780 | { | |
781 | return kstrtobool(arg, &trust_cpu); | |
782 | } | |
d97c68d1 JD |
783 | static int __init parse_trust_bootloader(char *arg) |
784 | { | |
785 | return kstrtobool(arg, &trust_bootloader); | |
786 | } | |
92c653cf | 787 | early_param("random.trust_cpu", parse_trust_cpu); |
d97c68d1 | 788 | early_param("random.trust_bootloader", parse_trust_bootloader); |
3655adc7 | 789 | |
b7b67d13 JD |
790 | static int random_pm_notification(struct notifier_block *nb, unsigned long action, void *data) |
791 | { | |
792 | unsigned long flags, entropy = random_get_entropy(); | |
793 | ||
794 | /* | |
795 | * Encode a representation of how long the system has been suspended, | |
796 | * in a way that is distinct from prior system suspends. | |
797 | */ | |
798 | ktime_t stamps[] = { ktime_get(), ktime_get_boottime(), ktime_get_real() }; | |
799 | ||
800 | spin_lock_irqsave(&input_pool.lock, flags); | |
801 | _mix_pool_bytes(&action, sizeof(action)); | |
802 | _mix_pool_bytes(stamps, sizeof(stamps)); | |
803 | _mix_pool_bytes(&entropy, sizeof(entropy)); | |
804 | spin_unlock_irqrestore(&input_pool.lock, flags); | |
805 | ||
806 | if (crng_ready() && (action == PM_RESTORE_PREPARE || | |
261e224d KS |
807 | (action == PM_POST_SUSPEND && !IS_ENABLED(CONFIG_PM_AUTOSLEEP) && |
808 | !IS_ENABLED(CONFIG_PM_USERSPACE_AUTOSLEEP)))) { | |
e85c0fc1 | 809 | crng_reseed(); |
b7b67d13 JD |
810 | pr_notice("crng reseeded on system resumption\n"); |
811 | } | |
812 | return 0; | |
813 | } | |
814 | ||
815 | static struct notifier_block pm_notifier = { .notifier_call = random_pm_notification }; | |
816 | ||
3655adc7 | 817 | /* |
f6238499 JD |
818 | * This is called extremely early, before time keeping functionality is |
819 | * available, but arch randomness is. Interrupts are not yet enabled. | |
3655adc7 | 820 | */ |
f6238499 | 821 | void __init random_init_early(const char *command_line) |
3655adc7 | 822 | { |
d349ab99 | 823 | unsigned long entropy[BLAKE2S_BLOCK_SIZE / sizeof(long)]; |
f6238499 | 824 | size_t i, longs, arch_bits; |
186873c5 | 825 | |
1754abb3 JD |
826 | #if defined(LATENT_ENTROPY_PLUGIN) |
827 | static const u8 compiletime_seed[BLAKE2S_BLOCK_SIZE] __initconst __latent_entropy; | |
828 | _mix_pool_bytes(compiletime_seed, sizeof(compiletime_seed)); | |
829 | #endif | |
830 | ||
d349ab99 | 831 | for (i = 0, arch_bits = sizeof(entropy) * 8; i < ARRAY_SIZE(entropy);) { |
f5e4ec15 | 832 | longs = arch_get_random_seed_longs_early(entropy, ARRAY_SIZE(entropy) - i); |
d349ab99 JD |
833 | if (longs) { |
834 | _mix_pool_bytes(entropy, sizeof(*entropy) * longs); | |
835 | i += longs; | |
836 | continue; | |
837 | } | |
f5e4ec15 | 838 | longs = arch_get_random_longs_early(entropy, ARRAY_SIZE(entropy) - i); |
d349ab99 JD |
839 | if (longs) { |
840 | _mix_pool_bytes(entropy, sizeof(*entropy) * longs); | |
841 | i += longs; | |
842 | continue; | |
92c653cf | 843 | } |
d349ab99 JD |
844 | arch_bits -= sizeof(*entropy) * 8; |
845 | ++i; | |
92c653cf | 846 | } |
f6238499 | 847 | |
dd54fd7d | 848 | _mix_pool_bytes(init_utsname(), sizeof(*(init_utsname()))); |
2f14062b | 849 | _mix_pool_bytes(command_line, strlen(command_line)); |
f6238499 JD |
850 | |
851 | /* Reseed if already seeded by earlier phases. */ | |
852 | if (crng_ready()) | |
853 | crng_reseed(); | |
854 | else if (trust_cpu) | |
855 | _credit_init_bits(arch_bits); | |
856 | } | |
857 | ||
858 | /* | |
859 | * This is called a little bit after the prior function, and now there is | |
860 | * access to timestamps counters. Interrupts are not yet enabled. | |
861 | */ | |
862 | void __init random_init(void) | |
863 | { | |
864 | unsigned long entropy = random_get_entropy(); | |
865 | ktime_t now = ktime_get_real(); | |
866 | ||
f6238499 JD |
867 | _mix_pool_bytes(&now, sizeof(now)); |
868 | _mix_pool_bytes(&entropy, sizeof(entropy)); | |
2f14062b | 869 | add_latent_entropy(); |
186873c5 | 870 | |
60e5b288 | 871 | /* |
f6238499 JD |
872 | * If we were initialized by the cpu or bootloader before jump labels |
873 | * are initialized, then we should enable the static branch here, where | |
60e5b288 JD |
874 | * it's guaranteed that jump labels have been initialized. |
875 | */ | |
876 | if (!static_branch_likely(&crng_is_ready) && crng_init >= CRNG_READY) | |
877 | crng_set_ready(NULL); | |
878 | ||
f6238499 | 879 | /* Reseed if already seeded by earlier phases. */ |
e85c0fc1 JD |
880 | if (crng_ready()) |
881 | crng_reseed(); | |
e192be9d | 882 | |
b7b67d13 JD |
883 | WARN_ON(register_pm_notifier(&pm_notifier)); |
884 | ||
f6238499 JD |
885 | WARN(!entropy, "Missing cycle counter and fallback timer; RNG " |
886 | "entropy collection will consequently suffer."); | |
3655adc7 | 887 | } |
e192be9d | 888 | |
a2080a67 | 889 | /* |
e192be9d TT |
890 | * Add device- or boot-specific data to the input pool to help |
891 | * initialize it. | |
a2080a67 | 892 | * |
e192be9d TT |
893 | * None of this adds any entropy; it is meant to avoid the problem of |
894 | * the entropy pool having similar initial state across largely | |
895 | * identical devices. | |
a2080a67 | 896 | */ |
a1940263 | 897 | void add_device_randomness(const void *buf, size_t len) |
a2080a67 | 898 | { |
4b758eda JD |
899 | unsigned long entropy = random_get_entropy(); |
900 | unsigned long flags; | |
a2080a67 | 901 | |
3ef4cb2d | 902 | spin_lock_irqsave(&input_pool.lock, flags); |
4b758eda | 903 | _mix_pool_bytes(&entropy, sizeof(entropy)); |
a1940263 | 904 | _mix_pool_bytes(buf, len); |
3ef4cb2d | 905 | spin_unlock_irqrestore(&input_pool.lock, flags); |
a2080a67 LT |
906 | } |
907 | EXPORT_SYMBOL(add_device_randomness); | |
908 | ||
92c653cf JD |
909 | /* |
910 | * Interface for in-kernel drivers of true hardware RNGs. | |
911 | * Those devices may produce endless random bits and will be throttled | |
912 | * when our pool is full. | |
913 | */ | |
a1940263 | 914 | void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy) |
92c653cf | 915 | { |
a1940263 | 916 | mix_pool_bytes(buf, len); |
e85c0fc1 JD |
917 | credit_init_bits(entropy); |
918 | ||
92c653cf | 919 | /* |
745558f9 | 920 | * Throttle writing to once every reseed interval, unless we're not yet |
d775335e | 921 | * initialized or no entropy is credited. |
92c653cf | 922 | */ |
d775335e | 923 | if (!kthread_should_stop() && (crng_ready() || !entropy)) |
745558f9 | 924 | schedule_timeout_interruptible(crng_reseed_interval()); |
92c653cf JD |
925 | } |
926 | EXPORT_SYMBOL_GPL(add_hwgenerator_randomness); | |
927 | ||
928 | /* | |
5c3b747e JD |
929 | * Handle random seed passed by bootloader, and credit it if |
930 | * CONFIG_RANDOM_TRUST_BOOTLOADER is set. | |
92c653cf | 931 | */ |
39e0f991 | 932 | void __init add_bootloader_randomness(const void *buf, size_t len) |
92c653cf | 933 | { |
a1940263 | 934 | mix_pool_bytes(buf, len); |
d97c68d1 | 935 | if (trust_bootloader) |
a1940263 | 936 | credit_init_bits(len * 8); |
92c653cf | 937 | } |
92c653cf | 938 | |
a4107d34 | 939 | #if IS_ENABLED(CONFIG_VMGENID) |
f3c2682b JD |
940 | static BLOCKING_NOTIFIER_HEAD(vmfork_chain); |
941 | ||
ae099e8e JD |
942 | /* |
943 | * Handle a new unique VM ID, which is unique, not secret, so we | |
944 | * don't credit it, but we do immediately force a reseed after so | |
945 | * that it's used by the crng posthaste. | |
946 | */ | |
560181c2 | 947 | void __cold add_vmfork_randomness(const void *unique_vm_id, size_t len) |
ae099e8e | 948 | { |
a1940263 | 949 | add_device_randomness(unique_vm_id, len); |
ae099e8e | 950 | if (crng_ready()) { |
e85c0fc1 | 951 | crng_reseed(); |
ae099e8e JD |
952 | pr_notice("crng reseeded due to virtual machine fork\n"); |
953 | } | |
f3c2682b | 954 | blocking_notifier_call_chain(&vmfork_chain, 0, NULL); |
ae099e8e | 955 | } |
a4107d34 | 956 | #if IS_MODULE(CONFIG_VMGENID) |
ae099e8e | 957 | EXPORT_SYMBOL_GPL(add_vmfork_randomness); |
a4107d34 | 958 | #endif |
f3c2682b | 959 | |
560181c2 | 960 | int __cold register_random_vmfork_notifier(struct notifier_block *nb) |
f3c2682b JD |
961 | { |
962 | return blocking_notifier_chain_register(&vmfork_chain, nb); | |
963 | } | |
964 | EXPORT_SYMBOL_GPL(register_random_vmfork_notifier); | |
965 | ||
560181c2 | 966 | int __cold unregister_random_vmfork_notifier(struct notifier_block *nb) |
f3c2682b JD |
967 | { |
968 | return blocking_notifier_chain_unregister(&vmfork_chain, nb); | |
969 | } | |
970 | EXPORT_SYMBOL_GPL(unregister_random_vmfork_notifier); | |
a4107d34 | 971 | #endif |
ae099e8e | 972 | |
92c653cf | 973 | struct fast_pool { |
f5eab0e2 | 974 | unsigned long pool[4]; |
92c653cf | 975 | unsigned long last; |
3191dd5a | 976 | unsigned int count; |
748bc4dd | 977 | struct timer_list mix; |
92c653cf JD |
978 | }; |
979 | ||
748bc4dd JD |
980 | static void mix_interrupt_randomness(struct timer_list *work); |
981 | ||
f5eab0e2 JD |
982 | static DEFINE_PER_CPU(struct fast_pool, irq_randomness) = { |
983 | #ifdef CONFIG_64BIT | |
e73aaae2 | 984 | #define FASTMIX_PERM SIPHASH_PERMUTATION |
748bc4dd | 985 | .pool = { SIPHASH_CONST_0, SIPHASH_CONST_1, SIPHASH_CONST_2, SIPHASH_CONST_3 }, |
f5eab0e2 | 986 | #else |
e73aaae2 | 987 | #define FASTMIX_PERM HSIPHASH_PERMUTATION |
748bc4dd | 988 | .pool = { HSIPHASH_CONST_0, HSIPHASH_CONST_1, HSIPHASH_CONST_2, HSIPHASH_CONST_3 }, |
f5eab0e2 | 989 | #endif |
748bc4dd | 990 | .mix = __TIMER_INITIALIZER(mix_interrupt_randomness, 0) |
f5eab0e2 JD |
991 | }; |
992 | ||
92c653cf | 993 | /* |
f5eab0e2 JD |
994 | * This is [Half]SipHash-1-x, starting from an empty key. Because |
995 | * the key is fixed, it assumes that its inputs are non-malicious, | |
996 | * and therefore this has no security on its own. s represents the | |
4b758eda | 997 | * four-word SipHash state, while v represents a two-word input. |
92c653cf | 998 | */ |
791332b3 | 999 | static void fast_mix(unsigned long s[4], unsigned long v1, unsigned long v2) |
92c653cf | 1000 | { |
791332b3 | 1001 | s[3] ^= v1; |
e73aaae2 | 1002 | FASTMIX_PERM(s[0], s[1], s[2], s[3]); |
791332b3 JD |
1003 | s[0] ^= v1; |
1004 | s[3] ^= v2; | |
e73aaae2 | 1005 | FASTMIX_PERM(s[0], s[1], s[2], s[3]); |
791332b3 | 1006 | s[0] ^= v2; |
92c653cf JD |
1007 | } |
1008 | ||
3191dd5a JD |
1009 | #ifdef CONFIG_SMP |
1010 | /* | |
1011 | * This function is called when the CPU has just come online, with | |
1012 | * entry CPUHP_AP_RANDOM_ONLINE, just after CPUHP_AP_WORKQUEUE_ONLINE. | |
1013 | */ | |
560181c2 | 1014 | int __cold random_online_cpu(unsigned int cpu) |
3191dd5a JD |
1015 | { |
1016 | /* | |
1017 | * During CPU shutdown and before CPU onlining, add_interrupt_ | |
1018 | * randomness() may schedule mix_interrupt_randomness(), and | |
1019 | * set the MIX_INFLIGHT flag. However, because the worker can | |
1020 | * be scheduled on a different CPU during this period, that | |
1021 | * flag will never be cleared. For that reason, we zero out | |
1022 | * the flag here, which runs just after workqueues are onlined | |
1023 | * for the CPU again. This also has the effect of setting the | |
1024 | * irq randomness count to zero so that new accumulated irqs | |
1025 | * are fresh. | |
1026 | */ | |
1027 | per_cpu_ptr(&irq_randomness, cpu)->count = 0; | |
1028 | return 0; | |
1029 | } | |
1030 | #endif | |
1031 | ||
748bc4dd | 1032 | static void mix_interrupt_randomness(struct timer_list *work) |
58340f8e JD |
1033 | { |
1034 | struct fast_pool *fast_pool = container_of(work, struct fast_pool, mix); | |
f5eab0e2 | 1035 | /* |
4b758eda JD |
1036 | * The size of the copied stack pool is explicitly 2 longs so that we |
1037 | * only ever ingest half of the siphash output each time, retaining | |
1038 | * the other half as the next "key" that carries over. The entropy is | |
1039 | * supposed to be sufficiently dispersed between bits so on average | |
1040 | * we don't wind up "losing" some. | |
f5eab0e2 | 1041 | */ |
4b758eda | 1042 | unsigned long pool[2]; |
e3e33fc2 | 1043 | unsigned int count; |
58340f8e JD |
1044 | |
1045 | /* Check to see if we're running on the wrong CPU due to hotplug. */ | |
1046 | local_irq_disable(); | |
1047 | if (fast_pool != this_cpu_ptr(&irq_randomness)) { | |
1048 | local_irq_enable(); | |
58340f8e JD |
1049 | return; |
1050 | } | |
1051 | ||
1052 | /* | |
1053 | * Copy the pool to the stack so that the mixer always has a | |
1054 | * consistent view, before we reenable irqs again. | |
1055 | */ | |
f5eab0e2 | 1056 | memcpy(pool, fast_pool->pool, sizeof(pool)); |
e3e33fc2 | 1057 | count = fast_pool->count; |
3191dd5a | 1058 | fast_pool->count = 0; |
58340f8e JD |
1059 | fast_pool->last = jiffies; |
1060 | local_irq_enable(); | |
1061 | ||
5c3b747e | 1062 | mix_pool_bytes(pool, sizeof(pool)); |
e78a802a | 1063 | credit_init_bits(clamp_t(unsigned int, (count & U16_MAX) / 64, 1, sizeof(pool) * 8)); |
c2a7de4f | 1064 | |
58340f8e JD |
1065 | memzero_explicit(pool, sizeof(pool)); |
1066 | } | |
1067 | ||
703f7066 | 1068 | void add_interrupt_randomness(int irq) |
1da177e4 | 1069 | { |
58340f8e | 1070 | enum { MIX_INFLIGHT = 1U << 31 }; |
4b758eda | 1071 | unsigned long entropy = random_get_entropy(); |
248045b8 JD |
1072 | struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness); |
1073 | struct pt_regs *regs = get_irq_regs(); | |
58340f8e | 1074 | unsigned int new_count; |
b2f408fe | 1075 | |
791332b3 JD |
1076 | fast_mix(fast_pool->pool, entropy, |
1077 | (regs ? instruction_pointer(regs) : _RET_IP_) ^ swab(irq)); | |
3191dd5a | 1078 | new_count = ++fast_pool->count; |
3060d6fe | 1079 | |
58340f8e | 1080 | if (new_count & MIX_INFLIGHT) |
1da177e4 LT |
1081 | return; |
1082 | ||
534d2eaf | 1083 | if (new_count < 1024 && !time_is_before_jiffies(fast_pool->last + HZ)) |
91fcb532 | 1084 | return; |
83664a69 | 1085 | |
3191dd5a | 1086 | fast_pool->count |= MIX_INFLIGHT; |
748bc4dd JD |
1087 | if (!timer_pending(&fast_pool->mix)) { |
1088 | fast_pool->mix.expires = jiffies; | |
1089 | add_timer_on(&fast_pool->mix, raw_smp_processor_id()); | |
1090 | } | |
1da177e4 | 1091 | } |
4b44f2d1 | 1092 | EXPORT_SYMBOL_GPL(add_interrupt_randomness); |
1da177e4 | 1093 | |
a4b5c26b JD |
1094 | /* There is one of these per entropy source */ |
1095 | struct timer_rand_state { | |
1096 | unsigned long last_time; | |
1097 | long last_delta, last_delta2; | |
1098 | }; | |
1099 | ||
1100 | /* | |
1101 | * This function adds entropy to the entropy "pool" by using timing | |
e3e33fc2 JD |
1102 | * delays. It uses the timer_rand_state structure to make an estimate |
1103 | * of how many bits of entropy this call has added to the pool. The | |
1104 | * value "num" is also added to the pool; it should somehow describe | |
1105 | * the type of event that just happened. | |
a4b5c26b JD |
1106 | */ |
1107 | static void add_timer_randomness(struct timer_rand_state *state, unsigned int num) | |
1108 | { | |
1109 | unsigned long entropy = random_get_entropy(), now = jiffies, flags; | |
1110 | long delta, delta2, delta3; | |
e3e33fc2 | 1111 | unsigned int bits; |
a4b5c26b | 1112 | |
e3e33fc2 JD |
1113 | /* |
1114 | * If we're in a hard IRQ, add_interrupt_randomness() will be called | |
1115 | * sometime after, so mix into the fast pool. | |
1116 | */ | |
1117 | if (in_hardirq()) { | |
791332b3 | 1118 | fast_mix(this_cpu_ptr(&irq_randomness)->pool, entropy, num); |
e3e33fc2 JD |
1119 | } else { |
1120 | spin_lock_irqsave(&input_pool.lock, flags); | |
1121 | _mix_pool_bytes(&entropy, sizeof(entropy)); | |
1122 | _mix_pool_bytes(&num, sizeof(num)); | |
1123 | spin_unlock_irqrestore(&input_pool.lock, flags); | |
1124 | } | |
a4b5c26b JD |
1125 | |
1126 | if (crng_ready()) | |
1127 | return; | |
1128 | ||
1129 | /* | |
1130 | * Calculate number of bits of randomness we probably added. | |
1131 | * We take into account the first, second and third-order deltas | |
1132 | * in order to make our estimate. | |
1133 | */ | |
1134 | delta = now - READ_ONCE(state->last_time); | |
1135 | WRITE_ONCE(state->last_time, now); | |
1136 | ||
1137 | delta2 = delta - READ_ONCE(state->last_delta); | |
1138 | WRITE_ONCE(state->last_delta, delta); | |
1139 | ||
1140 | delta3 = delta2 - READ_ONCE(state->last_delta2); | |
1141 | WRITE_ONCE(state->last_delta2, delta2); | |
1142 | ||
1143 | if (delta < 0) | |
1144 | delta = -delta; | |
1145 | if (delta2 < 0) | |
1146 | delta2 = -delta2; | |
1147 | if (delta3 < 0) | |
1148 | delta3 = -delta3; | |
1149 | if (delta > delta2) | |
1150 | delta = delta2; | |
1151 | if (delta > delta3) | |
1152 | delta = delta3; | |
1153 | ||
1154 | /* | |
e3e33fc2 JD |
1155 | * delta is now minimum absolute delta. Round down by 1 bit |
1156 | * on general principles, and limit entropy estimate to 11 bits. | |
1157 | */ | |
1158 | bits = min(fls(delta >> 1), 11); | |
1159 | ||
1160 | /* | |
1161 | * As mentioned above, if we're in a hard IRQ, add_interrupt_randomness() | |
1162 | * will run after this, which uses a different crediting scheme of 1 bit | |
1163 | * per every 64 interrupts. In order to let that function do accounting | |
1164 | * close to the one in this function, we credit a full 64/64 bit per bit, | |
1165 | * and then subtract one to account for the extra one added. | |
a4b5c26b | 1166 | */ |
e3e33fc2 JD |
1167 | if (in_hardirq()) |
1168 | this_cpu_ptr(&irq_randomness)->count += max(1u, bits * 64) - 1; | |
1169 | else | |
560181c2 | 1170 | _credit_init_bits(bits); |
a4b5c26b JD |
1171 | } |
1172 | ||
a1940263 | 1173 | void add_input_randomness(unsigned int type, unsigned int code, unsigned int value) |
a4b5c26b JD |
1174 | { |
1175 | static unsigned char last_value; | |
1176 | static struct timer_rand_state input_timer_state = { INITIAL_JIFFIES }; | |
1177 | ||
1178 | /* Ignore autorepeat and the like. */ | |
1179 | if (value == last_value) | |
1180 | return; | |
1181 | ||
1182 | last_value = value; | |
1183 | add_timer_randomness(&input_timer_state, | |
1184 | (type << 4) ^ code ^ (code >> 4) ^ value); | |
1185 | } | |
1186 | EXPORT_SYMBOL_GPL(add_input_randomness); | |
1187 | ||
1188 | #ifdef CONFIG_BLOCK | |
1189 | void add_disk_randomness(struct gendisk *disk) | |
1190 | { | |
1191 | if (!disk || !disk->random) | |
1192 | return; | |
1193 | /* First major is 1, so we get >= 0x200 here. */ | |
1194 | add_timer_randomness(disk->random, 0x100 + disk_devt(disk)); | |
1195 | } | |
1196 | EXPORT_SYMBOL_GPL(add_disk_randomness); | |
1197 | ||
560181c2 | 1198 | void __cold rand_initialize_disk(struct gendisk *disk) |
a4b5c26b JD |
1199 | { |
1200 | struct timer_rand_state *state; | |
1201 | ||
1202 | /* | |
1203 | * If kzalloc returns null, we just won't use that entropy | |
1204 | * source. | |
1205 | */ | |
1206 | state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL); | |
1207 | if (state) { | |
1208 | state->last_time = INITIAL_JIFFIES; | |
1209 | disk->random = state; | |
1210 | } | |
1211 | } | |
1212 | #endif | |
1213 | ||
78c768e6 JD |
1214 | struct entropy_timer_state { |
1215 | unsigned long entropy; | |
1216 | struct timer_list timer; | |
1217 | unsigned int samples, samples_per_bit; | |
1218 | }; | |
1219 | ||
50ee7529 LT |
1220 | /* |
1221 | * Each time the timer fires, we expect that we got an unpredictable | |
1222 | * jump in the cycle counter. Even if the timer is running on another | |
1223 | * CPU, the timer activity will be touching the stack of the CPU that is | |
1224 | * generating entropy.. | |
1225 | * | |
1226 | * Note that we don't re-arm the timer in the timer itself - we are | |
1227 | * happy to be scheduled away, since that just makes the load more | |
1228 | * complex, but we do not want the timer to keep ticking unless the | |
1229 | * entropy loop is running. | |
1230 | * | |
1231 | * So the re-arming always happens in the entropy loop itself. | |
1232 | */ | |
560181c2 | 1233 | static void __cold entropy_timer(struct timer_list *timer) |
50ee7529 | 1234 | { |
78c768e6 JD |
1235 | struct entropy_timer_state *state = container_of(timer, struct entropy_timer_state, timer); |
1236 | ||
1237 | if (++state->samples == state->samples_per_bit) { | |
e85c0fc1 | 1238 | credit_init_bits(1); |
78c768e6 JD |
1239 | state->samples = 0; |
1240 | } | |
50ee7529 LT |
1241 | } |
1242 | ||
1243 | /* | |
1244 | * If we have an actual cycle counter, see if we can | |
1245 | * generate enough entropy with timing noise | |
1246 | */ | |
560181c2 | 1247 | static void __cold try_to_generate_entropy(void) |
50ee7529 | 1248 | { |
12273347 | 1249 | enum { NUM_TRIAL_SAMPLES = 8192, MAX_SAMPLES_PER_BIT = HZ / 15 }; |
78c768e6 JD |
1250 | struct entropy_timer_state stack; |
1251 | unsigned int i, num_different = 0; | |
1252 | unsigned long last = random_get_entropy(); | |
50ee7529 | 1253 | |
78c768e6 JD |
1254 | for (i = 0; i < NUM_TRIAL_SAMPLES - 1; ++i) { |
1255 | stack.entropy = random_get_entropy(); | |
1256 | if (stack.entropy != last) | |
1257 | ++num_different; | |
1258 | last = stack.entropy; | |
1259 | } | |
1260 | stack.samples_per_bit = DIV_ROUND_UP(NUM_TRIAL_SAMPLES, num_different + 1); | |
1261 | if (stack.samples_per_bit > MAX_SAMPLES_PER_BIT) | |
50ee7529 LT |
1262 | return; |
1263 | ||
78c768e6 | 1264 | stack.samples = 0; |
50ee7529 | 1265 | timer_setup_on_stack(&stack.timer, entropy_timer, 0); |
3e504d20 | 1266 | while (!crng_ready() && !signal_pending(current)) { |
50ee7529 | 1267 | if (!timer_pending(&stack.timer)) |
12273347 | 1268 | mod_timer(&stack.timer, jiffies); |
4b758eda | 1269 | mix_pool_bytes(&stack.entropy, sizeof(stack.entropy)); |
50ee7529 | 1270 | schedule(); |
4b758eda | 1271 | stack.entropy = random_get_entropy(); |
50ee7529 LT |
1272 | } |
1273 | ||
1274 | del_timer_sync(&stack.timer); | |
1275 | destroy_timer_on_stack(&stack.timer); | |
4b758eda | 1276 | mix_pool_bytes(&stack.entropy, sizeof(stack.entropy)); |
50ee7529 LT |
1277 | } |
1278 | ||
a6adf8e7 JD |
1279 | |
1280 | /********************************************************************** | |
1281 | * | |
1282 | * Userspace reader/writer interfaces. | |
1283 | * | |
1284 | * getrandom(2) is the primary modern interface into the RNG and should | |
1285 | * be used in preference to anything else. | |
1286 | * | |
0313bc27 LT |
1287 | * Reading from /dev/random has the same functionality as calling |
1288 | * getrandom(2) with flags=0. In earlier versions, however, it had | |
1289 | * vastly different semantics and should therefore be avoided, to | |
1290 | * prevent backwards compatibility issues. | |
1291 | * | |
1292 | * Reading from /dev/urandom has the same functionality as calling | |
1293 | * getrandom(2) with flags=GRND_INSECURE. Because it does not block | |
1294 | * waiting for the RNG to be ready, it should not be used. | |
a6adf8e7 JD |
1295 | * |
1296 | * Writing to either /dev/random or /dev/urandom adds entropy to | |
1297 | * the input pool but does not credit it. | |
1298 | * | |
0313bc27 LT |
1299 | * Polling on /dev/random indicates when the RNG is initialized, on |
1300 | * the read side, and when it wants new entropy, on the write side. | |
a6adf8e7 JD |
1301 | * |
1302 | * Both /dev/random and /dev/urandom have the same set of ioctls for | |
1303 | * adding entropy, getting the entropy count, zeroing the count, and | |
1304 | * reseeding the crng. | |
1305 | * | |
1306 | **********************************************************************/ | |
1307 | ||
a1940263 | 1308 | SYSCALL_DEFINE3(getrandom, char __user *, ubuf, size_t, len, unsigned int, flags) |
1da177e4 | 1309 | { |
1b388e77 JA |
1310 | struct iov_iter iter; |
1311 | struct iovec iov; | |
1312 | int ret; | |
1313 | ||
a6adf8e7 JD |
1314 | if (flags & ~(GRND_NONBLOCK | GRND_RANDOM | GRND_INSECURE)) |
1315 | return -EINVAL; | |
301f0595 | 1316 | |
a6adf8e7 JD |
1317 | /* |
1318 | * Requesting insecure and blocking randomness at the same time makes | |
1319 | * no sense. | |
1320 | */ | |
1321 | if ((flags & (GRND_INSECURE | GRND_RANDOM)) == (GRND_INSECURE | GRND_RANDOM)) | |
1322 | return -EINVAL; | |
c6f1deb1 | 1323 | |
f5bda35f | 1324 | if (!crng_ready() && !(flags & GRND_INSECURE)) { |
a6adf8e7 JD |
1325 | if (flags & GRND_NONBLOCK) |
1326 | return -EAGAIN; | |
1327 | ret = wait_for_random_bytes(); | |
1328 | if (unlikely(ret)) | |
1329 | return ret; | |
1330 | } | |
1b388e77 JA |
1331 | |
1332 | ret = import_single_range(READ, ubuf, len, &iov, &iter); | |
1333 | if (unlikely(ret)) | |
1334 | return ret; | |
1335 | return get_random_bytes_user(&iter); | |
30c08efe AL |
1336 | } |
1337 | ||
248045b8 | 1338 | static __poll_t random_poll(struct file *file, poll_table *wait) |
1da177e4 | 1339 | { |
30c08efe | 1340 | poll_wait(file, &crng_init_wait, wait); |
e85c0fc1 | 1341 | return crng_ready() ? EPOLLIN | EPOLLRDNORM : EPOLLOUT | EPOLLWRNORM; |
1da177e4 LT |
1342 | } |
1343 | ||
1ce6c8d6 | 1344 | static ssize_t write_pool_user(struct iov_iter *iter) |
1da177e4 | 1345 | { |
04ec96b7 | 1346 | u8 block[BLAKE2S_BLOCK_SIZE]; |
22b0a222 JA |
1347 | ssize_t ret = 0; |
1348 | size_t copied; | |
1da177e4 | 1349 | |
22b0a222 JA |
1350 | if (unlikely(!iov_iter_count(iter))) |
1351 | return 0; | |
1352 | ||
1353 | for (;;) { | |
1354 | copied = copy_from_iter(block, sizeof(block), iter); | |
1355 | ret += copied; | |
1356 | mix_pool_bytes(block, copied); | |
1357 | if (!iov_iter_count(iter) || copied != sizeof(block)) | |
1358 | break; | |
1ce6c8d6 JD |
1359 | |
1360 | BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0); | |
1361 | if (ret % PAGE_SIZE == 0) { | |
1362 | if (signal_pending(current)) | |
1363 | break; | |
1364 | cond_resched(); | |
1365 | } | |
1da177e4 | 1366 | } |
7f397dcd | 1367 | |
7b5164fb | 1368 | memzero_explicit(block, sizeof(block)); |
22b0a222 | 1369 | return ret ? ret : -EFAULT; |
7f397dcd MM |
1370 | } |
1371 | ||
22b0a222 | 1372 | static ssize_t random_write_iter(struct kiocb *kiocb, struct iov_iter *iter) |
7f397dcd | 1373 | { |
1ce6c8d6 | 1374 | return write_pool_user(iter); |
1da177e4 LT |
1375 | } |
1376 | ||
1b388e77 | 1377 | static ssize_t urandom_read_iter(struct kiocb *kiocb, struct iov_iter *iter) |
0313bc27 LT |
1378 | { |
1379 | static int maxwarn = 10; | |
1380 | ||
48bff105 JD |
1381 | /* |
1382 | * Opportunistically attempt to initialize the RNG on platforms that | |
1383 | * have fast cycle counters, but don't (for now) require it to succeed. | |
1384 | */ | |
1385 | if (!crng_ready()) | |
1386 | try_to_generate_entropy(); | |
1387 | ||
cc1e127b JD |
1388 | if (!crng_ready()) { |
1389 | if (!ratelimit_disable && maxwarn <= 0) | |
1390 | ++urandom_warning.missed; | |
1391 | else if (ratelimit_disable || __ratelimit(&urandom_warning)) { | |
1392 | --maxwarn; | |
1b388e77 JA |
1393 | pr_notice("%s: uninitialized urandom read (%zu bytes read)\n", |
1394 | current->comm, iov_iter_count(iter)); | |
cc1e127b | 1395 | } |
0313bc27 LT |
1396 | } |
1397 | ||
1b388e77 | 1398 | return get_random_bytes_user(iter); |
0313bc27 LT |
1399 | } |
1400 | ||
1b388e77 | 1401 | static ssize_t random_read_iter(struct kiocb *kiocb, struct iov_iter *iter) |
a6adf8e7 JD |
1402 | { |
1403 | int ret; | |
1404 | ||
cd4f24ae JD |
1405 | if (!crng_ready() && |
1406 | ((kiocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO)) || | |
1407 | (kiocb->ki_filp->f_flags & O_NONBLOCK))) | |
1408 | return -EAGAIN; | |
1409 | ||
a6adf8e7 JD |
1410 | ret = wait_for_random_bytes(); |
1411 | if (ret != 0) | |
1412 | return ret; | |
1b388e77 | 1413 | return get_random_bytes_user(iter); |
a6adf8e7 JD |
1414 | } |
1415 | ||
43ae4860 | 1416 | static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg) |
1da177e4 | 1417 | { |
1da177e4 | 1418 | int __user *p = (int __user *)arg; |
22b0a222 | 1419 | int ent_count; |
1da177e4 LT |
1420 | |
1421 | switch (cmd) { | |
1422 | case RNDGETENTCNT: | |
a6adf8e7 | 1423 | /* Inherently racy, no point locking. */ |
e85c0fc1 | 1424 | if (put_user(input_pool.init_bits, p)) |
1da177e4 LT |
1425 | return -EFAULT; |
1426 | return 0; | |
1427 | case RNDADDTOENTCNT: | |
1428 | if (!capable(CAP_SYS_ADMIN)) | |
1429 | return -EPERM; | |
1430 | if (get_user(ent_count, p)) | |
1431 | return -EFAULT; | |
a49c010e JD |
1432 | if (ent_count < 0) |
1433 | return -EINVAL; | |
e85c0fc1 | 1434 | credit_init_bits(ent_count); |
a49c010e | 1435 | return 0; |
22b0a222 JA |
1436 | case RNDADDENTROPY: { |
1437 | struct iov_iter iter; | |
1438 | struct iovec iov; | |
1439 | ssize_t ret; | |
1440 | int len; | |
1441 | ||
1da177e4 LT |
1442 | if (!capable(CAP_SYS_ADMIN)) |
1443 | return -EPERM; | |
1444 | if (get_user(ent_count, p++)) | |
1445 | return -EFAULT; | |
1446 | if (ent_count < 0) | |
1447 | return -EINVAL; | |
22b0a222 JA |
1448 | if (get_user(len, p++)) |
1449 | return -EFAULT; | |
1450 | ret = import_single_range(WRITE, p, len, &iov, &iter); | |
1451 | if (unlikely(ret)) | |
1452 | return ret; | |
1ce6c8d6 | 1453 | ret = write_pool_user(&iter); |
22b0a222 JA |
1454 | if (unlikely(ret < 0)) |
1455 | return ret; | |
1456 | /* Since we're crediting, enforce that it was all written into the pool. */ | |
1457 | if (unlikely(ret != len)) | |
1da177e4 | 1458 | return -EFAULT; |
e85c0fc1 | 1459 | credit_init_bits(ent_count); |
a49c010e | 1460 | return 0; |
22b0a222 | 1461 | } |
1da177e4 LT |
1462 | case RNDZAPENTCNT: |
1463 | case RNDCLEARPOOL: | |
e85c0fc1 | 1464 | /* No longer has any effect. */ |
1da177e4 LT |
1465 | if (!capable(CAP_SYS_ADMIN)) |
1466 | return -EPERM; | |
1da177e4 | 1467 | return 0; |
d848e5f8 TT |
1468 | case RNDRESEEDCRNG: |
1469 | if (!capable(CAP_SYS_ADMIN)) | |
1470 | return -EPERM; | |
a96cfe2d | 1471 | if (!crng_ready()) |
d848e5f8 | 1472 | return -ENODATA; |
e85c0fc1 | 1473 | crng_reseed(); |
d848e5f8 | 1474 | return 0; |
1da177e4 LT |
1475 | default: |
1476 | return -EINVAL; | |
1477 | } | |
1478 | } | |
1479 | ||
9a6f70bb JD |
1480 | static int random_fasync(int fd, struct file *filp, int on) |
1481 | { | |
1482 | return fasync_helper(fd, filp, on, &fasync); | |
1483 | } | |
1484 | ||
2b8693c0 | 1485 | const struct file_operations random_fops = { |
1b388e77 | 1486 | .read_iter = random_read_iter, |
22b0a222 | 1487 | .write_iter = random_write_iter, |
248045b8 | 1488 | .poll = random_poll, |
43ae4860 | 1489 | .unlocked_ioctl = random_ioctl, |
507e4e2b | 1490 | .compat_ioctl = compat_ptr_ioctl, |
9a6f70bb | 1491 | .fasync = random_fasync, |
6038f373 | 1492 | .llseek = noop_llseek, |
79025e72 JA |
1493 | .splice_read = generic_file_splice_read, |
1494 | .splice_write = iter_file_splice_write, | |
1da177e4 LT |
1495 | }; |
1496 | ||
0313bc27 | 1497 | const struct file_operations urandom_fops = { |
1b388e77 | 1498 | .read_iter = urandom_read_iter, |
22b0a222 | 1499 | .write_iter = random_write_iter, |
0313bc27 LT |
1500 | .unlocked_ioctl = random_ioctl, |
1501 | .compat_ioctl = compat_ptr_ioctl, | |
1502 | .fasync = random_fasync, | |
1503 | .llseek = noop_llseek, | |
79025e72 JA |
1504 | .splice_read = generic_file_splice_read, |
1505 | .splice_write = iter_file_splice_write, | |
0313bc27 LT |
1506 | }; |
1507 | ||
0deff3c4 | 1508 | |
1da177e4 LT |
1509 | /******************************************************************** |
1510 | * | |
0deff3c4 JD |
1511 | * Sysctl interface. |
1512 | * | |
1513 | * These are partly unused legacy knobs with dummy values to not break | |
1514 | * userspace and partly still useful things. They are usually accessible | |
1515 | * in /proc/sys/kernel/random/ and are as follows: | |
1516 | * | |
1517 | * - boot_id - a UUID representing the current boot. | |
1518 | * | |
1519 | * - uuid - a random UUID, different each time the file is read. | |
1520 | * | |
1521 | * - poolsize - the number of bits of entropy that the input pool can | |
1522 | * hold, tied to the POOL_BITS constant. | |
1523 | * | |
1524 | * - entropy_avail - the number of bits of entropy currently in the | |
1525 | * input pool. Always <= poolsize. | |
1526 | * | |
1527 | * - write_wakeup_threshold - the amount of entropy in the input pool | |
1528 | * below which write polls to /dev/random will unblock, requesting | |
e3d2c5e7 | 1529 | * more entropy, tied to the POOL_READY_BITS constant. It is writable |
0deff3c4 JD |
1530 | * to avoid breaking old userspaces, but writing to it does not |
1531 | * change any behavior of the RNG. | |
1532 | * | |
d0efdf35 | 1533 | * - urandom_min_reseed_secs - fixed to the value CRNG_RESEED_INTERVAL. |
0deff3c4 JD |
1534 | * It is writable to avoid breaking old userspaces, but writing |
1535 | * to it does not change any behavior of the RNG. | |
1da177e4 LT |
1536 | * |
1537 | ********************************************************************/ | |
1538 | ||
1539 | #ifdef CONFIG_SYSCTL | |
1540 | ||
1541 | #include <linux/sysctl.h> | |
1542 | ||
d0efdf35 | 1543 | static int sysctl_random_min_urandom_seed = CRNG_RESEED_INTERVAL / HZ; |
e3d2c5e7 | 1544 | static int sysctl_random_write_wakeup_bits = POOL_READY_BITS; |
489c7fc4 | 1545 | static int sysctl_poolsize = POOL_BITS; |
64276a99 | 1546 | static u8 sysctl_bootid[UUID_SIZE]; |
1da177e4 LT |
1547 | |
1548 | /* | |
f22052b2 | 1549 | * This function is used to return both the bootid UUID, and random |
64276a99 | 1550 | * UUID. The difference is in whether table->data is NULL; if it is, |
1da177e4 | 1551 | * then a new UUID is generated and returned to the user. |
1da177e4 | 1552 | */ |
a1940263 | 1553 | static int proc_do_uuid(struct ctl_table *table, int write, void *buf, |
248045b8 | 1554 | size_t *lenp, loff_t *ppos) |
1da177e4 | 1555 | { |
64276a99 JD |
1556 | u8 tmp_uuid[UUID_SIZE], *uuid; |
1557 | char uuid_string[UUID_STRING_LEN + 1]; | |
1558 | struct ctl_table fake_table = { | |
1559 | .data = uuid_string, | |
1560 | .maxlen = UUID_STRING_LEN | |
1561 | }; | |
1562 | ||
1563 | if (write) | |
1564 | return -EPERM; | |
1da177e4 LT |
1565 | |
1566 | uuid = table->data; | |
1567 | if (!uuid) { | |
1568 | uuid = tmp_uuid; | |
1da177e4 | 1569 | generate_random_uuid(uuid); |
44e4360f MD |
1570 | } else { |
1571 | static DEFINE_SPINLOCK(bootid_spinlock); | |
1572 | ||
1573 | spin_lock(&bootid_spinlock); | |
1574 | if (!uuid[8]) | |
1575 | generate_random_uuid(uuid); | |
1576 | spin_unlock(&bootid_spinlock); | |
1577 | } | |
1da177e4 | 1578 | |
64276a99 | 1579 | snprintf(uuid_string, sizeof(uuid_string), "%pU", uuid); |
a1940263 | 1580 | return proc_dostring(&fake_table, 0, buf, lenp, ppos); |
1da177e4 LT |
1581 | } |
1582 | ||
77553cf8 | 1583 | /* The same as proc_dointvec, but writes don't change anything. */ |
a1940263 | 1584 | static int proc_do_rointvec(struct ctl_table *table, int write, void *buf, |
77553cf8 JD |
1585 | size_t *lenp, loff_t *ppos) |
1586 | { | |
a1940263 | 1587 | return write ? 0 : proc_dointvec(table, 0, buf, lenp, ppos); |
77553cf8 JD |
1588 | } |
1589 | ||
5475e8f0 | 1590 | static struct ctl_table random_table[] = { |
1da177e4 | 1591 | { |
1da177e4 LT |
1592 | .procname = "poolsize", |
1593 | .data = &sysctl_poolsize, | |
1594 | .maxlen = sizeof(int), | |
1595 | .mode = 0444, | |
6d456111 | 1596 | .proc_handler = proc_dointvec, |
1da177e4 LT |
1597 | }, |
1598 | { | |
1da177e4 | 1599 | .procname = "entropy_avail", |
e85c0fc1 | 1600 | .data = &input_pool.init_bits, |
1da177e4 LT |
1601 | .maxlen = sizeof(int), |
1602 | .mode = 0444, | |
c5704490 | 1603 | .proc_handler = proc_dointvec, |
1da177e4 | 1604 | }, |
1da177e4 | 1605 | { |
1da177e4 | 1606 | .procname = "write_wakeup_threshold", |
0deff3c4 | 1607 | .data = &sysctl_random_write_wakeup_bits, |
1da177e4 LT |
1608 | .maxlen = sizeof(int), |
1609 | .mode = 0644, | |
77553cf8 | 1610 | .proc_handler = proc_do_rointvec, |
1da177e4 | 1611 | }, |
f5c2742c TT |
1612 | { |
1613 | .procname = "urandom_min_reseed_secs", | |
0deff3c4 | 1614 | .data = &sysctl_random_min_urandom_seed, |
f5c2742c TT |
1615 | .maxlen = sizeof(int), |
1616 | .mode = 0644, | |
77553cf8 | 1617 | .proc_handler = proc_do_rointvec, |
f5c2742c | 1618 | }, |
1da177e4 | 1619 | { |
1da177e4 LT |
1620 | .procname = "boot_id", |
1621 | .data = &sysctl_bootid, | |
1da177e4 | 1622 | .mode = 0444, |
6d456111 | 1623 | .proc_handler = proc_do_uuid, |
1da177e4 LT |
1624 | }, |
1625 | { | |
1da177e4 | 1626 | .procname = "uuid", |
1da177e4 | 1627 | .mode = 0444, |
6d456111 | 1628 | .proc_handler = proc_do_uuid, |
1da177e4 | 1629 | }, |
894d2491 | 1630 | { } |
1da177e4 | 1631 | }; |
5475e8f0 XN |
1632 | |
1633 | /* | |
2f14062b JD |
1634 | * random_init() is called before sysctl_init(), |
1635 | * so we cannot call register_sysctl_init() in random_init() | |
5475e8f0 XN |
1636 | */ |
1637 | static int __init random_sysctls_init(void) | |
1638 | { | |
1639 | register_sysctl_init("kernel/random", random_table); | |
1640 | return 0; | |
1641 | } | |
1642 | device_initcall(random_sysctls_init); | |
0deff3c4 | 1643 | #endif |