1// SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause) 2/* 3 * Copyright (C) 2017-2022 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved. 4 * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005 5 * Copyright Theodore Ts'o, 1994, 1995, 1996, 1997, 1998, 1999. All rights reserved. 6 * 7 * This driver produces cryptographically secure pseudorandom data. It is divided 8 * into roughly six sections, each with a section header: 9 * 10 * - Initialization and readiness waiting. 11 * - Fast key erasure RNG, the "crng". 12 * - Entropy accumulation and extraction routines. 13 * - Entropy collection routines. 14 * - Userspace reader/writer interfaces. 15 * - Sysctl interface. 16 * 17 * The high level overview is that there is one input pool, into which 18 * various pieces of data are hashed. Prior to initialization, some of that 19 * data is then "credited" as having a certain number of bits of entropy. 20 * When enough bits of entropy are available, the hash is finalized and 21 * handed as a key to a stream cipher that expands it indefinitely for 22 * various consumers. This key is periodically refreshed as the various 23 * entropy collectors, described below, add data to the input pool. 24 */ 25 26#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 27 28#include <linux/utsname.h> 29#include <linux/module.h> 30#include <linux/kernel.h> 31#include <linux/major.h> 32#include <linux/string.h> 33#include <linux/fcntl.h> 34#include <linux/slab.h> 35#include <linux/random.h> 36#include <linux/poll.h> 37#include <linux/init.h> 38#include <linux/fs.h> 39#include <linux/genhd.h> 40#include <linux/interrupt.h> 41#include <linux/mm.h> 42#include <linux/nodemask.h> 43#include <linux/spinlock.h> 44#include <linux/kthread.h> 45#include <linux/percpu.h> 46#include <linux/ptrace.h> 47#include <linux/workqueue.h> 48#include <linux/irq.h> 49#include <linux/ratelimit.h> 50#include <linux/syscalls.h> 51#include <linux/completion.h> 52#include <linux/uuid.h> 53#include <linux/uaccess.h> 54#include <linux/siphash.h> 55#include <linux/uio.h> 56#include <crypto/chacha.h> 57#include <crypto/blake2s.h> 58#include <asm/processor.h> 59#include <asm/irq.h> 60#include <asm/irq_regs.h> 61#include <asm/io.h> 62 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 *********************************************************************/ 72 73/* 74 * crng_init is protected by base_crng->lock, and only increases 75 * its value (from empty->early->ready). 76 */ 77static 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 */ 81} crng_init __read_mostly = CRNG_EMPTY; 82#define crng_ready() (likely(crng_init >= CRNG_READY)) 83/* Various types of waiters for crng_init->CRNG_READY transition. */ 84static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait); 85static struct fasync_struct *fasync; 86static DEFINE_SPINLOCK(random_ready_chain_lock); 87static RAW_NOTIFIER_HEAD(random_ready_chain); 88 89/* Control how we warn userspace. */ 90static struct ratelimit_state urandom_warning = 91 RATELIMIT_STATE_INIT_FLAGS("urandom_warning", HZ, 3, RATELIMIT_MSG_ON_RELEASE); 92static int ratelimit_disable __read_mostly = 93 IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM); 94module_param_named(ratelimit_disable, ratelimit_disable, int, 0644); 95MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression"); 96 97/* 98 * Returns whether or not the input pool has been seeded and thus guaranteed 99 * to supply cryptographically secure random numbers. This applies to: the 100 * /dev/urandom device, the get_random_bytes function, and the get_random_{u32, 101 * ,u64,int,long} family of functions. 102 * 103 * Returns: true if the input pool has been seeded. 104 * false if the input pool has not been seeded. 105 */ 106bool rng_is_initialized(void) 107{ 108 return crng_ready(); 109} 110EXPORT_SYMBOL(rng_is_initialized); 111 112/* Used by wait_for_random_bytes(), and considered an entropy collector, below. */ 113static void try_to_generate_entropy(void); 114 115/* 116 * Wait for the input pool to be seeded and thus guaranteed to supply 117 * cryptographically secure random numbers. This applies to: the /dev/urandom 118 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long} 119 * family of functions. Using any of these functions without first calling 120 * this function forfeits the guarantee of security. 121 * 122 * Returns: 0 if the input pool has been seeded. 123 * -ERESTARTSYS if the function was interrupted by a signal. 124 */ 125int wait_for_random_bytes(void) 126{ 127 while (!crng_ready()) { 128 int ret; 129 130 try_to_generate_entropy(); 131 ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ); 132 if (ret) 133 return ret > 0 ? 0 : ret; 134 } 135 return 0; 136} 137EXPORT_SYMBOL(wait_for_random_bytes); 138 139/* 140 * Add a callback function that will be invoked when the input 141 * pool is initialised. 142 * 143 * returns: 0 if callback is successfully added 144 * -EALREADY if pool is already initialised (callback not called) 145 */ 146int __cold register_random_ready_notifier(struct notifier_block *nb) 147{ 148 unsigned long flags; 149 int ret = -EALREADY; 150 151 if (crng_ready()) 152 return ret; 153 154 spin_lock_irqsave(&random_ready_chain_lock, flags); 155 if (!crng_ready()) 156 ret = raw_notifier_chain_register(&random_ready_chain, nb); 157 spin_unlock_irqrestore(&random_ready_chain_lock, flags); 158 return ret; 159} 160 161/* 162 * Delete a previously registered readiness callback function. 163 */ 164int __cold unregister_random_ready_notifier(struct notifier_block *nb) 165{ 166 unsigned long flags; 167 int ret; 168 169 spin_lock_irqsave(&random_ready_chain_lock, flags); 170 ret = raw_notifier_chain_unregister(&random_ready_chain, nb); 171 spin_unlock_irqrestore(&random_ready_chain_lock, flags); 172 return ret; 173} 174 175static void __cold process_random_ready_list(void) 176{ 177 unsigned long flags; 178 179 spin_lock_irqsave(&random_ready_chain_lock, flags); 180 raw_notifier_call_chain(&random_ready_chain, 0, NULL); 181 spin_unlock_irqrestore(&random_ready_chain_lock, flags); 182} 183 184#define warn_unseeded_randomness() \ 185 if (IS_ENABLED(CONFIG_WARN_ALL_UNSEEDED_RANDOM) && !crng_ready()) \ 186 printk_deferred(KERN_NOTICE "random: %s called from %pS with crng_init=%d\n", \ 187 __func__, (void *)_RET_IP_, crng_init) 188 189 190/********************************************************************* 191 * 192 * Fast key erasure RNG, the "crng". 193 * 194 * These functions expand entropy from the entropy extractor into 195 * long streams for external consumption using the "fast key erasure" 196 * RNG described at <https://blog.cr.yp.to/20170723-random.html>. 197 * 198 * There are a few exported interfaces for use by other drivers: 199 * 200 * void get_random_bytes(void *buf, size_t len) 201 * u32 get_random_u32() 202 * u64 get_random_u64() 203 * unsigned int get_random_int() 204 * unsigned long get_random_long() 205 * 206 * These interfaces will return the requested number of random bytes 207 * into the given buffer or as a return value. This is equivalent to 208 * a read from /dev/urandom. The u32, u64, int, and long family of 209 * functions may be higher performance for one-off random integers, 210 * because they do a bit of buffering and do not invoke reseeding 211 * until the buffer is emptied. 212 * 213 *********************************************************************/ 214 215enum { 216 CRNG_RESEED_START_INTERVAL = HZ, 217 CRNG_RESEED_INTERVAL = 60 * HZ 218}; 219 220static struct { 221 u8 key[CHACHA_KEY_SIZE] __aligned(__alignof__(long)); 222 unsigned long birth; 223 unsigned long generation; 224 spinlock_t lock; 225} base_crng = { 226 .lock = __SPIN_LOCK_UNLOCKED(base_crng.lock) 227}; 228 229struct crng { 230 u8 key[CHACHA_KEY_SIZE]; 231 unsigned long generation; 232 local_lock_t lock; 233}; 234 235static DEFINE_PER_CPU(struct crng, crngs) = { 236 .generation = ULONG_MAX, 237 .lock = INIT_LOCAL_LOCK(crngs.lock), 238}; 239 240/* Used by crng_reseed() and crng_make_state() to extract a new seed from the input pool. */ 241static void extract_entropy(void *buf, size_t len); 242 243/* This extracts a new crng key from the input pool. */ 244static void crng_reseed(void) 245{ 246 unsigned long flags; 247 unsigned long next_gen; 248 u8 key[CHACHA_KEY_SIZE]; 249 250 extract_entropy(key, sizeof(key)); 251 252 /* 253 * We copy the new key into the base_crng, overwriting the old one, 254 * and update the generation counter. We avoid hitting ULONG_MAX, 255 * because the per-cpu crngs are initialized to ULONG_MAX, so this 256 * forces new CPUs that come online to always initialize. 257 */ 258 spin_lock_irqsave(&base_crng.lock, flags); 259 memcpy(base_crng.key, key, sizeof(base_crng.key)); 260 next_gen = base_crng.generation + 1; 261 if (next_gen == ULONG_MAX) 262 ++next_gen; 263 WRITE_ONCE(base_crng.generation, next_gen); 264 WRITE_ONCE(base_crng.birth, jiffies); 265 if (!crng_ready()) 266 crng_init = CRNG_READY; 267 spin_unlock_irqrestore(&base_crng.lock, flags); 268 memzero_explicit(key, sizeof(key)); 269} 270 271/* 272 * This generates a ChaCha block using the provided key, and then 273 * immediately overwites that key with half the block. It returns 274 * the resultant ChaCha state to the user, along with the second 275 * half of the block containing 32 bytes of random data that may 276 * be used; random_data_len may not be greater than 32. 277 * 278 * The returned ChaCha state contains within it a copy of the old 279 * key value, at index 4, so the state should always be zeroed out 280 * immediately after using in order to maintain forward secrecy. 281 * If the state cannot be erased in a timely manner, then it is 282 * safer to set the random_data parameter to &chacha_state[4] so 283 * that this function overwrites it before returning. 284 */ 285static void crng_fast_key_erasure(u8 key[CHACHA_KEY_SIZE], 286 u32 chacha_state[CHACHA_STATE_WORDS], 287 u8 *random_data, size_t random_data_len) 288{ 289 u8 first_block[CHACHA_BLOCK_SIZE]; 290 291 BUG_ON(random_data_len > 32); 292 293 chacha_init_consts(chacha_state); 294 memcpy(&chacha_state[4], key, CHACHA_KEY_SIZE); 295 memset(&chacha_state[12], 0, sizeof(u32) * 4); 296 chacha20_block(chacha_state, first_block); 297 298 memcpy(key, first_block, CHACHA_KEY_SIZE); 299 memcpy(random_data, first_block + CHACHA_KEY_SIZE, random_data_len); 300 memzero_explicit(first_block, sizeof(first_block)); 301} 302 303/* 304 * Return whether the crng seed is considered to be sufficiently old 305 * that a reseeding is needed. This happens if the last reseeding 306 * was CRNG_RESEED_INTERVAL ago, or during early boot, at an interval 307 * proportional to the uptime. 308 */ 309static bool crng_has_old_seed(void) 310{ 311 static bool early_boot = true; 312 unsigned long interval = CRNG_RESEED_INTERVAL; 313 314 if (unlikely(READ_ONCE(early_boot))) { 315 time64_t uptime = ktime_get_seconds(); 316 if (uptime >= CRNG_RESEED_INTERVAL / HZ * 2) 317 WRITE_ONCE(early_boot, false); 318 else 319 interval = max_t(unsigned int, CRNG_RESEED_START_INTERVAL, 320 (unsigned int)uptime / 2 * HZ); 321 } 322 return time_is_before_jiffies(READ_ONCE(base_crng.birth) + interval); 323} 324 325/* 326 * This function returns a ChaCha state that you may use for generating 327 * random data. It also returns up to 32 bytes on its own of random data 328 * that may be used; random_data_len may not be greater than 32. 329 */ 330static void crng_make_state(u32 chacha_state[CHACHA_STATE_WORDS], 331 u8 *random_data, size_t random_data_len) 332{ 333 unsigned long flags; 334 struct crng *crng; 335 336 BUG_ON(random_data_len > 32); 337 338 /* 339 * For the fast path, we check whether we're ready, unlocked first, and 340 * then re-check once locked later. In the case where we're really not 341 * ready, we do fast key erasure with the base_crng directly, extracting 342 * when crng_init is CRNG_EMPTY. 343 */ 344 if (!crng_ready()) { 345 bool ready; 346 347 spin_lock_irqsave(&base_crng.lock, flags); 348 ready = crng_ready(); 349 if (!ready) { 350 if (crng_init == CRNG_EMPTY) 351 extract_entropy(base_crng.key, sizeof(base_crng.key)); 352 crng_fast_key_erasure(base_crng.key, chacha_state, 353 random_data, random_data_len); 354 } 355 spin_unlock_irqrestore(&base_crng.lock, flags); 356 if (!ready) 357 return; 358 } 359 360 /* 361 * If the base_crng is old enough, we reseed, which in turn bumps the 362 * generation counter that we check below. 363 */ 364 if (unlikely(crng_has_old_seed())) 365 crng_reseed(); 366 367 local_lock_irqsave(&crngs.lock, flags); 368 crng = raw_cpu_ptr(&crngs); 369 370 /* 371 * If our per-cpu crng is older than the base_crng, then it means 372 * somebody reseeded the base_crng. In that case, we do fast key 373 * erasure on the base_crng, and use its output as the new key 374 * for our per-cpu crng. This brings us up to date with base_crng. 375 */ 376 if (unlikely(crng->generation != READ_ONCE(base_crng.generation))) { 377 spin_lock(&base_crng.lock); 378 crng_fast_key_erasure(base_crng.key, chacha_state, 379 crng->key, sizeof(crng->key)); 380 crng->generation = base_crng.generation; 381 spin_unlock(&base_crng.lock); 382 } 383 384 /* 385 * Finally, when we've made it this far, our per-cpu crng has an up 386 * to date key, and we can do fast key erasure with it to produce 387 * some random data and a ChaCha state for the caller. All other 388 * branches of this function are "unlikely", so most of the time we 389 * should wind up here immediately. 390 */ 391 crng_fast_key_erasure(crng->key, chacha_state, random_data, random_data_len); 392 local_unlock_irqrestore(&crngs.lock, flags); 393} 394 395static void _get_random_bytes(void *buf, size_t len) 396{ 397 u32 chacha_state[CHACHA_STATE_WORDS]; 398 u8 tmp[CHACHA_BLOCK_SIZE]; 399 size_t first_block_len; 400 401 if (!len) 402 return; 403 404 first_block_len = min_t(size_t, 32, len); 405 crng_make_state(chacha_state, buf, first_block_len); 406 len -= first_block_len; 407 buf += first_block_len; 408 409 while (len) { 410 if (len < CHACHA_BLOCK_SIZE) { 411 chacha20_block(chacha_state, tmp); 412 memcpy(buf, tmp, len); 413 memzero_explicit(tmp, sizeof(tmp)); 414 break; 415 } 416 417 chacha20_block(chacha_state, buf); 418 if (unlikely(chacha_state[12] == 0)) 419 ++chacha_state[13]; 420 len -= CHACHA_BLOCK_SIZE; 421 buf += CHACHA_BLOCK_SIZE; 422 } 423 424 memzero_explicit(chacha_state, sizeof(chacha_state)); 425} 426 427/* 428 * This function is the exported kernel interface. It returns some 429 * number of good random numbers, suitable for key generation, seeding 430 * TCP sequence numbers, etc. It does not rely on the hardware random 431 * number generator. For random bytes direct from the hardware RNG 432 * (when available), use get_random_bytes_arch(). In order to ensure 433 * that the randomness provided by this function is okay, the function 434 * wait_for_random_bytes() should be called and return 0 at least once 435 * at any point prior. 436 */ 437void get_random_bytes(void *buf, size_t len) 438{ 439 warn_unseeded_randomness(); 440 _get_random_bytes(buf, len); 441} 442EXPORT_SYMBOL(get_random_bytes); 443 444static ssize_t get_random_bytes_user(struct iov_iter *iter) 445{ 446 u32 chacha_state[CHACHA_STATE_WORDS]; 447 u8 block[CHACHA_BLOCK_SIZE]; 448 size_t ret = 0, copied; 449 450 if (unlikely(!iov_iter_count(iter))) 451 return 0; 452 453 /* 454 * Immediately overwrite the ChaCha key at index 4 with random 455 * bytes, in case userspace causes copy_to_iter() below to sleep 456 * forever, so that we still retain forward secrecy in that case. 457 */ 458 crng_make_state(chacha_state, (u8 *)&chacha_state[4], CHACHA_KEY_SIZE); 459 /* 460 * However, if we're doing a read of len <= 32, we don't need to 461 * use chacha_state after, so we can simply return those bytes to 462 * the user directly. 463 */ 464 if (iov_iter_count(iter) <= CHACHA_KEY_SIZE) { 465 ret = copy_to_iter(&chacha_state[4], CHACHA_KEY_SIZE, iter); 466 goto out_zero_chacha; 467 } 468 469 for (;;) { 470 chacha20_block(chacha_state, block); 471 if (unlikely(chacha_state[12] == 0)) 472 ++chacha_state[13]; 473 474 copied = copy_to_iter(block, sizeof(block), iter); 475 ret += copied; 476 if (!iov_iter_count(iter) || copied != sizeof(block)) 477 break; 478 479 BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0); 480 if (ret % PAGE_SIZE == 0) { 481 if (signal_pending(current)) 482 break; 483 cond_resched(); 484 } 485 } 486 487 memzero_explicit(block, sizeof(block)); 488out_zero_chacha: 489 memzero_explicit(chacha_state, sizeof(chacha_state)); 490 return ret ? ret : -EFAULT; 491} 492 493/* 494 * Batched entropy returns random integers. The quality of the random 495 * number is good as /dev/urandom. In order to ensure that the randomness 496 * provided by this function is okay, the function wait_for_random_bytes() 497 * should be called and return 0 at least once at any point prior. 498 */ 499 500#define DEFINE_BATCHED_ENTROPY(type) \ 501struct batch_ ##type { \ 502 /* \ 503 * We make this 1.5x a ChaCha block, so that we get the \ 504 * remaining 32 bytes from fast key erasure, plus one full \ 505 * block from the detached ChaCha state. We can increase \ 506 * the size of this later if needed so long as we keep the \ 507 * formula of (integer_blocks + 0.5) * CHACHA_BLOCK_SIZE. \ 508 */ \ 509 type entropy[CHACHA_BLOCK_SIZE * 3 / (2 * sizeof(type))]; \ 510 local_lock_t lock; \ 511 unsigned long generation; \ 512 unsigned int position; \ 513}; \ 514 \ 515static DEFINE_PER_CPU(struct batch_ ##type, batched_entropy_ ##type) = { \ 516 .lock = INIT_LOCAL_LOCK(batched_entropy_ ##type.lock), \ 517 .position = UINT_MAX \ 518}; \ 519 \ 520type get_random_ ##type(void) \ 521{ \ 522 type ret; \ 523 unsigned long flags; \ 524 struct batch_ ##type *batch; \ 525 unsigned long next_gen; \ 526 \ 527 warn_unseeded_randomness(); \ 528 \ 529 if (!crng_ready()) { \ 530 _get_random_bytes(&ret, sizeof(ret)); \ 531 return ret; \ 532 } \ 533 \ 534 local_lock_irqsave(&batched_entropy_ ##type.lock, flags); \ 535 batch = raw_cpu_ptr(&batched_entropy_##type); \ 536 \ 537 next_gen = READ_ONCE(base_crng.generation); \ 538 if (batch->position >= ARRAY_SIZE(batch->entropy) || \ 539 next_gen != batch->generation) { \ 540 _get_random_bytes(batch->entropy, sizeof(batch->entropy)); \ 541 batch->position = 0; \ 542 batch->generation = next_gen; \ 543 } \ 544 \ 545 ret = batch->entropy[batch->position]; \ 546 batch->entropy[batch->position] = 0; \ 547 ++batch->position; \ 548 local_unlock_irqrestore(&batched_entropy_ ##type.lock, flags); \ 549 return ret; \ 550} \ 551EXPORT_SYMBOL(get_random_ ##type); 552 553DEFINE_BATCHED_ENTROPY(u64) 554DEFINE_BATCHED_ENTROPY(u32) 555 556#ifdef CONFIG_SMP 557/* 558 * This function is called when the CPU is coming up, with entry 559 * CPUHP_RANDOM_PREPARE, which comes before CPUHP_WORKQUEUE_PREP. 560 */ 561int __cold random_prepare_cpu(unsigned int cpu) 562{ 563 /* 564 * When the cpu comes back online, immediately invalidate both 565 * the per-cpu crng and all batches, so that we serve fresh 566 * randomness. 567 */ 568 per_cpu_ptr(&crngs, cpu)->generation = ULONG_MAX; 569 per_cpu_ptr(&batched_entropy_u32, cpu)->position = UINT_MAX; 570 per_cpu_ptr(&batched_entropy_u64, cpu)->position = UINT_MAX; 571 return 0; 572} 573#endif 574 575/* 576 * This function will use the architecture-specific hardware random 577 * number generator if it is available. It is not recommended for 578 * use. Use get_random_bytes() instead. It returns the number of 579 * bytes filled in. 580 */ 581size_t __must_check get_random_bytes_arch(void *buf, size_t len) 582{ 583 size_t left = len; 584 u8 *p = buf; 585 586 while (left) { 587 unsigned long v; 588 size_t block_len = min_t(size_t, left, sizeof(unsigned long)); 589 590 if (!arch_get_random_long(&v)) 591 break; 592 593 memcpy(p, &v, block_len); 594 p += block_len; 595 left -= block_len; 596 } 597 598 return len - left; 599} 600EXPORT_SYMBOL(get_random_bytes_arch); 601 602 603/********************************************************************** 604 * 605 * Entropy accumulation and extraction routines. 606 * 607 * Callers may add entropy via: 608 * 609 * static void mix_pool_bytes(const void *buf, size_t len) 610 * 611 * After which, if added entropy should be credited: 612 * 613 * static void credit_init_bits(size_t bits) 614 * 615 * Finally, extract entropy via: 616 * 617 * static void extract_entropy(void *buf, size_t len) 618 * 619 **********************************************************************/ 620 621enum { 622 POOL_BITS = BLAKE2S_HASH_SIZE * 8, 623 POOL_READY_BITS = POOL_BITS, /* When crng_init->CRNG_READY */ 624 POOL_EARLY_BITS = POOL_READY_BITS / 2 /* When crng_init->CRNG_EARLY */ 625}; 626 627static struct { 628 struct blake2s_state hash; 629 spinlock_t lock; 630 unsigned int init_bits; 631} input_pool = { 632 .hash.h = { BLAKE2S_IV0 ^ (0x01010000 | BLAKE2S_HASH_SIZE), 633 BLAKE2S_IV1, BLAKE2S_IV2, BLAKE2S_IV3, BLAKE2S_IV4, 634 BLAKE2S_IV5, BLAKE2S_IV6, BLAKE2S_IV7 }, 635 .hash.outlen = BLAKE2S_HASH_SIZE, 636 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock), 637}; 638 639static void _mix_pool_bytes(const void *buf, size_t len) 640{ 641 blake2s_update(&input_pool.hash, buf, len); 642} 643 644/* 645 * This function adds bytes into the input pool. It does not 646 * update the initialization bit counter; the caller should call 647 * credit_init_bits if this is appropriate. 648 */ 649static void mix_pool_bytes(const void *buf, size_t len) 650{ 651 unsigned long flags; 652 653 spin_lock_irqsave(&input_pool.lock, flags); 654 _mix_pool_bytes(buf, len); 655 spin_unlock_irqrestore(&input_pool.lock, flags); 656} 657 658/* 659 * This is an HKDF-like construction for using the hashed collected entropy 660 * as a PRF key, that's then expanded block-by-block. 661 */ 662static void extract_entropy(void *buf, size_t len) 663{ 664 unsigned long flags; 665 u8 seed[BLAKE2S_HASH_SIZE], next_key[BLAKE2S_HASH_SIZE]; 666 struct { 667 unsigned long rdseed[32 / sizeof(long)]; 668 size_t counter; 669 } block; 670 size_t i; 671 672 for (i = 0; i < ARRAY_SIZE(block.rdseed); ++i) { 673 if (!arch_get_random_seed_long(&block.rdseed[i]) && 674 !arch_get_random_long(&block.rdseed[i])) 675 block.rdseed[i] = random_get_entropy(); 676 } 677 678 spin_lock_irqsave(&input_pool.lock, flags); 679 680 /* seed = HASHPRF(last_key, entropy_input) */ 681 blake2s_final(&input_pool.hash, seed); 682 683 /* next_key = HASHPRF(seed, RDSEED || 0) */ 684 block.counter = 0; 685 blake2s(next_key, (u8 *)&block, seed, sizeof(next_key), sizeof(block), sizeof(seed)); 686 blake2s_init_key(&input_pool.hash, BLAKE2S_HASH_SIZE, next_key, sizeof(next_key)); 687 688 spin_unlock_irqrestore(&input_pool.lock, flags); 689 memzero_explicit(next_key, sizeof(next_key)); 690 691 while (len) { 692 i = min_t(size_t, len, BLAKE2S_HASH_SIZE); 693 /* output = HASHPRF(seed, RDSEED || ++counter) */ 694 ++block.counter; 695 blake2s(buf, (u8 *)&block, seed, i, sizeof(block), sizeof(seed)); 696 len -= i; 697 buf += i; 698 } 699 700 memzero_explicit(seed, sizeof(seed)); 701 memzero_explicit(&block, sizeof(block)); 702} 703 704#define credit_init_bits(bits) if (!crng_ready()) _credit_init_bits(bits) 705 706static void __cold _credit_init_bits(size_t bits) 707{ 708 unsigned int new, orig, add; 709 unsigned long flags; 710 711 if (!bits) 712 return; 713 714 add = min_t(size_t, bits, POOL_BITS); 715 716 do { 717 orig = READ_ONCE(input_pool.init_bits); 718 new = min_t(unsigned int, POOL_BITS, orig + add); 719 } while (cmpxchg(&input_pool.init_bits, orig, new) != orig); 720 721 if (orig < POOL_READY_BITS && new >= POOL_READY_BITS) { 722 crng_reseed(); /* Sets crng_init to CRNG_READY under base_crng.lock. */ 723 process_random_ready_list(); 724 wake_up_interruptible(&crng_init_wait); 725 kill_fasync(&fasync, SIGIO, POLL_IN); 726 pr_notice("crng init done\n"); 727 if (urandom_warning.missed) 728 pr_notice("%d urandom warning(s) missed due to ratelimiting\n", 729 urandom_warning.missed); 730 } else if (orig < POOL_EARLY_BITS && new >= POOL_EARLY_BITS) { 731 spin_lock_irqsave(&base_crng.lock, flags); 732 /* Check if crng_init is CRNG_EMPTY, to avoid race with crng_reseed(). */ 733 if (crng_init == CRNG_EMPTY) { 734 extract_entropy(base_crng.key, sizeof(base_crng.key)); 735 crng_init = CRNG_EARLY; 736 } 737 spin_unlock_irqrestore(&base_crng.lock, flags); 738 } 739} 740 741 742/********************************************************************** 743 * 744 * Entropy collection routines. 745 * 746 * The following exported functions are used for pushing entropy into 747 * the above entropy accumulation routines: 748 * 749 * void add_device_randomness(const void *buf, size_t len); 750 * void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy); 751 * void add_bootloader_randomness(const void *buf, size_t len); 752 * void add_interrupt_randomness(int irq); 753 * void add_input_randomness(unsigned int type, unsigned int code, unsigned int value); 754 * void add_disk_randomness(struct gendisk *disk); 755 * 756 * add_device_randomness() adds data to the input pool that 757 * is likely to differ between two devices (or possibly even per boot). 758 * This would be things like MAC addresses or serial numbers, or the 759 * read-out of the RTC. This does *not* credit any actual entropy to 760 * the pool, but it initializes the pool to different values for devices 761 * that might otherwise be identical and have very little entropy 762 * available to them (particularly common in the embedded world). 763 * 764 * add_hwgenerator_randomness() is for true hardware RNGs, and will credit 765 * entropy as specified by the caller. If the entropy pool is full it will 766 * block until more entropy is needed. 767 * 768 * add_bootloader_randomness() is called by bootloader drivers, such as EFI 769 * and device tree, and credits its input depending on whether or not the 770 * configuration option CONFIG_RANDOM_TRUST_BOOTLOADER is set. 771 * 772 * add_interrupt_randomness() uses the interrupt timing as random 773 * inputs to the entropy pool. Using the cycle counters and the irq source 774 * as inputs, it feeds the input pool roughly once a second or after 64 775 * interrupts, crediting 1 bit of entropy for whichever comes first. 776 * 777 * add_input_randomness() uses the input layer interrupt timing, as well 778 * as the event type information from the hardware. 779 * 780 * add_disk_randomness() uses what amounts to the seek time of block 781 * layer request events, on a per-disk_devt basis, as input to the 782 * entropy pool. Note that high-speed solid state drives with very low 783 * seek times do not make for good sources of entropy, as their seek 784 * times are usually fairly consistent. 785 * 786 * The last two routines try to estimate how many bits of entropy 787 * to credit. They do this by keeping track of the first and second 788 * order deltas of the event timings. 789 * 790 **********************************************************************/ 791 792static bool trust_cpu __ro_after_init = IS_ENABLED(CONFIG_RANDOM_TRUST_CPU); 793static bool trust_bootloader __ro_after_init = IS_ENABLED(CONFIG_RANDOM_TRUST_BOOTLOADER); 794static int __init parse_trust_cpu(char *arg) 795{ 796 return kstrtobool(arg, &trust_cpu); 797} 798static int __init parse_trust_bootloader(char *arg) 799{ 800 return kstrtobool(arg, &trust_bootloader); 801} 802early_param("random.trust_cpu", parse_trust_cpu); 803early_param("random.trust_bootloader", parse_trust_bootloader); 804 805/* 806 * The first collection of entropy occurs at system boot while interrupts 807 * are still turned off. Here we push in latent entropy, RDSEED, a timestamp, 808 * utsname(), and the command line. Depending on the above configuration knob, 809 * RDSEED may be considered sufficient for initialization. Note that much 810 * earlier setup may already have pushed entropy into the input pool by the 811 * time we get here. 812 */ 813int __init random_init(const char *command_line) 814{ 815 ktime_t now = ktime_get_real(); 816 unsigned int i, arch_bytes; 817 unsigned long entropy; 818 819#if defined(LATENT_ENTROPY_PLUGIN) 820 static const u8 compiletime_seed[BLAKE2S_BLOCK_SIZE] __initconst __latent_entropy; 821 _mix_pool_bytes(compiletime_seed, sizeof(compiletime_seed)); 822#endif 823 824 for (i = 0, arch_bytes = BLAKE2S_BLOCK_SIZE; 825 i < BLAKE2S_BLOCK_SIZE; i += sizeof(entropy)) { 826 if (!arch_get_random_seed_long_early(&entropy) && 827 !arch_get_random_long_early(&entropy)) { 828 entropy = random_get_entropy(); 829 arch_bytes -= sizeof(entropy); 830 } 831 _mix_pool_bytes(&entropy, sizeof(entropy)); 832 } 833 _mix_pool_bytes(&now, sizeof(now)); 834 _mix_pool_bytes(utsname(), sizeof(*(utsname()))); 835 _mix_pool_bytes(command_line, strlen(command_line)); 836 add_latent_entropy(); 837 838 if (crng_ready()) 839 crng_reseed(); 840 else if (trust_cpu) 841 credit_init_bits(arch_bytes * 8); 842 843 return 0; 844} 845 846/* 847 * Add device- or boot-specific data to the input pool to help 848 * initialize it. 849 * 850 * None of this adds any entropy; it is meant to avoid the problem of 851 * the entropy pool having similar initial state across largely 852 * identical devices. 853 */ 854void add_device_randomness(const void *buf, size_t len) 855{ 856 unsigned long entropy = random_get_entropy(); 857 unsigned long flags; 858 859 spin_lock_irqsave(&input_pool.lock, flags); 860 _mix_pool_bytes(&entropy, sizeof(entropy)); 861 _mix_pool_bytes(buf, len); 862 spin_unlock_irqrestore(&input_pool.lock, flags); 863} 864EXPORT_SYMBOL(add_device_randomness); 865 866/* 867 * Interface for in-kernel drivers of true hardware RNGs. 868 * Those devices may produce endless random bits and will be throttled 869 * when our pool is full. 870 */ 871void add_hwgenerator_randomness(const void *buf, size_t len, size_t entropy) 872{ 873 mix_pool_bytes(buf, len); 874 credit_init_bits(entropy); 875 876 /* 877 * Throttle writing to once every CRNG_RESEED_INTERVAL, unless 878 * we're not yet initialized. 879 */ 880 if (!kthread_should_stop() && crng_ready()) 881 schedule_timeout_interruptible(CRNG_RESEED_INTERVAL); 882} 883EXPORT_SYMBOL_GPL(add_hwgenerator_randomness); 884 885/* 886 * Handle random seed passed by bootloader, and credit it if 887 * CONFIG_RANDOM_TRUST_BOOTLOADER is set. 888 */ 889void __cold add_bootloader_randomness(const void *buf, size_t len) 890{ 891 mix_pool_bytes(buf, len); 892 if (trust_bootloader) 893 credit_init_bits(len * 8); 894} 895EXPORT_SYMBOL_GPL(add_bootloader_randomness); 896 897struct fast_pool { 898 unsigned long pool[4]; 899 unsigned long last; 900 unsigned int count; 901 struct timer_list mix; 902}; 903 904static void mix_interrupt_randomness(struct timer_list *work); 905 906static DEFINE_PER_CPU(struct fast_pool, irq_randomness) = { 907#ifdef CONFIG_64BIT 908#define FASTMIX_PERM SIPHASH_PERMUTATION 909 .pool = { SIPHASH_CONST_0, SIPHASH_CONST_1, SIPHASH_CONST_2, SIPHASH_CONST_3 }, 910#else 911#define FASTMIX_PERM HSIPHASH_PERMUTATION 912 .pool = { HSIPHASH_CONST_0, HSIPHASH_CONST_1, HSIPHASH_CONST_2, HSIPHASH_CONST_3 }, 913#endif 914 .mix = __TIMER_INITIALIZER(mix_interrupt_randomness, 0) 915}; 916 917/* 918 * This is [Half]SipHash-1-x, starting from an empty key. Because 919 * the key is fixed, it assumes that its inputs are non-malicious, 920 * and therefore this has no security on its own. s represents the 921 * four-word SipHash state, while v represents a two-word input. 922 */ 923static void fast_mix(unsigned long s[4], unsigned long v1, unsigned long v2) 924{ 925 s[3] ^= v1; 926 FASTMIX_PERM(s[0], s[1], s[2], s[3]); 927 s[0] ^= v1; 928 s[3] ^= v2; 929 FASTMIX_PERM(s[0], s[1], s[2], s[3]); 930 s[0] ^= v2; 931} 932 933#ifdef CONFIG_SMP 934/* 935 * This function is called when the CPU has just come online, with 936 * entry CPUHP_AP_RANDOM_ONLINE, just after CPUHP_AP_WORKQUEUE_ONLINE. 937 */ 938int __cold random_online_cpu(unsigned int cpu) 939{ 940 /* 941 * During CPU shutdown and before CPU onlining, add_interrupt_ 942 * randomness() may schedule mix_interrupt_randomness(), and 943 * set the MIX_INFLIGHT flag. However, because the worker can 944 * be scheduled on a different CPU during this period, that 945 * flag will never be cleared. For that reason, we zero out 946 * the flag here, which runs just after workqueues are onlined 947 * for the CPU again. This also has the effect of setting the 948 * irq randomness count to zero so that new accumulated irqs 949 * are fresh. 950 */ 951 per_cpu_ptr(&irq_randomness, cpu)->count = 0; 952 return 0; 953} 954#endif 955 956static void mix_interrupt_randomness(struct timer_list *work) 957{ 958 struct fast_pool *fast_pool = container_of(work, struct fast_pool, mix); 959 /* 960 * The size of the copied stack pool is explicitly 2 longs so that we 961 * only ever ingest half of the siphash output each time, retaining 962 * the other half as the next "key" that carries over. The entropy is 963 * supposed to be sufficiently dispersed between bits so on average 964 * we don't wind up "losing" some. 965 */ 966 unsigned long pool[2]; 967 unsigned int count; 968 969 /* Check to see if we're running on the wrong CPU due to hotplug. */ 970 local_irq_disable(); 971 if (fast_pool != this_cpu_ptr(&irq_randomness)) { 972 local_irq_enable(); 973 return; 974 } 975 976 /* 977 * Copy the pool to the stack so that the mixer always has a 978 * consistent view, before we reenable irqs again. 979 */ 980 memcpy(pool, fast_pool->pool, sizeof(pool)); 981 count = fast_pool->count; 982 fast_pool->count = 0; 983 fast_pool->last = jiffies; 984 local_irq_enable(); 985 986 mix_pool_bytes(pool, sizeof(pool)); 987 credit_init_bits(clamp_t(unsigned int, (count & U16_MAX) / 64, 1, sizeof(pool) * 8)); 988 989 memzero_explicit(pool, sizeof(pool)); 990} 991 992void add_interrupt_randomness(int irq) 993{ 994 enum { MIX_INFLIGHT = 1U << 31 }; 995 unsigned long entropy = random_get_entropy(); 996 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness); 997 struct pt_regs *regs = get_irq_regs(); 998 unsigned int new_count; 999 1000 fast_mix(fast_pool->pool, entropy, 1001 (regs ? instruction_pointer(regs) : _RET_IP_) ^ swab(irq)); 1002 new_count = ++fast_pool->count; 1003 1004 if (new_count & MIX_INFLIGHT) 1005 return; 1006 1007 if (new_count < 1024 && !time_is_before_jiffies(fast_pool->last + HZ)) 1008 return; 1009 1010 fast_pool->count |= MIX_INFLIGHT; 1011 if (!timer_pending(&fast_pool->mix)) { 1012 fast_pool->mix.expires = jiffies; 1013 add_timer_on(&fast_pool->mix, raw_smp_processor_id()); 1014 } 1015} 1016EXPORT_SYMBOL_GPL(add_interrupt_randomness); 1017 1018/* There is one of these per entropy source */ 1019struct timer_rand_state { 1020 unsigned long last_time; 1021 long last_delta, last_delta2; 1022}; 1023 1024/* 1025 * This function adds entropy to the entropy "pool" by using timing 1026 * delays. It uses the timer_rand_state structure to make an estimate 1027 * of how many bits of entropy this call has added to the pool. The 1028 * value "num" is also added to the pool; it should somehow describe 1029 * the type of event that just happened. 1030 */ 1031static void add_timer_randomness(struct timer_rand_state *state, unsigned int num) 1032{ 1033 unsigned long entropy = random_get_entropy(), now = jiffies, flags; 1034 long delta, delta2, delta3; 1035 unsigned int bits; 1036 1037 /* 1038 * If we're in a hard IRQ, add_interrupt_randomness() will be called 1039 * sometime after, so mix into the fast pool. 1040 */ 1041 if (in_irq()) { 1042 fast_mix(this_cpu_ptr(&irq_randomness)->pool, entropy, num); 1043 } else { 1044 spin_lock_irqsave(&input_pool.lock, flags); 1045 _mix_pool_bytes(&entropy, sizeof(entropy)); 1046 _mix_pool_bytes(&num, sizeof(num)); 1047 spin_unlock_irqrestore(&input_pool.lock, flags); 1048 } 1049 1050 if (crng_ready()) 1051 return; 1052 1053 /* 1054 * Calculate number of bits of randomness we probably added. 1055 * We take into account the first, second and third-order deltas 1056 * in order to make our estimate. 1057 */ 1058 delta = now - READ_ONCE(state->last_time); 1059 WRITE_ONCE(state->last_time, now); 1060 1061 delta2 = delta - READ_ONCE(state->last_delta); 1062 WRITE_ONCE(state->last_delta, delta); 1063 1064 delta3 = delta2 - READ_ONCE(state->last_delta2); 1065 WRITE_ONCE(state->last_delta2, delta2); 1066 1067 if (delta < 0) 1068 delta = -delta; 1069 if (delta2 < 0) 1070 delta2 = -delta2; 1071 if (delta3 < 0) 1072 delta3 = -delta3; 1073 if (delta > delta2) 1074 delta = delta2; 1075 if (delta > delta3) 1076 delta = delta3; 1077 1078 /* 1079 * delta is now minimum absolute delta. Round down by 1 bit 1080 * on general principles, and limit entropy estimate to 11 bits. 1081 */ 1082 bits = min(fls(delta >> 1), 11); 1083 1084 /* 1085 * As mentioned above, if we're in a hard IRQ, add_interrupt_randomness() 1086 * will run after this, which uses a different crediting scheme of 1 bit 1087 * per every 64 interrupts. In order to let that function do accounting 1088 * close to the one in this function, we credit a full 64/64 bit per bit, 1089 * and then subtract one to account for the extra one added. 1090 */ 1091 if (in_irq()) 1092 this_cpu_ptr(&irq_randomness)->count += max(1u, bits * 64) - 1; 1093 else 1094 _credit_init_bits(bits); 1095} 1096 1097void add_input_randomness(unsigned int type, unsigned int code, unsigned int value) 1098{ 1099 static unsigned char last_value; 1100 static struct timer_rand_state input_timer_state = { INITIAL_JIFFIES }; 1101 1102 /* Ignore autorepeat and the like. */ 1103 if (value == last_value) 1104 return; 1105 1106 last_value = value; 1107 add_timer_randomness(&input_timer_state, 1108 (type << 4) ^ code ^ (code >> 4) ^ value); 1109} 1110EXPORT_SYMBOL_GPL(add_input_randomness); 1111 1112#ifdef CONFIG_BLOCK 1113void add_disk_randomness(struct gendisk *disk) 1114{ 1115 if (!disk || !disk->random) 1116 return; 1117 /* First major is 1, so we get >= 0x200 here. */ 1118 add_timer_randomness(disk->random, 0x100 + disk_devt(disk)); 1119} 1120EXPORT_SYMBOL_GPL(add_disk_randomness); 1121 1122void __cold rand_initialize_disk(struct gendisk *disk) 1123{ 1124 struct timer_rand_state *state; 1125 1126 /* 1127 * If kzalloc returns null, we just won't use that entropy 1128 * source. 1129 */ 1130 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL); 1131 if (state) { 1132 state->last_time = INITIAL_JIFFIES; 1133 disk->random = state; 1134 } 1135} 1136#endif 1137 1138/* 1139 * Each time the timer fires, we expect that we got an unpredictable 1140 * jump in the cycle counter. Even if the timer is running on another 1141 * CPU, the timer activity will be touching the stack of the CPU that is 1142 * generating entropy.. 1143 * 1144 * Note that we don't re-arm the timer in the timer itself - we are 1145 * happy to be scheduled away, since that just makes the load more 1146 * complex, but we do not want the timer to keep ticking unless the 1147 * entropy loop is running. 1148 * 1149 * So the re-arming always happens in the entropy loop itself. 1150 */ 1151static void __cold entropy_timer(struct timer_list *t) 1152{ 1153 credit_init_bits(1); 1154} 1155 1156/* 1157 * If we have an actual cycle counter, see if we can 1158 * generate enough entropy with timing noise 1159 */ 1160static void __cold try_to_generate_entropy(void) 1161{ 1162 struct { 1163 unsigned long entropy; 1164 struct timer_list timer; 1165 } stack; 1166 1167 stack.entropy = random_get_entropy(); 1168 1169 /* Slow counter - or none. Don't even bother */ 1170 if (stack.entropy == random_get_entropy()) 1171 return; 1172 1173 timer_setup_on_stack(&stack.timer, entropy_timer, 0); 1174 while (!crng_ready() && !signal_pending(current)) { 1175 if (!timer_pending(&stack.timer)) 1176 mod_timer(&stack.timer, jiffies + 1); 1177 mix_pool_bytes(&stack.entropy, sizeof(stack.entropy)); 1178 schedule(); 1179 stack.entropy = random_get_entropy(); 1180 } 1181 1182 del_timer_sync(&stack.timer); 1183 destroy_timer_on_stack(&stack.timer); 1184 mix_pool_bytes(&stack.entropy, sizeof(stack.entropy)); 1185} 1186 1187 1188/********************************************************************** 1189 * 1190 * Userspace reader/writer interfaces. 1191 * 1192 * getrandom(2) is the primary modern interface into the RNG and should 1193 * be used in preference to anything else. 1194 * 1195 * Reading from /dev/random has the same functionality as calling 1196 * getrandom(2) with flags=0. In earlier versions, however, it had 1197 * vastly different semantics and should therefore be avoided, to 1198 * prevent backwards compatibility issues. 1199 * 1200 * Reading from /dev/urandom has the same functionality as calling 1201 * getrandom(2) with flags=GRND_INSECURE. Because it does not block 1202 * waiting for the RNG to be ready, it should not be used. 1203 * 1204 * Writing to either /dev/random or /dev/urandom adds entropy to 1205 * the input pool but does not credit it. 1206 * 1207 * Polling on /dev/random indicates when the RNG is initialized, on 1208 * the read side, and when it wants new entropy, on the write side. 1209 * 1210 * Both /dev/random and /dev/urandom have the same set of ioctls for 1211 * adding entropy, getting the entropy count, zeroing the count, and 1212 * reseeding the crng. 1213 * 1214 **********************************************************************/ 1215 1216SYSCALL_DEFINE3(getrandom, char __user *, ubuf, size_t, len, unsigned int, flags) 1217{ 1218 struct iov_iter iter; 1219 struct iovec iov; 1220 int ret; 1221 1222 if (flags & ~(GRND_NONBLOCK | GRND_RANDOM | GRND_INSECURE)) 1223 return -EINVAL; 1224 1225 /* 1226 * Requesting insecure and blocking randomness at the same time makes 1227 * no sense. 1228 */ 1229 if ((flags & (GRND_INSECURE | GRND_RANDOM)) == (GRND_INSECURE | GRND_RANDOM)) 1230 return -EINVAL; 1231 1232 if (!crng_ready() && !(flags & GRND_INSECURE)) { 1233 if (flags & GRND_NONBLOCK) 1234 return -EAGAIN; 1235 ret = wait_for_random_bytes(); 1236 if (unlikely(ret)) 1237 return ret; 1238 } 1239 1240 ret = import_single_range(READ, ubuf, len, &iov, &iter); 1241 if (unlikely(ret)) 1242 return ret; 1243 return get_random_bytes_user(&iter); 1244} 1245 1246static __poll_t random_poll(struct file *file, poll_table *wait) 1247{ 1248 poll_wait(file, &crng_init_wait, wait); 1249 return crng_ready() ? EPOLLIN | EPOLLRDNORM : EPOLLOUT | EPOLLWRNORM; 1250} 1251 1252static ssize_t write_pool_user(struct iov_iter *iter) 1253{ 1254 u8 block[BLAKE2S_BLOCK_SIZE]; 1255 ssize_t ret = 0; 1256 size_t copied; 1257 1258 if (unlikely(!iov_iter_count(iter))) 1259 return 0; 1260 1261 for (;;) { 1262 copied = copy_from_iter(block, sizeof(block), iter); 1263 ret += copied; 1264 mix_pool_bytes(block, copied); 1265 if (!iov_iter_count(iter) || copied != sizeof(block)) 1266 break; 1267 1268 BUILD_BUG_ON(PAGE_SIZE % sizeof(block) != 0); 1269 if (ret % PAGE_SIZE == 0) { 1270 if (signal_pending(current)) 1271 break; 1272 cond_resched(); 1273 } 1274 } 1275 1276 memzero_explicit(block, sizeof(block)); 1277 return ret ? ret : -EFAULT; 1278} 1279 1280static ssize_t random_write_iter(struct kiocb *kiocb, struct iov_iter *iter) 1281{ 1282 return write_pool_user(iter); 1283} 1284 1285static ssize_t urandom_read_iter(struct kiocb *kiocb, struct iov_iter *iter) 1286{ 1287 static int maxwarn = 10; 1288 1289 if (!crng_ready()) { 1290 if (!ratelimit_disable && maxwarn <= 0) 1291 ++urandom_warning.missed; 1292 else if (ratelimit_disable || __ratelimit(&urandom_warning)) { 1293 --maxwarn; 1294 pr_notice("%s: uninitialized urandom read (%zu bytes read)\n", 1295 current->comm, iov_iter_count(iter)); 1296 } 1297 } 1298 1299 return get_random_bytes_user(iter); 1300} 1301 1302static ssize_t random_read_iter(struct kiocb *kiocb, struct iov_iter *iter) 1303{ 1304 int ret; 1305 1306 if (!crng_ready() && 1307 ((kiocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO)) || 1308 (kiocb->ki_filp->f_flags & O_NONBLOCK))) 1309 return -EAGAIN; 1310 1311 ret = wait_for_random_bytes(); 1312 if (ret != 0) 1313 return ret; 1314 return get_random_bytes_user(iter); 1315} 1316 1317static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg) 1318{ 1319 int __user *p = (int __user *)arg; 1320 int ent_count; 1321 1322 switch (cmd) { 1323 case RNDGETENTCNT: 1324 /* Inherently racy, no point locking. */ 1325 if (put_user(input_pool.init_bits, p)) 1326 return -EFAULT; 1327 return 0; 1328 case RNDADDTOENTCNT: 1329 if (!capable(CAP_SYS_ADMIN)) 1330 return -EPERM; 1331 if (get_user(ent_count, p)) 1332 return -EFAULT; 1333 if (ent_count < 0) 1334 return -EINVAL; 1335 credit_init_bits(ent_count); 1336 return 0; 1337 case RNDADDENTROPY: { 1338 struct iov_iter iter; 1339 struct iovec iov; 1340 ssize_t ret; 1341 int len; 1342 1343 if (!capable(CAP_SYS_ADMIN)) 1344 return -EPERM; 1345 if (get_user(ent_count, p++)) 1346 return -EFAULT; 1347 if (ent_count < 0) 1348 return -EINVAL; 1349 if (get_user(len, p++)) 1350 return -EFAULT; 1351 ret = import_single_range(WRITE, p, len, &iov, &iter); 1352 if (unlikely(ret)) 1353 return ret; 1354 ret = write_pool_user(&iter); 1355 if (unlikely(ret < 0)) 1356 return ret; 1357 /* Since we're crediting, enforce that it was all written into the pool. */ 1358 if (unlikely(ret != len)) 1359 return -EFAULT; 1360 credit_init_bits(ent_count); 1361 return 0; 1362 } 1363 case RNDZAPENTCNT: 1364 case RNDCLEARPOOL: 1365 /* No longer has any effect. */ 1366 if (!capable(CAP_SYS_ADMIN)) 1367 return -EPERM; 1368 return 0; 1369 case RNDRESEEDCRNG: 1370 if (!capable(CAP_SYS_ADMIN)) 1371 return -EPERM; 1372 if (!crng_ready()) 1373 return -ENODATA; 1374 crng_reseed(); 1375 return 0; 1376 default: 1377 return -EINVAL; 1378 } 1379} 1380 1381static int random_fasync(int fd, struct file *filp, int on) 1382{ 1383 return fasync_helper(fd, filp, on, &fasync); 1384} 1385 1386const struct file_operations random_fops = { 1387 .read_iter = random_read_iter, 1388 .write_iter = random_write_iter, 1389 .poll = random_poll, 1390 .unlocked_ioctl = random_ioctl, 1391 .compat_ioctl = compat_ptr_ioctl, 1392 .fasync = random_fasync, 1393 .llseek = noop_llseek, 1394 .splice_read = generic_file_splice_read, 1395 .splice_write = iter_file_splice_write, 1396}; 1397 1398const struct file_operations urandom_fops = { 1399 .read_iter = urandom_read_iter, 1400 .write_iter = random_write_iter, 1401 .unlocked_ioctl = random_ioctl, 1402 .compat_ioctl = compat_ptr_ioctl, 1403 .fasync = random_fasync, 1404 .llseek = noop_llseek, 1405 .splice_read = generic_file_splice_read, 1406 .splice_write = iter_file_splice_write, 1407}; 1408 1409 1410/******************************************************************** 1411 * 1412 * Sysctl interface. 1413 * 1414 * These are partly unused legacy knobs with dummy values to not break 1415 * userspace and partly still useful things. They are usually accessible 1416 * in /proc/sys/kernel/random/ and are as follows: 1417 * 1418 * - boot_id - a UUID representing the current boot. 1419 * 1420 * - uuid - a random UUID, different each time the file is read. 1421 * 1422 * - poolsize - the number of bits of entropy that the input pool can 1423 * hold, tied to the POOL_BITS constant. 1424 * 1425 * - entropy_avail - the number of bits of entropy currently in the 1426 * input pool. Always <= poolsize. 1427 * 1428 * - write_wakeup_threshold - the amount of entropy in the input pool 1429 * below which write polls to /dev/random will unblock, requesting 1430 * more entropy, tied to the POOL_READY_BITS constant. It is writable 1431 * to avoid breaking old userspaces, but writing to it does not 1432 * change any behavior of the RNG. 1433 * 1434 * - urandom_min_reseed_secs - fixed to the value CRNG_RESEED_INTERVAL. 1435 * It is writable to avoid breaking old userspaces, but writing 1436 * to it does not change any behavior of the RNG. 1437 * 1438 ********************************************************************/ 1439 1440#ifdef CONFIG_SYSCTL 1441 1442#include <linux/sysctl.h> 1443 1444static int sysctl_random_min_urandom_seed = CRNG_RESEED_INTERVAL / HZ; 1445static int sysctl_random_write_wakeup_bits = POOL_READY_BITS; 1446static int sysctl_poolsize = POOL_BITS; 1447static u8 sysctl_bootid[UUID_SIZE]; 1448 1449/* 1450 * This function is used to return both the bootid UUID, and random 1451 * UUID. The difference is in whether table->data is NULL; if it is, 1452 * then a new UUID is generated and returned to the user. 1453 */ 1454static int proc_do_uuid(struct ctl_table *table, int write, void *buf, 1455 size_t *lenp, loff_t *ppos) 1456{ 1457 u8 tmp_uuid[UUID_SIZE], *uuid; 1458 char uuid_string[UUID_STRING_LEN + 1]; 1459 struct ctl_table fake_table = { 1460 .data = uuid_string, 1461 .maxlen = UUID_STRING_LEN 1462 }; 1463 1464 if (write) 1465 return -EPERM; 1466 1467 uuid = table->data; 1468 if (!uuid) { 1469 uuid = tmp_uuid; 1470 generate_random_uuid(uuid); 1471 } else { 1472 static DEFINE_SPINLOCK(bootid_spinlock); 1473 1474 spin_lock(&bootid_spinlock); 1475 if (!uuid[8]) 1476 generate_random_uuid(uuid); 1477 spin_unlock(&bootid_spinlock); 1478 } 1479 1480 snprintf(uuid_string, sizeof(uuid_string), "%pU", uuid); 1481 return proc_dostring(&fake_table, 0, buf, lenp, ppos); 1482} 1483 1484/* The same as proc_dointvec, but writes don't change anything. */ 1485static int proc_do_rointvec(struct ctl_table *table, int write, void *buf, 1486 size_t *lenp, loff_t *ppos) 1487{ 1488 return write ? 0 : proc_dointvec(table, 0, buf, lenp, ppos); 1489} 1490 1491extern struct ctl_table random_table[]; 1492struct ctl_table random_table[] = { 1493 { 1494 .procname = "poolsize", 1495 .data = &sysctl_poolsize, 1496 .maxlen = sizeof(int), 1497 .mode = 0444, 1498 .proc_handler = proc_dointvec, 1499 }, 1500 { 1501 .procname = "entropy_avail", 1502 .data = &input_pool.init_bits, 1503 .maxlen = sizeof(int), 1504 .mode = 0444, 1505 .proc_handler = proc_dointvec, 1506 }, 1507 { 1508 .procname = "write_wakeup_threshold", 1509 .data = &sysctl_random_write_wakeup_bits, 1510 .maxlen = sizeof(int), 1511 .mode = 0644, 1512 .proc_handler = proc_do_rointvec, 1513 }, 1514 { 1515 .procname = "urandom_min_reseed_secs", 1516 .data = &sysctl_random_min_urandom_seed, 1517 .maxlen = sizeof(int), 1518 .mode = 0644, 1519 .proc_handler = proc_do_rointvec, 1520 }, 1521 { 1522 .procname = "boot_id", 1523 .data = &sysctl_bootid, 1524 .mode = 0444, 1525 .proc_handler = proc_do_uuid, 1526 }, 1527 { 1528 .procname = "uuid", 1529 .mode = 0444, 1530 .proc_handler = proc_do_uuid, 1531 }, 1532 { } 1533}; 1534#endif /* CONFIG_SYSCTL */ 1535