2 * random.c -- A strong random number generator
4 * Copyright (C) 2017-2022 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
6 * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005
8 * Copyright Theodore Ts'o, 1994, 1995, 1996, 1997, 1998, 1999. All
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, and the entire permission notice in its entirety,
16 * including the disclaimer of warranties.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. The name of the author may not be used to endorse or promote
21 * products derived from this software without specific prior
24 * ALTERNATIVELY, this product may be distributed under the terms of
25 * the GNU General Public License, in which case the provisions of the GPL are
26 * required INSTEAD OF the above restrictions. (This clause is
27 * necessary due to a potential bad interaction between the GPL and
28 * the restrictions contained in a BSD-style copyright.)
30 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
31 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
32 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ALL OF
33 * WHICH ARE HEREBY DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE
34 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
35 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT
36 * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
37 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
38 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
39 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
40 * USE OF THIS SOFTWARE, EVEN IF NOT ADVISED OF THE POSSIBILITY OF SUCH
45 * (now, with legal B.S. out of the way.....)
47 * This routine gathers environmental noise from device drivers, etc.,
48 * and returns good random numbers, suitable for cryptographic use.
49 * Besides the obvious cryptographic uses, these numbers are also good
50 * for seeding TCP sequence numbers, and other places where it is
51 * desirable to have numbers which are not only random, but hard to
52 * predict by an attacker.
57 * Computers are very predictable devices. Hence it is extremely hard
58 * to produce truly random numbers on a computer --- as opposed to
59 * pseudo-random numbers, which can easily generated by using a
60 * algorithm. Unfortunately, it is very easy for attackers to guess
61 * the sequence of pseudo-random number generators, and for some
62 * applications this is not acceptable. So instead, we must try to
63 * gather "environmental noise" from the computer's environment, which
64 * must be hard for outside attackers to observe, and use that to
65 * generate random numbers. In a Unix environment, this is best done
66 * from inside the kernel.
68 * Sources of randomness from the environment include inter-keyboard
69 * timings, inter-interrupt timings from some interrupts, and other
70 * events which are both (a) non-deterministic and (b) hard for an
71 * outside observer to measure. Randomness from these sources are
72 * added to an "entropy pool", which is mixed using a CRC-like function.
73 * This is not cryptographically strong, but it is adequate assuming
74 * the randomness is not chosen maliciously, and it is fast enough that
75 * the overhead of doing it on every interrupt is very reasonable.
76 * As random bytes are mixed into the entropy pool, the routines keep
77 * an *estimate* of how many bits of randomness have been stored into
78 * the random number generator's internal state.
80 * When random bytes are desired, they are obtained by taking the BLAKE2s
81 * hash of the contents of the "entropy pool". The BLAKE2s hash avoids
82 * exposing the internal state of the entropy pool. It is believed to
83 * be computationally infeasible to derive any useful information
84 * about the input of BLAKE2s from its output. Even if it is possible to
85 * analyze BLAKE2s in some clever way, as long as the amount of data
86 * returned from the generator is less than the inherent entropy in
87 * the pool, the output data is totally unpredictable. For this
88 * reason, the routine decreases its internal estimate of how many
89 * bits of "true randomness" are contained in the entropy pool as it
90 * outputs random numbers.
92 * If this estimate goes to zero, the routine can still generate
93 * random numbers; however, an attacker may (at least in theory) be
94 * able to infer the future output of the generator from prior
95 * outputs. This requires successful cryptanalysis of BLAKE2s, which is
96 * not believed to be feasible, but there is a remote possibility.
97 * Nonetheless, these numbers should be useful for the vast majority
100 * Exported interfaces ---- output
101 * ===============================
103 * There are four exported interfaces; two for use within the kernel,
104 * and two for use from userspace.
106 * Exported interfaces ---- userspace output
107 * -----------------------------------------
109 * The userspace interfaces are two character devices /dev/random and
110 * /dev/urandom. /dev/random is suitable for use when very high
111 * quality randomness is desired (for example, for key generation or
112 * one-time pads), as it will only return a maximum of the number of
113 * bits of randomness (as estimated by the random number generator)
114 * contained in the entropy pool.
116 * The /dev/urandom device does not have this limit, and will return
117 * as many bytes as are requested. As more and more random bytes are
118 * requested without giving time for the entropy pool to recharge,
119 * this will result in random numbers that are merely cryptographically
120 * strong. For many applications, however, this is acceptable.
122 * Exported interfaces ---- kernel output
123 * --------------------------------------
125 * The primary kernel interface is
127 * void get_random_bytes(void *buf, int nbytes);
129 * This interface will return the requested number of random bytes,
130 * and place it in the requested buffer. This is equivalent to a
131 * read from /dev/urandom.
133 * For less critical applications, there are the functions:
135 * u32 get_random_u32()
136 * u64 get_random_u64()
137 * unsigned int get_random_int()
138 * unsigned long get_random_long()
140 * These are produced by a cryptographic RNG seeded from get_random_bytes,
141 * and so do not deplete the entropy pool as much. These are recommended
142 * for most in-kernel operations *if the result is going to be stored in
145 * Specifically, the get_random_int() family do not attempt to do
146 * "anti-backtracking". If you capture the state of the kernel (e.g.
147 * by snapshotting the VM), you can figure out previous get_random_int()
148 * return values. But if the value is stored in the kernel anyway,
149 * this is not a problem.
151 * It *is* safe to expose get_random_int() output to attackers (e.g. as
152 * network cookies); given outputs 1..n, it's not feasible to predict
153 * outputs 0 or n+1. The only concern is an attacker who breaks into
154 * the kernel later; the get_random_int() engine is not reseeded as
155 * often as the get_random_bytes() one.
157 * get_random_bytes() is needed for keys that need to stay secret after
158 * they are erased from the kernel. For example, any key that will
159 * be wrapped and stored encrypted. And session encryption keys: we'd
160 * like to know that after the session is closed and the keys erased,
161 * the plaintext is unrecoverable to someone who recorded the ciphertext.
163 * But for network ports/cookies, stack canaries, PRNG seeds, address
164 * space layout randomization, session *authentication* keys, or other
165 * applications where the sensitive data is stored in the kernel in
166 * plaintext for as long as it's sensitive, the get_random_int() family
169 * Consider ASLR. We want to keep the address space secret from an
170 * outside attacker while the process is running, but once the address
171 * space is torn down, it's of no use to an attacker any more. And it's
172 * stored in kernel data structures as long as it's alive, so worrying
173 * about an attacker's ability to extrapolate it from the get_random_int()
176 * Even some cryptographic keys are safe to generate with get_random_int().
177 * In particular, keys for SipHash are generally fine. Here, knowledge
178 * of the key authorizes you to do something to a kernel object (inject
179 * packets to a network connection, or flood a hash table), and the
180 * key is stored with the object being protected. Once it goes away,
181 * we no longer care if anyone knows the key.
186 * For even weaker applications, see the pseudorandom generator
187 * prandom_u32(), prandom_max(), and prandom_bytes(). If the random
188 * numbers aren't security-critical at all, these are *far* cheaper.
189 * Useful for self-tests, random error simulation, randomized backoffs,
190 * and any other application where you trust that nobody is trying to
191 * maliciously mess with you by guessing the "random" numbers.
193 * Exported interfaces ---- input
194 * ==============================
196 * The current exported interfaces for gathering environmental noise
197 * from the devices are:
199 * void add_device_randomness(const void *buf, unsigned int size);
200 * void add_input_randomness(unsigned int type, unsigned int code,
201 * unsigned int value);
202 * void add_interrupt_randomness(int irq);
203 * void add_disk_randomness(struct gendisk *disk);
204 * void add_hwgenerator_randomness(const char *buffer, size_t count,
206 * void add_bootloader_randomness(const void *buf, unsigned int size);
208 * add_device_randomness() is for adding data to the random pool that
209 * is likely to differ between two devices (or possibly even per boot).
210 * This would be things like MAC addresses or serial numbers, or the
211 * read-out of the RTC. This does *not* add any actual entropy to the
212 * pool, but it initializes the pool to different values for devices
213 * that might otherwise be identical and have very little entropy
214 * available to them (particularly common in the embedded world).
216 * add_input_randomness() uses the input layer interrupt timing, as well as
217 * the event type information from the hardware.
219 * add_interrupt_randomness() uses the interrupt timing as random
220 * inputs to the entropy pool. Using the cycle counters and the irq source
221 * as inputs, it feeds the randomness roughly once a second.
223 * add_disk_randomness() uses what amounts to the seek time of block
224 * layer request events, on a per-disk_devt basis, as input to the
225 * entropy pool. Note that high-speed solid state drives with very low
226 * seek times do not make for good sources of entropy, as their seek
227 * times are usually fairly consistent.
229 * All of these routines try to estimate how many bits of randomness a
230 * particular randomness source. They do this by keeping track of the
231 * first and second order deltas of the event timings.
233 * add_hwgenerator_randomness() is for true hardware RNGs, and will credit
234 * entropy as specified by the caller. If the entropy pool is full it will
235 * block until more entropy is needed.
237 * add_bootloader_randomness() is the same as add_hwgenerator_randomness() or
238 * add_device_randomness(), depending on whether or not the configuration
239 * option CONFIG_RANDOM_TRUST_BOOTLOADER is set.
241 * Ensuring unpredictability at system startup
242 * ============================================
244 * When any operating system starts up, it will go through a sequence
245 * of actions that are fairly predictable by an adversary, especially
246 * if the start-up does not involve interaction with a human operator.
247 * This reduces the actual number of bits of unpredictability in the
248 * entropy pool below the value in entropy_count. In order to
249 * counteract this effect, it helps to carry information in the
250 * entropy pool across shut-downs and start-ups. To do this, put the
251 * following lines an appropriate script which is run during the boot
254 * echo "Initializing random number generator..."
255 * random_seed=/var/run/random-seed
256 * # Carry a random seed from start-up to start-up
257 * # Load and then save the whole entropy pool
258 * if [ -f $random_seed ]; then
259 * cat $random_seed >/dev/urandom
263 * chmod 600 $random_seed
264 * dd if=/dev/urandom of=$random_seed count=1 bs=512
266 * and the following lines in an appropriate script which is run as
267 * the system is shutdown:
269 * # Carry a random seed from shut-down to start-up
270 * # Save the whole entropy pool
271 * echo "Saving random seed..."
272 * random_seed=/var/run/random-seed
274 * chmod 600 $random_seed
275 * dd if=/dev/urandom of=$random_seed count=1 bs=512
277 * For example, on most modern systems using the System V init
278 * scripts, such code fragments would be found in
279 * /etc/rc.d/init.d/random. On older Linux systems, the correct script
280 * location might be in /etc/rcb.d/rc.local or /etc/rc.d/rc.0.
282 * Effectively, these commands cause the contents of the entropy pool
283 * to be saved at shut-down time and reloaded into the entropy pool at
284 * start-up. (The 'dd' in the addition to the bootup script is to
285 * make sure that /etc/random-seed is different for every start-up,
286 * even if the system crashes without executing rc.0.) Even with
287 * complete knowledge of the start-up activities, predicting the state
288 * of the entropy pool requires knowledge of the previous history of
291 * Configuring the /dev/random driver under Linux
292 * ==============================================
294 * The /dev/random driver under Linux uses minor numbers 8 and 9 of
295 * the /dev/mem major number (#1). So if your system does not have
296 * /dev/random and /dev/urandom created already, they can be created
297 * by using the commands:
299 * mknod /dev/random c 1 8
300 * mknod /dev/urandom c 1 9
305 * Ideas for constructing this random number generator were derived
306 * from Pretty Good Privacy's random number generator, and from private
307 * discussions with Phil Karn. Colin Plumb provided a faster random
308 * number generator, which speed up the mixing function of the entropy
309 * pool, taken from PGPfone. Dale Worley has also contributed many
310 * useful ideas and suggestions to improve this driver.
312 * Any flaws in the design are solely my responsibility, and should
313 * not be attributed to the Phil, Colin, or any of authors of PGP.
315 * Further background information on this topic may be obtained from
316 * RFC 1750, "Randomness Recommendations for Security", by Donald
317 * Eastlake, Steve Crocker, and Jeff Schiller.
320 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
322 #include <linux/utsname.h>
323 #include <linux/module.h>
324 #include <linux/kernel.h>
325 #include <linux/major.h>
326 #include <linux/string.h>
327 #include <linux/fcntl.h>
328 #include <linux/slab.h>
329 #include <linux/random.h>
330 #include <linux/poll.h>
331 #include <linux/init.h>
332 #include <linux/fs.h>
333 #include <linux/genhd.h>
334 #include <linux/interrupt.h>
335 #include <linux/mm.h>
336 #include <linux/nodemask.h>
337 #include <linux/spinlock.h>
338 #include <linux/kthread.h>
339 #include <linux/percpu.h>
340 #include <linux/ptrace.h>
341 #include <linux/workqueue.h>
342 #include <linux/irq.h>
343 #include <linux/ratelimit.h>
344 #include <linux/syscalls.h>
345 #include <linux/completion.h>
346 #include <linux/uuid.h>
347 #include <crypto/chacha.h>
348 #include <crypto/blake2s.h>
350 #include <asm/processor.h>
351 #include <linux/uaccess.h>
353 #include <asm/irq_regs.h>
356 #define CREATE_TRACE_POINTS
357 #include <trace/events/random.h>
359 /* #define ADD_INTERRUPT_BENCH */
362 * If the entropy count falls under this number of bits, then we
363 * should wake up processes which are selecting or polling on write
364 * access to /dev/random.
366 static int random_write_wakeup_bits = 28 * (1 << 5);
369 * Originally, we used a primitive polynomial of degree .poolwords
370 * over GF(2). The taps for various sizes are defined below. They
371 * were chosen to be evenly spaced except for the last tap, which is 1
372 * to get the twisting happening as fast as possible.
374 * For the purposes of better mixing, we use the CRC-32 polynomial as
375 * well to make a (modified) twisted Generalized Feedback Shift
376 * Register. (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR
377 * generators. ACM Transactions on Modeling and Computer Simulation
378 * 2(3):179-194. Also see M. Matsumoto & Y. Kurita, 1994. Twisted
379 * GFSR generators II. ACM Transactions on Modeling and Computer
380 * Simulation 4:254-266)
382 * Thanks to Colin Plumb for suggesting this.
384 * The mixing operation is much less sensitive than the output hash,
385 * where we use BLAKE2s. All that we want of mixing operation is that
386 * it be a good non-cryptographic hash; i.e. it not produce collisions
387 * when fed "random" data of the sort we expect to see. As long as
388 * the pool state differs for different inputs, we have preserved the
389 * input entropy and done a good job. The fact that an intelligent
390 * attacker can construct inputs that will produce controlled
391 * alterations to the pool's state is not important because we don't
392 * consider such inputs to contribute any randomness. The only
393 * property we need with respect to them is that the attacker can't
394 * increase his/her knowledge of the pool's state. Since all
395 * additions are reversible (knowing the final state and the input,
396 * you can reconstruct the initial state), if an attacker has any
397 * uncertainty about the initial state, he/she can only shuffle that
398 * uncertainty about, but never cause any collisions (which would
399 * decrease the uncertainty).
401 * Our mixing functions were analyzed by Lacharme, Roeck, Strubel, and
402 * Videau in their paper, "The Linux Pseudorandom Number Generator
403 * Revisited" (see: http://eprint.iacr.org/2012/251.pdf). In their
404 * paper, they point out that we are not using a true Twisted GFSR,
405 * since Matsumoto & Kurita used a trinomial feedback polynomial (that
406 * is, with only three taps, instead of the six that we are using).
407 * As a result, the resulting polynomial is neither primitive nor
408 * irreducible, and hence does not have a maximal period over
409 * GF(2**32). They suggest a slight change to the generator
410 * polynomial which improves the resulting TGFSR polynomial to be
411 * irreducible, which we have made here.
415 POOL_WORDMASK = POOL_WORDS - 1,
416 POOL_BYTES = POOL_WORDS * sizeof(u32),
417 POOL_BITS = POOL_BYTES * 8,
418 POOL_BITSHIFT = ilog2(POOL_BITS),
420 /* To allow fractional bits to be tracked, the entropy_count field is
421 * denominated in units of 1/8th bits. */
422 POOL_ENTROPY_SHIFT = 3,
423 #define POOL_ENTROPY_BITS() (input_pool.entropy_count >> POOL_ENTROPY_SHIFT)
424 POOL_FRACBITS = POOL_BITS << POOL_ENTROPY_SHIFT,
426 /* x^128 + x^104 + x^76 + x^51 +x^25 + x + 1 */
433 EXTRACT_SIZE = BLAKE2S_HASH_SIZE / 2
437 * Static global variables
439 static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
440 static struct fasync_struct *fasync;
442 static DEFINE_SPINLOCK(random_ready_list_lock);
443 static LIST_HEAD(random_ready_list);
447 unsigned long init_time;
451 static struct crng_state primary_crng = {
452 .lock = __SPIN_LOCK_UNLOCKED(primary_crng.lock),
453 .state[0] = CHACHA_CONSTANT_EXPA,
454 .state[1] = CHACHA_CONSTANT_ND_3,
455 .state[2] = CHACHA_CONSTANT_2_BY,
456 .state[3] = CHACHA_CONSTANT_TE_K,
460 * crng_init = 0 --> Uninitialized
462 * 2 --> Initialized from input_pool
464 * crng_init is protected by primary_crng->lock, and only increases
465 * its value (from 0->1->2).
467 static int crng_init = 0;
468 static bool crng_need_final_init = false;
469 #define crng_ready() (likely(crng_init > 1))
470 static int crng_init_cnt = 0;
471 static unsigned long crng_global_init_time = 0;
472 #define CRNG_INIT_CNT_THRESH (2 * CHACHA_KEY_SIZE)
473 static void _extract_crng(struct crng_state *crng, u8 out[CHACHA_BLOCK_SIZE]);
474 static void _crng_backtrack_protect(struct crng_state *crng,
475 u8 tmp[CHACHA_BLOCK_SIZE], int used);
476 static void process_random_ready_list(void);
477 static void _get_random_bytes(void *buf, int nbytes);
479 static struct ratelimit_state unseeded_warning =
480 RATELIMIT_STATE_INIT("warn_unseeded_randomness", HZ, 3);
481 static struct ratelimit_state urandom_warning =
482 RATELIMIT_STATE_INIT("warn_urandom_randomness", HZ, 3);
484 static int ratelimit_disable __read_mostly;
486 module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
487 MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
489 /**********************************************************************
491 * OS independent entropy store. Here are the functions which handle
492 * storing entropy in an entropy pool.
494 **********************************************************************/
496 static u32 input_pool_data[POOL_WORDS] __latent_entropy;
504 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
507 static ssize_t extract_entropy(void *buf, size_t nbytes, int min);
508 static ssize_t _extract_entropy(void *buf, size_t nbytes);
510 static void crng_reseed(struct crng_state *crng, bool use_input_pool);
512 static const u32 twist_table[8] = {
513 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
514 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
517 * This function adds bytes into the entropy "pool". It does not
518 * update the entropy estimate. The caller should call
519 * credit_entropy_bits if this is appropriate.
521 * The pool is stirred with a primitive polynomial of the appropriate
522 * degree, and then twisted. We twist by three bits at a time because
523 * it's cheap to do so and helps slightly in the expected case where
524 * the entropy is concentrated in the low-order bits.
526 static void _mix_pool_bytes(const void *in, int nbytes)
530 const u8 *bytes = in;
533 input_rotate = input_pool.input_rotate;
534 i = input_pool.add_ptr;
536 /* mix one byte at a time to simplify size handling and churn faster */
538 w = rol32(*bytes++, input_rotate);
539 i = (i - 1) & POOL_WORDMASK;
541 /* XOR in the various taps */
542 w ^= input_pool_data[i];
543 w ^= input_pool_data[(i + POOL_TAP1) & POOL_WORDMASK];
544 w ^= input_pool_data[(i + POOL_TAP2) & POOL_WORDMASK];
545 w ^= input_pool_data[(i + POOL_TAP3) & POOL_WORDMASK];
546 w ^= input_pool_data[(i + POOL_TAP4) & POOL_WORDMASK];
547 w ^= input_pool_data[(i + POOL_TAP5) & POOL_WORDMASK];
549 /* Mix the result back in with a twist */
550 input_pool_data[i] = (w >> 3) ^ twist_table[w & 7];
553 * Normally, we add 7 bits of rotation to the pool.
554 * At the beginning of the pool, add an extra 7 bits
555 * rotation, so that successive passes spread the
556 * input bits across the pool evenly.
558 input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
561 input_pool.input_rotate = input_rotate;
562 input_pool.add_ptr = i;
565 static void __mix_pool_bytes(const void *in, int nbytes)
567 trace_mix_pool_bytes_nolock(nbytes, _RET_IP_);
568 _mix_pool_bytes(in, nbytes);
571 static void mix_pool_bytes(const void *in, int nbytes)
575 trace_mix_pool_bytes(nbytes, _RET_IP_);
576 spin_lock_irqsave(&input_pool.lock, flags);
577 _mix_pool_bytes(in, nbytes);
578 spin_unlock_irqrestore(&input_pool.lock, flags);
589 * This is a fast mixing routine used by the interrupt randomness
590 * collector. It's hardcoded for an 128 bit pool and assumes that any
591 * locks that might be needed are taken by the caller.
593 static void fast_mix(struct fast_pool *f)
595 u32 a = f->pool[0], b = f->pool[1];
596 u32 c = f->pool[2], d = f->pool[3];
599 b = rol32(b, 6); d = rol32(d, 27);
603 b = rol32(b, 16); d = rol32(d, 14);
607 b = rol32(b, 6); d = rol32(d, 27);
611 b = rol32(b, 16); d = rol32(d, 14);
614 f->pool[0] = a; f->pool[1] = b;
615 f->pool[2] = c; f->pool[3] = d;
619 static void process_random_ready_list(void)
622 struct random_ready_callback *rdy, *tmp;
624 spin_lock_irqsave(&random_ready_list_lock, flags);
625 list_for_each_entry_safe(rdy, tmp, &random_ready_list, list) {
626 struct module *owner = rdy->owner;
628 list_del_init(&rdy->list);
632 spin_unlock_irqrestore(&random_ready_list_lock, flags);
636 * Credit (or debit) the entropy store with n bits of entropy.
637 * Use credit_entropy_bits_safe() if the value comes from userspace
638 * or otherwise should be checked for extreme values.
640 static void credit_entropy_bits(int nbits)
642 int entropy_count, entropy_bits, orig;
643 int nfrac = nbits << POOL_ENTROPY_SHIFT;
645 /* Ensure that the multiplication can avoid being 64 bits wide. */
646 BUILD_BUG_ON(2 * (POOL_ENTROPY_SHIFT + POOL_BITSHIFT) > 31);
652 entropy_count = orig = READ_ONCE(input_pool.entropy_count);
655 entropy_count += nfrac;
658 * Credit: we have to account for the possibility of
659 * overwriting already present entropy. Even in the
660 * ideal case of pure Shannon entropy, new contributions
661 * approach the full value asymptotically:
663 * entropy <- entropy + (pool_size - entropy) *
664 * (1 - exp(-add_entropy/pool_size))
666 * For add_entropy <= pool_size/2 then
667 * (1 - exp(-add_entropy/pool_size)) >=
668 * (add_entropy/pool_size)*0.7869...
669 * so we can approximate the exponential with
670 * 3/4*add_entropy/pool_size and still be on the
671 * safe side by adding at most pool_size/2 at a time.
673 * The use of pool_size-2 in the while statement is to
674 * prevent rounding artifacts from making the loop
675 * arbitrarily long; this limits the loop to log2(pool_size)*2
676 * turns no matter how large nbits is.
679 const int s = POOL_BITSHIFT + POOL_ENTROPY_SHIFT + 2;
680 /* The +2 corresponds to the /4 in the denominator */
683 unsigned int anfrac = min(pnfrac, POOL_FRACBITS / 2);
685 ((POOL_FRACBITS - entropy_count) * anfrac * 3) >> s;
687 entropy_count += add;
689 } while (unlikely(entropy_count < POOL_FRACBITS - 2 && pnfrac));
692 if (WARN_ON(entropy_count < 0)) {
693 pr_warn("negative entropy/overflow: count %d\n", entropy_count);
695 } else if (entropy_count > POOL_FRACBITS)
696 entropy_count = POOL_FRACBITS;
697 if (cmpxchg(&input_pool.entropy_count, orig, entropy_count) != orig)
700 trace_credit_entropy_bits(nbits, entropy_count >> POOL_ENTROPY_SHIFT, _RET_IP_);
702 entropy_bits = entropy_count >> POOL_ENTROPY_SHIFT;
703 if (crng_init < 2 && entropy_bits >= 128)
704 crng_reseed(&primary_crng, true);
707 static int credit_entropy_bits_safe(int nbits)
712 /* Cap the value to avoid overflows */
713 nbits = min(nbits, POOL_BITS);
715 credit_entropy_bits(nbits);
719 /*********************************************************************
721 * CRNG using CHACHA20
723 *********************************************************************/
725 #define CRNG_RESEED_INTERVAL (300 * HZ)
727 static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
730 * Hack to deal with crazy userspace progams when they are all trying
731 * to access /dev/urandom in parallel. The programs are almost
732 * certainly doing something terribly wrong, but we'll work around
733 * their brain damage.
735 static struct crng_state **crng_node_pool __read_mostly;
737 static void invalidate_batched_entropy(void);
738 static void numa_crng_init(void);
740 static bool trust_cpu __ro_after_init = IS_ENABLED(CONFIG_RANDOM_TRUST_CPU);
741 static int __init parse_trust_cpu(char *arg)
743 return kstrtobool(arg, &trust_cpu);
745 early_param("random.trust_cpu", parse_trust_cpu);
747 static bool crng_init_try_arch(struct crng_state *crng)
750 bool arch_init = true;
753 for (i = 4; i < 16; i++) {
754 if (!arch_get_random_seed_long(&rv) &&
755 !arch_get_random_long(&rv)) {
756 rv = random_get_entropy();
759 crng->state[i] ^= rv;
765 static bool __init crng_init_try_arch_early(struct crng_state *crng)
768 bool arch_init = true;
771 for (i = 4; i < 16; i++) {
772 if (!arch_get_random_seed_long_early(&rv) &&
773 !arch_get_random_long_early(&rv)) {
774 rv = random_get_entropy();
777 crng->state[i] ^= rv;
783 static void crng_initialize_secondary(struct crng_state *crng)
785 chacha_init_consts(crng->state);
786 _get_random_bytes(&crng->state[4], sizeof(u32) * 12);
787 crng_init_try_arch(crng);
788 crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
791 static void __init crng_initialize_primary(struct crng_state *crng)
793 _extract_entropy(&crng->state[4], sizeof(u32) * 12);
794 if (crng_init_try_arch_early(crng) && trust_cpu && crng_init < 2) {
795 invalidate_batched_entropy();
798 pr_notice("crng init done (trusting CPU's manufacturer)\n");
800 crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
803 static void crng_finalize_init(struct crng_state *crng)
805 if (crng != &primary_crng || crng_init >= 2)
808 /* We can't call numa_crng_init until we have workqueues,
809 * so mark this for processing later. */
810 crng_need_final_init = true;
814 invalidate_batched_entropy();
817 process_random_ready_list();
818 wake_up_interruptible(&crng_init_wait);
819 kill_fasync(&fasync, SIGIO, POLL_IN);
820 pr_notice("crng init done\n");
821 if (unseeded_warning.missed) {
822 pr_notice("%d get_random_xx warning(s) missed due to ratelimiting\n",
823 unseeded_warning.missed);
824 unseeded_warning.missed = 0;
826 if (urandom_warning.missed) {
827 pr_notice("%d urandom warning(s) missed due to ratelimiting\n",
828 urandom_warning.missed);
829 urandom_warning.missed = 0;
833 static void do_numa_crng_init(struct work_struct *work)
836 struct crng_state *crng;
837 struct crng_state **pool;
839 pool = kcalloc(nr_node_ids, sizeof(*pool), GFP_KERNEL | __GFP_NOFAIL);
840 for_each_online_node(i) {
841 crng = kmalloc_node(sizeof(struct crng_state),
842 GFP_KERNEL | __GFP_NOFAIL, i);
843 spin_lock_init(&crng->lock);
844 crng_initialize_secondary(crng);
847 /* pairs with READ_ONCE() in select_crng() */
848 if (cmpxchg_release(&crng_node_pool, NULL, pool) != NULL) {
855 static DECLARE_WORK(numa_crng_init_work, do_numa_crng_init);
857 static void numa_crng_init(void)
859 if (IS_ENABLED(CONFIG_NUMA))
860 schedule_work(&numa_crng_init_work);
863 static struct crng_state *select_crng(void)
865 if (IS_ENABLED(CONFIG_NUMA)) {
866 struct crng_state **pool;
867 int nid = numa_node_id();
869 /* pairs with cmpxchg_release() in do_numa_crng_init() */
870 pool = READ_ONCE(crng_node_pool);
871 if (pool && pool[nid])
875 return &primary_crng;
879 * crng_fast_load() can be called by code in the interrupt service
880 * path. So we can't afford to dilly-dally. Returns the number of
881 * bytes processed from cp.
883 static size_t crng_fast_load(const u8 *cp, size_t len)
889 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
891 if (crng_init != 0) {
892 spin_unlock_irqrestore(&primary_crng.lock, flags);
895 p = (u8 *)&primary_crng.state[4];
896 while (len > 0 && crng_init_cnt < CRNG_INIT_CNT_THRESH) {
897 p[crng_init_cnt % CHACHA_KEY_SIZE] ^= *cp;
898 cp++; crng_init_cnt++; len--; ret++;
900 spin_unlock_irqrestore(&primary_crng.lock, flags);
901 if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
902 invalidate_batched_entropy();
904 pr_notice("fast init done\n");
910 * crng_slow_load() is called by add_device_randomness, which has two
911 * attributes. (1) We can't trust the buffer passed to it is
912 * guaranteed to be unpredictable (so it might not have any entropy at
913 * all), and (2) it doesn't have the performance constraints of
916 * So we do something more comprehensive which is guaranteed to touch
917 * all of the primary_crng's state, and which uses a LFSR with a
918 * period of 255 as part of the mixing algorithm. Finally, we do
919 * *not* advance crng_init_cnt since buffer we may get may be something
920 * like a fixed DMI table (for example), which might very well be
921 * unique to the machine, but is otherwise unvarying.
923 static int crng_slow_load(const u8 *cp, size_t len)
928 unsigned int i, max = CHACHA_KEY_SIZE;
929 const u8 *src_buf = cp;
930 u8 *dest_buf = (u8 *)&primary_crng.state[4];
932 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
934 if (crng_init != 0) {
935 spin_unlock_irqrestore(&primary_crng.lock, flags);
941 for (i = 0; i < max; i++) {
946 tmp = dest_buf[i % CHACHA_KEY_SIZE];
947 dest_buf[i % CHACHA_KEY_SIZE] ^= src_buf[i % len] ^ lfsr;
948 lfsr += (tmp << 3) | (tmp >> 5);
950 spin_unlock_irqrestore(&primary_crng.lock, flags);
954 static void crng_reseed(struct crng_state *crng, bool use_input_pool)
959 u8 block[CHACHA_BLOCK_SIZE];
963 if (use_input_pool) {
964 num = extract_entropy(&buf, 32, 16);
968 _extract_crng(&primary_crng, buf.block);
969 _crng_backtrack_protect(&primary_crng, buf.block,
972 spin_lock_irqsave(&crng->lock, flags);
973 for (i = 0; i < 8; i++) {
975 if (!arch_get_random_seed_long(&rv) &&
976 !arch_get_random_long(&rv))
977 rv = random_get_entropy();
978 crng->state[i + 4] ^= buf.key[i] ^ rv;
980 memzero_explicit(&buf, sizeof(buf));
981 WRITE_ONCE(crng->init_time, jiffies);
982 spin_unlock_irqrestore(&crng->lock, flags);
983 crng_finalize_init(crng);
986 static void _extract_crng(struct crng_state *crng, u8 out[CHACHA_BLOCK_SIZE])
988 unsigned long flags, init_time;
991 init_time = READ_ONCE(crng->init_time);
992 if (time_after(READ_ONCE(crng_global_init_time), init_time) ||
993 time_after(jiffies, init_time + CRNG_RESEED_INTERVAL))
994 crng_reseed(crng, crng == &primary_crng);
996 spin_lock_irqsave(&crng->lock, flags);
997 chacha20_block(&crng->state[0], out);
998 if (crng->state[12] == 0)
1000 spin_unlock_irqrestore(&crng->lock, flags);
1003 static void extract_crng(u8 out[CHACHA_BLOCK_SIZE])
1005 _extract_crng(select_crng(), out);
1009 * Use the leftover bytes from the CRNG block output (if there is
1010 * enough) to mutate the CRNG key to provide backtracking protection.
1012 static void _crng_backtrack_protect(struct crng_state *crng,
1013 u8 tmp[CHACHA_BLOCK_SIZE], int used)
1015 unsigned long flags;
1019 used = round_up(used, sizeof(u32));
1020 if (used + CHACHA_KEY_SIZE > CHACHA_BLOCK_SIZE) {
1024 spin_lock_irqsave(&crng->lock, flags);
1025 s = (u32 *)&tmp[used];
1026 d = &crng->state[4];
1027 for (i = 0; i < 8; i++)
1029 spin_unlock_irqrestore(&crng->lock, flags);
1032 static void crng_backtrack_protect(u8 tmp[CHACHA_BLOCK_SIZE], int used)
1034 _crng_backtrack_protect(select_crng(), tmp, used);
1037 static ssize_t extract_crng_user(void __user *buf, size_t nbytes)
1039 ssize_t ret = 0, i = CHACHA_BLOCK_SIZE;
1040 u8 tmp[CHACHA_BLOCK_SIZE] __aligned(4);
1041 int large_request = (nbytes > 256);
1044 if (large_request && need_resched()) {
1045 if (signal_pending(current)) {
1054 i = min_t(int, nbytes, CHACHA_BLOCK_SIZE);
1055 if (copy_to_user(buf, tmp, i)) {
1064 crng_backtrack_protect(tmp, i);
1066 /* Wipe data just written to memory */
1067 memzero_explicit(tmp, sizeof(tmp));
1072 /*********************************************************************
1074 * Entropy input management
1076 *********************************************************************/
1078 /* There is one of these per entropy source */
1079 struct timer_rand_state {
1081 long last_delta, last_delta2;
1084 #define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
1087 * Add device- or boot-specific data to the input pool to help
1090 * None of this adds any entropy; it is meant to avoid the problem of
1091 * the entropy pool having similar initial state across largely
1092 * identical devices.
1094 void add_device_randomness(const void *buf, unsigned int size)
1096 unsigned long time = random_get_entropy() ^ jiffies;
1097 unsigned long flags;
1099 if (!crng_ready() && size)
1100 crng_slow_load(buf, size);
1102 trace_add_device_randomness(size, _RET_IP_);
1103 spin_lock_irqsave(&input_pool.lock, flags);
1104 _mix_pool_bytes(buf, size);
1105 _mix_pool_bytes(&time, sizeof(time));
1106 spin_unlock_irqrestore(&input_pool.lock, flags);
1108 EXPORT_SYMBOL(add_device_randomness);
1110 static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
1113 * This function adds entropy to the entropy "pool" by using timing
1114 * delays. It uses the timer_rand_state structure to make an estimate
1115 * of how many bits of entropy this call has added to the pool.
1117 * The number "num" is also added to the pool - it should somehow describe
1118 * the type of event which just happened. This is currently 0-255 for
1119 * keyboard scan codes, and 256 upwards for interrupts.
1122 static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
1126 unsigned int cycles;
1129 long delta, delta2, delta3;
1131 sample.jiffies = jiffies;
1132 sample.cycles = random_get_entropy();
1134 mix_pool_bytes(&sample, sizeof(sample));
1137 * Calculate number of bits of randomness we probably added.
1138 * We take into account the first, second and third-order deltas
1139 * in order to make our estimate.
1141 delta = sample.jiffies - READ_ONCE(state->last_time);
1142 WRITE_ONCE(state->last_time, sample.jiffies);
1144 delta2 = delta - READ_ONCE(state->last_delta);
1145 WRITE_ONCE(state->last_delta, delta);
1147 delta3 = delta2 - READ_ONCE(state->last_delta2);
1148 WRITE_ONCE(state->last_delta2, delta2);
1162 * delta is now minimum absolute delta.
1163 * Round down by 1 bit on general principles,
1164 * and limit entropy estimate to 12 bits.
1166 credit_entropy_bits(min_t(int, fls(delta >> 1), 11));
1169 void add_input_randomness(unsigned int type, unsigned int code,
1172 static unsigned char last_value;
1174 /* ignore autorepeat and the like */
1175 if (value == last_value)
1179 add_timer_randomness(&input_timer_state,
1180 (type << 4) ^ code ^ (code >> 4) ^ value);
1181 trace_add_input_randomness(POOL_ENTROPY_BITS());
1183 EXPORT_SYMBOL_GPL(add_input_randomness);
1185 static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
1187 #ifdef ADD_INTERRUPT_BENCH
1188 static unsigned long avg_cycles, avg_deviation;
1190 #define AVG_SHIFT 8 /* Exponential average factor k=1/256 */
1191 #define FIXED_1_2 (1 << (AVG_SHIFT - 1))
1193 static void add_interrupt_bench(cycles_t start)
1195 long delta = random_get_entropy() - start;
1197 /* Use a weighted moving average */
1198 delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
1199 avg_cycles += delta;
1200 /* And average deviation */
1201 delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
1202 avg_deviation += delta;
1205 #define add_interrupt_bench(x)
1208 static u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
1210 u32 *ptr = (u32 *)regs;
1215 idx = READ_ONCE(f->reg_idx);
1216 if (idx >= sizeof(struct pt_regs) / sizeof(u32))
1219 WRITE_ONCE(f->reg_idx, idx);
1223 void add_interrupt_randomness(int irq)
1225 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1226 struct pt_regs *regs = get_irq_regs();
1227 unsigned long now = jiffies;
1228 cycles_t cycles = random_get_entropy();
1233 cycles = get_reg(fast_pool, regs);
1234 c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
1235 j_high = (sizeof(now) > 4) ? now >> 32 : 0;
1236 fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
1237 fast_pool->pool[1] ^= now ^ c_high;
1238 ip = regs ? instruction_pointer(regs) : _RET_IP_;
1239 fast_pool->pool[2] ^= ip;
1240 fast_pool->pool[3] ^=
1241 (sizeof(ip) > 4) ? ip >> 32 : get_reg(fast_pool, regs);
1243 fast_mix(fast_pool);
1244 add_interrupt_bench(cycles);
1246 if (unlikely(crng_init == 0)) {
1247 if ((fast_pool->count >= 64) &&
1248 crng_fast_load((u8 *)fast_pool->pool, sizeof(fast_pool->pool)) > 0) {
1249 fast_pool->count = 0;
1250 fast_pool->last = now;
1255 if ((fast_pool->count < 64) && !time_after(now, fast_pool->last + HZ))
1258 if (!spin_trylock(&input_pool.lock))
1261 fast_pool->last = now;
1262 __mix_pool_bytes(&fast_pool->pool, sizeof(fast_pool->pool));
1263 spin_unlock(&input_pool.lock);
1265 fast_pool->count = 0;
1267 /* award one bit for the contents of the fast pool */
1268 credit_entropy_bits(1);
1270 EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1273 void add_disk_randomness(struct gendisk *disk)
1275 if (!disk || !disk->random)
1277 /* first major is 1, so we get >= 0x200 here */
1278 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1279 trace_add_disk_randomness(disk_devt(disk), POOL_ENTROPY_BITS());
1281 EXPORT_SYMBOL_GPL(add_disk_randomness);
1284 /*********************************************************************
1286 * Entropy extraction routines
1288 *********************************************************************/
1291 * This function decides how many bytes to actually take from the
1292 * given pool, and also debits the entropy count accordingly.
1294 static size_t account(size_t nbytes, int min)
1296 int entropy_count, orig;
1297 size_t ibytes, nfrac;
1299 BUG_ON(input_pool.entropy_count > POOL_FRACBITS);
1301 /* Can we pull enough? */
1303 entropy_count = orig = READ_ONCE(input_pool.entropy_count);
1304 if (WARN_ON(entropy_count < 0)) {
1305 pr_warn("negative entropy count: count %d\n", entropy_count);
1309 /* never pull more than available */
1310 ibytes = min_t(size_t, nbytes, entropy_count >> (POOL_ENTROPY_SHIFT + 3));
1313 nfrac = ibytes << (POOL_ENTROPY_SHIFT + 3);
1314 if ((size_t)entropy_count > nfrac)
1315 entropy_count -= nfrac;
1319 if (cmpxchg(&input_pool.entropy_count, orig, entropy_count) != orig)
1322 trace_debit_entropy(8 * ibytes);
1323 if (ibytes && POOL_ENTROPY_BITS() < random_write_wakeup_bits) {
1324 wake_up_interruptible(&random_write_wait);
1325 kill_fasync(&fasync, SIGIO, POLL_OUT);
1332 * This function does the actual extraction for extract_entropy.
1334 * Note: we assume that .poolwords is a multiple of 16 words.
1336 static void extract_buf(u8 *out)
1338 struct blake2s_state state __aligned(__alignof__(unsigned long));
1339 u8 hash[BLAKE2S_HASH_SIZE];
1340 unsigned long *salt;
1341 unsigned long flags;
1343 blake2s_init(&state, sizeof(hash));
1346 * If we have an architectural hardware random number
1347 * generator, use it for BLAKE2's salt & personal fields.
1349 for (salt = (unsigned long *)&state.h[4];
1350 salt < (unsigned long *)&state.h[8]; ++salt) {
1352 if (!arch_get_random_long(&v))
1357 /* Generate a hash across the pool */
1358 spin_lock_irqsave(&input_pool.lock, flags);
1359 blake2s_update(&state, (const u8 *)input_pool_data, POOL_BYTES);
1360 blake2s_final(&state, hash); /* final zeros out state */
1363 * We mix the hash back into the pool to prevent backtracking
1364 * attacks (where the attacker knows the state of the pool
1365 * plus the current outputs, and attempts to find previous
1366 * outputs), unless the hash function can be inverted. By
1367 * mixing at least a hash worth of hash data back, we make
1368 * brute-forcing the feedback as hard as brute-forcing the
1371 __mix_pool_bytes(hash, sizeof(hash));
1372 spin_unlock_irqrestore(&input_pool.lock, flags);
1374 /* Note that EXTRACT_SIZE is half of hash size here, because above
1375 * we've dumped the full length back into mixer. By reducing the
1376 * amount that we emit, we retain a level of forward secrecy.
1378 memcpy(out, hash, EXTRACT_SIZE);
1379 memzero_explicit(hash, sizeof(hash));
1382 static ssize_t _extract_entropy(void *buf, size_t nbytes)
1385 u8 tmp[EXTRACT_SIZE];
1389 i = min_t(int, nbytes, EXTRACT_SIZE);
1390 memcpy(buf, tmp, i);
1396 /* Wipe data just returned from memory */
1397 memzero_explicit(tmp, sizeof(tmp));
1403 * This function extracts randomness from the "entropy pool", and
1404 * returns it in a buffer.
1406 * The min parameter specifies the minimum amount we can pull before
1407 * failing to avoid races that defeat catastrophic reseeding.
1409 static ssize_t extract_entropy(void *buf, size_t nbytes, int min)
1411 trace_extract_entropy(nbytes, POOL_ENTROPY_BITS(), _RET_IP_);
1412 nbytes = account(nbytes, min);
1413 return _extract_entropy(buf, nbytes);
1416 #define warn_unseeded_randomness(previous) \
1417 _warn_unseeded_randomness(__func__, (void *)_RET_IP_, (previous))
1419 static void _warn_unseeded_randomness(const char *func_name, void *caller, void **previous)
1421 #ifdef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1422 const bool print_once = false;
1424 static bool print_once __read_mostly;
1427 if (print_once || crng_ready() ||
1428 (previous && (caller == READ_ONCE(*previous))))
1430 WRITE_ONCE(*previous, caller);
1431 #ifndef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1434 if (__ratelimit(&unseeded_warning))
1435 printk_deferred(KERN_NOTICE "random: %s called from %pS with crng_init=%d\n",
1436 func_name, caller, crng_init);
1440 * This function is the exported kernel interface. It returns some
1441 * number of good random numbers, suitable for key generation, seeding
1442 * TCP sequence numbers, etc. It does not rely on the hardware random
1443 * number generator. For random bytes direct from the hardware RNG
1444 * (when available), use get_random_bytes_arch(). In order to ensure
1445 * that the randomness provided by this function is okay, the function
1446 * wait_for_random_bytes() should be called and return 0 at least once
1447 * at any point prior.
1449 static void _get_random_bytes(void *buf, int nbytes)
1451 u8 tmp[CHACHA_BLOCK_SIZE] __aligned(4);
1453 trace_get_random_bytes(nbytes, _RET_IP_);
1455 while (nbytes >= CHACHA_BLOCK_SIZE) {
1457 buf += CHACHA_BLOCK_SIZE;
1458 nbytes -= CHACHA_BLOCK_SIZE;
1463 memcpy(buf, tmp, nbytes);
1464 crng_backtrack_protect(tmp, nbytes);
1466 crng_backtrack_protect(tmp, CHACHA_BLOCK_SIZE);
1467 memzero_explicit(tmp, sizeof(tmp));
1470 void get_random_bytes(void *buf, int nbytes)
1472 static void *previous;
1474 warn_unseeded_randomness(&previous);
1475 _get_random_bytes(buf, nbytes);
1477 EXPORT_SYMBOL(get_random_bytes);
1480 * Each time the timer fires, we expect that we got an unpredictable
1481 * jump in the cycle counter. Even if the timer is running on another
1482 * CPU, the timer activity will be touching the stack of the CPU that is
1483 * generating entropy..
1485 * Note that we don't re-arm the timer in the timer itself - we are
1486 * happy to be scheduled away, since that just makes the load more
1487 * complex, but we do not want the timer to keep ticking unless the
1488 * entropy loop is running.
1490 * So the re-arming always happens in the entropy loop itself.
1492 static void entropy_timer(struct timer_list *t)
1494 credit_entropy_bits(1);
1498 * If we have an actual cycle counter, see if we can
1499 * generate enough entropy with timing noise
1501 static void try_to_generate_entropy(void)
1505 struct timer_list timer;
1508 stack.now = random_get_entropy();
1510 /* Slow counter - or none. Don't even bother */
1511 if (stack.now == random_get_entropy())
1514 timer_setup_on_stack(&stack.timer, entropy_timer, 0);
1515 while (!crng_ready()) {
1516 if (!timer_pending(&stack.timer))
1517 mod_timer(&stack.timer, jiffies + 1);
1518 mix_pool_bytes(&stack.now, sizeof(stack.now));
1520 stack.now = random_get_entropy();
1523 del_timer_sync(&stack.timer);
1524 destroy_timer_on_stack(&stack.timer);
1525 mix_pool_bytes(&stack.now, sizeof(stack.now));
1529 * Wait for the urandom pool to be seeded and thus guaranteed to supply
1530 * cryptographically secure random numbers. This applies to: the /dev/urandom
1531 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
1532 * family of functions. Using any of these functions without first calling
1533 * this function forfeits the guarantee of security.
1535 * Returns: 0 if the urandom pool has been seeded.
1536 * -ERESTARTSYS if the function was interrupted by a signal.
1538 int wait_for_random_bytes(void)
1540 if (likely(crng_ready()))
1545 ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ);
1547 return ret > 0 ? 0 : ret;
1549 try_to_generate_entropy();
1550 } while (!crng_ready());
1554 EXPORT_SYMBOL(wait_for_random_bytes);
1557 * Returns whether or not the urandom pool has been seeded and thus guaranteed
1558 * to supply cryptographically secure random numbers. This applies to: the
1559 * /dev/urandom device, the get_random_bytes function, and the get_random_{u32,
1560 * ,u64,int,long} family of functions.
1562 * Returns: true if the urandom pool has been seeded.
1563 * false if the urandom pool has not been seeded.
1565 bool rng_is_initialized(void)
1567 return crng_ready();
1569 EXPORT_SYMBOL(rng_is_initialized);
1572 * Add a callback function that will be invoked when the nonblocking
1573 * pool is initialised.
1575 * returns: 0 if callback is successfully added
1576 * -EALREADY if pool is already initialised (callback not called)
1577 * -ENOENT if module for callback is not alive
1579 int add_random_ready_callback(struct random_ready_callback *rdy)
1581 struct module *owner;
1582 unsigned long flags;
1583 int err = -EALREADY;
1589 if (!try_module_get(owner))
1592 spin_lock_irqsave(&random_ready_list_lock, flags);
1598 list_add(&rdy->list, &random_ready_list);
1602 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1608 EXPORT_SYMBOL(add_random_ready_callback);
1611 * Delete a previously registered readiness callback function.
1613 void del_random_ready_callback(struct random_ready_callback *rdy)
1615 unsigned long flags;
1616 struct module *owner = NULL;
1618 spin_lock_irqsave(&random_ready_list_lock, flags);
1619 if (!list_empty(&rdy->list)) {
1620 list_del_init(&rdy->list);
1623 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1627 EXPORT_SYMBOL(del_random_ready_callback);
1630 * This function will use the architecture-specific hardware random
1631 * number generator if it is available. The arch-specific hw RNG will
1632 * almost certainly be faster than what we can do in software, but it
1633 * is impossible to verify that it is implemented securely (as
1634 * opposed, to, say, the AES encryption of a sequence number using a
1635 * key known by the NSA). So it's useful if we need the speed, but
1636 * only if we're willing to trust the hardware manufacturer not to
1637 * have put in a back door.
1639 * Return number of bytes filled in.
1641 int __must_check get_random_bytes_arch(void *buf, int nbytes)
1646 trace_get_random_bytes_arch(left, _RET_IP_);
1649 int chunk = min_t(int, left, sizeof(unsigned long));
1651 if (!arch_get_random_long(&v))
1654 memcpy(p, &v, chunk);
1659 return nbytes - left;
1661 EXPORT_SYMBOL(get_random_bytes_arch);
1664 * init_std_data - initialize pool with system data
1666 * This function clears the pool's entropy count and mixes some system
1667 * data into the pool to prepare it for use. The pool is not cleared
1668 * as that can only decrease the entropy in the pool.
1670 static void __init init_std_data(void)
1673 ktime_t now = ktime_get_real();
1676 mix_pool_bytes(&now, sizeof(now));
1677 for (i = POOL_BYTES; i > 0; i -= sizeof(rv)) {
1678 if (!arch_get_random_seed_long(&rv) &&
1679 !arch_get_random_long(&rv))
1680 rv = random_get_entropy();
1681 mix_pool_bytes(&rv, sizeof(rv));
1683 mix_pool_bytes(utsname(), sizeof(*(utsname())));
1687 * Note that setup_arch() may call add_device_randomness()
1688 * long before we get here. This allows seeding of the pools
1689 * with some platform dependent data very early in the boot
1690 * process. But it limits our options here. We must use
1691 * statically allocated structures that already have all
1692 * initializations complete at compile time. We should also
1693 * take care not to overwrite the precious per platform data
1696 int __init rand_initialize(void)
1699 if (crng_need_final_init)
1700 crng_finalize_init(&primary_crng);
1701 crng_initialize_primary(&primary_crng);
1702 crng_global_init_time = jiffies;
1703 if (ratelimit_disable) {
1704 urandom_warning.interval = 0;
1705 unseeded_warning.interval = 0;
1711 void rand_initialize_disk(struct gendisk *disk)
1713 struct timer_rand_state *state;
1716 * If kzalloc returns null, we just won't use that entropy
1719 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1721 state->last_time = INITIAL_JIFFIES;
1722 disk->random = state;
1727 static ssize_t urandom_read_nowarn(struct file *file, char __user *buf,
1728 size_t nbytes, loff_t *ppos)
1732 nbytes = min_t(size_t, nbytes, INT_MAX >> (POOL_ENTROPY_SHIFT + 3));
1733 ret = extract_crng_user(buf, nbytes);
1734 trace_urandom_read(8 * nbytes, 0, POOL_ENTROPY_BITS());
1738 static ssize_t urandom_read(struct file *file, char __user *buf, size_t nbytes,
1741 static int maxwarn = 10;
1743 if (!crng_ready() && maxwarn > 0) {
1745 if (__ratelimit(&urandom_warning))
1746 pr_notice("%s: uninitialized urandom read (%zd bytes read)\n",
1747 current->comm, nbytes);
1750 return urandom_read_nowarn(file, buf, nbytes, ppos);
1753 static ssize_t random_read(struct file *file, char __user *buf, size_t nbytes,
1758 ret = wait_for_random_bytes();
1761 return urandom_read_nowarn(file, buf, nbytes, ppos);
1764 static __poll_t random_poll(struct file *file, poll_table *wait)
1768 poll_wait(file, &crng_init_wait, wait);
1769 poll_wait(file, &random_write_wait, wait);
1772 mask |= EPOLLIN | EPOLLRDNORM;
1773 if (POOL_ENTROPY_BITS() < random_write_wakeup_bits)
1774 mask |= EPOLLOUT | EPOLLWRNORM;
1778 static int write_pool(const char __user *buffer, size_t count)
1782 const char __user *p = buffer;
1787 bytes = min(count, sizeof(buf));
1788 if (copy_from_user(&buf, p, bytes))
1791 for (b = bytes; b > 0; b -= sizeof(u32), i++) {
1792 if (!arch_get_random_int(&t))
1800 mix_pool_bytes(buf, bytes);
1807 static ssize_t random_write(struct file *file, const char __user *buffer,
1808 size_t count, loff_t *ppos)
1812 ret = write_pool(buffer, count);
1816 return (ssize_t)count;
1819 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1821 int size, ent_count;
1822 int __user *p = (int __user *)arg;
1827 /* inherently racy, no point locking */
1828 ent_count = POOL_ENTROPY_BITS();
1829 if (put_user(ent_count, p))
1832 case RNDADDTOENTCNT:
1833 if (!capable(CAP_SYS_ADMIN))
1835 if (get_user(ent_count, p))
1837 return credit_entropy_bits_safe(ent_count);
1839 if (!capable(CAP_SYS_ADMIN))
1841 if (get_user(ent_count, p++))
1845 if (get_user(size, p++))
1847 retval = write_pool((const char __user *)p, size);
1850 return credit_entropy_bits_safe(ent_count);
1854 * Clear the entropy pool counters. We no longer clear
1855 * the entropy pool, as that's silly.
1857 if (!capable(CAP_SYS_ADMIN))
1859 input_pool.entropy_count = 0;
1862 if (!capable(CAP_SYS_ADMIN))
1866 crng_reseed(&primary_crng, true);
1867 WRITE_ONCE(crng_global_init_time, jiffies - 1);
1874 static int random_fasync(int fd, struct file *filp, int on)
1876 return fasync_helper(fd, filp, on, &fasync);
1879 const struct file_operations random_fops = {
1880 .read = random_read,
1881 .write = random_write,
1882 .poll = random_poll,
1883 .unlocked_ioctl = random_ioctl,
1884 .compat_ioctl = compat_ptr_ioctl,
1885 .fasync = random_fasync,
1886 .llseek = noop_llseek,
1889 const struct file_operations urandom_fops = {
1890 .read = urandom_read,
1891 .write = random_write,
1892 .unlocked_ioctl = random_ioctl,
1893 .compat_ioctl = compat_ptr_ioctl,
1894 .fasync = random_fasync,
1895 .llseek = noop_llseek,
1898 SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count, unsigned int,
1903 if (flags & ~(GRND_NONBLOCK | GRND_RANDOM | GRND_INSECURE))
1907 * Requesting insecure and blocking randomness at the same time makes
1910 if ((flags & (GRND_INSECURE | GRND_RANDOM)) == (GRND_INSECURE | GRND_RANDOM))
1913 if (count > INT_MAX)
1916 if (!(flags & GRND_INSECURE) && !crng_ready()) {
1917 if (flags & GRND_NONBLOCK)
1919 ret = wait_for_random_bytes();
1923 return urandom_read_nowarn(NULL, buf, count, NULL);
1926 /********************************************************************
1930 ********************************************************************/
1932 #ifdef CONFIG_SYSCTL
1934 #include <linux/sysctl.h>
1936 static int min_write_thresh;
1937 static int max_write_thresh = POOL_BITS;
1938 static int random_min_urandom_seed = 60;
1939 static char sysctl_bootid[16];
1942 * This function is used to return both the bootid UUID, and random
1943 * UUID. The difference is in whether table->data is NULL; if it is,
1944 * then a new UUID is generated and returned to the user.
1946 * If the user accesses this via the proc interface, the UUID will be
1947 * returned as an ASCII string in the standard UUID format; if via the
1948 * sysctl system call, as 16 bytes of binary data.
1950 static int proc_do_uuid(struct ctl_table *table, int write, void *buffer,
1951 size_t *lenp, loff_t *ppos)
1953 struct ctl_table fake_table;
1954 unsigned char buf[64], tmp_uuid[16], *uuid;
1959 generate_random_uuid(uuid);
1961 static DEFINE_SPINLOCK(bootid_spinlock);
1963 spin_lock(&bootid_spinlock);
1965 generate_random_uuid(uuid);
1966 spin_unlock(&bootid_spinlock);
1969 sprintf(buf, "%pU", uuid);
1971 fake_table.data = buf;
1972 fake_table.maxlen = sizeof(buf);
1974 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
1978 * Return entropy available scaled to integral bits
1980 static int proc_do_entropy(struct ctl_table *table, int write, void *buffer,
1981 size_t *lenp, loff_t *ppos)
1983 struct ctl_table fake_table;
1986 entropy_count = *(int *)table->data >> POOL_ENTROPY_SHIFT;
1988 fake_table.data = &entropy_count;
1989 fake_table.maxlen = sizeof(entropy_count);
1991 return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
1994 static int sysctl_poolsize = POOL_BITS;
1995 static struct ctl_table random_table[] = {
1997 .procname = "poolsize",
1998 .data = &sysctl_poolsize,
1999 .maxlen = sizeof(int),
2001 .proc_handler = proc_dointvec,
2004 .procname = "entropy_avail",
2005 .maxlen = sizeof(int),
2007 .proc_handler = proc_do_entropy,
2008 .data = &input_pool.entropy_count,
2011 .procname = "write_wakeup_threshold",
2012 .data = &random_write_wakeup_bits,
2013 .maxlen = sizeof(int),
2015 .proc_handler = proc_dointvec_minmax,
2016 .extra1 = &min_write_thresh,
2017 .extra2 = &max_write_thresh,
2020 .procname = "urandom_min_reseed_secs",
2021 .data = &random_min_urandom_seed,
2022 .maxlen = sizeof(int),
2024 .proc_handler = proc_dointvec,
2027 .procname = "boot_id",
2028 .data = &sysctl_bootid,
2031 .proc_handler = proc_do_uuid,
2037 .proc_handler = proc_do_uuid,
2039 #ifdef ADD_INTERRUPT_BENCH
2041 .procname = "add_interrupt_avg_cycles",
2042 .data = &avg_cycles,
2043 .maxlen = sizeof(avg_cycles),
2045 .proc_handler = proc_doulongvec_minmax,
2048 .procname = "add_interrupt_avg_deviation",
2049 .data = &avg_deviation,
2050 .maxlen = sizeof(avg_deviation),
2052 .proc_handler = proc_doulongvec_minmax,
2059 * rand_initialize() is called before sysctl_init(),
2060 * so we cannot call register_sysctl_init() in rand_initialize()
2062 static int __init random_sysctls_init(void)
2064 register_sysctl_init("kernel/random", random_table);
2067 device_initcall(random_sysctls_init);
2068 #endif /* CONFIG_SYSCTL */
2070 struct batched_entropy {
2072 u64 entropy_u64[CHACHA_BLOCK_SIZE / sizeof(u64)];
2073 u32 entropy_u32[CHACHA_BLOCK_SIZE / sizeof(u32)];
2075 unsigned int position;
2076 spinlock_t batch_lock;
2080 * Get a random word for internal kernel use only. The quality of the random
2081 * number is good as /dev/urandom, but there is no backtrack protection, with
2082 * the goal of being quite fast and not depleting entropy. In order to ensure
2083 * that the randomness provided by this function is okay, the function
2084 * wait_for_random_bytes() should be called and return 0 at least once at any
2087 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64) = {
2088 .batch_lock = __SPIN_LOCK_UNLOCKED(batched_entropy_u64.lock),
2091 u64 get_random_u64(void)
2094 unsigned long flags;
2095 struct batched_entropy *batch;
2096 static void *previous;
2098 warn_unseeded_randomness(&previous);
2100 batch = raw_cpu_ptr(&batched_entropy_u64);
2101 spin_lock_irqsave(&batch->batch_lock, flags);
2102 if (batch->position % ARRAY_SIZE(batch->entropy_u64) == 0) {
2103 extract_crng((u8 *)batch->entropy_u64);
2104 batch->position = 0;
2106 ret = batch->entropy_u64[batch->position++];
2107 spin_unlock_irqrestore(&batch->batch_lock, flags);
2110 EXPORT_SYMBOL(get_random_u64);
2112 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32) = {
2113 .batch_lock = __SPIN_LOCK_UNLOCKED(batched_entropy_u32.lock),
2115 u32 get_random_u32(void)
2118 unsigned long flags;
2119 struct batched_entropy *batch;
2120 static void *previous;
2122 warn_unseeded_randomness(&previous);
2124 batch = raw_cpu_ptr(&batched_entropy_u32);
2125 spin_lock_irqsave(&batch->batch_lock, flags);
2126 if (batch->position % ARRAY_SIZE(batch->entropy_u32) == 0) {
2127 extract_crng((u8 *)batch->entropy_u32);
2128 batch->position = 0;
2130 ret = batch->entropy_u32[batch->position++];
2131 spin_unlock_irqrestore(&batch->batch_lock, flags);
2134 EXPORT_SYMBOL(get_random_u32);
2136 /* It's important to invalidate all potential batched entropy that might
2137 * be stored before the crng is initialized, which we can do lazily by
2138 * simply resetting the counter to zero so that it's re-extracted on the
2140 static void invalidate_batched_entropy(void)
2143 unsigned long flags;
2145 for_each_possible_cpu(cpu) {
2146 struct batched_entropy *batched_entropy;
2148 batched_entropy = per_cpu_ptr(&batched_entropy_u32, cpu);
2149 spin_lock_irqsave(&batched_entropy->batch_lock, flags);
2150 batched_entropy->position = 0;
2151 spin_unlock(&batched_entropy->batch_lock);
2153 batched_entropy = per_cpu_ptr(&batched_entropy_u64, cpu);
2154 spin_lock(&batched_entropy->batch_lock);
2155 batched_entropy->position = 0;
2156 spin_unlock_irqrestore(&batched_entropy->batch_lock, flags);
2161 * randomize_page - Generate a random, page aligned address
2162 * @start: The smallest acceptable address the caller will take.
2163 * @range: The size of the area, starting at @start, within which the
2164 * random address must fall.
2166 * If @start + @range would overflow, @range is capped.
2168 * NOTE: Historical use of randomize_range, which this replaces, presumed that
2169 * @start was already page aligned. We now align it regardless.
2171 * Return: A page aligned address within [start, start + range). On error,
2172 * @start is returned.
2174 unsigned long randomize_page(unsigned long start, unsigned long range)
2176 if (!PAGE_ALIGNED(start)) {
2177 range -= PAGE_ALIGN(start) - start;
2178 start = PAGE_ALIGN(start);
2181 if (start > ULONG_MAX - range)
2182 range = ULONG_MAX - start;
2184 range >>= PAGE_SHIFT;
2189 return start + (get_random_long() % range << PAGE_SHIFT);
2192 /* Interface for in-kernel drivers of true hardware RNGs.
2193 * Those devices may produce endless random bits and will be throttled
2194 * when our pool is full.
2196 void add_hwgenerator_randomness(const char *buffer, size_t count,
2199 if (unlikely(crng_init == 0)) {
2200 size_t ret = crng_fast_load(buffer, count);
2201 mix_pool_bytes(buffer, ret);
2204 if (!count || crng_init == 0)
2208 /* Suspend writing if we're above the trickle threshold.
2209 * We'll be woken up again once below random_write_wakeup_thresh,
2210 * or when the calling thread is about to terminate.
2212 wait_event_interruptible(random_write_wait,
2213 !system_wq || kthread_should_stop() ||
2214 POOL_ENTROPY_BITS() <= random_write_wakeup_bits);
2215 mix_pool_bytes(buffer, count);
2216 credit_entropy_bits(entropy);
2218 EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
2220 /* Handle random seed passed by bootloader.
2221 * If the seed is trustworthy, it would be regarded as hardware RNGs. Otherwise
2222 * it would be regarded as device data.
2223 * The decision is controlled by CONFIG_RANDOM_TRUST_BOOTLOADER.
2225 void add_bootloader_randomness(const void *buf, unsigned int size)
2227 if (IS_ENABLED(CONFIG_RANDOM_TRUST_BOOTLOADER))
2228 add_hwgenerator_randomness(buf, size, size * 8);
2230 add_device_randomness(buf, size);
2232 EXPORT_SYMBOL_GPL(add_bootloader_randomness);