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 or 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/fips.h>
341 #include <linux/ptrace.h>
342 #include <linux/workqueue.h>
343 #include <linux/irq.h>
344 #include <linux/ratelimit.h>
345 #include <linux/syscalls.h>
346 #include <linux/completion.h>
347 #include <linux/uuid.h>
348 #include <crypto/chacha.h>
349 #include <crypto/blake2s.h>
351 #include <asm/processor.h>
352 #include <linux/uaccess.h>
354 #include <asm/irq_regs.h>
357 #define CREATE_TRACE_POINTS
358 #include <trace/events/random.h>
360 /* #define ADD_INTERRUPT_BENCH */
363 * Configuration information
365 #define INPUT_POOL_SHIFT 12
366 #define INPUT_POOL_WORDS (1 << (INPUT_POOL_SHIFT-5))
367 #define OUTPUT_POOL_SHIFT 10
368 #define OUTPUT_POOL_WORDS (1 << (OUTPUT_POOL_SHIFT-5))
369 #define EXTRACT_SIZE (BLAKE2S_HASH_SIZE / 2)
372 * To allow fractional bits to be tracked, the entropy_count field is
373 * denominated in units of 1/8th bits.
375 * 2*(ENTROPY_SHIFT + poolbitshift) must <= 31, or the multiply in
376 * credit_entropy_bits() needs to be 64 bits wide.
378 #define ENTROPY_SHIFT 3
379 #define ENTROPY_BITS(r) ((r)->entropy_count >> ENTROPY_SHIFT)
382 * If the entropy count falls under this number of bits, then we
383 * should wake up processes which are selecting or polling on write
384 * access to /dev/random.
386 static int random_write_wakeup_bits = 28 * OUTPUT_POOL_WORDS;
389 * Originally, we used a primitive polynomial of degree .poolwords
390 * over GF(2). The taps for various sizes are defined below. They
391 * were chosen to be evenly spaced except for the last tap, which is 1
392 * to get the twisting happening as fast as possible.
394 * For the purposes of better mixing, we use the CRC-32 polynomial as
395 * well to make a (modified) twisted Generalized Feedback Shift
396 * Register. (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR
397 * generators. ACM Transactions on Modeling and Computer Simulation
398 * 2(3):179-194. Also see M. Matsumoto & Y. Kurita, 1994. Twisted
399 * GFSR generators II. ACM Transactions on Modeling and Computer
400 * Simulation 4:254-266)
402 * Thanks to Colin Plumb for suggesting this.
404 * The mixing operation is much less sensitive than the output hash,
405 * where we use BLAKE2s. All that we want of mixing operation is that
406 * it be a good non-cryptographic hash; i.e. it not produce collisions
407 * when fed "random" data of the sort we expect to see. As long as
408 * the pool state differs for different inputs, we have preserved the
409 * input entropy and done a good job. The fact that an intelligent
410 * attacker can construct inputs that will produce controlled
411 * alterations to the pool's state is not important because we don't
412 * consider such inputs to contribute any randomness. The only
413 * property we need with respect to them is that the attacker can't
414 * increase his/her knowledge of the pool's state. Since all
415 * additions are reversible (knowing the final state and the input,
416 * you can reconstruct the initial state), if an attacker has any
417 * uncertainty about the initial state, he/she can only shuffle that
418 * uncertainty about, but never cause any collisions (which would
419 * decrease the uncertainty).
421 * Our mixing functions were analyzed by Lacharme, Roeck, Strubel, and
422 * Videau in their paper, "The Linux Pseudorandom Number Generator
423 * Revisited" (see: http://eprint.iacr.org/2012/251.pdf). In their
424 * paper, they point out that we are not using a true Twisted GFSR,
425 * since Matsumoto & Kurita used a trinomial feedback polynomial (that
426 * is, with only three taps, instead of the six that we are using).
427 * As a result, the resulting polynomial is neither primitive nor
428 * irreducible, and hence does not have a maximal period over
429 * GF(2**32). They suggest a slight change to the generator
430 * polynomial which improves the resulting TGFSR polynomial to be
431 * irreducible, which we have made here.
433 static const struct poolinfo {
434 int poolbitshift, poolwords, poolbytes, poolfracbits;
435 #define S(x) ilog2(x)+5, (x), (x)*4, (x) << (ENTROPY_SHIFT+5)
436 int tap1, tap2, tap3, tap4, tap5;
437 } poolinfo_table[] = {
438 /* was: x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 */
439 /* x^128 + x^104 + x^76 + x^51 +x^25 + x + 1 */
440 { S(128), 104, 76, 51, 25, 1 },
444 * Static global variables
446 static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
447 static struct fasync_struct *fasync;
449 static DEFINE_SPINLOCK(random_ready_list_lock);
450 static LIST_HEAD(random_ready_list);
454 unsigned long init_time;
458 static struct crng_state primary_crng = {
459 .lock = __SPIN_LOCK_UNLOCKED(primary_crng.lock),
460 .state[0] = CHACHA_CONSTANT_EXPA,
461 .state[1] = CHACHA_CONSTANT_ND_3,
462 .state[2] = CHACHA_CONSTANT_2_BY,
463 .state[3] = CHACHA_CONSTANT_TE_K,
467 * crng_init = 0 --> Uninitialized
469 * 2 --> Initialized from input_pool
471 * crng_init is protected by primary_crng->lock, and only increases
472 * its value (from 0->1->2).
474 static int crng_init = 0;
475 static bool crng_need_final_init = false;
476 #define crng_ready() (likely(crng_init > 1))
477 static int crng_init_cnt = 0;
478 static unsigned long crng_global_init_time = 0;
479 #define CRNG_INIT_CNT_THRESH (2*CHACHA_KEY_SIZE)
480 static void _extract_crng(struct crng_state *crng, __u8 out[CHACHA_BLOCK_SIZE]);
481 static void _crng_backtrack_protect(struct crng_state *crng,
482 __u8 tmp[CHACHA_BLOCK_SIZE], int used);
483 static void process_random_ready_list(void);
484 static void _get_random_bytes(void *buf, int nbytes);
486 static struct ratelimit_state unseeded_warning =
487 RATELIMIT_STATE_INIT("warn_unseeded_randomness", HZ, 3);
488 static struct ratelimit_state urandom_warning =
489 RATELIMIT_STATE_INIT("warn_urandom_randomness", HZ, 3);
491 static int ratelimit_disable __read_mostly;
493 module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
494 MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
496 /**********************************************************************
498 * OS independent entropy store. Here are the functions which handle
499 * storing entropy in an entropy pool.
501 **********************************************************************/
503 struct entropy_store;
504 struct entropy_store {
505 /* read-only data: */
506 const struct poolinfo *poolinfo;
510 /* read-write data: */
512 unsigned short add_ptr;
513 unsigned short input_rotate;
515 unsigned int last_data_init:1;
516 __u8 last_data[EXTRACT_SIZE];
519 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
520 size_t nbytes, int min, int rsvd);
521 static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
522 size_t nbytes, int fips);
524 static void crng_reseed(struct crng_state *crng, struct entropy_store *r);
525 static __u32 input_pool_data[INPUT_POOL_WORDS] __latent_entropy;
527 static struct entropy_store input_pool = {
528 .poolinfo = &poolinfo_table[0],
530 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
531 .pool = input_pool_data
534 static __u32 const twist_table[8] = {
535 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
536 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
539 * This function adds bytes into the entropy "pool". It does not
540 * update the entropy estimate. The caller should call
541 * credit_entropy_bits if this is appropriate.
543 * The pool is stirred with a primitive polynomial of the appropriate
544 * degree, and then twisted. We twist by three bits at a time because
545 * it's cheap to do so and helps slightly in the expected case where
546 * the entropy is concentrated in the low-order bits.
548 static void _mix_pool_bytes(struct entropy_store *r, const void *in,
551 unsigned long i, tap1, tap2, tap3, tap4, tap5;
553 int wordmask = r->poolinfo->poolwords - 1;
554 const unsigned char *bytes = in;
557 tap1 = r->poolinfo->tap1;
558 tap2 = r->poolinfo->tap2;
559 tap3 = r->poolinfo->tap3;
560 tap4 = r->poolinfo->tap4;
561 tap5 = r->poolinfo->tap5;
563 input_rotate = r->input_rotate;
566 /* mix one byte at a time to simplify size handling and churn faster */
568 w = rol32(*bytes++, input_rotate);
569 i = (i - 1) & wordmask;
571 /* XOR in the various taps */
573 w ^= r->pool[(i + tap1) & wordmask];
574 w ^= r->pool[(i + tap2) & wordmask];
575 w ^= r->pool[(i + tap3) & wordmask];
576 w ^= r->pool[(i + tap4) & wordmask];
577 w ^= r->pool[(i + tap5) & wordmask];
579 /* Mix the result back in with a twist */
580 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
583 * Normally, we add 7 bits of rotation to the pool.
584 * At the beginning of the pool, add an extra 7 bits
585 * rotation, so that successive passes spread the
586 * input bits across the pool evenly.
588 input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
591 r->input_rotate = input_rotate;
595 static void __mix_pool_bytes(struct entropy_store *r, const void *in,
598 trace_mix_pool_bytes_nolock(r->name, nbytes, _RET_IP_);
599 _mix_pool_bytes(r, in, nbytes);
602 static void mix_pool_bytes(struct entropy_store *r, const void *in,
607 trace_mix_pool_bytes(r->name, nbytes, _RET_IP_);
608 spin_lock_irqsave(&r->lock, flags);
609 _mix_pool_bytes(r, in, nbytes);
610 spin_unlock_irqrestore(&r->lock, flags);
616 unsigned short reg_idx;
621 * This is a fast mixing routine used by the interrupt randomness
622 * collector. It's hardcoded for an 128 bit pool and assumes that any
623 * locks that might be needed are taken by the caller.
625 static void fast_mix(struct fast_pool *f)
627 __u32 a = f->pool[0], b = f->pool[1];
628 __u32 c = f->pool[2], d = f->pool[3];
631 b = rol32(b, 6); d = rol32(d, 27);
635 b = rol32(b, 16); d = rol32(d, 14);
639 b = rol32(b, 6); d = rol32(d, 27);
643 b = rol32(b, 16); d = rol32(d, 14);
646 f->pool[0] = a; f->pool[1] = b;
647 f->pool[2] = c; f->pool[3] = d;
651 static void process_random_ready_list(void)
654 struct random_ready_callback *rdy, *tmp;
656 spin_lock_irqsave(&random_ready_list_lock, flags);
657 list_for_each_entry_safe(rdy, tmp, &random_ready_list, list) {
658 struct module *owner = rdy->owner;
660 list_del_init(&rdy->list);
664 spin_unlock_irqrestore(&random_ready_list_lock, flags);
668 * Credit (or debit) the entropy store with n bits of entropy.
669 * Use credit_entropy_bits_safe() if the value comes from userspace
670 * or otherwise should be checked for extreme values.
672 static void credit_entropy_bits(struct entropy_store *r, int nbits)
674 int entropy_count, orig;
675 const int pool_size = r->poolinfo->poolfracbits;
676 int nfrac = nbits << ENTROPY_SHIFT;
682 entropy_count = orig = READ_ONCE(r->entropy_count);
685 entropy_count += nfrac;
688 * Credit: we have to account for the possibility of
689 * overwriting already present entropy. Even in the
690 * ideal case of pure Shannon entropy, new contributions
691 * approach the full value asymptotically:
693 * entropy <- entropy + (pool_size - entropy) *
694 * (1 - exp(-add_entropy/pool_size))
696 * For add_entropy <= pool_size/2 then
697 * (1 - exp(-add_entropy/pool_size)) >=
698 * (add_entropy/pool_size)*0.7869...
699 * so we can approximate the exponential with
700 * 3/4*add_entropy/pool_size and still be on the
701 * safe side by adding at most pool_size/2 at a time.
703 * The use of pool_size-2 in the while statement is to
704 * prevent rounding artifacts from making the loop
705 * arbitrarily long; this limits the loop to log2(pool_size)*2
706 * turns no matter how large nbits is.
709 const int s = r->poolinfo->poolbitshift + ENTROPY_SHIFT + 2;
710 /* The +2 corresponds to the /4 in the denominator */
713 unsigned int anfrac = min(pnfrac, pool_size/2);
715 ((pool_size - entropy_count)*anfrac*3) >> s;
717 entropy_count += add;
719 } while (unlikely(entropy_count < pool_size-2 && pnfrac));
722 if (WARN_ON(entropy_count < 0)) {
723 pr_warn("negative entropy/overflow: pool %s count %d\n",
724 r->name, entropy_count);
726 } else if (entropy_count > pool_size)
727 entropy_count = pool_size;
728 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
731 trace_credit_entropy_bits(r->name, nbits,
732 entropy_count >> ENTROPY_SHIFT, _RET_IP_);
734 if (r == &input_pool) {
735 int entropy_bits = entropy_count >> ENTROPY_SHIFT;
737 if (crng_init < 2 && entropy_bits >= 128)
738 crng_reseed(&primary_crng, r);
742 static int credit_entropy_bits_safe(struct entropy_store *r, int nbits)
744 const int nbits_max = r->poolinfo->poolwords * 32;
749 /* Cap the value to avoid overflows */
750 nbits = min(nbits, nbits_max);
752 credit_entropy_bits(r, nbits);
756 /*********************************************************************
758 * CRNG using CHACHA20
760 *********************************************************************/
762 #define CRNG_RESEED_INTERVAL (300*HZ)
764 static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
767 * Hack to deal with crazy userspace progams when they are all trying
768 * to access /dev/urandom in parallel. The programs are almost
769 * certainly doing something terribly wrong, but we'll work around
770 * their brain damage.
772 static struct crng_state **crng_node_pool __read_mostly;
774 static void invalidate_batched_entropy(void);
775 static void numa_crng_init(void);
777 static bool trust_cpu __ro_after_init = IS_ENABLED(CONFIG_RANDOM_TRUST_CPU);
778 static int __init parse_trust_cpu(char *arg)
780 return kstrtobool(arg, &trust_cpu);
782 early_param("random.trust_cpu", parse_trust_cpu);
784 static bool crng_init_try_arch(struct crng_state *crng)
787 bool arch_init = true;
790 for (i = 4; i < 16; i++) {
791 if (!arch_get_random_seed_long(&rv) &&
792 !arch_get_random_long(&rv)) {
793 rv = random_get_entropy();
796 crng->state[i] ^= rv;
802 static bool __init crng_init_try_arch_early(struct crng_state *crng)
805 bool arch_init = true;
808 for (i = 4; i < 16; i++) {
809 if (!arch_get_random_seed_long_early(&rv) &&
810 !arch_get_random_long_early(&rv)) {
811 rv = random_get_entropy();
814 crng->state[i] ^= rv;
820 static void crng_initialize_secondary(struct crng_state *crng)
822 chacha_init_consts(crng->state);
823 _get_random_bytes(&crng->state[4], sizeof(__u32) * 12);
824 crng_init_try_arch(crng);
825 crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
828 static void __init crng_initialize_primary(struct crng_state *crng)
830 _extract_entropy(&input_pool, &crng->state[4], sizeof(__u32) * 12, 0);
831 if (crng_init_try_arch_early(crng) && trust_cpu && crng_init < 2) {
832 invalidate_batched_entropy();
835 pr_notice("crng init done (trusting CPU's manufacturer)\n");
837 crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
840 static void crng_finalize_init(struct crng_state *crng)
842 if (crng != &primary_crng || crng_init >= 2)
845 /* We can't call numa_crng_init until we have workqueues,
846 * so mark this for processing later. */
847 crng_need_final_init = true;
851 invalidate_batched_entropy();
854 process_random_ready_list();
855 wake_up_interruptible(&crng_init_wait);
856 kill_fasync(&fasync, SIGIO, POLL_IN);
857 pr_notice("crng init done\n");
858 if (unseeded_warning.missed) {
859 pr_notice("%d get_random_xx warning(s) missed due to ratelimiting\n",
860 unseeded_warning.missed);
861 unseeded_warning.missed = 0;
863 if (urandom_warning.missed) {
864 pr_notice("%d urandom warning(s) missed due to ratelimiting\n",
865 urandom_warning.missed);
866 urandom_warning.missed = 0;
870 static void do_numa_crng_init(struct work_struct *work)
873 struct crng_state *crng;
874 struct crng_state **pool;
876 pool = kcalloc(nr_node_ids, sizeof(*pool), GFP_KERNEL|__GFP_NOFAIL);
877 for_each_online_node(i) {
878 crng = kmalloc_node(sizeof(struct crng_state),
879 GFP_KERNEL | __GFP_NOFAIL, i);
880 spin_lock_init(&crng->lock);
881 crng_initialize_secondary(crng);
884 /* pairs with READ_ONCE() in select_crng() */
885 if (cmpxchg_release(&crng_node_pool, NULL, pool) != NULL) {
892 static DECLARE_WORK(numa_crng_init_work, do_numa_crng_init);
894 static void numa_crng_init(void)
896 if (IS_ENABLED(CONFIG_NUMA))
897 schedule_work(&numa_crng_init_work);
900 static struct crng_state *select_crng(void)
902 if (IS_ENABLED(CONFIG_NUMA)) {
903 struct crng_state **pool;
904 int nid = numa_node_id();
906 /* pairs with cmpxchg_release() in do_numa_crng_init() */
907 pool = READ_ONCE(crng_node_pool);
908 if (pool && pool[nid])
912 return &primary_crng;
916 * crng_fast_load() can be called by code in the interrupt service
917 * path. So we can't afford to dilly-dally. Returns the number of
918 * bytes processed from cp.
920 static size_t crng_fast_load(const char *cp, size_t len)
926 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
928 if (crng_init != 0) {
929 spin_unlock_irqrestore(&primary_crng.lock, flags);
932 p = (unsigned char *) &primary_crng.state[4];
933 while (len > 0 && crng_init_cnt < CRNG_INIT_CNT_THRESH) {
934 p[crng_init_cnt % CHACHA_KEY_SIZE] ^= *cp;
935 cp++; crng_init_cnt++; len--; ret++;
937 spin_unlock_irqrestore(&primary_crng.lock, flags);
938 if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
939 invalidate_batched_entropy();
941 pr_notice("fast init done\n");
947 * crng_slow_load() is called by add_device_randomness, which has two
948 * attributes. (1) We can't trust the buffer passed to it is
949 * guaranteed to be unpredictable (so it might not have any entropy at
950 * all), and (2) it doesn't have the performance constraints of
953 * So we do something more comprehensive which is guaranteed to touch
954 * all of the primary_crng's state, and which uses a LFSR with a
955 * period of 255 as part of the mixing algorithm. Finally, we do
956 * *not* advance crng_init_cnt since buffer we may get may be something
957 * like a fixed DMI table (for example), which might very well be
958 * unique to the machine, but is otherwise unvarying.
960 static int crng_slow_load(const char *cp, size_t len)
963 static unsigned char lfsr = 1;
965 unsigned i, max = CHACHA_KEY_SIZE;
966 const char * src_buf = cp;
967 char * dest_buf = (char *) &primary_crng.state[4];
969 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
971 if (crng_init != 0) {
972 spin_unlock_irqrestore(&primary_crng.lock, flags);
978 for (i = 0; i < max ; i++) {
983 tmp = dest_buf[i % CHACHA_KEY_SIZE];
984 dest_buf[i % CHACHA_KEY_SIZE] ^= src_buf[i % len] ^ lfsr;
985 lfsr += (tmp << 3) | (tmp >> 5);
987 spin_unlock_irqrestore(&primary_crng.lock, flags);
991 static void crng_reseed(struct crng_state *crng, struct entropy_store *r)
996 __u8 block[CHACHA_BLOCK_SIZE];
1001 num = extract_entropy(r, &buf, 32, 16, 0);
1005 _extract_crng(&primary_crng, buf.block);
1006 _crng_backtrack_protect(&primary_crng, buf.block,
1009 spin_lock_irqsave(&crng->lock, flags);
1010 for (i = 0; i < 8; i++) {
1012 if (!arch_get_random_seed_long(&rv) &&
1013 !arch_get_random_long(&rv))
1014 rv = random_get_entropy();
1015 crng->state[i+4] ^= buf.key[i] ^ rv;
1017 memzero_explicit(&buf, sizeof(buf));
1018 WRITE_ONCE(crng->init_time, jiffies);
1019 spin_unlock_irqrestore(&crng->lock, flags);
1020 crng_finalize_init(crng);
1023 static void _extract_crng(struct crng_state *crng,
1024 __u8 out[CHACHA_BLOCK_SIZE])
1026 unsigned long v, flags, init_time;
1029 init_time = READ_ONCE(crng->init_time);
1030 if (time_after(READ_ONCE(crng_global_init_time), init_time) ||
1031 time_after(jiffies, init_time + CRNG_RESEED_INTERVAL))
1032 crng_reseed(crng, crng == &primary_crng ?
1033 &input_pool : NULL);
1035 spin_lock_irqsave(&crng->lock, flags);
1036 if (arch_get_random_long(&v))
1037 crng->state[14] ^= v;
1038 chacha20_block(&crng->state[0], out);
1039 if (crng->state[12] == 0)
1041 spin_unlock_irqrestore(&crng->lock, flags);
1044 static void extract_crng(__u8 out[CHACHA_BLOCK_SIZE])
1046 _extract_crng(select_crng(), out);
1050 * Use the leftover bytes from the CRNG block output (if there is
1051 * enough) to mutate the CRNG key to provide backtracking protection.
1053 static void _crng_backtrack_protect(struct crng_state *crng,
1054 __u8 tmp[CHACHA_BLOCK_SIZE], int used)
1056 unsigned long flags;
1060 used = round_up(used, sizeof(__u32));
1061 if (used + CHACHA_KEY_SIZE > CHACHA_BLOCK_SIZE) {
1065 spin_lock_irqsave(&crng->lock, flags);
1066 s = (__u32 *) &tmp[used];
1067 d = &crng->state[4];
1068 for (i=0; i < 8; i++)
1070 spin_unlock_irqrestore(&crng->lock, flags);
1073 static void crng_backtrack_protect(__u8 tmp[CHACHA_BLOCK_SIZE], int used)
1075 _crng_backtrack_protect(select_crng(), tmp, used);
1078 static ssize_t extract_crng_user(void __user *buf, size_t nbytes)
1080 ssize_t ret = 0, i = CHACHA_BLOCK_SIZE;
1081 __u8 tmp[CHACHA_BLOCK_SIZE] __aligned(4);
1082 int large_request = (nbytes > 256);
1085 if (large_request && need_resched()) {
1086 if (signal_pending(current)) {
1095 i = min_t(int, nbytes, CHACHA_BLOCK_SIZE);
1096 if (copy_to_user(buf, tmp, i)) {
1105 crng_backtrack_protect(tmp, i);
1107 /* Wipe data just written to memory */
1108 memzero_explicit(tmp, sizeof(tmp));
1114 /*********************************************************************
1116 * Entropy input management
1118 *********************************************************************/
1120 /* There is one of these per entropy source */
1121 struct timer_rand_state {
1123 long last_delta, last_delta2;
1126 #define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
1129 * Add device- or boot-specific data to the input pool to help
1132 * None of this adds any entropy; it is meant to avoid the problem of
1133 * the entropy pool having similar initial state across largely
1134 * identical devices.
1136 void add_device_randomness(const void *buf, unsigned int size)
1138 unsigned long time = random_get_entropy() ^ jiffies;
1139 unsigned long flags;
1141 if (!crng_ready() && size)
1142 crng_slow_load(buf, size);
1144 trace_add_device_randomness(size, _RET_IP_);
1145 spin_lock_irqsave(&input_pool.lock, flags);
1146 _mix_pool_bytes(&input_pool, buf, size);
1147 _mix_pool_bytes(&input_pool, &time, sizeof(time));
1148 spin_unlock_irqrestore(&input_pool.lock, flags);
1150 EXPORT_SYMBOL(add_device_randomness);
1152 static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
1155 * This function adds entropy to the entropy "pool" by using timing
1156 * delays. It uses the timer_rand_state structure to make an estimate
1157 * of how many bits of entropy this call has added to the pool.
1159 * The number "num" is also added to the pool - it should somehow describe
1160 * the type of event which just happened. This is currently 0-255 for
1161 * keyboard scan codes, and 256 upwards for interrupts.
1164 static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
1166 struct entropy_store *r;
1172 long delta, delta2, delta3;
1174 sample.jiffies = jiffies;
1175 sample.cycles = random_get_entropy();
1178 mix_pool_bytes(r, &sample, sizeof(sample));
1181 * Calculate number of bits of randomness we probably added.
1182 * We take into account the first, second and third-order deltas
1183 * in order to make our estimate.
1185 delta = sample.jiffies - READ_ONCE(state->last_time);
1186 WRITE_ONCE(state->last_time, sample.jiffies);
1188 delta2 = delta - READ_ONCE(state->last_delta);
1189 WRITE_ONCE(state->last_delta, delta);
1191 delta3 = delta2 - READ_ONCE(state->last_delta2);
1192 WRITE_ONCE(state->last_delta2, delta2);
1206 * delta is now minimum absolute delta.
1207 * Round down by 1 bit on general principles,
1208 * and limit entropy estimate to 12 bits.
1210 credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
1213 void add_input_randomness(unsigned int type, unsigned int code,
1216 static unsigned char last_value;
1218 /* ignore autorepeat and the like */
1219 if (value == last_value)
1223 add_timer_randomness(&input_timer_state,
1224 (type << 4) ^ code ^ (code >> 4) ^ value);
1225 trace_add_input_randomness(ENTROPY_BITS(&input_pool));
1227 EXPORT_SYMBOL_GPL(add_input_randomness);
1229 static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
1231 #ifdef ADD_INTERRUPT_BENCH
1232 static unsigned long avg_cycles, avg_deviation;
1234 #define AVG_SHIFT 8 /* Exponential average factor k=1/256 */
1235 #define FIXED_1_2 (1 << (AVG_SHIFT-1))
1237 static void add_interrupt_bench(cycles_t start)
1239 long delta = random_get_entropy() - start;
1241 /* Use a weighted moving average */
1242 delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
1243 avg_cycles += delta;
1244 /* And average deviation */
1245 delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
1246 avg_deviation += delta;
1249 #define add_interrupt_bench(x)
1252 static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
1254 __u32 *ptr = (__u32 *) regs;
1259 idx = READ_ONCE(f->reg_idx);
1260 if (idx >= sizeof(struct pt_regs) / sizeof(__u32))
1263 WRITE_ONCE(f->reg_idx, idx);
1267 void add_interrupt_randomness(int irq)
1269 struct entropy_store *r;
1270 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1271 struct pt_regs *regs = get_irq_regs();
1272 unsigned long now = jiffies;
1273 cycles_t cycles = random_get_entropy();
1274 __u32 c_high, j_high;
1278 cycles = get_reg(fast_pool, regs);
1279 c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
1280 j_high = (sizeof(now) > 4) ? now >> 32 : 0;
1281 fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
1282 fast_pool->pool[1] ^= now ^ c_high;
1283 ip = regs ? instruction_pointer(regs) : _RET_IP_;
1284 fast_pool->pool[2] ^= ip;
1285 fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
1286 get_reg(fast_pool, regs);
1288 fast_mix(fast_pool);
1289 add_interrupt_bench(cycles);
1291 if (unlikely(crng_init == 0)) {
1292 if ((fast_pool->count >= 64) &&
1293 crng_fast_load((char *) fast_pool->pool,
1294 sizeof(fast_pool->pool)) > 0) {
1295 fast_pool->count = 0;
1296 fast_pool->last = now;
1301 if ((fast_pool->count < 64) &&
1302 !time_after(now, fast_pool->last + HZ))
1306 if (!spin_trylock(&r->lock))
1309 fast_pool->last = now;
1310 __mix_pool_bytes(r, &fast_pool->pool, sizeof(fast_pool->pool));
1311 spin_unlock(&r->lock);
1313 fast_pool->count = 0;
1315 /* award one bit for the contents of the fast pool */
1316 credit_entropy_bits(r, 1);
1318 EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1321 void add_disk_randomness(struct gendisk *disk)
1323 if (!disk || !disk->random)
1325 /* first major is 1, so we get >= 0x200 here */
1326 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1327 trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
1329 EXPORT_SYMBOL_GPL(add_disk_randomness);
1332 /*********************************************************************
1334 * Entropy extraction routines
1336 *********************************************************************/
1339 * This function decides how many bytes to actually take from the
1340 * given pool, and also debits the entropy count accordingly.
1342 static size_t account(struct entropy_store *r, size_t nbytes, int min,
1345 int entropy_count, orig, have_bytes;
1346 size_t ibytes, nfrac;
1348 BUG_ON(r->entropy_count > r->poolinfo->poolfracbits);
1350 /* Can we pull enough? */
1352 entropy_count = orig = READ_ONCE(r->entropy_count);
1354 /* never pull more than available */
1355 have_bytes = entropy_count >> (ENTROPY_SHIFT + 3);
1357 if ((have_bytes -= reserved) < 0)
1359 ibytes = min_t(size_t, ibytes, have_bytes);
1363 if (WARN_ON(entropy_count < 0)) {
1364 pr_warn("negative entropy count: pool %s count %d\n",
1365 r->name, entropy_count);
1368 nfrac = ibytes << (ENTROPY_SHIFT + 3);
1369 if ((size_t) entropy_count > nfrac)
1370 entropy_count -= nfrac;
1374 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
1377 trace_debit_entropy(r->name, 8 * ibytes);
1378 if (ibytes && ENTROPY_BITS(r) < random_write_wakeup_bits) {
1379 wake_up_interruptible(&random_write_wait);
1380 kill_fasync(&fasync, SIGIO, POLL_OUT);
1387 * This function does the actual extraction for extract_entropy.
1389 * Note: we assume that .poolwords is a multiple of 16 words.
1391 static void extract_buf(struct entropy_store *r, __u8 *out)
1393 struct blake2s_state state __aligned(__alignof__(unsigned long));
1394 u8 hash[BLAKE2S_HASH_SIZE];
1395 unsigned long *salt;
1396 unsigned long flags;
1398 blake2s_init(&state, sizeof(hash));
1401 * If we have an architectural hardware random number
1402 * generator, use it for BLAKE2's salt & personal fields.
1404 for (salt = (unsigned long *)&state.h[4];
1405 salt < (unsigned long *)&state.h[8]; ++salt) {
1407 if (!arch_get_random_long(&v))
1412 /* Generate a hash across the pool */
1413 spin_lock_irqsave(&r->lock, flags);
1414 blake2s_update(&state, (const u8 *)r->pool,
1415 r->poolinfo->poolwords * sizeof(*r->pool));
1416 blake2s_final(&state, hash); /* final zeros out state */
1419 * We mix the hash back into the pool to prevent backtracking
1420 * attacks (where the attacker knows the state of the pool
1421 * plus the current outputs, and attempts to find previous
1422 * outputs), unless the hash function can be inverted. By
1423 * mixing at least a hash worth of hash data back, we make
1424 * brute-forcing the feedback as hard as brute-forcing the
1427 __mix_pool_bytes(r, hash, sizeof(hash));
1428 spin_unlock_irqrestore(&r->lock, flags);
1430 /* Note that EXTRACT_SIZE is half of hash size here, because above
1431 * we've dumped the full length back into mixer. By reducing the
1432 * amount that we emit, we retain a level of forward secrecy.
1434 memcpy(out, hash, EXTRACT_SIZE);
1435 memzero_explicit(hash, sizeof(hash));
1438 static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
1439 size_t nbytes, int fips)
1442 __u8 tmp[EXTRACT_SIZE];
1443 unsigned long flags;
1446 extract_buf(r, tmp);
1449 spin_lock_irqsave(&r->lock, flags);
1450 if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
1451 panic("Hardware RNG duplicated output!\n");
1452 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1453 spin_unlock_irqrestore(&r->lock, flags);
1455 i = min_t(int, nbytes, EXTRACT_SIZE);
1456 memcpy(buf, tmp, i);
1462 /* Wipe data just returned from memory */
1463 memzero_explicit(tmp, sizeof(tmp));
1469 * This function extracts randomness from the "entropy pool", and
1470 * returns it in a buffer.
1472 * The min parameter specifies the minimum amount we can pull before
1473 * failing to avoid races that defeat catastrophic reseeding while the
1474 * reserved parameter indicates how much entropy we must leave in the
1475 * pool after each pull to avoid starving other readers.
1477 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1478 size_t nbytes, int min, int reserved)
1480 __u8 tmp[EXTRACT_SIZE];
1481 unsigned long flags;
1483 /* if last_data isn't primed, we need EXTRACT_SIZE extra bytes */
1485 spin_lock_irqsave(&r->lock, flags);
1486 if (!r->last_data_init) {
1487 r->last_data_init = 1;
1488 spin_unlock_irqrestore(&r->lock, flags);
1489 trace_extract_entropy(r->name, EXTRACT_SIZE,
1490 ENTROPY_BITS(r), _RET_IP_);
1491 extract_buf(r, tmp);
1492 spin_lock_irqsave(&r->lock, flags);
1493 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1495 spin_unlock_irqrestore(&r->lock, flags);
1498 trace_extract_entropy(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1499 nbytes = account(r, nbytes, min, reserved);
1501 return _extract_entropy(r, buf, nbytes, fips_enabled);
1504 #define warn_unseeded_randomness(previous) \
1505 _warn_unseeded_randomness(__func__, (void *) _RET_IP_, (previous))
1507 static void _warn_unseeded_randomness(const char *func_name, void *caller,
1510 #ifdef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1511 const bool print_once = false;
1513 static bool print_once __read_mostly;
1518 (previous && (caller == READ_ONCE(*previous))))
1520 WRITE_ONCE(*previous, caller);
1521 #ifndef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1524 if (__ratelimit(&unseeded_warning))
1525 printk_deferred(KERN_NOTICE "random: %s called from %pS "
1526 "with crng_init=%d\n", func_name, caller,
1531 * This function is the exported kernel interface. It returns some
1532 * number of good random numbers, suitable for key generation, seeding
1533 * TCP sequence numbers, etc. It does not rely on the hardware random
1534 * number generator. For random bytes direct from the hardware RNG
1535 * (when available), use get_random_bytes_arch(). In order to ensure
1536 * that the randomness provided by this function is okay, the function
1537 * wait_for_random_bytes() should be called and return 0 at least once
1538 * at any point prior.
1540 static void _get_random_bytes(void *buf, int nbytes)
1542 __u8 tmp[CHACHA_BLOCK_SIZE] __aligned(4);
1544 trace_get_random_bytes(nbytes, _RET_IP_);
1546 while (nbytes >= CHACHA_BLOCK_SIZE) {
1548 buf += CHACHA_BLOCK_SIZE;
1549 nbytes -= CHACHA_BLOCK_SIZE;
1554 memcpy(buf, tmp, nbytes);
1555 crng_backtrack_protect(tmp, nbytes);
1557 crng_backtrack_protect(tmp, CHACHA_BLOCK_SIZE);
1558 memzero_explicit(tmp, sizeof(tmp));
1561 void get_random_bytes(void *buf, int nbytes)
1563 static void *previous;
1565 warn_unseeded_randomness(&previous);
1566 _get_random_bytes(buf, nbytes);
1568 EXPORT_SYMBOL(get_random_bytes);
1572 * Each time the timer fires, we expect that we got an unpredictable
1573 * jump in the cycle counter. Even if the timer is running on another
1574 * CPU, the timer activity will be touching the stack of the CPU that is
1575 * generating entropy..
1577 * Note that we don't re-arm the timer in the timer itself - we are
1578 * happy to be scheduled away, since that just makes the load more
1579 * complex, but we do not want the timer to keep ticking unless the
1580 * entropy loop is running.
1582 * So the re-arming always happens in the entropy loop itself.
1584 static void entropy_timer(struct timer_list *t)
1586 credit_entropy_bits(&input_pool, 1);
1590 * If we have an actual cycle counter, see if we can
1591 * generate enough entropy with timing noise
1593 static void try_to_generate_entropy(void)
1597 struct timer_list timer;
1600 stack.now = random_get_entropy();
1602 /* Slow counter - or none. Don't even bother */
1603 if (stack.now == random_get_entropy())
1606 timer_setup_on_stack(&stack.timer, entropy_timer, 0);
1607 while (!crng_ready()) {
1608 if (!timer_pending(&stack.timer))
1609 mod_timer(&stack.timer, jiffies+1);
1610 mix_pool_bytes(&input_pool, &stack.now, sizeof(stack.now));
1612 stack.now = random_get_entropy();
1615 del_timer_sync(&stack.timer);
1616 destroy_timer_on_stack(&stack.timer);
1617 mix_pool_bytes(&input_pool, &stack.now, sizeof(stack.now));
1621 * Wait for the urandom pool to be seeded and thus guaranteed to supply
1622 * cryptographically secure random numbers. This applies to: the /dev/urandom
1623 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
1624 * family of functions. Using any of these functions without first calling
1625 * this function forfeits the guarantee of security.
1627 * Returns: 0 if the urandom pool has been seeded.
1628 * -ERESTARTSYS if the function was interrupted by a signal.
1630 int wait_for_random_bytes(void)
1632 if (likely(crng_ready()))
1637 ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ);
1639 return ret > 0 ? 0 : ret;
1641 try_to_generate_entropy();
1642 } while (!crng_ready());
1646 EXPORT_SYMBOL(wait_for_random_bytes);
1649 * Returns whether or not the urandom pool has been seeded and thus guaranteed
1650 * to supply cryptographically secure random numbers. This applies to: the
1651 * /dev/urandom device, the get_random_bytes function, and the get_random_{u32,
1652 * ,u64,int,long} family of functions.
1654 * Returns: true if the urandom pool has been seeded.
1655 * false if the urandom pool has not been seeded.
1657 bool rng_is_initialized(void)
1659 return crng_ready();
1661 EXPORT_SYMBOL(rng_is_initialized);
1664 * Add a callback function that will be invoked when the nonblocking
1665 * pool is initialised.
1667 * returns: 0 if callback is successfully added
1668 * -EALREADY if pool is already initialised (callback not called)
1669 * -ENOENT if module for callback is not alive
1671 int add_random_ready_callback(struct random_ready_callback *rdy)
1673 struct module *owner;
1674 unsigned long flags;
1675 int err = -EALREADY;
1681 if (!try_module_get(owner))
1684 spin_lock_irqsave(&random_ready_list_lock, flags);
1690 list_add(&rdy->list, &random_ready_list);
1694 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1700 EXPORT_SYMBOL(add_random_ready_callback);
1703 * Delete a previously registered readiness callback function.
1705 void del_random_ready_callback(struct random_ready_callback *rdy)
1707 unsigned long flags;
1708 struct module *owner = NULL;
1710 spin_lock_irqsave(&random_ready_list_lock, flags);
1711 if (!list_empty(&rdy->list)) {
1712 list_del_init(&rdy->list);
1715 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1719 EXPORT_SYMBOL(del_random_ready_callback);
1722 * This function will use the architecture-specific hardware random
1723 * number generator if it is available. The arch-specific hw RNG will
1724 * almost certainly be faster than what we can do in software, but it
1725 * is impossible to verify that it is implemented securely (as
1726 * opposed, to, say, the AES encryption of a sequence number using a
1727 * key known by the NSA). So it's useful if we need the speed, but
1728 * only if we're willing to trust the hardware manufacturer not to
1729 * have put in a back door.
1731 * Return number of bytes filled in.
1733 int __must_check get_random_bytes_arch(void *buf, int nbytes)
1738 trace_get_random_bytes_arch(left, _RET_IP_);
1741 int chunk = min_t(int, left, sizeof(unsigned long));
1743 if (!arch_get_random_long(&v))
1746 memcpy(p, &v, chunk);
1751 return nbytes - left;
1753 EXPORT_SYMBOL(get_random_bytes_arch);
1756 * init_std_data - initialize pool with system data
1758 * @r: pool to initialize
1760 * This function clears the pool's entropy count and mixes some system
1761 * data into the pool to prepare it for use. The pool is not cleared
1762 * as that can only decrease the entropy in the pool.
1764 static void __init init_std_data(struct entropy_store *r)
1767 ktime_t now = ktime_get_real();
1770 mix_pool_bytes(r, &now, sizeof(now));
1771 for (i = r->poolinfo->poolbytes; i > 0; i -= sizeof(rv)) {
1772 if (!arch_get_random_seed_long(&rv) &&
1773 !arch_get_random_long(&rv))
1774 rv = random_get_entropy();
1775 mix_pool_bytes(r, &rv, sizeof(rv));
1777 mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1781 * Note that setup_arch() may call add_device_randomness()
1782 * long before we get here. This allows seeding of the pools
1783 * with some platform dependent data very early in the boot
1784 * process. But it limits our options here. We must use
1785 * statically allocated structures that already have all
1786 * initializations complete at compile time. We should also
1787 * take care not to overwrite the precious per platform data
1790 int __init rand_initialize(void)
1792 init_std_data(&input_pool);
1793 if (crng_need_final_init)
1794 crng_finalize_init(&primary_crng);
1795 crng_initialize_primary(&primary_crng);
1796 crng_global_init_time = jiffies;
1797 if (ratelimit_disable) {
1798 urandom_warning.interval = 0;
1799 unseeded_warning.interval = 0;
1805 void rand_initialize_disk(struct gendisk *disk)
1807 struct timer_rand_state *state;
1810 * If kzalloc returns null, we just won't use that entropy
1813 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1815 state->last_time = INITIAL_JIFFIES;
1816 disk->random = state;
1822 urandom_read_nowarn(struct file *file, char __user *buf, size_t nbytes,
1827 nbytes = min_t(size_t, nbytes, INT_MAX >> (ENTROPY_SHIFT + 3));
1828 ret = extract_crng_user(buf, nbytes);
1829 trace_urandom_read(8 * nbytes, 0, ENTROPY_BITS(&input_pool));
1834 urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1836 unsigned long flags;
1837 static int maxwarn = 10;
1839 if (!crng_ready() && maxwarn > 0) {
1841 if (__ratelimit(&urandom_warning))
1842 pr_notice("%s: uninitialized urandom read (%zd bytes read)\n",
1843 current->comm, nbytes);
1844 spin_lock_irqsave(&primary_crng.lock, flags);
1846 spin_unlock_irqrestore(&primary_crng.lock, flags);
1849 return urandom_read_nowarn(file, buf, nbytes, ppos);
1853 random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1857 ret = wait_for_random_bytes();
1860 return urandom_read_nowarn(file, buf, nbytes, ppos);
1864 random_poll(struct file *file, poll_table * wait)
1868 poll_wait(file, &crng_init_wait, wait);
1869 poll_wait(file, &random_write_wait, wait);
1872 mask |= EPOLLIN | EPOLLRDNORM;
1873 if (ENTROPY_BITS(&input_pool) < random_write_wakeup_bits)
1874 mask |= EPOLLOUT | EPOLLWRNORM;
1879 write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1883 const char __user *p = buffer;
1888 bytes = min(count, sizeof(buf));
1889 if (copy_from_user(&buf, p, bytes))
1892 for (b = bytes ; b > 0 ; b -= sizeof(__u32), i++) {
1893 if (!arch_get_random_int(&t))
1901 mix_pool_bytes(r, buf, bytes);
1908 static ssize_t random_write(struct file *file, const char __user *buffer,
1909 size_t count, loff_t *ppos)
1913 ret = write_pool(&input_pool, buffer, count);
1917 return (ssize_t)count;
1920 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1922 int size, ent_count;
1923 int __user *p = (int __user *)arg;
1928 /* inherently racy, no point locking */
1929 ent_count = ENTROPY_BITS(&input_pool);
1930 if (put_user(ent_count, p))
1933 case RNDADDTOENTCNT:
1934 if (!capable(CAP_SYS_ADMIN))
1936 if (get_user(ent_count, p))
1938 return credit_entropy_bits_safe(&input_pool, ent_count);
1940 if (!capable(CAP_SYS_ADMIN))
1942 if (get_user(ent_count, p++))
1946 if (get_user(size, p++))
1948 retval = write_pool(&input_pool, (const char __user *)p,
1952 return credit_entropy_bits_safe(&input_pool, ent_count);
1956 * Clear the entropy pool counters. We no longer clear
1957 * the entropy pool, as that's silly.
1959 if (!capable(CAP_SYS_ADMIN))
1961 input_pool.entropy_count = 0;
1964 if (!capable(CAP_SYS_ADMIN))
1968 crng_reseed(&primary_crng, &input_pool);
1969 WRITE_ONCE(crng_global_init_time, jiffies - 1);
1976 static int random_fasync(int fd, struct file *filp, int on)
1978 return fasync_helper(fd, filp, on, &fasync);
1981 const struct file_operations random_fops = {
1982 .read = random_read,
1983 .write = random_write,
1984 .poll = random_poll,
1985 .unlocked_ioctl = random_ioctl,
1986 .compat_ioctl = compat_ptr_ioctl,
1987 .fasync = random_fasync,
1988 .llseek = noop_llseek,
1991 const struct file_operations urandom_fops = {
1992 .read = urandom_read,
1993 .write = random_write,
1994 .unlocked_ioctl = random_ioctl,
1995 .compat_ioctl = compat_ptr_ioctl,
1996 .fasync = random_fasync,
1997 .llseek = noop_llseek,
2000 SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count,
2001 unsigned int, flags)
2005 if (flags & ~(GRND_NONBLOCK|GRND_RANDOM|GRND_INSECURE))
2009 * Requesting insecure and blocking randomness at the same time makes
2012 if ((flags & (GRND_INSECURE|GRND_RANDOM)) == (GRND_INSECURE|GRND_RANDOM))
2015 if (count > INT_MAX)
2018 if (!(flags & GRND_INSECURE) && !crng_ready()) {
2019 if (flags & GRND_NONBLOCK)
2021 ret = wait_for_random_bytes();
2025 return urandom_read_nowarn(NULL, buf, count, NULL);
2028 /********************************************************************
2032 ********************************************************************/
2034 #ifdef CONFIG_SYSCTL
2036 #include <linux/sysctl.h>
2038 static int min_write_thresh;
2039 static int max_write_thresh = INPUT_POOL_WORDS * 32;
2040 static int random_min_urandom_seed = 60;
2041 static char sysctl_bootid[16];
2044 * This function is used to return both the bootid UUID, and random
2045 * UUID. The difference is in whether table->data is NULL; if it is,
2046 * then a new UUID is generated and returned to the user.
2048 * If the user accesses this via the proc interface, the UUID will be
2049 * returned as an ASCII string in the standard UUID format; if via the
2050 * sysctl system call, as 16 bytes of binary data.
2052 static int proc_do_uuid(struct ctl_table *table, int write,
2053 void *buffer, size_t *lenp, loff_t *ppos)
2055 struct ctl_table fake_table;
2056 unsigned char buf[64], tmp_uuid[16], *uuid;
2061 generate_random_uuid(uuid);
2063 static DEFINE_SPINLOCK(bootid_spinlock);
2065 spin_lock(&bootid_spinlock);
2067 generate_random_uuid(uuid);
2068 spin_unlock(&bootid_spinlock);
2071 sprintf(buf, "%pU", uuid);
2073 fake_table.data = buf;
2074 fake_table.maxlen = sizeof(buf);
2076 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
2080 * Return entropy available scaled to integral bits
2082 static int proc_do_entropy(struct ctl_table *table, int write,
2083 void *buffer, size_t *lenp, loff_t *ppos)
2085 struct ctl_table fake_table;
2088 entropy_count = *(int *)table->data >> ENTROPY_SHIFT;
2090 fake_table.data = &entropy_count;
2091 fake_table.maxlen = sizeof(entropy_count);
2093 return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
2096 static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
2097 extern struct ctl_table random_table[];
2098 struct ctl_table random_table[] = {
2100 .procname = "poolsize",
2101 .data = &sysctl_poolsize,
2102 .maxlen = sizeof(int),
2104 .proc_handler = proc_dointvec,
2107 .procname = "entropy_avail",
2108 .maxlen = sizeof(int),
2110 .proc_handler = proc_do_entropy,
2111 .data = &input_pool.entropy_count,
2114 .procname = "write_wakeup_threshold",
2115 .data = &random_write_wakeup_bits,
2116 .maxlen = sizeof(int),
2118 .proc_handler = proc_dointvec_minmax,
2119 .extra1 = &min_write_thresh,
2120 .extra2 = &max_write_thresh,
2123 .procname = "urandom_min_reseed_secs",
2124 .data = &random_min_urandom_seed,
2125 .maxlen = sizeof(int),
2127 .proc_handler = proc_dointvec,
2130 .procname = "boot_id",
2131 .data = &sysctl_bootid,
2134 .proc_handler = proc_do_uuid,
2140 .proc_handler = proc_do_uuid,
2142 #ifdef ADD_INTERRUPT_BENCH
2144 .procname = "add_interrupt_avg_cycles",
2145 .data = &avg_cycles,
2146 .maxlen = sizeof(avg_cycles),
2148 .proc_handler = proc_doulongvec_minmax,
2151 .procname = "add_interrupt_avg_deviation",
2152 .data = &avg_deviation,
2153 .maxlen = sizeof(avg_deviation),
2155 .proc_handler = proc_doulongvec_minmax,
2160 #endif /* CONFIG_SYSCTL */
2162 struct batched_entropy {
2164 u64 entropy_u64[CHACHA_BLOCK_SIZE / sizeof(u64)];
2165 u32 entropy_u32[CHACHA_BLOCK_SIZE / sizeof(u32)];
2167 unsigned int position;
2168 spinlock_t batch_lock;
2172 * Get a random word for internal kernel use only. The quality of the random
2173 * number is good as /dev/urandom, but there is no backtrack protection, with
2174 * the goal of being quite fast and not depleting entropy. In order to ensure
2175 * that the randomness provided by this function is okay, the function
2176 * wait_for_random_bytes() should be called and return 0 at least once at any
2179 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64) = {
2180 .batch_lock = __SPIN_LOCK_UNLOCKED(batched_entropy_u64.lock),
2183 u64 get_random_u64(void)
2186 unsigned long flags;
2187 struct batched_entropy *batch;
2188 static void *previous;
2190 warn_unseeded_randomness(&previous);
2192 batch = raw_cpu_ptr(&batched_entropy_u64);
2193 spin_lock_irqsave(&batch->batch_lock, flags);
2194 if (batch->position % ARRAY_SIZE(batch->entropy_u64) == 0) {
2195 extract_crng((u8 *)batch->entropy_u64);
2196 batch->position = 0;
2198 ret = batch->entropy_u64[batch->position++];
2199 spin_unlock_irqrestore(&batch->batch_lock, flags);
2202 EXPORT_SYMBOL(get_random_u64);
2204 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32) = {
2205 .batch_lock = __SPIN_LOCK_UNLOCKED(batched_entropy_u32.lock),
2207 u32 get_random_u32(void)
2210 unsigned long flags;
2211 struct batched_entropy *batch;
2212 static void *previous;
2214 warn_unseeded_randomness(&previous);
2216 batch = raw_cpu_ptr(&batched_entropy_u32);
2217 spin_lock_irqsave(&batch->batch_lock, flags);
2218 if (batch->position % ARRAY_SIZE(batch->entropy_u32) == 0) {
2219 extract_crng((u8 *)batch->entropy_u32);
2220 batch->position = 0;
2222 ret = batch->entropy_u32[batch->position++];
2223 spin_unlock_irqrestore(&batch->batch_lock, flags);
2226 EXPORT_SYMBOL(get_random_u32);
2228 /* It's important to invalidate all potential batched entropy that might
2229 * be stored before the crng is initialized, which we can do lazily by
2230 * simply resetting the counter to zero so that it's re-extracted on the
2232 static void invalidate_batched_entropy(void)
2235 unsigned long flags;
2237 for_each_possible_cpu (cpu) {
2238 struct batched_entropy *batched_entropy;
2240 batched_entropy = per_cpu_ptr(&batched_entropy_u32, cpu);
2241 spin_lock_irqsave(&batched_entropy->batch_lock, flags);
2242 batched_entropy->position = 0;
2243 spin_unlock(&batched_entropy->batch_lock);
2245 batched_entropy = per_cpu_ptr(&batched_entropy_u64, cpu);
2246 spin_lock(&batched_entropy->batch_lock);
2247 batched_entropy->position = 0;
2248 spin_unlock_irqrestore(&batched_entropy->batch_lock, flags);
2253 * randomize_page - Generate a random, page aligned address
2254 * @start: The smallest acceptable address the caller will take.
2255 * @range: The size of the area, starting at @start, within which the
2256 * random address must fall.
2258 * If @start + @range would overflow, @range is capped.
2260 * NOTE: Historical use of randomize_range, which this replaces, presumed that
2261 * @start was already page aligned. We now align it regardless.
2263 * Return: A page aligned address within [start, start + range). On error,
2264 * @start is returned.
2267 randomize_page(unsigned long start, unsigned long range)
2269 if (!PAGE_ALIGNED(start)) {
2270 range -= PAGE_ALIGN(start) - start;
2271 start = PAGE_ALIGN(start);
2274 if (start > ULONG_MAX - range)
2275 range = ULONG_MAX - start;
2277 range >>= PAGE_SHIFT;
2282 return start + (get_random_long() % range << PAGE_SHIFT);
2285 /* Interface for in-kernel drivers of true hardware RNGs.
2286 * Those devices may produce endless random bits and will be throttled
2287 * when our pool is full.
2289 void add_hwgenerator_randomness(const char *buffer, size_t count,
2292 struct entropy_store *poolp = &input_pool;
2294 if (unlikely(crng_init == 0)) {
2295 size_t ret = crng_fast_load(buffer, count);
2296 mix_pool_bytes(poolp, buffer, ret);
2299 if (!count || crng_init == 0)
2303 /* Suspend writing if we're above the trickle threshold.
2304 * We'll be woken up again once below random_write_wakeup_thresh,
2305 * or when the calling thread is about to terminate.
2307 wait_event_interruptible(random_write_wait,
2308 !system_wq || kthread_should_stop() ||
2309 ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
2310 mix_pool_bytes(poolp, buffer, count);
2311 credit_entropy_bits(poolp, entropy);
2313 EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
2315 /* Handle random seed passed by bootloader.
2316 * If the seed is trustworthy, it would be regarded as hardware RNGs. Otherwise
2317 * it would be regarded as device data.
2318 * The decision is controlled by CONFIG_RANDOM_TRUST_BOOTLOADER.
2320 void add_bootloader_randomness(const void *buf, unsigned int size)
2322 if (IS_ENABLED(CONFIG_RANDOM_TRUST_BOOTLOADER))
2323 add_hwgenerator_randomness(buf, size, size * 8);
2325 add_device_randomness(buf, size);
2327 EXPORT_SYMBOL_GPL(add_bootloader_randomness);