2 * random.c -- A strong random number generator
4 * Copyright (C) 2017 Jason A. Donenfeld <Jason@zx2c4.com>. All
7 * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005
9 * Copyright Theodore Ts'o, 1994, 1995, 1996, 1997, 1998, 1999. All
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, and the entire permission notice in its entirety,
17 * including the disclaimer of warranties.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. The name of the author may not be used to endorse or promote
22 * products derived from this software without specific prior
25 * ALTERNATIVELY, this product may be distributed under the terms of
26 * the GNU General Public License, in which case the provisions of the GPL are
27 * required INSTEAD OF the above restrictions. (This clause is
28 * necessary due to a potential bad interaction between the GPL and
29 * the restrictions contained in a BSD-style copyright.)
31 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
32 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
33 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ALL OF
34 * WHICH ARE HEREBY DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE
35 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
36 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT
37 * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
38 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
39 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
40 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
41 * USE OF THIS SOFTWARE, EVEN IF NOT ADVISED OF THE POSSIBILITY OF SUCH
46 * (now, with legal B.S. out of the way.....)
48 * This routine gathers environmental noise from device drivers, etc.,
49 * and returns good random numbers, suitable for cryptographic use.
50 * Besides the obvious cryptographic uses, these numbers are also good
51 * for seeding TCP sequence numbers, and other places where it is
52 * desirable to have numbers which are not only random, but hard to
53 * predict by an attacker.
58 * Computers are very predictable devices. Hence it is extremely hard
59 * to produce truly random numbers on a computer --- as opposed to
60 * pseudo-random numbers, which can easily generated by using a
61 * algorithm. Unfortunately, it is very easy for attackers to guess
62 * the sequence of pseudo-random number generators, and for some
63 * applications this is not acceptable. So instead, we must try to
64 * gather "environmental noise" from the computer's environment, which
65 * must be hard for outside attackers to observe, and use that to
66 * generate random numbers. In a Unix environment, this is best done
67 * from inside the kernel.
69 * Sources of randomness from the environment include inter-keyboard
70 * timings, inter-interrupt timings from some interrupts, and other
71 * events which are both (a) non-deterministic and (b) hard for an
72 * outside observer to measure. Randomness from these sources are
73 * added to an "entropy pool", which is mixed using a CRC-like function.
74 * This is not cryptographically strong, but it is adequate assuming
75 * the randomness is not chosen maliciously, and it is fast enough that
76 * the overhead of doing it on every interrupt is very reasonable.
77 * As random bytes are mixed into the entropy pool, the routines keep
78 * an *estimate* of how many bits of randomness have been stored into
79 * the random number generator's internal state.
81 * When random bytes are desired, they are obtained by taking the SHA
82 * hash of the contents of the "entropy pool". The SHA hash avoids
83 * exposing the internal state of the entropy pool. It is believed to
84 * be computationally infeasible to derive any useful information
85 * about the input of SHA from its output. Even if it is possible to
86 * analyze SHA in some clever way, as long as the amount of data
87 * returned from the generator is less than the inherent entropy in
88 * the pool, the output data is totally unpredictable. For this
89 * reason, the routine decreases its internal estimate of how many
90 * bits of "true randomness" are contained in the entropy pool as it
91 * outputs random numbers.
93 * If this estimate goes to zero, the routine can still generate
94 * random numbers; however, an attacker may (at least in theory) be
95 * able to infer the future output of the generator from prior
96 * outputs. This requires successful cryptanalysis of SHA, which is
97 * not believed to be feasible, but there is a remote possibility.
98 * Nonetheless, these numbers should be useful for the vast majority
101 * Exported interfaces ---- output
102 * ===============================
104 * There are four exported interfaces; two for use within the kernel,
105 * and two or use from userspace.
107 * Exported interfaces ---- userspace output
108 * -----------------------------------------
110 * The userspace interfaces are two character devices /dev/random and
111 * /dev/urandom. /dev/random is suitable for use when very high
112 * quality randomness is desired (for example, for key generation or
113 * one-time pads), as it will only return a maximum of the number of
114 * bits of randomness (as estimated by the random number generator)
115 * contained in the entropy pool.
117 * The /dev/urandom device does not have this limit, and will return
118 * as many bytes as are requested. As more and more random bytes are
119 * requested without giving time for the entropy pool to recharge,
120 * this will result in random numbers that are merely cryptographically
121 * strong. For many applications, however, this is acceptable.
123 * Exported interfaces ---- kernel output
124 * --------------------------------------
126 * The primary kernel interface is
128 * void get_random_bytes(void *buf, int nbytes);
130 * This interface will return the requested number of random bytes,
131 * and place it in the requested buffer. This is equivalent to a
132 * read from /dev/urandom.
134 * For less critical applications, there are the functions:
136 * u32 get_random_u32()
137 * u64 get_random_u64()
138 * unsigned int get_random_int()
139 * unsigned long get_random_long()
141 * These are produced by a cryptographic RNG seeded from get_random_bytes,
142 * and so do not deplete the entropy pool as much. These are recommended
143 * for most in-kernel operations *if the result is going to be stored in
146 * Specifically, the get_random_int() family do not attempt to do
147 * "anti-backtracking". If you capture the state of the kernel (e.g.
148 * by snapshotting the VM), you can figure out previous get_random_int()
149 * return values. But if the value is stored in the kernel anyway,
150 * this is not a problem.
152 * It *is* safe to expose get_random_int() output to attackers (e.g. as
153 * network cookies); given outputs 1..n, it's not feasible to predict
154 * outputs 0 or n+1. The only concern is an attacker who breaks into
155 * the kernel later; the get_random_int() engine is not reseeded as
156 * often as the get_random_bytes() one.
158 * get_random_bytes() is needed for keys that need to stay secret after
159 * they are erased from the kernel. For example, any key that will
160 * be wrapped and stored encrypted. And session encryption keys: we'd
161 * like to know that after the session is closed and the keys erased,
162 * the plaintext is unrecoverable to someone who recorded the ciphertext.
164 * But for network ports/cookies, stack canaries, PRNG seeds, address
165 * space layout randomization, session *authentication* keys, or other
166 * applications where the sensitive data is stored in the kernel in
167 * plaintext for as long as it's sensitive, the get_random_int() family
170 * Consider ASLR. We want to keep the address space secret from an
171 * outside attacker while the process is running, but once the address
172 * space is torn down, it's of no use to an attacker any more. And it's
173 * stored in kernel data structures as long as it's alive, so worrying
174 * about an attacker's ability to extrapolate it from the get_random_int()
177 * Even some cryptographic keys are safe to generate with get_random_int().
178 * In particular, keys for SipHash are generally fine. Here, knowledge
179 * of the key authorizes you to do something to a kernel object (inject
180 * packets to a network connection, or flood a hash table), and the
181 * key is stored with the object being protected. Once it goes away,
182 * we no longer care if anyone knows the key.
187 * For even weaker applications, see the pseudorandom generator
188 * prandom_u32(), prandom_max(), and prandom_bytes(). If the random
189 * numbers aren't security-critical at all, these are *far* cheaper.
190 * Useful for self-tests, random error simulation, randomized backoffs,
191 * and any other application where you trust that nobody is trying to
192 * maliciously mess with you by guessing the "random" numbers.
194 * Exported interfaces ---- input
195 * ==============================
197 * The current exported interfaces for gathering environmental noise
198 * from the devices are:
200 * void add_device_randomness(const void *buf, unsigned int size);
201 * void add_input_randomness(unsigned int type, unsigned int code,
202 * unsigned int value);
203 * void add_interrupt_randomness(int irq, int irq_flags);
204 * void add_disk_randomness(struct gendisk *disk);
206 * add_device_randomness() is for adding data to the random pool that
207 * is likely to differ between two devices (or possibly even per boot).
208 * This would be things like MAC addresses or serial numbers, or the
209 * read-out of the RTC. This does *not* add any actual entropy to the
210 * pool, but it initializes the pool to different values for devices
211 * that might otherwise be identical and have very little entropy
212 * available to them (particularly common in the embedded world).
214 * add_input_randomness() uses the input layer interrupt timing, as well as
215 * the event type information from the hardware.
217 * add_interrupt_randomness() uses the interrupt timing as random
218 * inputs to the entropy pool. Using the cycle counters and the irq source
219 * as inputs, it feeds the randomness roughly once a second.
221 * add_disk_randomness() uses what amounts to the seek time of block
222 * layer request events, on a per-disk_devt basis, as input to the
223 * entropy pool. Note that high-speed solid state drives with very low
224 * seek times do not make for good sources of entropy, as their seek
225 * times are usually fairly consistent.
227 * All of these routines try to estimate how many bits of randomness a
228 * particular randomness source. They do this by keeping track of the
229 * first and second order deltas of the event timings.
231 * Ensuring unpredictability at system startup
232 * ============================================
234 * When any operating system starts up, it will go through a sequence
235 * of actions that are fairly predictable by an adversary, especially
236 * if the start-up does not involve interaction with a human operator.
237 * This reduces the actual number of bits of unpredictability in the
238 * entropy pool below the value in entropy_count. In order to
239 * counteract this effect, it helps to carry information in the
240 * entropy pool across shut-downs and start-ups. To do this, put the
241 * following lines an appropriate script which is run during the boot
244 * echo "Initializing random number generator..."
245 * random_seed=/var/run/random-seed
246 * # Carry a random seed from start-up to start-up
247 * # Load and then save the whole entropy pool
248 * if [ -f $random_seed ]; then
249 * cat $random_seed >/dev/urandom
253 * chmod 600 $random_seed
254 * dd if=/dev/urandom of=$random_seed count=1 bs=512
256 * and the following lines in an appropriate script which is run as
257 * the system is shutdown:
259 * # Carry a random seed from shut-down to start-up
260 * # Save the whole entropy pool
261 * echo "Saving random seed..."
262 * random_seed=/var/run/random-seed
264 * chmod 600 $random_seed
265 * dd if=/dev/urandom of=$random_seed count=1 bs=512
267 * For example, on most modern systems using the System V init
268 * scripts, such code fragments would be found in
269 * /etc/rc.d/init.d/random. On older Linux systems, the correct script
270 * location might be in /etc/rcb.d/rc.local or /etc/rc.d/rc.0.
272 * Effectively, these commands cause the contents of the entropy pool
273 * to be saved at shut-down time and reloaded into the entropy pool at
274 * start-up. (The 'dd' in the addition to the bootup script is to
275 * make sure that /etc/random-seed is different for every start-up,
276 * even if the system crashes without executing rc.0.) Even with
277 * complete knowledge of the start-up activities, predicting the state
278 * of the entropy pool requires knowledge of the previous history of
281 * Configuring the /dev/random driver under Linux
282 * ==============================================
284 * The /dev/random driver under Linux uses minor numbers 8 and 9 of
285 * the /dev/mem major number (#1). So if your system does not have
286 * /dev/random and /dev/urandom created already, they can be created
287 * by using the commands:
289 * mknod /dev/random c 1 8
290 * mknod /dev/urandom c 1 9
295 * Ideas for constructing this random number generator were derived
296 * from Pretty Good Privacy's random number generator, and from private
297 * discussions with Phil Karn. Colin Plumb provided a faster random
298 * number generator, which speed up the mixing function of the entropy
299 * pool, taken from PGPfone. Dale Worley has also contributed many
300 * useful ideas and suggestions to improve this driver.
302 * Any flaws in the design are solely my responsibility, and should
303 * not be attributed to the Phil, Colin, or any of authors of PGP.
305 * Further background information on this topic may be obtained from
306 * RFC 1750, "Randomness Recommendations for Security", by Donald
307 * Eastlake, Steve Crocker, and Jeff Schiller.
310 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
312 #include <linux/utsname.h>
313 #include <linux/module.h>
314 #include <linux/kernel.h>
315 #include <linux/major.h>
316 #include <linux/string.h>
317 #include <linux/fcntl.h>
318 #include <linux/slab.h>
319 #include <linux/random.h>
320 #include <linux/poll.h>
321 #include <linux/init.h>
322 #include <linux/fs.h>
323 #include <linux/genhd.h>
324 #include <linux/interrupt.h>
325 #include <linux/mm.h>
326 #include <linux/nodemask.h>
327 #include <linux/spinlock.h>
328 #include <linux/kthread.h>
329 #include <linux/percpu.h>
330 #include <linux/fips.h>
331 #include <linux/ptrace.h>
332 #include <linux/workqueue.h>
333 #include <linux/irq.h>
334 #include <linux/ratelimit.h>
335 #include <linux/syscalls.h>
336 #include <linux/completion.h>
337 #include <linux/uuid.h>
338 #include <crypto/chacha.h>
339 #include <crypto/sha1.h>
341 #include <asm/processor.h>
342 #include <linux/uaccess.h>
344 #include <asm/irq_regs.h>
347 #define CREATE_TRACE_POINTS
348 #include <trace/events/random.h>
350 /* #define ADD_INTERRUPT_BENCH */
353 * Configuration information
355 #define INPUT_POOL_SHIFT 12
356 #define INPUT_POOL_WORDS (1 << (INPUT_POOL_SHIFT-5))
357 #define OUTPUT_POOL_SHIFT 10
358 #define OUTPUT_POOL_WORDS (1 << (OUTPUT_POOL_SHIFT-5))
359 #define EXTRACT_SIZE 10
362 #define LONGS(x) (((x) + sizeof(unsigned long) - 1)/sizeof(unsigned long))
365 * To allow fractional bits to be tracked, the entropy_count field is
366 * denominated in units of 1/8th bits.
368 * 2*(ENTROPY_SHIFT + poolbitshift) must <= 31, or the multiply in
369 * credit_entropy_bits() needs to be 64 bits wide.
371 #define ENTROPY_SHIFT 3
372 #define ENTROPY_BITS(r) ((r)->entropy_count >> ENTROPY_SHIFT)
375 * If the entropy count falls under this number of bits, then we
376 * should wake up processes which are selecting or polling on write
377 * access to /dev/random.
379 static int random_write_wakeup_bits = 28 * OUTPUT_POOL_WORDS;
382 * Originally, we used a primitive polynomial of degree .poolwords
383 * over GF(2). The taps for various sizes are defined below. They
384 * were chosen to be evenly spaced except for the last tap, which is 1
385 * to get the twisting happening as fast as possible.
387 * For the purposes of better mixing, we use the CRC-32 polynomial as
388 * well to make a (modified) twisted Generalized Feedback Shift
389 * Register. (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR
390 * generators. ACM Transactions on Modeling and Computer Simulation
391 * 2(3):179-194. Also see M. Matsumoto & Y. Kurita, 1994. Twisted
392 * GFSR generators II. ACM Transactions on Modeling and Computer
393 * Simulation 4:254-266)
395 * Thanks to Colin Plumb for suggesting this.
397 * The mixing operation is much less sensitive than the output hash,
398 * where we use SHA-1. All that we want of mixing operation is that
399 * it be a good non-cryptographic hash; i.e. it not produce collisions
400 * when fed "random" data of the sort we expect to see. As long as
401 * the pool state differs for different inputs, we have preserved the
402 * input entropy and done a good job. The fact that an intelligent
403 * attacker can construct inputs that will produce controlled
404 * alterations to the pool's state is not important because we don't
405 * consider such inputs to contribute any randomness. The only
406 * property we need with respect to them is that the attacker can't
407 * increase his/her knowledge of the pool's state. Since all
408 * additions are reversible (knowing the final state and the input,
409 * you can reconstruct the initial state), if an attacker has any
410 * uncertainty about the initial state, he/she can only shuffle that
411 * uncertainty about, but never cause any collisions (which would
412 * decrease the uncertainty).
414 * Our mixing functions were analyzed by Lacharme, Roeck, Strubel, and
415 * Videau in their paper, "The Linux Pseudorandom Number Generator
416 * Revisited" (see: http://eprint.iacr.org/2012/251.pdf). In their
417 * paper, they point out that we are not using a true Twisted GFSR,
418 * since Matsumoto & Kurita used a trinomial feedback polynomial (that
419 * is, with only three taps, instead of the six that we are using).
420 * As a result, the resulting polynomial is neither primitive nor
421 * irreducible, and hence does not have a maximal period over
422 * GF(2**32). They suggest a slight change to the generator
423 * polynomial which improves the resulting TGFSR polynomial to be
424 * irreducible, which we have made here.
426 static const struct poolinfo {
427 int poolbitshift, poolwords, poolbytes, poolfracbits;
428 #define S(x) ilog2(x)+5, (x), (x)*4, (x) << (ENTROPY_SHIFT+5)
429 int tap1, tap2, tap3, tap4, tap5;
430 } poolinfo_table[] = {
431 /* was: x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 */
432 /* x^128 + x^104 + x^76 + x^51 +x^25 + x + 1 */
433 { S(128), 104, 76, 51, 25, 1 },
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),
456 * crng_init = 0 --> Uninitialized
458 * 2 --> Initialized from input_pool
460 * crng_init is protected by primary_crng->lock, and only increases
461 * its value (from 0->1->2).
463 static int crng_init = 0;
464 #define crng_ready() (likely(crng_init > 1))
465 static int crng_init_cnt = 0;
466 static unsigned long crng_global_init_time = 0;
467 #define CRNG_INIT_CNT_THRESH (2*CHACHA_KEY_SIZE)
468 static void _extract_crng(struct crng_state *crng, __u8 out[CHACHA_BLOCK_SIZE]);
469 static void _crng_backtrack_protect(struct crng_state *crng,
470 __u8 tmp[CHACHA_BLOCK_SIZE], int used);
471 static void process_random_ready_list(void);
472 static void _get_random_bytes(void *buf, int nbytes);
474 static struct ratelimit_state unseeded_warning =
475 RATELIMIT_STATE_INIT("warn_unseeded_randomness", HZ, 3);
476 static struct ratelimit_state urandom_warning =
477 RATELIMIT_STATE_INIT("warn_urandom_randomness", HZ, 3);
479 static int ratelimit_disable __read_mostly;
481 module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
482 MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
484 /**********************************************************************
486 * OS independent entropy store. Here are the functions which handle
487 * storing entropy in an entropy pool.
489 **********************************************************************/
491 struct entropy_store;
492 struct entropy_store {
493 /* read-only data: */
494 const struct poolinfo *poolinfo;
498 /* read-write data: */
500 unsigned short add_ptr;
501 unsigned short input_rotate;
503 unsigned int last_data_init:1;
504 __u8 last_data[EXTRACT_SIZE];
507 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
508 size_t nbytes, int min, int rsvd);
509 static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
510 size_t nbytes, int fips);
512 static void crng_reseed(struct crng_state *crng, struct entropy_store *r);
513 static __u32 input_pool_data[INPUT_POOL_WORDS] __latent_entropy;
515 static struct entropy_store input_pool = {
516 .poolinfo = &poolinfo_table[0],
518 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
519 .pool = input_pool_data
522 static __u32 const twist_table[8] = {
523 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
524 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
527 * This function adds bytes into the entropy "pool". It does not
528 * update the entropy estimate. The caller should call
529 * credit_entropy_bits if this is appropriate.
531 * The pool is stirred with a primitive polynomial of the appropriate
532 * degree, and then twisted. We twist by three bits at a time because
533 * it's cheap to do so and helps slightly in the expected case where
534 * the entropy is concentrated in the low-order bits.
536 static void _mix_pool_bytes(struct entropy_store *r, const void *in,
539 unsigned long i, tap1, tap2, tap3, tap4, tap5;
541 int wordmask = r->poolinfo->poolwords - 1;
542 const char *bytes = in;
545 tap1 = r->poolinfo->tap1;
546 tap2 = r->poolinfo->tap2;
547 tap3 = r->poolinfo->tap3;
548 tap4 = r->poolinfo->tap4;
549 tap5 = r->poolinfo->tap5;
551 input_rotate = r->input_rotate;
554 /* mix one byte at a time to simplify size handling and churn faster */
556 w = rol32(*bytes++, input_rotate);
557 i = (i - 1) & wordmask;
559 /* XOR in the various taps */
561 w ^= r->pool[(i + tap1) & wordmask];
562 w ^= r->pool[(i + tap2) & wordmask];
563 w ^= r->pool[(i + tap3) & wordmask];
564 w ^= r->pool[(i + tap4) & wordmask];
565 w ^= r->pool[(i + tap5) & wordmask];
567 /* Mix the result back in with a twist */
568 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
571 * Normally, we add 7 bits of rotation to the pool.
572 * At the beginning of the pool, add an extra 7 bits
573 * rotation, so that successive passes spread the
574 * input bits across the pool evenly.
576 input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
579 r->input_rotate = input_rotate;
583 static void __mix_pool_bytes(struct entropy_store *r, const void *in,
586 trace_mix_pool_bytes_nolock(r->name, nbytes, _RET_IP_);
587 _mix_pool_bytes(r, in, nbytes);
590 static void mix_pool_bytes(struct entropy_store *r, const void *in,
595 trace_mix_pool_bytes(r->name, nbytes, _RET_IP_);
596 spin_lock_irqsave(&r->lock, flags);
597 _mix_pool_bytes(r, in, nbytes);
598 spin_unlock_irqrestore(&r->lock, flags);
604 unsigned short reg_idx;
609 * This is a fast mixing routine used by the interrupt randomness
610 * collector. It's hardcoded for an 128 bit pool and assumes that any
611 * locks that might be needed are taken by the caller.
613 static void fast_mix(struct fast_pool *f)
615 __u32 a = f->pool[0], b = f->pool[1];
616 __u32 c = f->pool[2], d = f->pool[3];
619 b = rol32(b, 6); d = rol32(d, 27);
623 b = rol32(b, 16); d = rol32(d, 14);
627 b = rol32(b, 6); d = rol32(d, 27);
631 b = rol32(b, 16); d = rol32(d, 14);
634 f->pool[0] = a; f->pool[1] = b;
635 f->pool[2] = c; f->pool[3] = d;
639 static void process_random_ready_list(void)
642 struct random_ready_callback *rdy, *tmp;
644 spin_lock_irqsave(&random_ready_list_lock, flags);
645 list_for_each_entry_safe(rdy, tmp, &random_ready_list, list) {
646 struct module *owner = rdy->owner;
648 list_del_init(&rdy->list);
652 spin_unlock_irqrestore(&random_ready_list_lock, flags);
656 * Credit (or debit) the entropy store with n bits of entropy.
657 * Use credit_entropy_bits_safe() if the value comes from userspace
658 * or otherwise should be checked for extreme values.
660 static void credit_entropy_bits(struct entropy_store *r, int nbits)
662 int entropy_count, orig;
663 const int pool_size = r->poolinfo->poolfracbits;
664 int nfrac = nbits << ENTROPY_SHIFT;
670 entropy_count = orig = READ_ONCE(r->entropy_count);
673 entropy_count += nfrac;
676 * Credit: we have to account for the possibility of
677 * overwriting already present entropy. Even in the
678 * ideal case of pure Shannon entropy, new contributions
679 * approach the full value asymptotically:
681 * entropy <- entropy + (pool_size - entropy) *
682 * (1 - exp(-add_entropy/pool_size))
684 * For add_entropy <= pool_size/2 then
685 * (1 - exp(-add_entropy/pool_size)) >=
686 * (add_entropy/pool_size)*0.7869...
687 * so we can approximate the exponential with
688 * 3/4*add_entropy/pool_size and still be on the
689 * safe side by adding at most pool_size/2 at a time.
691 * The use of pool_size-2 in the while statement is to
692 * prevent rounding artifacts from making the loop
693 * arbitrarily long; this limits the loop to log2(pool_size)*2
694 * turns no matter how large nbits is.
697 const int s = r->poolinfo->poolbitshift + ENTROPY_SHIFT + 2;
698 /* The +2 corresponds to the /4 in the denominator */
701 unsigned int anfrac = min(pnfrac, pool_size/2);
703 ((pool_size - entropy_count)*anfrac*3) >> s;
705 entropy_count += add;
707 } while (unlikely(entropy_count < pool_size-2 && pnfrac));
710 if (WARN_ON(entropy_count < 0)) {
711 pr_warn("negative entropy/overflow: pool %s count %d\n",
712 r->name, entropy_count);
714 } else if (entropy_count > pool_size)
715 entropy_count = pool_size;
716 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
719 trace_credit_entropy_bits(r->name, nbits,
720 entropy_count >> ENTROPY_SHIFT, _RET_IP_);
722 if (r == &input_pool) {
723 int entropy_bits = entropy_count >> ENTROPY_SHIFT;
725 if (crng_init < 2 && entropy_bits >= 128)
726 crng_reseed(&primary_crng, r);
730 static int credit_entropy_bits_safe(struct entropy_store *r, int nbits)
732 const int nbits_max = r->poolinfo->poolwords * 32;
737 /* Cap the value to avoid overflows */
738 nbits = min(nbits, nbits_max);
740 credit_entropy_bits(r, nbits);
744 /*********************************************************************
746 * CRNG using CHACHA20
748 *********************************************************************/
750 #define CRNG_RESEED_INTERVAL (300*HZ)
752 static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
756 * Hack to deal with crazy userspace progams when they are all trying
757 * to access /dev/urandom in parallel. The programs are almost
758 * certainly doing something terribly wrong, but we'll work around
759 * their brain damage.
761 static struct crng_state **crng_node_pool __read_mostly;
764 static void invalidate_batched_entropy(void);
765 static void numa_crng_init(void);
767 static bool trust_cpu __ro_after_init = IS_ENABLED(CONFIG_RANDOM_TRUST_CPU);
768 static int __init parse_trust_cpu(char *arg)
770 return kstrtobool(arg, &trust_cpu);
772 early_param("random.trust_cpu", parse_trust_cpu);
774 static bool crng_init_try_arch(struct crng_state *crng)
777 bool arch_init = true;
780 for (i = 4; i < 16; i++) {
781 if (!arch_get_random_seed_long(&rv) &&
782 !arch_get_random_long(&rv)) {
783 rv = random_get_entropy();
786 crng->state[i] ^= rv;
792 static bool __init crng_init_try_arch_early(struct crng_state *crng)
795 bool arch_init = true;
798 for (i = 4; i < 16; i++) {
799 if (!arch_get_random_seed_long_early(&rv) &&
800 !arch_get_random_long_early(&rv)) {
801 rv = random_get_entropy();
804 crng->state[i] ^= rv;
810 static void __maybe_unused crng_initialize_secondary(struct crng_state *crng)
812 chacha_init_consts(crng->state);
813 _get_random_bytes(&crng->state[4], sizeof(__u32) * 12);
814 crng_init_try_arch(crng);
815 crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
818 static void __init crng_initialize_primary(struct crng_state *crng)
820 chacha_init_consts(crng->state);
821 _extract_entropy(&input_pool, &crng->state[4], sizeof(__u32) * 12, 0);
822 if (crng_init_try_arch_early(crng) && trust_cpu) {
823 invalidate_batched_entropy();
826 pr_notice("crng done (trusting CPU's manufacturer)\n");
828 crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
832 static void do_numa_crng_init(struct work_struct *work)
835 struct crng_state *crng;
836 struct crng_state **pool;
838 pool = kcalloc(nr_node_ids, sizeof(*pool), GFP_KERNEL|__GFP_NOFAIL);
839 for_each_online_node(i) {
840 crng = kmalloc_node(sizeof(struct crng_state),
841 GFP_KERNEL | __GFP_NOFAIL, i);
842 spin_lock_init(&crng->lock);
843 crng_initialize_secondary(crng);
847 if (cmpxchg(&crng_node_pool, NULL, pool)) {
854 static DECLARE_WORK(numa_crng_init_work, do_numa_crng_init);
856 static void numa_crng_init(void)
858 schedule_work(&numa_crng_init_work);
861 static void numa_crng_init(void) {}
865 * crng_fast_load() can be called by code in the interrupt service
866 * path. So we can't afford to dilly-dally.
868 static int crng_fast_load(const char *cp, size_t len)
873 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
875 if (crng_init != 0) {
876 spin_unlock_irqrestore(&primary_crng.lock, flags);
879 p = (unsigned char *) &primary_crng.state[4];
880 while (len > 0 && crng_init_cnt < CRNG_INIT_CNT_THRESH) {
881 p[crng_init_cnt % CHACHA_KEY_SIZE] ^= *cp;
882 cp++; crng_init_cnt++; len--;
884 spin_unlock_irqrestore(&primary_crng.lock, flags);
885 if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
886 invalidate_batched_entropy();
888 pr_notice("fast init done\n");
894 * crng_slow_load() is called by add_device_randomness, which has two
895 * attributes. (1) We can't trust the buffer passed to it is
896 * guaranteed to be unpredictable (so it might not have any entropy at
897 * all), and (2) it doesn't have the performance constraints of
900 * So we do something more comprehensive which is guaranteed to touch
901 * all of the primary_crng's state, and which uses a LFSR with a
902 * period of 255 as part of the mixing algorithm. Finally, we do
903 * *not* advance crng_init_cnt since buffer we may get may be something
904 * like a fixed DMI table (for example), which might very well be
905 * unique to the machine, but is otherwise unvarying.
907 static int crng_slow_load(const char *cp, size_t len)
910 static unsigned char lfsr = 1;
912 unsigned i, max = CHACHA_KEY_SIZE;
913 const char * src_buf = cp;
914 char * dest_buf = (char *) &primary_crng.state[4];
916 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
918 if (crng_init != 0) {
919 spin_unlock_irqrestore(&primary_crng.lock, flags);
925 for (i = 0; i < max ; i++) {
930 tmp = dest_buf[i % CHACHA_KEY_SIZE];
931 dest_buf[i % CHACHA_KEY_SIZE] ^= src_buf[i % len] ^ lfsr;
932 lfsr += (tmp << 3) | (tmp >> 5);
934 spin_unlock_irqrestore(&primary_crng.lock, flags);
938 static void crng_reseed(struct crng_state *crng, struct entropy_store *r)
943 __u8 block[CHACHA_BLOCK_SIZE];
948 num = extract_entropy(r, &buf, 32, 16, 0);
952 _extract_crng(&primary_crng, buf.block);
953 _crng_backtrack_protect(&primary_crng, buf.block,
956 spin_lock_irqsave(&crng->lock, flags);
957 for (i = 0; i < 8; i++) {
959 if (!arch_get_random_seed_long(&rv) &&
960 !arch_get_random_long(&rv))
961 rv = random_get_entropy();
962 crng->state[i+4] ^= buf.key[i] ^ rv;
964 memzero_explicit(&buf, sizeof(buf));
965 crng->init_time = jiffies;
966 spin_unlock_irqrestore(&crng->lock, flags);
967 if (crng == &primary_crng && crng_init < 2) {
968 invalidate_batched_entropy();
971 process_random_ready_list();
972 wake_up_interruptible(&crng_init_wait);
973 kill_fasync(&fasync, SIGIO, POLL_IN);
974 pr_notice("crng init done\n");
975 if (unseeded_warning.missed) {
976 pr_notice("%d get_random_xx warning(s) missed due to ratelimiting\n",
977 unseeded_warning.missed);
978 unseeded_warning.missed = 0;
980 if (urandom_warning.missed) {
981 pr_notice("%d urandom warning(s) missed due to ratelimiting\n",
982 urandom_warning.missed);
983 urandom_warning.missed = 0;
988 static void _extract_crng(struct crng_state *crng,
989 __u8 out[CHACHA_BLOCK_SIZE])
991 unsigned long v, flags;
994 (time_after(crng_global_init_time, crng->init_time) ||
995 time_after(jiffies, crng->init_time + CRNG_RESEED_INTERVAL)))
996 crng_reseed(crng, crng == &primary_crng ? &input_pool : NULL);
997 spin_lock_irqsave(&crng->lock, flags);
998 if (arch_get_random_long(&v))
999 crng->state[14] ^= v;
1000 chacha20_block(&crng->state[0], out);
1001 if (crng->state[12] == 0)
1003 spin_unlock_irqrestore(&crng->lock, flags);
1006 static void extract_crng(__u8 out[CHACHA_BLOCK_SIZE])
1008 struct crng_state *crng = NULL;
1012 crng = crng_node_pool[numa_node_id()];
1015 crng = &primary_crng;
1016 _extract_crng(crng, out);
1020 * Use the leftover bytes from the CRNG block output (if there is
1021 * enough) to mutate the CRNG key to provide backtracking protection.
1023 static void _crng_backtrack_protect(struct crng_state *crng,
1024 __u8 tmp[CHACHA_BLOCK_SIZE], int used)
1026 unsigned long flags;
1030 used = round_up(used, sizeof(__u32));
1031 if (used + CHACHA_KEY_SIZE > CHACHA_BLOCK_SIZE) {
1035 spin_lock_irqsave(&crng->lock, flags);
1036 s = (__u32 *) &tmp[used];
1037 d = &crng->state[4];
1038 for (i=0; i < 8; i++)
1040 spin_unlock_irqrestore(&crng->lock, flags);
1043 static void crng_backtrack_protect(__u8 tmp[CHACHA_BLOCK_SIZE], int used)
1045 struct crng_state *crng = NULL;
1049 crng = crng_node_pool[numa_node_id()];
1052 crng = &primary_crng;
1053 _crng_backtrack_protect(crng, tmp, used);
1056 static ssize_t extract_crng_user(void __user *buf, size_t nbytes)
1058 ssize_t ret = 0, i = CHACHA_BLOCK_SIZE;
1059 __u8 tmp[CHACHA_BLOCK_SIZE] __aligned(4);
1060 int large_request = (nbytes > 256);
1063 if (large_request && need_resched()) {
1064 if (signal_pending(current)) {
1073 i = min_t(int, nbytes, CHACHA_BLOCK_SIZE);
1074 if (copy_to_user(buf, tmp, i)) {
1083 crng_backtrack_protect(tmp, i);
1085 /* Wipe data just written to memory */
1086 memzero_explicit(tmp, sizeof(tmp));
1092 /*********************************************************************
1094 * Entropy input management
1096 *********************************************************************/
1098 /* There is one of these per entropy source */
1099 struct timer_rand_state {
1101 long last_delta, last_delta2;
1104 #define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
1107 * Add device- or boot-specific data to the input pool to help
1110 * None of this adds any entropy; it is meant to avoid the problem of
1111 * the entropy pool having similar initial state across largely
1112 * identical devices.
1114 void add_device_randomness(const void *buf, unsigned int size)
1116 unsigned long time = random_get_entropy() ^ jiffies;
1117 unsigned long flags;
1119 if (!crng_ready() && size)
1120 crng_slow_load(buf, size);
1122 trace_add_device_randomness(size, _RET_IP_);
1123 spin_lock_irqsave(&input_pool.lock, flags);
1124 _mix_pool_bytes(&input_pool, buf, size);
1125 _mix_pool_bytes(&input_pool, &time, sizeof(time));
1126 spin_unlock_irqrestore(&input_pool.lock, flags);
1128 EXPORT_SYMBOL(add_device_randomness);
1130 static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
1133 * This function adds entropy to the entropy "pool" by using timing
1134 * delays. It uses the timer_rand_state structure to make an estimate
1135 * of how many bits of entropy this call has added to the pool.
1137 * The number "num" is also added to the pool - it should somehow describe
1138 * the type of event which just happened. This is currently 0-255 for
1139 * keyboard scan codes, and 256 upwards for interrupts.
1142 static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
1144 struct entropy_store *r;
1150 long delta, delta2, delta3;
1152 sample.jiffies = jiffies;
1153 sample.cycles = random_get_entropy();
1156 mix_pool_bytes(r, &sample, sizeof(sample));
1159 * Calculate number of bits of randomness we probably added.
1160 * We take into account the first, second and third-order deltas
1161 * in order to make our estimate.
1163 delta = sample.jiffies - READ_ONCE(state->last_time);
1164 WRITE_ONCE(state->last_time, sample.jiffies);
1166 delta2 = delta - READ_ONCE(state->last_delta);
1167 WRITE_ONCE(state->last_delta, delta);
1169 delta3 = delta2 - READ_ONCE(state->last_delta2);
1170 WRITE_ONCE(state->last_delta2, delta2);
1184 * delta is now minimum absolute delta.
1185 * Round down by 1 bit on general principles,
1186 * and limit entropy estimate to 12 bits.
1188 credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
1191 void add_input_randomness(unsigned int type, unsigned int code,
1194 static unsigned char last_value;
1196 /* ignore autorepeat and the like */
1197 if (value == last_value)
1201 add_timer_randomness(&input_timer_state,
1202 (type << 4) ^ code ^ (code >> 4) ^ value);
1203 trace_add_input_randomness(ENTROPY_BITS(&input_pool));
1205 EXPORT_SYMBOL_GPL(add_input_randomness);
1207 static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
1209 #ifdef ADD_INTERRUPT_BENCH
1210 static unsigned long avg_cycles, avg_deviation;
1212 #define AVG_SHIFT 8 /* Exponential average factor k=1/256 */
1213 #define FIXED_1_2 (1 << (AVG_SHIFT-1))
1215 static void add_interrupt_bench(cycles_t start)
1217 long delta = random_get_entropy() - start;
1219 /* Use a weighted moving average */
1220 delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
1221 avg_cycles += delta;
1222 /* And average deviation */
1223 delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
1224 avg_deviation += delta;
1227 #define add_interrupt_bench(x)
1230 static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
1232 __u32 *ptr = (__u32 *) regs;
1237 idx = READ_ONCE(f->reg_idx);
1238 if (idx >= sizeof(struct pt_regs) / sizeof(__u32))
1241 WRITE_ONCE(f->reg_idx, idx);
1245 void add_interrupt_randomness(int irq, int irq_flags)
1247 struct entropy_store *r;
1248 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1249 struct pt_regs *regs = get_irq_regs();
1250 unsigned long now = jiffies;
1251 cycles_t cycles = random_get_entropy();
1252 __u32 c_high, j_high;
1256 cycles = get_reg(fast_pool, regs);
1257 c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
1258 j_high = (sizeof(now) > 4) ? now >> 32 : 0;
1259 fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
1260 fast_pool->pool[1] ^= now ^ c_high;
1261 ip = regs ? instruction_pointer(regs) : _RET_IP_;
1262 fast_pool->pool[2] ^= ip;
1263 fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
1264 get_reg(fast_pool, regs);
1266 fast_mix(fast_pool);
1267 add_interrupt_bench(cycles);
1269 if (unlikely(crng_init == 0)) {
1270 if ((fast_pool->count >= 64) &&
1271 crng_fast_load((char *) fast_pool->pool,
1272 sizeof(fast_pool->pool))) {
1273 fast_pool->count = 0;
1274 fast_pool->last = now;
1279 if ((fast_pool->count < 64) &&
1280 !time_after(now, fast_pool->last + HZ))
1284 if (!spin_trylock(&r->lock))
1287 fast_pool->last = now;
1288 __mix_pool_bytes(r, &fast_pool->pool, sizeof(fast_pool->pool));
1289 spin_unlock(&r->lock);
1291 fast_pool->count = 0;
1293 /* award one bit for the contents of the fast pool */
1294 credit_entropy_bits(r, 1);
1296 EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1299 void add_disk_randomness(struct gendisk *disk)
1301 if (!disk || !disk->random)
1303 /* first major is 1, so we get >= 0x200 here */
1304 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1305 trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
1307 EXPORT_SYMBOL_GPL(add_disk_randomness);
1310 /*********************************************************************
1312 * Entropy extraction routines
1314 *********************************************************************/
1317 * This function decides how many bytes to actually take from the
1318 * given pool, and also debits the entropy count accordingly.
1320 static size_t account(struct entropy_store *r, size_t nbytes, int min,
1323 int entropy_count, orig, have_bytes;
1324 size_t ibytes, nfrac;
1326 BUG_ON(r->entropy_count > r->poolinfo->poolfracbits);
1328 /* Can we pull enough? */
1330 entropy_count = orig = READ_ONCE(r->entropy_count);
1332 /* never pull more than available */
1333 have_bytes = entropy_count >> (ENTROPY_SHIFT + 3);
1335 if ((have_bytes -= reserved) < 0)
1337 ibytes = min_t(size_t, ibytes, have_bytes);
1341 if (WARN_ON(entropy_count < 0)) {
1342 pr_warn("negative entropy count: pool %s count %d\n",
1343 r->name, entropy_count);
1346 nfrac = ibytes << (ENTROPY_SHIFT + 3);
1347 if ((size_t) entropy_count > nfrac)
1348 entropy_count -= nfrac;
1352 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
1355 trace_debit_entropy(r->name, 8 * ibytes);
1356 if (ibytes && ENTROPY_BITS(r) < random_write_wakeup_bits) {
1357 wake_up_interruptible(&random_write_wait);
1358 kill_fasync(&fasync, SIGIO, POLL_OUT);
1365 * This function does the actual extraction for extract_entropy.
1367 * Note: we assume that .poolwords is a multiple of 16 words.
1369 static void extract_buf(struct entropy_store *r, __u8 *out)
1374 unsigned long l[LONGS(20)];
1376 __u32 workspace[SHA1_WORKSPACE_WORDS];
1377 unsigned long flags;
1380 * If we have an architectural hardware random number
1381 * generator, use it for SHA's initial vector
1384 for (i = 0; i < LONGS(20); i++) {
1386 if (!arch_get_random_long(&v))
1391 /* Generate a hash across the pool, 16 words (512 bits) at a time */
1392 spin_lock_irqsave(&r->lock, flags);
1393 for (i = 0; i < r->poolinfo->poolwords; i += 16)
1394 sha1_transform(hash.w, (__u8 *)(r->pool + i), workspace);
1397 * We mix the hash back into the pool to prevent backtracking
1398 * attacks (where the attacker knows the state of the pool
1399 * plus the current outputs, and attempts to find previous
1400 * ouputs), unless the hash function can be inverted. By
1401 * mixing at least a SHA1 worth of hash data back, we make
1402 * brute-forcing the feedback as hard as brute-forcing the
1405 __mix_pool_bytes(r, hash.w, sizeof(hash.w));
1406 spin_unlock_irqrestore(&r->lock, flags);
1408 memzero_explicit(workspace, sizeof(workspace));
1411 * In case the hash function has some recognizable output
1412 * pattern, we fold it in half. Thus, we always feed back
1413 * twice as much data as we output.
1415 hash.w[0] ^= hash.w[3];
1416 hash.w[1] ^= hash.w[4];
1417 hash.w[2] ^= rol32(hash.w[2], 16);
1419 memcpy(out, &hash, EXTRACT_SIZE);
1420 memzero_explicit(&hash, sizeof(hash));
1423 static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
1424 size_t nbytes, int fips)
1427 __u8 tmp[EXTRACT_SIZE];
1428 unsigned long flags;
1431 extract_buf(r, tmp);
1434 spin_lock_irqsave(&r->lock, flags);
1435 if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
1436 panic("Hardware RNG duplicated output!\n");
1437 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1438 spin_unlock_irqrestore(&r->lock, flags);
1440 i = min_t(int, nbytes, EXTRACT_SIZE);
1441 memcpy(buf, tmp, i);
1447 /* Wipe data just returned from memory */
1448 memzero_explicit(tmp, sizeof(tmp));
1454 * This function extracts randomness from the "entropy pool", and
1455 * returns it in a buffer.
1457 * The min parameter specifies the minimum amount we can pull before
1458 * failing to avoid races that defeat catastrophic reseeding while the
1459 * reserved parameter indicates how much entropy we must leave in the
1460 * pool after each pull to avoid starving other readers.
1462 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1463 size_t nbytes, int min, int reserved)
1465 __u8 tmp[EXTRACT_SIZE];
1466 unsigned long flags;
1468 /* if last_data isn't primed, we need EXTRACT_SIZE extra bytes */
1470 spin_lock_irqsave(&r->lock, flags);
1471 if (!r->last_data_init) {
1472 r->last_data_init = 1;
1473 spin_unlock_irqrestore(&r->lock, flags);
1474 trace_extract_entropy(r->name, EXTRACT_SIZE,
1475 ENTROPY_BITS(r), _RET_IP_);
1476 extract_buf(r, tmp);
1477 spin_lock_irqsave(&r->lock, flags);
1478 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1480 spin_unlock_irqrestore(&r->lock, flags);
1483 trace_extract_entropy(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1484 nbytes = account(r, nbytes, min, reserved);
1486 return _extract_entropy(r, buf, nbytes, fips_enabled);
1489 #define warn_unseeded_randomness(previous) \
1490 _warn_unseeded_randomness(__func__, (void *) _RET_IP_, (previous))
1492 static void _warn_unseeded_randomness(const char *func_name, void *caller,
1495 #ifdef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1496 const bool print_once = false;
1498 static bool print_once __read_mostly;
1503 (previous && (caller == READ_ONCE(*previous))))
1505 WRITE_ONCE(*previous, caller);
1506 #ifndef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1509 if (__ratelimit(&unseeded_warning))
1510 printk_deferred(KERN_NOTICE "random: %s called from %pS "
1511 "with crng_init=%d\n", func_name, caller,
1516 * This function is the exported kernel interface. It returns some
1517 * number of good random numbers, suitable for key generation, seeding
1518 * TCP sequence numbers, etc. It does not rely on the hardware random
1519 * number generator. For random bytes direct from the hardware RNG
1520 * (when available), use get_random_bytes_arch(). In order to ensure
1521 * that the randomness provided by this function is okay, the function
1522 * wait_for_random_bytes() should be called and return 0 at least once
1523 * at any point prior.
1525 static void _get_random_bytes(void *buf, int nbytes)
1527 __u8 tmp[CHACHA_BLOCK_SIZE] __aligned(4);
1529 trace_get_random_bytes(nbytes, _RET_IP_);
1531 while (nbytes >= CHACHA_BLOCK_SIZE) {
1533 buf += CHACHA_BLOCK_SIZE;
1534 nbytes -= CHACHA_BLOCK_SIZE;
1539 memcpy(buf, tmp, nbytes);
1540 crng_backtrack_protect(tmp, nbytes);
1542 crng_backtrack_protect(tmp, CHACHA_BLOCK_SIZE);
1543 memzero_explicit(tmp, sizeof(tmp));
1546 void get_random_bytes(void *buf, int nbytes)
1548 static void *previous;
1550 warn_unseeded_randomness(&previous);
1551 _get_random_bytes(buf, nbytes);
1553 EXPORT_SYMBOL(get_random_bytes);
1557 * Each time the timer fires, we expect that we got an unpredictable
1558 * jump in the cycle counter. Even if the timer is running on another
1559 * CPU, the timer activity will be touching the stack of the CPU that is
1560 * generating entropy..
1562 * Note that we don't re-arm the timer in the timer itself - we are
1563 * happy to be scheduled away, since that just makes the load more
1564 * complex, but we do not want the timer to keep ticking unless the
1565 * entropy loop is running.
1567 * So the re-arming always happens in the entropy loop itself.
1569 static void entropy_timer(struct timer_list *t)
1571 credit_entropy_bits(&input_pool, 1);
1575 * If we have an actual cycle counter, see if we can
1576 * generate enough entropy with timing noise
1578 static void try_to_generate_entropy(void)
1582 struct timer_list timer;
1585 stack.now = random_get_entropy();
1587 /* Slow counter - or none. Don't even bother */
1588 if (stack.now == random_get_entropy())
1591 timer_setup_on_stack(&stack.timer, entropy_timer, 0);
1592 while (!crng_ready()) {
1593 if (!timer_pending(&stack.timer))
1594 mod_timer(&stack.timer, jiffies+1);
1595 mix_pool_bytes(&input_pool, &stack.now, sizeof(stack.now));
1597 stack.now = random_get_entropy();
1600 del_timer_sync(&stack.timer);
1601 destroy_timer_on_stack(&stack.timer);
1602 mix_pool_bytes(&input_pool, &stack.now, sizeof(stack.now));
1606 * Wait for the urandom pool to be seeded and thus guaranteed to supply
1607 * cryptographically secure random numbers. This applies to: the /dev/urandom
1608 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
1609 * family of functions. Using any of these functions without first calling
1610 * this function forfeits the guarantee of security.
1612 * Returns: 0 if the urandom pool has been seeded.
1613 * -ERESTARTSYS if the function was interrupted by a signal.
1615 int wait_for_random_bytes(void)
1617 if (likely(crng_ready()))
1622 ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ);
1624 return ret > 0 ? 0 : ret;
1626 try_to_generate_entropy();
1627 } while (!crng_ready());
1631 EXPORT_SYMBOL(wait_for_random_bytes);
1634 * Returns whether or not the urandom pool has been seeded and thus guaranteed
1635 * to supply cryptographically secure random numbers. This applies to: the
1636 * /dev/urandom device, the get_random_bytes function, and the get_random_{u32,
1637 * ,u64,int,long} family of functions.
1639 * Returns: true if the urandom pool has been seeded.
1640 * false if the urandom pool has not been seeded.
1642 bool rng_is_initialized(void)
1644 return crng_ready();
1646 EXPORT_SYMBOL(rng_is_initialized);
1649 * Add a callback function that will be invoked when the nonblocking
1650 * pool is initialised.
1652 * returns: 0 if callback is successfully added
1653 * -EALREADY if pool is already initialised (callback not called)
1654 * -ENOENT if module for callback is not alive
1656 int add_random_ready_callback(struct random_ready_callback *rdy)
1658 struct module *owner;
1659 unsigned long flags;
1660 int err = -EALREADY;
1666 if (!try_module_get(owner))
1669 spin_lock_irqsave(&random_ready_list_lock, flags);
1675 list_add(&rdy->list, &random_ready_list);
1679 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1685 EXPORT_SYMBOL(add_random_ready_callback);
1688 * Delete a previously registered readiness callback function.
1690 void del_random_ready_callback(struct random_ready_callback *rdy)
1692 unsigned long flags;
1693 struct module *owner = NULL;
1695 spin_lock_irqsave(&random_ready_list_lock, flags);
1696 if (!list_empty(&rdy->list)) {
1697 list_del_init(&rdy->list);
1700 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1704 EXPORT_SYMBOL(del_random_ready_callback);
1707 * This function will use the architecture-specific hardware random
1708 * number generator if it is available. The arch-specific hw RNG will
1709 * almost certainly be faster than what we can do in software, but it
1710 * is impossible to verify that it is implemented securely (as
1711 * opposed, to, say, the AES encryption of a sequence number using a
1712 * key known by the NSA). So it's useful if we need the speed, but
1713 * only if we're willing to trust the hardware manufacturer not to
1714 * have put in a back door.
1716 * Return number of bytes filled in.
1718 int __must_check get_random_bytes_arch(void *buf, int nbytes)
1723 trace_get_random_bytes_arch(left, _RET_IP_);
1726 int chunk = min_t(int, left, sizeof(unsigned long));
1728 if (!arch_get_random_long(&v))
1731 memcpy(p, &v, chunk);
1736 return nbytes - left;
1738 EXPORT_SYMBOL(get_random_bytes_arch);
1741 * init_std_data - initialize pool with system data
1743 * @r: pool to initialize
1745 * This function clears the pool's entropy count and mixes some system
1746 * data into the pool to prepare it for use. The pool is not cleared
1747 * as that can only decrease the entropy in the pool.
1749 static void __init init_std_data(struct entropy_store *r)
1752 ktime_t now = ktime_get_real();
1755 mix_pool_bytes(r, &now, sizeof(now));
1756 for (i = r->poolinfo->poolbytes; i > 0; i -= sizeof(rv)) {
1757 if (!arch_get_random_seed_long(&rv) &&
1758 !arch_get_random_long(&rv))
1759 rv = random_get_entropy();
1760 mix_pool_bytes(r, &rv, sizeof(rv));
1762 mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1766 * Note that setup_arch() may call add_device_randomness()
1767 * long before we get here. This allows seeding of the pools
1768 * with some platform dependent data very early in the boot
1769 * process. But it limits our options here. We must use
1770 * statically allocated structures that already have all
1771 * initializations complete at compile time. We should also
1772 * take care not to overwrite the precious per platform data
1775 int __init rand_initialize(void)
1777 init_std_data(&input_pool);
1778 crng_initialize_primary(&primary_crng);
1779 crng_global_init_time = jiffies;
1780 if (ratelimit_disable) {
1781 urandom_warning.interval = 0;
1782 unseeded_warning.interval = 0;
1788 void rand_initialize_disk(struct gendisk *disk)
1790 struct timer_rand_state *state;
1793 * If kzalloc returns null, we just won't use that entropy
1796 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1798 state->last_time = INITIAL_JIFFIES;
1799 disk->random = state;
1805 urandom_read_nowarn(struct file *file, char __user *buf, size_t nbytes,
1810 nbytes = min_t(size_t, nbytes, INT_MAX >> (ENTROPY_SHIFT + 3));
1811 ret = extract_crng_user(buf, nbytes);
1812 trace_urandom_read(8 * nbytes, 0, ENTROPY_BITS(&input_pool));
1817 urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1819 unsigned long flags;
1820 static int maxwarn = 10;
1822 if (!crng_ready() && maxwarn > 0) {
1824 if (__ratelimit(&urandom_warning))
1825 pr_notice("%s: uninitialized urandom read (%zd bytes read)\n",
1826 current->comm, nbytes);
1827 spin_lock_irqsave(&primary_crng.lock, flags);
1829 spin_unlock_irqrestore(&primary_crng.lock, flags);
1832 return urandom_read_nowarn(file, buf, nbytes, ppos);
1836 random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1840 ret = wait_for_random_bytes();
1843 return urandom_read_nowarn(file, buf, nbytes, ppos);
1847 random_poll(struct file *file, poll_table * wait)
1851 poll_wait(file, &crng_init_wait, wait);
1852 poll_wait(file, &random_write_wait, wait);
1855 mask |= EPOLLIN | EPOLLRDNORM;
1856 if (ENTROPY_BITS(&input_pool) < random_write_wakeup_bits)
1857 mask |= EPOLLOUT | EPOLLWRNORM;
1862 write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1866 const char __user *p = buffer;
1871 bytes = min(count, sizeof(buf));
1872 if (copy_from_user(&buf, p, bytes))
1875 for (b = bytes ; b > 0 ; b -= sizeof(__u32), i++) {
1876 if (!arch_get_random_int(&t))
1884 mix_pool_bytes(r, buf, bytes);
1891 static ssize_t random_write(struct file *file, const char __user *buffer,
1892 size_t count, loff_t *ppos)
1896 ret = write_pool(&input_pool, buffer, count);
1900 return (ssize_t)count;
1903 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1905 int size, ent_count;
1906 int __user *p = (int __user *)arg;
1911 /* inherently racy, no point locking */
1912 ent_count = ENTROPY_BITS(&input_pool);
1913 if (put_user(ent_count, p))
1916 case RNDADDTOENTCNT:
1917 if (!capable(CAP_SYS_ADMIN))
1919 if (get_user(ent_count, p))
1921 return credit_entropy_bits_safe(&input_pool, ent_count);
1923 if (!capable(CAP_SYS_ADMIN))
1925 if (get_user(ent_count, p++))
1929 if (get_user(size, p++))
1931 retval = write_pool(&input_pool, (const char __user *)p,
1935 return credit_entropy_bits_safe(&input_pool, ent_count);
1939 * Clear the entropy pool counters. We no longer clear
1940 * the entropy pool, as that's silly.
1942 if (!capable(CAP_SYS_ADMIN))
1944 input_pool.entropy_count = 0;
1947 if (!capable(CAP_SYS_ADMIN))
1951 crng_reseed(&primary_crng, &input_pool);
1952 crng_global_init_time = jiffies - 1;
1959 static int random_fasync(int fd, struct file *filp, int on)
1961 return fasync_helper(fd, filp, on, &fasync);
1964 const struct file_operations random_fops = {
1965 .read = random_read,
1966 .write = random_write,
1967 .poll = random_poll,
1968 .unlocked_ioctl = random_ioctl,
1969 .compat_ioctl = compat_ptr_ioctl,
1970 .fasync = random_fasync,
1971 .llseek = noop_llseek,
1974 const struct file_operations urandom_fops = {
1975 .read = urandom_read,
1976 .write = random_write,
1977 .unlocked_ioctl = random_ioctl,
1978 .compat_ioctl = compat_ptr_ioctl,
1979 .fasync = random_fasync,
1980 .llseek = noop_llseek,
1983 SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count,
1984 unsigned int, flags)
1988 if (flags & ~(GRND_NONBLOCK|GRND_RANDOM|GRND_INSECURE))
1992 * Requesting insecure and blocking randomness at the same time makes
1995 if ((flags & (GRND_INSECURE|GRND_RANDOM)) == (GRND_INSECURE|GRND_RANDOM))
1998 if (count > INT_MAX)
2001 if (!(flags & GRND_INSECURE) && !crng_ready()) {
2002 if (flags & GRND_NONBLOCK)
2004 ret = wait_for_random_bytes();
2008 return urandom_read_nowarn(NULL, buf, count, NULL);
2011 /********************************************************************
2015 ********************************************************************/
2017 #ifdef CONFIG_SYSCTL
2019 #include <linux/sysctl.h>
2021 static int min_write_thresh;
2022 static int max_write_thresh = INPUT_POOL_WORDS * 32;
2023 static int random_min_urandom_seed = 60;
2024 static char sysctl_bootid[16];
2027 * This function is used to return both the bootid UUID, and random
2028 * UUID. The difference is in whether table->data is NULL; if it is,
2029 * then a new UUID is generated and returned to the user.
2031 * If the user accesses this via the proc interface, the UUID will be
2032 * returned as an ASCII string in the standard UUID format; if via the
2033 * sysctl system call, as 16 bytes of binary data.
2035 static int proc_do_uuid(struct ctl_table *table, int write,
2036 void *buffer, size_t *lenp, loff_t *ppos)
2038 struct ctl_table fake_table;
2039 unsigned char buf[64], tmp_uuid[16], *uuid;
2044 generate_random_uuid(uuid);
2046 static DEFINE_SPINLOCK(bootid_spinlock);
2048 spin_lock(&bootid_spinlock);
2050 generate_random_uuid(uuid);
2051 spin_unlock(&bootid_spinlock);
2054 sprintf(buf, "%pU", uuid);
2056 fake_table.data = buf;
2057 fake_table.maxlen = sizeof(buf);
2059 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
2063 * Return entropy available scaled to integral bits
2065 static int proc_do_entropy(struct ctl_table *table, int write,
2066 void *buffer, size_t *lenp, loff_t *ppos)
2068 struct ctl_table fake_table;
2071 entropy_count = *(int *)table->data >> ENTROPY_SHIFT;
2073 fake_table.data = &entropy_count;
2074 fake_table.maxlen = sizeof(entropy_count);
2076 return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
2079 static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
2080 extern struct ctl_table random_table[];
2081 struct ctl_table random_table[] = {
2083 .procname = "poolsize",
2084 .data = &sysctl_poolsize,
2085 .maxlen = sizeof(int),
2087 .proc_handler = proc_dointvec,
2090 .procname = "entropy_avail",
2091 .maxlen = sizeof(int),
2093 .proc_handler = proc_do_entropy,
2094 .data = &input_pool.entropy_count,
2097 .procname = "write_wakeup_threshold",
2098 .data = &random_write_wakeup_bits,
2099 .maxlen = sizeof(int),
2101 .proc_handler = proc_dointvec_minmax,
2102 .extra1 = &min_write_thresh,
2103 .extra2 = &max_write_thresh,
2106 .procname = "urandom_min_reseed_secs",
2107 .data = &random_min_urandom_seed,
2108 .maxlen = sizeof(int),
2110 .proc_handler = proc_dointvec,
2113 .procname = "boot_id",
2114 .data = &sysctl_bootid,
2117 .proc_handler = proc_do_uuid,
2123 .proc_handler = proc_do_uuid,
2125 #ifdef ADD_INTERRUPT_BENCH
2127 .procname = "add_interrupt_avg_cycles",
2128 .data = &avg_cycles,
2129 .maxlen = sizeof(avg_cycles),
2131 .proc_handler = proc_doulongvec_minmax,
2134 .procname = "add_interrupt_avg_deviation",
2135 .data = &avg_deviation,
2136 .maxlen = sizeof(avg_deviation),
2138 .proc_handler = proc_doulongvec_minmax,
2143 #endif /* CONFIG_SYSCTL */
2145 struct batched_entropy {
2147 u64 entropy_u64[CHACHA_BLOCK_SIZE / sizeof(u64)];
2148 u32 entropy_u32[CHACHA_BLOCK_SIZE / sizeof(u32)];
2150 unsigned int position;
2151 spinlock_t batch_lock;
2155 * Get a random word for internal kernel use only. The quality of the random
2156 * number is good as /dev/urandom, but there is no backtrack protection, with
2157 * the goal of being quite fast and not depleting entropy. In order to ensure
2158 * that the randomness provided by this function is okay, the function
2159 * wait_for_random_bytes() should be called and return 0 at least once at any
2162 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64) = {
2163 .batch_lock = __SPIN_LOCK_UNLOCKED(batched_entropy_u64.lock),
2166 u64 get_random_u64(void)
2169 unsigned long flags;
2170 struct batched_entropy *batch;
2171 static void *previous;
2173 warn_unseeded_randomness(&previous);
2175 batch = raw_cpu_ptr(&batched_entropy_u64);
2176 spin_lock_irqsave(&batch->batch_lock, flags);
2177 if (batch->position % ARRAY_SIZE(batch->entropy_u64) == 0) {
2178 extract_crng((u8 *)batch->entropy_u64);
2179 batch->position = 0;
2181 ret = batch->entropy_u64[batch->position++];
2182 spin_unlock_irqrestore(&batch->batch_lock, flags);
2185 EXPORT_SYMBOL(get_random_u64);
2187 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32) = {
2188 .batch_lock = __SPIN_LOCK_UNLOCKED(batched_entropy_u32.lock),
2190 u32 get_random_u32(void)
2193 unsigned long flags;
2194 struct batched_entropy *batch;
2195 static void *previous;
2197 warn_unseeded_randomness(&previous);
2199 batch = raw_cpu_ptr(&batched_entropy_u32);
2200 spin_lock_irqsave(&batch->batch_lock, flags);
2201 if (batch->position % ARRAY_SIZE(batch->entropy_u32) == 0) {
2202 extract_crng((u8 *)batch->entropy_u32);
2203 batch->position = 0;
2205 ret = batch->entropy_u32[batch->position++];
2206 spin_unlock_irqrestore(&batch->batch_lock, flags);
2209 EXPORT_SYMBOL(get_random_u32);
2211 /* It's important to invalidate all potential batched entropy that might
2212 * be stored before the crng is initialized, which we can do lazily by
2213 * simply resetting the counter to zero so that it's re-extracted on the
2215 static void invalidate_batched_entropy(void)
2218 unsigned long flags;
2220 for_each_possible_cpu (cpu) {
2221 struct batched_entropy *batched_entropy;
2223 batched_entropy = per_cpu_ptr(&batched_entropy_u32, cpu);
2224 spin_lock_irqsave(&batched_entropy->batch_lock, flags);
2225 batched_entropy->position = 0;
2226 spin_unlock(&batched_entropy->batch_lock);
2228 batched_entropy = per_cpu_ptr(&batched_entropy_u64, cpu);
2229 spin_lock(&batched_entropy->batch_lock);
2230 batched_entropy->position = 0;
2231 spin_unlock_irqrestore(&batched_entropy->batch_lock, flags);
2236 * randomize_page - Generate a random, page aligned address
2237 * @start: The smallest acceptable address the caller will take.
2238 * @range: The size of the area, starting at @start, within which the
2239 * random address must fall.
2241 * If @start + @range would overflow, @range is capped.
2243 * NOTE: Historical use of randomize_range, which this replaces, presumed that
2244 * @start was already page aligned. We now align it regardless.
2246 * Return: A page aligned address within [start, start + range). On error,
2247 * @start is returned.
2250 randomize_page(unsigned long start, unsigned long range)
2252 if (!PAGE_ALIGNED(start)) {
2253 range -= PAGE_ALIGN(start) - start;
2254 start = PAGE_ALIGN(start);
2257 if (start > ULONG_MAX - range)
2258 range = ULONG_MAX - start;
2260 range >>= PAGE_SHIFT;
2265 return start + (get_random_long() % range << PAGE_SHIFT);
2268 /* Interface for in-kernel drivers of true hardware RNGs.
2269 * Those devices may produce endless random bits and will be throttled
2270 * when our pool is full.
2272 void add_hwgenerator_randomness(const char *buffer, size_t count,
2275 struct entropy_store *poolp = &input_pool;
2277 if (unlikely(crng_init == 0)) {
2278 crng_fast_load(buffer, count);
2282 /* Suspend writing if we're above the trickle threshold.
2283 * We'll be woken up again once below random_write_wakeup_thresh,
2284 * or when the calling thread is about to terminate.
2286 wait_event_interruptible(random_write_wait, kthread_should_stop() ||
2287 ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
2288 mix_pool_bytes(poolp, buffer, count);
2289 credit_entropy_bits(poolp, entropy);
2291 EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
2293 /* Handle random seed passed by bootloader.
2294 * If the seed is trustworthy, it would be regarded as hardware RNGs. Otherwise
2295 * it would be regarded as device data.
2296 * The decision is controlled by CONFIG_RANDOM_TRUST_BOOTLOADER.
2298 void add_bootloader_randomness(const void *buf, unsigned int size)
2300 if (IS_ENABLED(CONFIG_RANDOM_TRUST_BOOTLOADER))
2301 add_hwgenerator_randomness(buf, size, size * 8);
2303 add_device_randomness(buf, size);
2305 EXPORT_SYMBOL_GPL(add_bootloader_randomness);