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 three exported interfaces; the first is one designed to
105 * be used from within the kernel:
107 * void get_random_bytes(void *buf, int nbytes);
109 * This interface will return the requested number of random bytes,
110 * and place it in the requested buffer.
112 * The two other interfaces are two character devices /dev/random and
113 * /dev/urandom. /dev/random is suitable for use when very high
114 * quality randomness is desired (for example, for key generation or
115 * one-time pads), as it will only return a maximum of the number of
116 * bits of randomness (as estimated by the random number generator)
117 * contained in the entropy pool.
119 * The /dev/urandom device does not have this limit, and will return
120 * as many bytes as are requested. As more and more random bytes are
121 * requested without giving time for the entropy pool to recharge,
122 * this will result in random numbers that are merely cryptographically
123 * strong. For many applications, however, this is acceptable.
125 * Exported interfaces ---- input
126 * ==============================
128 * The current exported interfaces for gathering environmental noise
129 * from the devices are:
131 * void add_device_randomness(const void *buf, unsigned int size);
132 * void add_input_randomness(unsigned int type, unsigned int code,
133 * unsigned int value);
134 * void add_interrupt_randomness(int irq, int irq_flags);
135 * void add_disk_randomness(struct gendisk *disk);
137 * add_device_randomness() is for adding data to the random pool that
138 * is likely to differ between two devices (or possibly even per boot).
139 * This would be things like MAC addresses or serial numbers, or the
140 * read-out of the RTC. This does *not* add any actual entropy to the
141 * pool, but it initializes the pool to different values for devices
142 * that might otherwise be identical and have very little entropy
143 * available to them (particularly common in the embedded world).
145 * add_input_randomness() uses the input layer interrupt timing, as well as
146 * the event type information from the hardware.
148 * add_interrupt_randomness() uses the interrupt timing as random
149 * inputs to the entropy pool. Using the cycle counters and the irq source
150 * as inputs, it feeds the randomness roughly once a second.
152 * add_disk_randomness() uses what amounts to the seek time of block
153 * layer request events, on a per-disk_devt basis, as input to the
154 * entropy pool. Note that high-speed solid state drives with very low
155 * seek times do not make for good sources of entropy, as their seek
156 * times are usually fairly consistent.
158 * All of these routines try to estimate how many bits of randomness a
159 * particular randomness source. They do this by keeping track of the
160 * first and second order deltas of the event timings.
162 * Ensuring unpredictability at system startup
163 * ============================================
165 * When any operating system starts up, it will go through a sequence
166 * of actions that are fairly predictable by an adversary, especially
167 * if the start-up does not involve interaction with a human operator.
168 * This reduces the actual number of bits of unpredictability in the
169 * entropy pool below the value in entropy_count. In order to
170 * counteract this effect, it helps to carry information in the
171 * entropy pool across shut-downs and start-ups. To do this, put the
172 * following lines an appropriate script which is run during the boot
175 * echo "Initializing random number generator..."
176 * random_seed=/var/run/random-seed
177 * # Carry a random seed from start-up to start-up
178 * # Load and then save the whole entropy pool
179 * if [ -f $random_seed ]; then
180 * cat $random_seed >/dev/urandom
184 * chmod 600 $random_seed
185 * dd if=/dev/urandom of=$random_seed count=1 bs=512
187 * and the following lines in an appropriate script which is run as
188 * the system is shutdown:
190 * # Carry a random seed from shut-down to start-up
191 * # Save the whole entropy pool
192 * echo "Saving random seed..."
193 * random_seed=/var/run/random-seed
195 * chmod 600 $random_seed
196 * dd if=/dev/urandom of=$random_seed count=1 bs=512
198 * For example, on most modern systems using the System V init
199 * scripts, such code fragments would be found in
200 * /etc/rc.d/init.d/random. On older Linux systems, the correct script
201 * location might be in /etc/rcb.d/rc.local or /etc/rc.d/rc.0.
203 * Effectively, these commands cause the contents of the entropy pool
204 * to be saved at shut-down time and reloaded into the entropy pool at
205 * start-up. (The 'dd' in the addition to the bootup script is to
206 * make sure that /etc/random-seed is different for every start-up,
207 * even if the system crashes without executing rc.0.) Even with
208 * complete knowledge of the start-up activities, predicting the state
209 * of the entropy pool requires knowledge of the previous history of
212 * Configuring the /dev/random driver under Linux
213 * ==============================================
215 * The /dev/random driver under Linux uses minor numbers 8 and 9 of
216 * the /dev/mem major number (#1). So if your system does not have
217 * /dev/random and /dev/urandom created already, they can be created
218 * by using the commands:
220 * mknod /dev/random c 1 8
221 * mknod /dev/urandom c 1 9
226 * Ideas for constructing this random number generator were derived
227 * from Pretty Good Privacy's random number generator, and from private
228 * discussions with Phil Karn. Colin Plumb provided a faster random
229 * number generator, which speed up the mixing function of the entropy
230 * pool, taken from PGPfone. Dale Worley has also contributed many
231 * useful ideas and suggestions to improve this driver.
233 * Any flaws in the design are solely my responsibility, and should
234 * not be attributed to the Phil, Colin, or any of authors of PGP.
236 * Further background information on this topic may be obtained from
237 * RFC 1750, "Randomness Recommendations for Security", by Donald
238 * Eastlake, Steve Crocker, and Jeff Schiller.
241 #include <linux/utsname.h>
242 #include <linux/module.h>
243 #include <linux/kernel.h>
244 #include <linux/major.h>
245 #include <linux/string.h>
246 #include <linux/fcntl.h>
247 #include <linux/slab.h>
248 #include <linux/random.h>
249 #include <linux/poll.h>
250 #include <linux/init.h>
251 #include <linux/fs.h>
252 #include <linux/genhd.h>
253 #include <linux/interrupt.h>
254 #include <linux/mm.h>
255 #include <linux/nodemask.h>
256 #include <linux/spinlock.h>
257 #include <linux/kthread.h>
258 #include <linux/percpu.h>
259 #include <linux/cryptohash.h>
260 #include <linux/fips.h>
261 #include <linux/ptrace.h>
262 #include <linux/kmemcheck.h>
263 #include <linux/workqueue.h>
264 #include <linux/irq.h>
265 #include <linux/syscalls.h>
266 #include <linux/completion.h>
267 #include <linux/uuid.h>
268 #include <crypto/chacha20.h>
270 #include <asm/processor.h>
271 #include <linux/uaccess.h>
273 #include <asm/irq_regs.h>
276 #define CREATE_TRACE_POINTS
277 #include <trace/events/random.h>
279 /* #define ADD_INTERRUPT_BENCH */
282 * Configuration information
284 #define INPUT_POOL_SHIFT 12
285 #define INPUT_POOL_WORDS (1 << (INPUT_POOL_SHIFT-5))
286 #define OUTPUT_POOL_SHIFT 10
287 #define OUTPUT_POOL_WORDS (1 << (OUTPUT_POOL_SHIFT-5))
288 #define SEC_XFER_SIZE 512
289 #define EXTRACT_SIZE 10
292 #define LONGS(x) (((x) + sizeof(unsigned long) - 1)/sizeof(unsigned long))
295 * To allow fractional bits to be tracked, the entropy_count field is
296 * denominated in units of 1/8th bits.
298 * 2*(ENTROPY_SHIFT + log2(poolbits)) must <= 31, or the multiply in
299 * credit_entropy_bits() needs to be 64 bits wide.
301 #define ENTROPY_SHIFT 3
302 #define ENTROPY_BITS(r) ((r)->entropy_count >> ENTROPY_SHIFT)
305 * The minimum number of bits of entropy before we wake up a read on
306 * /dev/random. Should be enough to do a significant reseed.
308 static int random_read_wakeup_bits = 64;
311 * If the entropy count falls under this number of bits, then we
312 * should wake up processes which are selecting or polling on write
313 * access to /dev/random.
315 static int random_write_wakeup_bits = 28 * OUTPUT_POOL_WORDS;
318 * Originally, we used a primitive polynomial of degree .poolwords
319 * over GF(2). The taps for various sizes are defined below. They
320 * were chosen to be evenly spaced except for the last tap, which is 1
321 * to get the twisting happening as fast as possible.
323 * For the purposes of better mixing, we use the CRC-32 polynomial as
324 * well to make a (modified) twisted Generalized Feedback Shift
325 * Register. (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR
326 * generators. ACM Transactions on Modeling and Computer Simulation
327 * 2(3):179-194. Also see M. Matsumoto & Y. Kurita, 1994. Twisted
328 * GFSR generators II. ACM Transactions on Modeling and Computer
329 * Simulation 4:254-266)
331 * Thanks to Colin Plumb for suggesting this.
333 * The mixing operation is much less sensitive than the output hash,
334 * where we use SHA-1. All that we want of mixing operation is that
335 * it be a good non-cryptographic hash; i.e. it not produce collisions
336 * when fed "random" data of the sort we expect to see. As long as
337 * the pool state differs for different inputs, we have preserved the
338 * input entropy and done a good job. The fact that an intelligent
339 * attacker can construct inputs that will produce controlled
340 * alterations to the pool's state is not important because we don't
341 * consider such inputs to contribute any randomness. The only
342 * property we need with respect to them is that the attacker can't
343 * increase his/her knowledge of the pool's state. Since all
344 * additions are reversible (knowing the final state and the input,
345 * you can reconstruct the initial state), if an attacker has any
346 * uncertainty about the initial state, he/she can only shuffle that
347 * uncertainty about, but never cause any collisions (which would
348 * decrease the uncertainty).
350 * Our mixing functions were analyzed by Lacharme, Roeck, Strubel, and
351 * Videau in their paper, "The Linux Pseudorandom Number Generator
352 * Revisited" (see: http://eprint.iacr.org/2012/251.pdf). In their
353 * paper, they point out that we are not using a true Twisted GFSR,
354 * since Matsumoto & Kurita used a trinomial feedback polynomial (that
355 * is, with only three taps, instead of the six that we are using).
356 * As a result, the resulting polynomial is neither primitive nor
357 * irreducible, and hence does not have a maximal period over
358 * GF(2**32). They suggest a slight change to the generator
359 * polynomial which improves the resulting TGFSR polynomial to be
360 * irreducible, which we have made here.
362 static struct poolinfo {
363 int poolbitshift, poolwords, poolbytes, poolbits, poolfracbits;
364 #define S(x) ilog2(x)+5, (x), (x)*4, (x)*32, (x) << (ENTROPY_SHIFT+5)
365 int tap1, tap2, tap3, tap4, tap5;
366 } poolinfo_table[] = {
367 /* was: x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 */
368 /* x^128 + x^104 + x^76 + x^51 +x^25 + x + 1 */
369 { S(128), 104, 76, 51, 25, 1 },
370 /* was: x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 */
371 /* x^32 + x^26 + x^19 + x^14 + x^7 + x + 1 */
372 { S(32), 26, 19, 14, 7, 1 },
374 /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1 -- 115 */
375 { S(2048), 1638, 1231, 819, 411, 1 },
377 /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
378 { S(1024), 817, 615, 412, 204, 1 },
380 /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
381 { S(1024), 819, 616, 410, 207, 2 },
383 /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
384 { S(512), 411, 308, 208, 104, 1 },
386 /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
387 { S(512), 409, 307, 206, 102, 2 },
388 /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
389 { S(512), 409, 309, 205, 103, 2 },
391 /* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
392 { S(256), 205, 155, 101, 52, 1 },
394 /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
395 { S(128), 103, 78, 51, 27, 2 },
397 /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
398 { S(64), 52, 39, 26, 14, 1 },
403 * Static global variables
405 static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
406 static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
407 static struct fasync_struct *fasync;
409 static DEFINE_SPINLOCK(random_ready_list_lock);
410 static LIST_HEAD(random_ready_list);
414 unsigned long init_time;
418 struct crng_state primary_crng = {
419 .lock = __SPIN_LOCK_UNLOCKED(primary_crng.lock),
423 * crng_init = 0 --> Uninitialized
425 * 2 --> Initialized from input_pool
427 * crng_init is protected by primary_crng->lock, and only increases
428 * its value (from 0->1->2).
430 static int crng_init = 0;
431 #define crng_ready() (likely(crng_init > 0))
432 static int crng_init_cnt = 0;
433 #define CRNG_INIT_CNT_THRESH (2*CHACHA20_KEY_SIZE)
434 static void _extract_crng(struct crng_state *crng,
435 __u8 out[CHACHA20_BLOCK_SIZE]);
436 static void _crng_backtrack_protect(struct crng_state *crng,
437 __u8 tmp[CHACHA20_BLOCK_SIZE], int used);
438 static void process_random_ready_list(void);
440 /**********************************************************************
442 * OS independent entropy store. Here are the functions which handle
443 * storing entropy in an entropy pool.
445 **********************************************************************/
447 struct entropy_store;
448 struct entropy_store {
449 /* read-only data: */
450 const struct poolinfo *poolinfo;
453 struct entropy_store *pull;
454 struct work_struct push_work;
456 /* read-write data: */
457 unsigned long last_pulled;
459 unsigned short add_ptr;
460 unsigned short input_rotate;
463 unsigned int initialized:1;
464 unsigned int last_data_init:1;
465 __u8 last_data[EXTRACT_SIZE];
468 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
469 size_t nbytes, int min, int rsvd);
470 static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
471 size_t nbytes, int fips);
473 static void crng_reseed(struct crng_state *crng, struct entropy_store *r);
474 static void push_to_pool(struct work_struct *work);
475 static __u32 input_pool_data[INPUT_POOL_WORDS] __latent_entropy;
476 static __u32 blocking_pool_data[OUTPUT_POOL_WORDS] __latent_entropy;
478 static struct entropy_store input_pool = {
479 .poolinfo = &poolinfo_table[0],
481 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
482 .pool = input_pool_data
485 static struct entropy_store blocking_pool = {
486 .poolinfo = &poolinfo_table[1],
489 .lock = __SPIN_LOCK_UNLOCKED(blocking_pool.lock),
490 .pool = blocking_pool_data,
491 .push_work = __WORK_INITIALIZER(blocking_pool.push_work,
495 static __u32 const twist_table[8] = {
496 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
497 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
500 * This function adds bytes into the entropy "pool". It does not
501 * update the entropy estimate. The caller should call
502 * credit_entropy_bits if this is appropriate.
504 * The pool is stirred with a primitive polynomial of the appropriate
505 * degree, and then twisted. We twist by three bits at a time because
506 * it's cheap to do so and helps slightly in the expected case where
507 * the entropy is concentrated in the low-order bits.
509 static void _mix_pool_bytes(struct entropy_store *r, const void *in,
512 unsigned long i, tap1, tap2, tap3, tap4, tap5;
514 int wordmask = r->poolinfo->poolwords - 1;
515 const char *bytes = in;
518 tap1 = r->poolinfo->tap1;
519 tap2 = r->poolinfo->tap2;
520 tap3 = r->poolinfo->tap3;
521 tap4 = r->poolinfo->tap4;
522 tap5 = r->poolinfo->tap5;
524 input_rotate = r->input_rotate;
527 /* mix one byte at a time to simplify size handling and churn faster */
529 w = rol32(*bytes++, input_rotate);
530 i = (i - 1) & wordmask;
532 /* XOR in the various taps */
534 w ^= r->pool[(i + tap1) & wordmask];
535 w ^= r->pool[(i + tap2) & wordmask];
536 w ^= r->pool[(i + tap3) & wordmask];
537 w ^= r->pool[(i + tap4) & wordmask];
538 w ^= r->pool[(i + tap5) & wordmask];
540 /* Mix the result back in with a twist */
541 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
544 * Normally, we add 7 bits of rotation to the pool.
545 * At the beginning of the pool, add an extra 7 bits
546 * rotation, so that successive passes spread the
547 * input bits across the pool evenly.
549 input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
552 r->input_rotate = input_rotate;
556 static void __mix_pool_bytes(struct entropy_store *r, const void *in,
559 trace_mix_pool_bytes_nolock(r->name, nbytes, _RET_IP_);
560 _mix_pool_bytes(r, in, nbytes);
563 static void mix_pool_bytes(struct entropy_store *r, const void *in,
568 trace_mix_pool_bytes(r->name, nbytes, _RET_IP_);
569 spin_lock_irqsave(&r->lock, flags);
570 _mix_pool_bytes(r, in, nbytes);
571 spin_unlock_irqrestore(&r->lock, flags);
577 unsigned short reg_idx;
582 * This is a fast mixing routine used by the interrupt randomness
583 * collector. It's hardcoded for an 128 bit pool and assumes that any
584 * locks that might be needed are taken by the caller.
586 static void fast_mix(struct fast_pool *f)
588 __u32 a = f->pool[0], b = f->pool[1];
589 __u32 c = f->pool[2], d = f->pool[3];
592 b = rol32(b, 6); d = rol32(d, 27);
596 b = rol32(b, 16); d = rol32(d, 14);
600 b = rol32(b, 6); d = rol32(d, 27);
604 b = rol32(b, 16); d = rol32(d, 14);
607 f->pool[0] = a; f->pool[1] = b;
608 f->pool[2] = c; f->pool[3] = d;
612 static void process_random_ready_list(void)
615 struct random_ready_callback *rdy, *tmp;
617 spin_lock_irqsave(&random_ready_list_lock, flags);
618 list_for_each_entry_safe(rdy, tmp, &random_ready_list, list) {
619 struct module *owner = rdy->owner;
621 list_del_init(&rdy->list);
625 spin_unlock_irqrestore(&random_ready_list_lock, flags);
629 * Credit (or debit) the entropy store with n bits of entropy.
630 * Use credit_entropy_bits_safe() if the value comes from userspace
631 * or otherwise should be checked for extreme values.
633 static void credit_entropy_bits(struct entropy_store *r, int nbits)
635 int entropy_count, orig;
636 const int pool_size = r->poolinfo->poolfracbits;
637 int nfrac = nbits << ENTROPY_SHIFT;
643 entropy_count = orig = ACCESS_ONCE(r->entropy_count);
646 entropy_count += nfrac;
649 * Credit: we have to account for the possibility of
650 * overwriting already present entropy. Even in the
651 * ideal case of pure Shannon entropy, new contributions
652 * approach the full value asymptotically:
654 * entropy <- entropy + (pool_size - entropy) *
655 * (1 - exp(-add_entropy/pool_size))
657 * For add_entropy <= pool_size/2 then
658 * (1 - exp(-add_entropy/pool_size)) >=
659 * (add_entropy/pool_size)*0.7869...
660 * so we can approximate the exponential with
661 * 3/4*add_entropy/pool_size and still be on the
662 * safe side by adding at most pool_size/2 at a time.
664 * The use of pool_size-2 in the while statement is to
665 * prevent rounding artifacts from making the loop
666 * arbitrarily long; this limits the loop to log2(pool_size)*2
667 * turns no matter how large nbits is.
670 const int s = r->poolinfo->poolbitshift + ENTROPY_SHIFT + 2;
671 /* The +2 corresponds to the /4 in the denominator */
674 unsigned int anfrac = min(pnfrac, pool_size/2);
676 ((pool_size - entropy_count)*anfrac*3) >> s;
678 entropy_count += add;
680 } while (unlikely(entropy_count < pool_size-2 && pnfrac));
683 if (unlikely(entropy_count < 0)) {
684 pr_warn("random: negative entropy/overflow: pool %s count %d\n",
685 r->name, entropy_count);
688 } else if (entropy_count > pool_size)
689 entropy_count = pool_size;
690 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
693 r->entropy_total += nbits;
694 if (!r->initialized && r->entropy_total > 128) {
696 r->entropy_total = 0;
699 trace_credit_entropy_bits(r->name, nbits,
700 entropy_count >> ENTROPY_SHIFT,
701 r->entropy_total, _RET_IP_);
703 if (r == &input_pool) {
704 int entropy_bits = entropy_count >> ENTROPY_SHIFT;
706 if (crng_init < 2 && entropy_bits >= 128) {
707 crng_reseed(&primary_crng, r);
708 entropy_bits = r->entropy_count >> ENTROPY_SHIFT;
711 /* should we wake readers? */
712 if (entropy_bits >= random_read_wakeup_bits) {
713 wake_up_interruptible(&random_read_wait);
714 kill_fasync(&fasync, SIGIO, POLL_IN);
716 /* If the input pool is getting full, send some
717 * entropy to the blocking pool until it is 75% full.
719 if (entropy_bits > random_write_wakeup_bits &&
721 r->entropy_total >= 2*random_read_wakeup_bits) {
722 struct entropy_store *other = &blocking_pool;
724 if (other->entropy_count <=
725 3 * other->poolinfo->poolfracbits / 4) {
726 schedule_work(&other->push_work);
727 r->entropy_total = 0;
733 static int credit_entropy_bits_safe(struct entropy_store *r, int nbits)
735 const int nbits_max = (int)(~0U >> (ENTROPY_SHIFT + 1));
740 /* Cap the value to avoid overflows */
741 nbits = min(nbits, nbits_max);
743 credit_entropy_bits(r, nbits);
747 /*********************************************************************
749 * CRNG using CHACHA20
751 *********************************************************************/
753 #define CRNG_RESEED_INTERVAL (300*HZ)
755 static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
759 * Hack to deal with crazy userspace progams when they are all trying
760 * to access /dev/urandom in parallel. The programs are almost
761 * certainly doing something terribly wrong, but we'll work around
762 * their brain damage.
764 static struct crng_state **crng_node_pool __read_mostly;
767 static void invalidate_batched_entropy(void);
769 static void crng_initialize(struct crng_state *crng)
774 memcpy(&crng->state[0], "expand 32-byte k", 16);
775 if (crng == &primary_crng)
776 _extract_entropy(&input_pool, &crng->state[4],
777 sizeof(__u32) * 12, 0);
779 get_random_bytes(&crng->state[4], sizeof(__u32) * 12);
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();
784 crng->state[i] ^= rv;
786 crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
789 static int crng_fast_load(const char *cp, size_t len)
794 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
797 spin_unlock_irqrestore(&primary_crng.lock, flags);
800 p = (unsigned char *) &primary_crng.state[4];
801 while (len > 0 && crng_init_cnt < CRNG_INIT_CNT_THRESH) {
802 p[crng_init_cnt % CHACHA20_KEY_SIZE] ^= *cp;
803 cp++; crng_init_cnt++; len--;
805 spin_unlock_irqrestore(&primary_crng.lock, flags);
806 if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
807 invalidate_batched_entropy();
809 wake_up_interruptible(&crng_init_wait);
810 pr_notice("random: fast init done\n");
815 static void crng_reseed(struct crng_state *crng, struct entropy_store *r)
820 __u8 block[CHACHA20_BLOCK_SIZE];
825 num = extract_entropy(r, &buf, 32, 16, 0);
829 _extract_crng(&primary_crng, buf.block);
830 _crng_backtrack_protect(&primary_crng, buf.block,
833 spin_lock_irqsave(&primary_crng.lock, flags);
834 for (i = 0; i < 8; i++) {
836 if (!arch_get_random_seed_long(&rv) &&
837 !arch_get_random_long(&rv))
838 rv = random_get_entropy();
839 crng->state[i+4] ^= buf.key[i] ^ rv;
841 memzero_explicit(&buf, sizeof(buf));
842 crng->init_time = jiffies;
843 spin_unlock_irqrestore(&primary_crng.lock, flags);
844 if (crng == &primary_crng && crng_init < 2) {
845 invalidate_batched_entropy();
847 process_random_ready_list();
848 wake_up_interruptible(&crng_init_wait);
849 pr_notice("random: crng init done\n");
853 static void _extract_crng(struct crng_state *crng,
854 __u8 out[CHACHA20_BLOCK_SIZE])
856 unsigned long v, flags;
859 time_after(jiffies, crng->init_time + CRNG_RESEED_INTERVAL))
860 crng_reseed(crng, crng == &primary_crng ? &input_pool : NULL);
861 spin_lock_irqsave(&crng->lock, flags);
862 if (arch_get_random_long(&v))
863 crng->state[14] ^= v;
864 chacha20_block(&crng->state[0], out);
865 if (crng->state[12] == 0)
867 spin_unlock_irqrestore(&crng->lock, flags);
870 static void extract_crng(__u8 out[CHACHA20_BLOCK_SIZE])
872 struct crng_state *crng = NULL;
876 crng = crng_node_pool[numa_node_id()];
879 crng = &primary_crng;
880 _extract_crng(crng, out);
884 * Use the leftover bytes from the CRNG block output (if there is
885 * enough) to mutate the CRNG key to provide backtracking protection.
887 static void _crng_backtrack_protect(struct crng_state *crng,
888 __u8 tmp[CHACHA20_BLOCK_SIZE], int used)
894 used = round_up(used, sizeof(__u32));
895 if (used + CHACHA20_KEY_SIZE > CHACHA20_BLOCK_SIZE) {
899 spin_lock_irqsave(&crng->lock, flags);
900 s = (__u32 *) &tmp[used];
902 for (i=0; i < 8; i++)
904 spin_unlock_irqrestore(&crng->lock, flags);
907 static void crng_backtrack_protect(__u8 tmp[CHACHA20_BLOCK_SIZE], int used)
909 struct crng_state *crng = NULL;
913 crng = crng_node_pool[numa_node_id()];
916 crng = &primary_crng;
917 _crng_backtrack_protect(crng, tmp, used);
920 static ssize_t extract_crng_user(void __user *buf, size_t nbytes)
922 ssize_t ret = 0, i = CHACHA20_BLOCK_SIZE;
923 __u8 tmp[CHACHA20_BLOCK_SIZE];
924 int large_request = (nbytes > 256);
927 if (large_request && need_resched()) {
928 if (signal_pending(current)) {
937 i = min_t(int, nbytes, CHACHA20_BLOCK_SIZE);
938 if (copy_to_user(buf, tmp, i)) {
947 crng_backtrack_protect(tmp, i);
949 /* Wipe data just written to memory */
950 memzero_explicit(tmp, sizeof(tmp));
956 /*********************************************************************
958 * Entropy input management
960 *********************************************************************/
962 /* There is one of these per entropy source */
963 struct timer_rand_state {
965 long last_delta, last_delta2;
966 unsigned dont_count_entropy:1;
969 #define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
972 * Add device- or boot-specific data to the input pool to help
975 * None of this adds any entropy; it is meant to avoid the problem of
976 * the entropy pool having similar initial state across largely
979 void add_device_randomness(const void *buf, unsigned int size)
981 unsigned long time = random_get_entropy() ^ jiffies;
984 trace_add_device_randomness(size, _RET_IP_);
985 spin_lock_irqsave(&input_pool.lock, flags);
986 _mix_pool_bytes(&input_pool, buf, size);
987 _mix_pool_bytes(&input_pool, &time, sizeof(time));
988 spin_unlock_irqrestore(&input_pool.lock, flags);
990 EXPORT_SYMBOL(add_device_randomness);
992 static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
995 * This function adds entropy to the entropy "pool" by using timing
996 * delays. It uses the timer_rand_state structure to make an estimate
997 * of how many bits of entropy this call has added to the pool.
999 * The number "num" is also added to the pool - it should somehow describe
1000 * the type of event which just happened. This is currently 0-255 for
1001 * keyboard scan codes, and 256 upwards for interrupts.
1004 static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
1006 struct entropy_store *r;
1012 long delta, delta2, delta3;
1016 sample.jiffies = jiffies;
1017 sample.cycles = random_get_entropy();
1020 mix_pool_bytes(r, &sample, sizeof(sample));
1023 * Calculate number of bits of randomness we probably added.
1024 * We take into account the first, second and third-order deltas
1025 * in order to make our estimate.
1028 if (!state->dont_count_entropy) {
1029 delta = sample.jiffies - state->last_time;
1030 state->last_time = sample.jiffies;
1032 delta2 = delta - state->last_delta;
1033 state->last_delta = delta;
1035 delta3 = delta2 - state->last_delta2;
1036 state->last_delta2 = delta2;
1050 * delta is now minimum absolute delta.
1051 * Round down by 1 bit on general principles,
1052 * and limit entropy entimate to 12 bits.
1054 credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
1059 void add_input_randomness(unsigned int type, unsigned int code,
1062 static unsigned char last_value;
1064 /* ignore autorepeat and the like */
1065 if (value == last_value)
1069 add_timer_randomness(&input_timer_state,
1070 (type << 4) ^ code ^ (code >> 4) ^ value);
1071 trace_add_input_randomness(ENTROPY_BITS(&input_pool));
1073 EXPORT_SYMBOL_GPL(add_input_randomness);
1075 static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
1077 #ifdef ADD_INTERRUPT_BENCH
1078 static unsigned long avg_cycles, avg_deviation;
1080 #define AVG_SHIFT 8 /* Exponential average factor k=1/256 */
1081 #define FIXED_1_2 (1 << (AVG_SHIFT-1))
1083 static void add_interrupt_bench(cycles_t start)
1085 long delta = random_get_entropy() - start;
1087 /* Use a weighted moving average */
1088 delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
1089 avg_cycles += delta;
1090 /* And average deviation */
1091 delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
1092 avg_deviation += delta;
1095 #define add_interrupt_bench(x)
1098 static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
1100 __u32 *ptr = (__u32 *) regs;
1105 idx = READ_ONCE(f->reg_idx);
1106 if (idx >= sizeof(struct pt_regs) / sizeof(__u32))
1109 WRITE_ONCE(f->reg_idx, idx);
1113 void add_interrupt_randomness(int irq, int irq_flags)
1115 struct entropy_store *r;
1116 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1117 struct pt_regs *regs = get_irq_regs();
1118 unsigned long now = jiffies;
1119 cycles_t cycles = random_get_entropy();
1120 __u32 c_high, j_high;
1126 cycles = get_reg(fast_pool, regs);
1127 c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
1128 j_high = (sizeof(now) > 4) ? now >> 32 : 0;
1129 fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
1130 fast_pool->pool[1] ^= now ^ c_high;
1131 ip = regs ? instruction_pointer(regs) : _RET_IP_;
1132 fast_pool->pool[2] ^= ip;
1133 fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
1134 get_reg(fast_pool, regs);
1136 fast_mix(fast_pool);
1137 add_interrupt_bench(cycles);
1139 if (!crng_ready()) {
1140 if ((fast_pool->count >= 64) &&
1141 crng_fast_load((char *) fast_pool->pool,
1142 sizeof(fast_pool->pool))) {
1143 fast_pool->count = 0;
1144 fast_pool->last = now;
1149 if ((fast_pool->count < 64) &&
1150 !time_after(now, fast_pool->last + HZ))
1154 if (!spin_trylock(&r->lock))
1157 fast_pool->last = now;
1158 __mix_pool_bytes(r, &fast_pool->pool, sizeof(fast_pool->pool));
1161 * If we have architectural seed generator, produce a seed and
1162 * add it to the pool. For the sake of paranoia don't let the
1163 * architectural seed generator dominate the input from the
1166 if (arch_get_random_seed_long(&seed)) {
1167 __mix_pool_bytes(r, &seed, sizeof(seed));
1170 spin_unlock(&r->lock);
1172 fast_pool->count = 0;
1174 /* award one bit for the contents of the fast pool */
1175 credit_entropy_bits(r, credit + 1);
1177 EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1180 void add_disk_randomness(struct gendisk *disk)
1182 if (!disk || !disk->random)
1184 /* first major is 1, so we get >= 0x200 here */
1185 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1186 trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
1188 EXPORT_SYMBOL_GPL(add_disk_randomness);
1191 /*********************************************************************
1193 * Entropy extraction routines
1195 *********************************************************************/
1198 * This utility inline function is responsible for transferring entropy
1199 * from the primary pool to the secondary extraction pool. We make
1200 * sure we pull enough for a 'catastrophic reseed'.
1202 static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes);
1203 static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1206 r->entropy_count >= (nbytes << (ENTROPY_SHIFT + 3)) ||
1207 r->entropy_count > r->poolinfo->poolfracbits)
1210 _xfer_secondary_pool(r, nbytes);
1213 static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1215 __u32 tmp[OUTPUT_POOL_WORDS];
1219 /* pull at least as much as a wakeup */
1220 bytes = max_t(int, bytes, random_read_wakeup_bits / 8);
1221 /* but never more than the buffer size */
1222 bytes = min_t(int, bytes, sizeof(tmp));
1224 trace_xfer_secondary_pool(r->name, bytes * 8, nbytes * 8,
1225 ENTROPY_BITS(r), ENTROPY_BITS(r->pull));
1226 bytes = extract_entropy(r->pull, tmp, bytes,
1227 random_read_wakeup_bits / 8, 0);
1228 mix_pool_bytes(r, tmp, bytes);
1229 credit_entropy_bits(r, bytes*8);
1233 * Used as a workqueue function so that when the input pool is getting
1234 * full, we can "spill over" some entropy to the output pools. That
1235 * way the output pools can store some of the excess entropy instead
1236 * of letting it go to waste.
1238 static void push_to_pool(struct work_struct *work)
1240 struct entropy_store *r = container_of(work, struct entropy_store,
1243 _xfer_secondary_pool(r, random_read_wakeup_bits/8);
1244 trace_push_to_pool(r->name, r->entropy_count >> ENTROPY_SHIFT,
1245 r->pull->entropy_count >> ENTROPY_SHIFT);
1249 * This function decides how many bytes to actually take from the
1250 * given pool, and also debits the entropy count accordingly.
1252 static size_t account(struct entropy_store *r, size_t nbytes, int min,
1255 int entropy_count, orig, have_bytes;
1256 size_t ibytes, nfrac;
1258 BUG_ON(r->entropy_count > r->poolinfo->poolfracbits);
1260 /* Can we pull enough? */
1262 entropy_count = orig = ACCESS_ONCE(r->entropy_count);
1264 /* never pull more than available */
1265 have_bytes = entropy_count >> (ENTROPY_SHIFT + 3);
1267 if ((have_bytes -= reserved) < 0)
1269 ibytes = min_t(size_t, ibytes, have_bytes);
1273 if (unlikely(entropy_count < 0)) {
1274 pr_warn("random: negative entropy count: pool %s count %d\n",
1275 r->name, entropy_count);
1279 nfrac = ibytes << (ENTROPY_SHIFT + 3);
1280 if ((size_t) entropy_count > nfrac)
1281 entropy_count -= nfrac;
1285 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
1288 trace_debit_entropy(r->name, 8 * ibytes);
1290 (r->entropy_count >> ENTROPY_SHIFT) < random_write_wakeup_bits) {
1291 wake_up_interruptible(&random_write_wait);
1292 kill_fasync(&fasync, SIGIO, POLL_OUT);
1299 * This function does the actual extraction for extract_entropy and
1300 * extract_entropy_user.
1302 * Note: we assume that .poolwords is a multiple of 16 words.
1304 static void extract_buf(struct entropy_store *r, __u8 *out)
1309 unsigned long l[LONGS(20)];
1311 __u32 workspace[SHA_WORKSPACE_WORDS];
1312 unsigned long flags;
1315 * If we have an architectural hardware random number
1316 * generator, use it for SHA's initial vector
1319 for (i = 0; i < LONGS(20); i++) {
1321 if (!arch_get_random_long(&v))
1326 /* Generate a hash across the pool, 16 words (512 bits) at a time */
1327 spin_lock_irqsave(&r->lock, flags);
1328 for (i = 0; i < r->poolinfo->poolwords; i += 16)
1329 sha_transform(hash.w, (__u8 *)(r->pool + i), workspace);
1332 * We mix the hash back into the pool to prevent backtracking
1333 * attacks (where the attacker knows the state of the pool
1334 * plus the current outputs, and attempts to find previous
1335 * ouputs), unless the hash function can be inverted. By
1336 * mixing at least a SHA1 worth of hash data back, we make
1337 * brute-forcing the feedback as hard as brute-forcing the
1340 __mix_pool_bytes(r, hash.w, sizeof(hash.w));
1341 spin_unlock_irqrestore(&r->lock, flags);
1343 memzero_explicit(workspace, sizeof(workspace));
1346 * In case the hash function has some recognizable output
1347 * pattern, we fold it in half. Thus, we always feed back
1348 * twice as much data as we output.
1350 hash.w[0] ^= hash.w[3];
1351 hash.w[1] ^= hash.w[4];
1352 hash.w[2] ^= rol32(hash.w[2], 16);
1354 memcpy(out, &hash, EXTRACT_SIZE);
1355 memzero_explicit(&hash, sizeof(hash));
1358 static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
1359 size_t nbytes, int fips)
1362 __u8 tmp[EXTRACT_SIZE];
1363 unsigned long flags;
1366 extract_buf(r, tmp);
1369 spin_lock_irqsave(&r->lock, flags);
1370 if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
1371 panic("Hardware RNG duplicated output!\n");
1372 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1373 spin_unlock_irqrestore(&r->lock, flags);
1375 i = min_t(int, nbytes, EXTRACT_SIZE);
1376 memcpy(buf, tmp, i);
1382 /* Wipe data just returned from memory */
1383 memzero_explicit(tmp, sizeof(tmp));
1389 * This function extracts randomness from the "entropy pool", and
1390 * returns it in a buffer.
1392 * The min parameter specifies the minimum amount we can pull before
1393 * failing to avoid races that defeat catastrophic reseeding while the
1394 * reserved parameter indicates how much entropy we must leave in the
1395 * pool after each pull to avoid starving other readers.
1397 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1398 size_t nbytes, int min, int reserved)
1400 __u8 tmp[EXTRACT_SIZE];
1401 unsigned long flags;
1403 /* if last_data isn't primed, we need EXTRACT_SIZE extra bytes */
1405 spin_lock_irqsave(&r->lock, flags);
1406 if (!r->last_data_init) {
1407 r->last_data_init = 1;
1408 spin_unlock_irqrestore(&r->lock, flags);
1409 trace_extract_entropy(r->name, EXTRACT_SIZE,
1410 ENTROPY_BITS(r), _RET_IP_);
1411 xfer_secondary_pool(r, EXTRACT_SIZE);
1412 extract_buf(r, tmp);
1413 spin_lock_irqsave(&r->lock, flags);
1414 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1416 spin_unlock_irqrestore(&r->lock, flags);
1419 trace_extract_entropy(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1420 xfer_secondary_pool(r, nbytes);
1421 nbytes = account(r, nbytes, min, reserved);
1423 return _extract_entropy(r, buf, nbytes, fips_enabled);
1427 * This function extracts randomness from the "entropy pool", and
1428 * returns it in a userspace buffer.
1430 static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
1434 __u8 tmp[EXTRACT_SIZE];
1435 int large_request = (nbytes > 256);
1437 trace_extract_entropy_user(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1438 xfer_secondary_pool(r, nbytes);
1439 nbytes = account(r, nbytes, 0, 0);
1442 if (large_request && need_resched()) {
1443 if (signal_pending(current)) {
1451 extract_buf(r, tmp);
1452 i = min_t(int, nbytes, EXTRACT_SIZE);
1453 if (copy_to_user(buf, tmp, i)) {
1463 /* Wipe data just returned from memory */
1464 memzero_explicit(tmp, sizeof(tmp));
1470 * This function is the exported kernel interface. It returns some
1471 * number of good random numbers, suitable for key generation, seeding
1472 * TCP sequence numbers, etc. It does not rely on the hardware random
1473 * number generator. For random bytes direct from the hardware RNG
1474 * (when available), use get_random_bytes_arch(). In order to ensure
1475 * that the randomness provided by this function is okay, the function
1476 * wait_for_random_bytes() should be called and return 0 at least once
1477 * at any point prior.
1479 void get_random_bytes(void *buf, int nbytes)
1481 __u8 tmp[CHACHA20_BLOCK_SIZE];
1483 #ifdef CONFIG_WARN_UNSEEDED_RANDOM
1485 printk(KERN_NOTICE "random: %pF get_random_bytes called "
1486 "with crng_init = %d\n", (void *) _RET_IP_, crng_init);
1488 trace_get_random_bytes(nbytes, _RET_IP_);
1490 while (nbytes >= CHACHA20_BLOCK_SIZE) {
1492 buf += CHACHA20_BLOCK_SIZE;
1493 nbytes -= CHACHA20_BLOCK_SIZE;
1498 memcpy(buf, tmp, nbytes);
1499 crng_backtrack_protect(tmp, nbytes);
1501 crng_backtrack_protect(tmp, CHACHA20_BLOCK_SIZE);
1502 memzero_explicit(tmp, sizeof(tmp));
1504 EXPORT_SYMBOL(get_random_bytes);
1507 * Wait for the urandom pool to be seeded and thus guaranteed to supply
1508 * cryptographically secure random numbers. This applies to: the /dev/urandom
1509 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
1510 * family of functions. Using any of these functions without first calling
1511 * this function forfeits the guarantee of security.
1513 * Returns: 0 if the urandom pool has been seeded.
1514 * -ERESTARTSYS if the function was interrupted by a signal.
1516 int wait_for_random_bytes(void)
1518 if (likely(crng_ready()))
1520 return wait_event_interruptible(crng_init_wait, crng_ready());
1522 EXPORT_SYMBOL(wait_for_random_bytes);
1525 * Add a callback function that will be invoked when the nonblocking
1526 * pool is initialised.
1528 * returns: 0 if callback is successfully added
1529 * -EALREADY if pool is already initialised (callback not called)
1530 * -ENOENT if module for callback is not alive
1532 int add_random_ready_callback(struct random_ready_callback *rdy)
1534 struct module *owner;
1535 unsigned long flags;
1536 int err = -EALREADY;
1542 if (!try_module_get(owner))
1545 spin_lock_irqsave(&random_ready_list_lock, flags);
1551 list_add(&rdy->list, &random_ready_list);
1555 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1561 EXPORT_SYMBOL(add_random_ready_callback);
1564 * Delete a previously registered readiness callback function.
1566 void del_random_ready_callback(struct random_ready_callback *rdy)
1568 unsigned long flags;
1569 struct module *owner = NULL;
1571 spin_lock_irqsave(&random_ready_list_lock, flags);
1572 if (!list_empty(&rdy->list)) {
1573 list_del_init(&rdy->list);
1576 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1580 EXPORT_SYMBOL(del_random_ready_callback);
1583 * This function will use the architecture-specific hardware random
1584 * number generator if it is available. The arch-specific hw RNG will
1585 * almost certainly be faster than what we can do in software, but it
1586 * is impossible to verify that it is implemented securely (as
1587 * opposed, to, say, the AES encryption of a sequence number using a
1588 * key known by the NSA). So it's useful if we need the speed, but
1589 * only if we're willing to trust the hardware manufacturer not to
1590 * have put in a back door.
1592 void get_random_bytes_arch(void *buf, int nbytes)
1596 trace_get_random_bytes_arch(nbytes, _RET_IP_);
1599 int chunk = min(nbytes, (int)sizeof(unsigned long));
1601 if (!arch_get_random_long(&v))
1604 memcpy(p, &v, chunk);
1610 get_random_bytes(p, nbytes);
1612 EXPORT_SYMBOL(get_random_bytes_arch);
1616 * init_std_data - initialize pool with system data
1618 * @r: pool to initialize
1620 * This function clears the pool's entropy count and mixes some system
1621 * data into the pool to prepare it for use. The pool is not cleared
1622 * as that can only decrease the entropy in the pool.
1624 static void init_std_data(struct entropy_store *r)
1627 ktime_t now = ktime_get_real();
1630 r->last_pulled = jiffies;
1631 mix_pool_bytes(r, &now, sizeof(now));
1632 for (i = r->poolinfo->poolbytes; i > 0; i -= sizeof(rv)) {
1633 if (!arch_get_random_seed_long(&rv) &&
1634 !arch_get_random_long(&rv))
1635 rv = random_get_entropy();
1636 mix_pool_bytes(r, &rv, sizeof(rv));
1638 mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1642 * Note that setup_arch() may call add_device_randomness()
1643 * long before we get here. This allows seeding of the pools
1644 * with some platform dependent data very early in the boot
1645 * process. But it limits our options here. We must use
1646 * statically allocated structures that already have all
1647 * initializations complete at compile time. We should also
1648 * take care not to overwrite the precious per platform data
1651 static int rand_initialize(void)
1655 struct crng_state *crng;
1656 struct crng_state **pool;
1659 init_std_data(&input_pool);
1660 init_std_data(&blocking_pool);
1661 crng_initialize(&primary_crng);
1664 pool = kcalloc(nr_node_ids, sizeof(*pool), GFP_KERNEL|__GFP_NOFAIL);
1665 for_each_online_node(i) {
1666 crng = kmalloc_node(sizeof(struct crng_state),
1667 GFP_KERNEL | __GFP_NOFAIL, i);
1668 spin_lock_init(&crng->lock);
1669 crng_initialize(crng);
1673 crng_node_pool = pool;
1677 early_initcall(rand_initialize);
1680 void rand_initialize_disk(struct gendisk *disk)
1682 struct timer_rand_state *state;
1685 * If kzalloc returns null, we just won't use that entropy
1688 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1690 state->last_time = INITIAL_JIFFIES;
1691 disk->random = state;
1697 _random_read(int nonblock, char __user *buf, size_t nbytes)
1704 nbytes = min_t(size_t, nbytes, SEC_XFER_SIZE);
1706 n = extract_entropy_user(&blocking_pool, buf, nbytes);
1709 trace_random_read(n*8, (nbytes-n)*8,
1710 ENTROPY_BITS(&blocking_pool),
1711 ENTROPY_BITS(&input_pool));
1715 /* Pool is (near) empty. Maybe wait and retry. */
1719 wait_event_interruptible(random_read_wait,
1720 ENTROPY_BITS(&input_pool) >=
1721 random_read_wakeup_bits);
1722 if (signal_pending(current))
1723 return -ERESTARTSYS;
1728 random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1730 return _random_read(file->f_flags & O_NONBLOCK, buf, nbytes);
1734 urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1736 unsigned long flags;
1737 static int maxwarn = 10;
1740 if (!crng_ready() && maxwarn > 0) {
1742 printk(KERN_NOTICE "random: %s: uninitialized urandom read "
1743 "(%zd bytes read)\n",
1744 current->comm, nbytes);
1745 spin_lock_irqsave(&primary_crng.lock, flags);
1747 spin_unlock_irqrestore(&primary_crng.lock, flags);
1749 nbytes = min_t(size_t, nbytes, INT_MAX >> (ENTROPY_SHIFT + 3));
1750 ret = extract_crng_user(buf, nbytes);
1751 trace_urandom_read(8 * nbytes, 0, ENTROPY_BITS(&input_pool));
1756 random_poll(struct file *file, poll_table * wait)
1760 poll_wait(file, &random_read_wait, wait);
1761 poll_wait(file, &random_write_wait, wait);
1763 if (ENTROPY_BITS(&input_pool) >= random_read_wakeup_bits)
1764 mask |= POLLIN | POLLRDNORM;
1765 if (ENTROPY_BITS(&input_pool) < random_write_wakeup_bits)
1766 mask |= POLLOUT | POLLWRNORM;
1771 write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1775 const char __user *p = buffer;
1778 bytes = min(count, sizeof(buf));
1779 if (copy_from_user(&buf, p, bytes))
1785 mix_pool_bytes(r, buf, bytes);
1792 static ssize_t random_write(struct file *file, const char __user *buffer,
1793 size_t count, loff_t *ppos)
1797 ret = write_pool(&input_pool, buffer, count);
1801 return (ssize_t)count;
1804 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1806 int size, ent_count;
1807 int __user *p = (int __user *)arg;
1812 /* inherently racy, no point locking */
1813 ent_count = ENTROPY_BITS(&input_pool);
1814 if (put_user(ent_count, p))
1817 case RNDADDTOENTCNT:
1818 if (!capable(CAP_SYS_ADMIN))
1820 if (get_user(ent_count, p))
1822 return credit_entropy_bits_safe(&input_pool, ent_count);
1824 if (!capable(CAP_SYS_ADMIN))
1826 if (get_user(ent_count, p++))
1830 if (get_user(size, p++))
1832 retval = write_pool(&input_pool, (const char __user *)p,
1836 return credit_entropy_bits_safe(&input_pool, ent_count);
1840 * Clear the entropy pool counters. We no longer clear
1841 * the entropy pool, as that's silly.
1843 if (!capable(CAP_SYS_ADMIN))
1845 input_pool.entropy_count = 0;
1846 blocking_pool.entropy_count = 0;
1853 static int random_fasync(int fd, struct file *filp, int on)
1855 return fasync_helper(fd, filp, on, &fasync);
1858 const struct file_operations random_fops = {
1859 .read = random_read,
1860 .write = random_write,
1861 .poll = random_poll,
1862 .unlocked_ioctl = random_ioctl,
1863 .fasync = random_fasync,
1864 .llseek = noop_llseek,
1867 const struct file_operations urandom_fops = {
1868 .read = urandom_read,
1869 .write = random_write,
1870 .unlocked_ioctl = random_ioctl,
1871 .fasync = random_fasync,
1872 .llseek = noop_llseek,
1875 SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count,
1876 unsigned int, flags)
1880 if (flags & ~(GRND_NONBLOCK|GRND_RANDOM))
1883 if (count > INT_MAX)
1886 if (flags & GRND_RANDOM)
1887 return _random_read(flags & GRND_NONBLOCK, buf, count);
1889 if (!crng_ready()) {
1890 if (flags & GRND_NONBLOCK)
1892 ret = wait_for_random_bytes();
1896 return urandom_read(NULL, buf, count, NULL);
1899 /********************************************************************
1903 ********************************************************************/
1905 #ifdef CONFIG_SYSCTL
1907 #include <linux/sysctl.h>
1909 static int min_read_thresh = 8, min_write_thresh;
1910 static int max_read_thresh = OUTPUT_POOL_WORDS * 32;
1911 static int max_write_thresh = INPUT_POOL_WORDS * 32;
1912 static int random_min_urandom_seed = 60;
1913 static char sysctl_bootid[16];
1916 * This function is used to return both the bootid UUID, and random
1917 * UUID. The difference is in whether table->data is NULL; if it is,
1918 * then a new UUID is generated and returned to the user.
1920 * If the user accesses this via the proc interface, the UUID will be
1921 * returned as an ASCII string in the standard UUID format; if via the
1922 * sysctl system call, as 16 bytes of binary data.
1924 static int proc_do_uuid(struct ctl_table *table, int write,
1925 void __user *buffer, size_t *lenp, loff_t *ppos)
1927 struct ctl_table fake_table;
1928 unsigned char buf[64], tmp_uuid[16], *uuid;
1933 generate_random_uuid(uuid);
1935 static DEFINE_SPINLOCK(bootid_spinlock);
1937 spin_lock(&bootid_spinlock);
1939 generate_random_uuid(uuid);
1940 spin_unlock(&bootid_spinlock);
1943 sprintf(buf, "%pU", uuid);
1945 fake_table.data = buf;
1946 fake_table.maxlen = sizeof(buf);
1948 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
1952 * Return entropy available scaled to integral bits
1954 static int proc_do_entropy(struct ctl_table *table, int write,
1955 void __user *buffer, size_t *lenp, loff_t *ppos)
1957 struct ctl_table fake_table;
1960 entropy_count = *(int *)table->data >> ENTROPY_SHIFT;
1962 fake_table.data = &entropy_count;
1963 fake_table.maxlen = sizeof(entropy_count);
1965 return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
1968 static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
1969 extern struct ctl_table random_table[];
1970 struct ctl_table random_table[] = {
1972 .procname = "poolsize",
1973 .data = &sysctl_poolsize,
1974 .maxlen = sizeof(int),
1976 .proc_handler = proc_dointvec,
1979 .procname = "entropy_avail",
1980 .maxlen = sizeof(int),
1982 .proc_handler = proc_do_entropy,
1983 .data = &input_pool.entropy_count,
1986 .procname = "read_wakeup_threshold",
1987 .data = &random_read_wakeup_bits,
1988 .maxlen = sizeof(int),
1990 .proc_handler = proc_dointvec_minmax,
1991 .extra1 = &min_read_thresh,
1992 .extra2 = &max_read_thresh,
1995 .procname = "write_wakeup_threshold",
1996 .data = &random_write_wakeup_bits,
1997 .maxlen = sizeof(int),
1999 .proc_handler = proc_dointvec_minmax,
2000 .extra1 = &min_write_thresh,
2001 .extra2 = &max_write_thresh,
2004 .procname = "urandom_min_reseed_secs",
2005 .data = &random_min_urandom_seed,
2006 .maxlen = sizeof(int),
2008 .proc_handler = proc_dointvec,
2011 .procname = "boot_id",
2012 .data = &sysctl_bootid,
2015 .proc_handler = proc_do_uuid,
2021 .proc_handler = proc_do_uuid,
2023 #ifdef ADD_INTERRUPT_BENCH
2025 .procname = "add_interrupt_avg_cycles",
2026 .data = &avg_cycles,
2027 .maxlen = sizeof(avg_cycles),
2029 .proc_handler = proc_doulongvec_minmax,
2032 .procname = "add_interrupt_avg_deviation",
2033 .data = &avg_deviation,
2034 .maxlen = sizeof(avg_deviation),
2036 .proc_handler = proc_doulongvec_minmax,
2041 #endif /* CONFIG_SYSCTL */
2043 struct batched_entropy {
2045 u64 entropy_u64[CHACHA20_BLOCK_SIZE / sizeof(u64)];
2046 u32 entropy_u32[CHACHA20_BLOCK_SIZE / sizeof(u32)];
2048 unsigned int position;
2050 static rwlock_t batched_entropy_reset_lock = __RW_LOCK_UNLOCKED(batched_entropy_reset_lock);
2053 * Get a random word for internal kernel use only. The quality of the random
2054 * number is either as good as RDRAND or as good as /dev/urandom, with the
2055 * goal of being quite fast and not depleting entropy. In order to ensure
2056 * that the randomness provided by this function is okay, the function
2057 * wait_for_random_bytes() should be called and return 0 at least once
2058 * at any point prior.
2060 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64);
2061 u64 get_random_u64(void)
2064 bool use_lock = READ_ONCE(crng_init) < 2;
2065 unsigned long flags = 0;
2066 struct batched_entropy *batch;
2068 #if BITS_PER_LONG == 64
2069 if (arch_get_random_long((unsigned long *)&ret))
2072 if (arch_get_random_long((unsigned long *)&ret) &&
2073 arch_get_random_long((unsigned long *)&ret + 1))
2077 #ifdef CONFIG_WARN_UNSEEDED_RANDOM
2079 printk(KERN_NOTICE "random: %pF get_random_u64 called "
2080 "with crng_init = %d\n", (void *) _RET_IP_, crng_init);
2083 batch = &get_cpu_var(batched_entropy_u64);
2085 read_lock_irqsave(&batched_entropy_reset_lock, flags);
2086 if (batch->position % ARRAY_SIZE(batch->entropy_u64) == 0) {
2087 extract_crng((u8 *)batch->entropy_u64);
2088 batch->position = 0;
2090 ret = batch->entropy_u64[batch->position++];
2092 read_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2093 put_cpu_var(batched_entropy_u64);
2096 EXPORT_SYMBOL(get_random_u64);
2098 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32);
2099 u32 get_random_u32(void)
2102 bool use_lock = READ_ONCE(crng_init) < 2;
2103 unsigned long flags = 0;
2104 struct batched_entropy *batch;
2106 if (arch_get_random_int(&ret))
2109 #ifdef CONFIG_WARN_UNSEEDED_RANDOM
2111 printk(KERN_NOTICE "random: %pF get_random_u32 called "
2112 "with crng_init = %d\n", (void *) _RET_IP_, crng_init);
2115 batch = &get_cpu_var(batched_entropy_u32);
2117 read_lock_irqsave(&batched_entropy_reset_lock, flags);
2118 if (batch->position % ARRAY_SIZE(batch->entropy_u32) == 0) {
2119 extract_crng((u8 *)batch->entropy_u32);
2120 batch->position = 0;
2122 ret = batch->entropy_u32[batch->position++];
2124 read_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2125 put_cpu_var(batched_entropy_u32);
2128 EXPORT_SYMBOL(get_random_u32);
2130 /* It's important to invalidate all potential batched entropy that might
2131 * be stored before the crng is initialized, which we can do lazily by
2132 * simply resetting the counter to zero so that it's re-extracted on the
2134 static void invalidate_batched_entropy(void)
2137 unsigned long flags;
2139 write_lock_irqsave(&batched_entropy_reset_lock, flags);
2140 for_each_possible_cpu (cpu) {
2141 per_cpu_ptr(&batched_entropy_u32, cpu)->position = 0;
2142 per_cpu_ptr(&batched_entropy_u64, cpu)->position = 0;
2144 write_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2148 * randomize_page - Generate a random, page aligned address
2149 * @start: The smallest acceptable address the caller will take.
2150 * @range: The size of the area, starting at @start, within which the
2151 * random address must fall.
2153 * If @start + @range would overflow, @range is capped.
2155 * NOTE: Historical use of randomize_range, which this replaces, presumed that
2156 * @start was already page aligned. We now align it regardless.
2158 * Return: A page aligned address within [start, start + range). On error,
2159 * @start is returned.
2162 randomize_page(unsigned long start, unsigned long range)
2164 if (!PAGE_ALIGNED(start)) {
2165 range -= PAGE_ALIGN(start) - start;
2166 start = PAGE_ALIGN(start);
2169 if (start > ULONG_MAX - range)
2170 range = ULONG_MAX - start;
2172 range >>= PAGE_SHIFT;
2177 return start + (get_random_long() % range << PAGE_SHIFT);
2180 /* Interface for in-kernel drivers of true hardware RNGs.
2181 * Those devices may produce endless random bits and will be throttled
2182 * when our pool is full.
2184 void add_hwgenerator_randomness(const char *buffer, size_t count,
2187 struct entropy_store *poolp = &input_pool;
2189 if (!crng_ready()) {
2190 crng_fast_load(buffer, count);
2194 /* Suspend writing if we're above the trickle threshold.
2195 * We'll be woken up again once below random_write_wakeup_thresh,
2196 * or when the calling thread is about to terminate.
2198 wait_event_interruptible(random_write_wait, kthread_should_stop() ||
2199 ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
2200 mix_pool_bytes(poolp, buffer, count);
2201 credit_entropy_bits(poolp, entropy);
2203 EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);