2 * random.c -- A strong random number generator
4 * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005
6 * Copyright Theodore Ts'o, 1994, 1995, 1996, 1997, 1998, 1999. All
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, and the entire permission notice in its entirety,
14 * including the disclaimer of warranties.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. The name of the author may not be used to endorse or promote
19 * products derived from this software without specific prior
22 * ALTERNATIVELY, this product may be distributed under the terms of
23 * the GNU General Public License, in which case the provisions of the GPL are
24 * required INSTEAD OF the above restrictions. (This clause is
25 * necessary due to a potential bad interaction between the GPL and
26 * the restrictions contained in a BSD-style copyright.)
28 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
29 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
30 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ALL OF
31 * WHICH ARE HEREBY DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE
32 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
33 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT
34 * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
35 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
36 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
37 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
38 * USE OF THIS SOFTWARE, EVEN IF NOT ADVISED OF THE POSSIBILITY OF SUCH
43 * (now, with legal B.S. out of the way.....)
45 * This routine gathers environmental noise from device drivers, etc.,
46 * and returns good random numbers, suitable for cryptographic use.
47 * Besides the obvious cryptographic uses, these numbers are also good
48 * for seeding TCP sequence numbers, and other places where it is
49 * desirable to have numbers which are not only random, but hard to
50 * predict by an attacker.
55 * Computers are very predictable devices. Hence it is extremely hard
56 * to produce truly random numbers on a computer --- as opposed to
57 * pseudo-random numbers, which can easily generated by using a
58 * algorithm. Unfortunately, it is very easy for attackers to guess
59 * the sequence of pseudo-random number generators, and for some
60 * applications this is not acceptable. So instead, we must try to
61 * gather "environmental noise" from the computer's environment, which
62 * must be hard for outside attackers to observe, and use that to
63 * generate random numbers. In a Unix environment, this is best done
64 * from inside the kernel.
66 * Sources of randomness from the environment include inter-keyboard
67 * timings, inter-interrupt timings from some interrupts, and other
68 * events which are both (a) non-deterministic and (b) hard for an
69 * outside observer to measure. Randomness from these sources are
70 * added to an "entropy pool", which is mixed using a CRC-like function.
71 * This is not cryptographically strong, but it is adequate assuming
72 * the randomness is not chosen maliciously, and it is fast enough that
73 * the overhead of doing it on every interrupt is very reasonable.
74 * As random bytes are mixed into the entropy pool, the routines keep
75 * an *estimate* of how many bits of randomness have been stored into
76 * the random number generator's internal state.
78 * When random bytes are desired, they are obtained by taking the SHA
79 * hash of the contents of the "entropy pool". The SHA hash avoids
80 * exposing the internal state of the entropy pool. It is believed to
81 * be computationally infeasible to derive any useful information
82 * about the input of SHA from its output. Even if it is possible to
83 * analyze SHA in some clever way, as long as the amount of data
84 * returned from the generator is less than the inherent entropy in
85 * the pool, the output data is totally unpredictable. For this
86 * reason, the routine decreases its internal estimate of how many
87 * bits of "true randomness" are contained in the entropy pool as it
88 * outputs random numbers.
90 * If this estimate goes to zero, the routine can still generate
91 * random numbers; however, an attacker may (at least in theory) be
92 * able to infer the future output of the generator from prior
93 * outputs. This requires successful cryptanalysis of SHA, which is
94 * not believed to be feasible, but there is a remote possibility.
95 * Nonetheless, these numbers should be useful for the vast majority
98 * Exported interfaces ---- output
99 * ===============================
101 * There are three exported interfaces; the first is one designed to
102 * be used from within the kernel:
104 * void get_random_bytes(void *buf, int nbytes);
106 * This interface will return the requested number of random bytes,
107 * and place it in the requested buffer.
109 * The two other interfaces are two character devices /dev/random and
110 * /dev/urandom. /dev/random is suitable for use when very high
111 * quality randomness is desired (for example, for key generation or
112 * one-time pads), as it will only return a maximum of the number of
113 * bits of randomness (as estimated by the random number generator)
114 * contained in the entropy pool.
116 * The /dev/urandom device does not have this limit, and will return
117 * as many bytes as are requested. As more and more random bytes are
118 * requested without giving time for the entropy pool to recharge,
119 * this will result in random numbers that are merely cryptographically
120 * strong. For many applications, however, this is acceptable.
122 * Exported interfaces ---- input
123 * ==============================
125 * The current exported interfaces for gathering environmental noise
126 * from the devices are:
128 * void add_device_randomness(const void *buf, unsigned int size);
129 * void add_input_randomness(unsigned int type, unsigned int code,
130 * unsigned int value);
131 * void add_interrupt_randomness(int irq, int irq_flags);
132 * void add_disk_randomness(struct gendisk *disk);
134 * add_device_randomness() is for adding data to the random pool that
135 * is likely to differ between two devices (or possibly even per boot).
136 * This would be things like MAC addresses or serial numbers, or the
137 * read-out of the RTC. This does *not* add any actual entropy to the
138 * pool, but it initializes the pool to different values for devices
139 * that might otherwise be identical and have very little entropy
140 * available to them (particularly common in the embedded world).
142 * add_input_randomness() uses the input layer interrupt timing, as well as
143 * the event type information from the hardware.
145 * add_interrupt_randomness() uses the interrupt timing as random
146 * inputs to the entropy pool. Using the cycle counters and the irq source
147 * as inputs, it feeds the randomness roughly once a second.
149 * add_disk_randomness() uses what amounts to the seek time of block
150 * layer request events, on a per-disk_devt basis, as input to the
151 * entropy pool. Note that high-speed solid state drives with very low
152 * seek times do not make for good sources of entropy, as their seek
153 * times are usually fairly consistent.
155 * All of these routines try to estimate how many bits of randomness a
156 * particular randomness source. They do this by keeping track of the
157 * first and second order deltas of the event timings.
159 * Ensuring unpredictability at system startup
160 * ============================================
162 * When any operating system starts up, it will go through a sequence
163 * of actions that are fairly predictable by an adversary, especially
164 * if the start-up does not involve interaction with a human operator.
165 * This reduces the actual number of bits of unpredictability in the
166 * entropy pool below the value in entropy_count. In order to
167 * counteract this effect, it helps to carry information in the
168 * entropy pool across shut-downs and start-ups. To do this, put the
169 * following lines an appropriate script which is run during the boot
172 * echo "Initializing random number generator..."
173 * random_seed=/var/run/random-seed
174 * # Carry a random seed from start-up to start-up
175 * # Load and then save the whole entropy pool
176 * if [ -f $random_seed ]; then
177 * cat $random_seed >/dev/urandom
181 * chmod 600 $random_seed
182 * dd if=/dev/urandom of=$random_seed count=1 bs=512
184 * and the following lines in an appropriate script which is run as
185 * the system is shutdown:
187 * # Carry a random seed from shut-down to start-up
188 * # Save the whole entropy pool
189 * echo "Saving random seed..."
190 * random_seed=/var/run/random-seed
192 * chmod 600 $random_seed
193 * dd if=/dev/urandom of=$random_seed count=1 bs=512
195 * For example, on most modern systems using the System V init
196 * scripts, such code fragments would be found in
197 * /etc/rc.d/init.d/random. On older Linux systems, the correct script
198 * location might be in /etc/rcb.d/rc.local or /etc/rc.d/rc.0.
200 * Effectively, these commands cause the contents of the entropy pool
201 * to be saved at shut-down time and reloaded into the entropy pool at
202 * start-up. (The 'dd' in the addition to the bootup script is to
203 * make sure that /etc/random-seed is different for every start-up,
204 * even if the system crashes without executing rc.0.) Even with
205 * complete knowledge of the start-up activities, predicting the state
206 * of the entropy pool requires knowledge of the previous history of
209 * Configuring the /dev/random driver under Linux
210 * ==============================================
212 * The /dev/random driver under Linux uses minor numbers 8 and 9 of
213 * the /dev/mem major number (#1). So if your system does not have
214 * /dev/random and /dev/urandom created already, they can be created
215 * by using the commands:
217 * mknod /dev/random c 1 8
218 * mknod /dev/urandom c 1 9
223 * Ideas for constructing this random number generator were derived
224 * from Pretty Good Privacy's random number generator, and from private
225 * discussions with Phil Karn. Colin Plumb provided a faster random
226 * number generator, which speed up the mixing function of the entropy
227 * pool, taken from PGPfone. Dale Worley has also contributed many
228 * useful ideas and suggestions to improve this driver.
230 * Any flaws in the design are solely my responsibility, and should
231 * not be attributed to the Phil, Colin, or any of authors of PGP.
233 * Further background information on this topic may be obtained from
234 * RFC 1750, "Randomness Recommendations for Security", by Donald
235 * Eastlake, Steve Crocker, and Jeff Schiller.
238 #include <linux/utsname.h>
239 #include <linux/module.h>
240 #include <linux/kernel.h>
241 #include <linux/major.h>
242 #include <linux/string.h>
243 #include <linux/fcntl.h>
244 #include <linux/slab.h>
245 #include <linux/random.h>
246 #include <linux/poll.h>
247 #include <linux/init.h>
248 #include <linux/fs.h>
249 #include <linux/genhd.h>
250 #include <linux/interrupt.h>
251 #include <linux/mm.h>
252 #include <linux/nodemask.h>
253 #include <linux/spinlock.h>
254 #include <linux/kthread.h>
255 #include <linux/percpu.h>
256 #include <linux/cryptohash.h>
257 #include <linux/fips.h>
258 #include <linux/ptrace.h>
259 #include <linux/kmemcheck.h>
260 #include <linux/workqueue.h>
261 #include <linux/irq.h>
262 #include <linux/ratelimit.h>
263 #include <linux/syscalls.h>
264 #include <linux/completion.h>
265 #include <linux/uuid.h>
266 #include <crypto/chacha20.h>
268 #include <asm/processor.h>
269 #include <asm/uaccess.h>
271 #include <asm/irq_regs.h>
274 #define CREATE_TRACE_POINTS
275 #include <trace/events/random.h>
277 /* #define ADD_INTERRUPT_BENCH */
280 * Configuration information
282 #define INPUT_POOL_SHIFT 12
283 #define INPUT_POOL_WORDS (1 << (INPUT_POOL_SHIFT-5))
284 #define OUTPUT_POOL_SHIFT 10
285 #define OUTPUT_POOL_WORDS (1 << (OUTPUT_POOL_SHIFT-5))
286 #define SEC_XFER_SIZE 512
287 #define EXTRACT_SIZE 10
289 #define DEBUG_RANDOM_BOOT 0
291 #define LONGS(x) (((x) + sizeof(unsigned long) - 1)/sizeof(unsigned long))
294 * To allow fractional bits to be tracked, the entropy_count field is
295 * denominated in units of 1/8th bits.
297 * 2*(ENTROPY_SHIFT + log2(poolbits)) must <= 31, or the multiply in
298 * credit_entropy_bits() needs to be 64 bits wide.
300 #define ENTROPY_SHIFT 3
301 #define ENTROPY_BITS(r) ((r)->entropy_count >> ENTROPY_SHIFT)
304 * The minimum number of bits of entropy before we wake up a read on
305 * /dev/random. Should be enough to do a significant reseed.
307 static int random_read_wakeup_bits = 64;
310 * If the entropy count falls under this number of bits, then we
311 * should wake up processes which are selecting or polling on write
312 * access to /dev/random.
314 static int random_write_wakeup_bits = 28 * OUTPUT_POOL_WORDS;
317 * The minimum number of seconds between urandom pool reseeding. We
318 * do this to limit the amount of entropy that can be drained from the
319 * input pool even if there are heavy demands on /dev/urandom.
321 static int random_min_urandom_seed = 60;
324 * Originally, we used a primitive polynomial of degree .poolwords
325 * over GF(2). The taps for various sizes are defined below. They
326 * were chosen to be evenly spaced except for the last tap, which is 1
327 * to get the twisting happening as fast as possible.
329 * For the purposes of better mixing, we use the CRC-32 polynomial as
330 * well to make a (modified) twisted Generalized Feedback Shift
331 * Register. (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR
332 * generators. ACM Transactions on Modeling and Computer Simulation
333 * 2(3):179-194. Also see M. Matsumoto & Y. Kurita, 1994. Twisted
334 * GFSR generators II. ACM Transactions on Modeling and Computer
335 * Simulation 4:254-266)
337 * Thanks to Colin Plumb for suggesting this.
339 * The mixing operation is much less sensitive than the output hash,
340 * where we use SHA-1. All that we want of mixing operation is that
341 * it be a good non-cryptographic hash; i.e. it not produce collisions
342 * when fed "random" data of the sort we expect to see. As long as
343 * the pool state differs for different inputs, we have preserved the
344 * input entropy and done a good job. The fact that an intelligent
345 * attacker can construct inputs that will produce controlled
346 * alterations to the pool's state is not important because we don't
347 * consider such inputs to contribute any randomness. The only
348 * property we need with respect to them is that the attacker can't
349 * increase his/her knowledge of the pool's state. Since all
350 * additions are reversible (knowing the final state and the input,
351 * you can reconstruct the initial state), if an attacker has any
352 * uncertainty about the initial state, he/she can only shuffle that
353 * uncertainty about, but never cause any collisions (which would
354 * decrease the uncertainty).
356 * Our mixing functions were analyzed by Lacharme, Roeck, Strubel, and
357 * Videau in their paper, "The Linux Pseudorandom Number Generator
358 * Revisited" (see: http://eprint.iacr.org/2012/251.pdf). In their
359 * paper, they point out that we are not using a true Twisted GFSR,
360 * since Matsumoto & Kurita used a trinomial feedback polynomial (that
361 * is, with only three taps, instead of the six that we are using).
362 * As a result, the resulting polynomial is neither primitive nor
363 * irreducible, and hence does not have a maximal period over
364 * GF(2**32). They suggest a slight change to the generator
365 * polynomial which improves the resulting TGFSR polynomial to be
366 * irreducible, which we have made here.
368 static struct poolinfo {
369 int poolbitshift, poolwords, poolbytes, poolbits, poolfracbits;
370 #define S(x) ilog2(x)+5, (x), (x)*4, (x)*32, (x) << (ENTROPY_SHIFT+5)
371 int tap1, tap2, tap3, tap4, tap5;
372 } poolinfo_table[] = {
373 /* was: x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 */
374 /* x^128 + x^104 + x^76 + x^51 +x^25 + x + 1 */
375 { S(128), 104, 76, 51, 25, 1 },
376 /* was: x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 */
377 /* x^32 + x^26 + x^19 + x^14 + x^7 + x + 1 */
378 { S(32), 26, 19, 14, 7, 1 },
380 /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1 -- 115 */
381 { S(2048), 1638, 1231, 819, 411, 1 },
383 /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
384 { S(1024), 817, 615, 412, 204, 1 },
386 /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
387 { S(1024), 819, 616, 410, 207, 2 },
389 /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
390 { S(512), 411, 308, 208, 104, 1 },
392 /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
393 { S(512), 409, 307, 206, 102, 2 },
394 /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
395 { S(512), 409, 309, 205, 103, 2 },
397 /* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
398 { S(256), 205, 155, 101, 52, 1 },
400 /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
401 { S(128), 103, 78, 51, 27, 2 },
403 /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
404 { S(64), 52, 39, 26, 14, 1 },
409 * Static global variables
411 static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
412 static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
413 static DECLARE_WAIT_QUEUE_HEAD(urandom_init_wait);
414 static struct fasync_struct *fasync;
416 static DEFINE_SPINLOCK(random_ready_list_lock);
417 static LIST_HEAD(random_ready_list);
421 unsigned long init_time;
425 struct crng_state primary_crng = {
426 .lock = __SPIN_LOCK_UNLOCKED(primary_crng.lock),
430 * crng_init = 0 --> Uninitialized
432 * 2 --> Initialized from input_pool
434 * crng_init is protected by primary_crng->lock, and only increases
435 * its value (from 0->1->2).
437 static int crng_init = 0;
438 #define crng_ready() (likely(crng_init > 1))
439 static int crng_init_cnt = 0;
440 static unsigned long crng_global_init_time = 0;
441 #define CRNG_INIT_CNT_THRESH (2*CHACHA20_KEY_SIZE)
442 static void _extract_crng(struct crng_state *crng,
443 __u8 out[CHACHA20_BLOCK_SIZE]);
444 static void _crng_backtrack_protect(struct crng_state *crng,
445 __u8 tmp[CHACHA20_BLOCK_SIZE], int used);
446 static void process_random_ready_list(void);
448 static struct ratelimit_state unseeded_warning =
449 RATELIMIT_STATE_INIT("warn_unseeded_randomness", HZ, 3);
450 static struct ratelimit_state urandom_warning =
451 RATELIMIT_STATE_INIT("warn_urandom_randomness", HZ, 3);
453 static int ratelimit_disable __read_mostly;
455 module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
456 MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
458 /**********************************************************************
460 * OS independent entropy store. Here are the functions which handle
461 * storing entropy in an entropy pool.
463 **********************************************************************/
465 struct entropy_store;
466 struct entropy_store {
467 /* read-only data: */
468 const struct poolinfo *poolinfo;
471 struct entropy_store *pull;
472 struct work_struct push_work;
474 /* read-write data: */
475 unsigned long last_pulled;
477 unsigned short add_ptr;
478 unsigned short input_rotate;
481 unsigned int initialized:1;
482 unsigned int limit:1;
483 unsigned int last_data_init:1;
484 __u8 last_data[EXTRACT_SIZE];
487 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
488 size_t nbytes, int min, int rsvd);
489 static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
490 size_t nbytes, int fips);
492 static void crng_reseed(struct crng_state *crng, struct entropy_store *r);
493 static void push_to_pool(struct work_struct *work);
494 static __u32 input_pool_data[INPUT_POOL_WORDS] __latent_entropy;
495 static __u32 blocking_pool_data[OUTPUT_POOL_WORDS] __latent_entropy;
497 static struct entropy_store input_pool = {
498 .poolinfo = &poolinfo_table[0],
501 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
502 .pool = input_pool_data
505 static struct entropy_store blocking_pool = {
506 .poolinfo = &poolinfo_table[1],
510 .lock = __SPIN_LOCK_UNLOCKED(blocking_pool.lock),
511 .pool = blocking_pool_data,
512 .push_work = __WORK_INITIALIZER(blocking_pool.push_work,
516 static __u32 const twist_table[8] = {
517 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
518 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
521 * This function adds bytes into the entropy "pool". It does not
522 * update the entropy estimate. The caller should call
523 * credit_entropy_bits if this is appropriate.
525 * The pool is stirred with a primitive polynomial of the appropriate
526 * degree, and then twisted. We twist by three bits at a time because
527 * it's cheap to do so and helps slightly in the expected case where
528 * the entropy is concentrated in the low-order bits.
530 static void _mix_pool_bytes(struct entropy_store *r, const void *in,
533 unsigned long i, tap1, tap2, tap3, tap4, tap5;
535 int wordmask = r->poolinfo->poolwords - 1;
536 const char *bytes = in;
539 tap1 = r->poolinfo->tap1;
540 tap2 = r->poolinfo->tap2;
541 tap3 = r->poolinfo->tap3;
542 tap4 = r->poolinfo->tap4;
543 tap5 = r->poolinfo->tap5;
545 input_rotate = r->input_rotate;
548 /* mix one byte at a time to simplify size handling and churn faster */
550 w = rol32(*bytes++, input_rotate);
551 i = (i - 1) & wordmask;
553 /* XOR in the various taps */
555 w ^= r->pool[(i + tap1) & wordmask];
556 w ^= r->pool[(i + tap2) & wordmask];
557 w ^= r->pool[(i + tap3) & wordmask];
558 w ^= r->pool[(i + tap4) & wordmask];
559 w ^= r->pool[(i + tap5) & wordmask];
561 /* Mix the result back in with a twist */
562 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
565 * Normally, we add 7 bits of rotation to the pool.
566 * At the beginning of the pool, add an extra 7 bits
567 * rotation, so that successive passes spread the
568 * input bits across the pool evenly.
570 input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
573 r->input_rotate = input_rotate;
577 static void __mix_pool_bytes(struct entropy_store *r, const void *in,
580 trace_mix_pool_bytes_nolock(r->name, nbytes, _RET_IP_);
581 _mix_pool_bytes(r, in, nbytes);
584 static void mix_pool_bytes(struct entropy_store *r, const void *in,
589 trace_mix_pool_bytes(r->name, nbytes, _RET_IP_);
590 spin_lock_irqsave(&r->lock, flags);
591 _mix_pool_bytes(r, in, nbytes);
592 spin_unlock_irqrestore(&r->lock, flags);
598 unsigned short reg_idx;
603 * This is a fast mixing routine used by the interrupt randomness
604 * collector. It's hardcoded for an 128 bit pool and assumes that any
605 * locks that might be needed are taken by the caller.
607 static void fast_mix(struct fast_pool *f)
609 __u32 a = f->pool[0], b = f->pool[1];
610 __u32 c = f->pool[2], d = f->pool[3];
613 b = rol32(b, 6); d = rol32(d, 27);
617 b = rol32(b, 16); d = rol32(d, 14);
621 b = rol32(b, 6); d = rol32(d, 27);
625 b = rol32(b, 16); d = rol32(d, 14);
628 f->pool[0] = a; f->pool[1] = b;
629 f->pool[2] = c; f->pool[3] = d;
633 static void process_random_ready_list(void)
636 struct random_ready_callback *rdy, *tmp;
638 spin_lock_irqsave(&random_ready_list_lock, flags);
639 list_for_each_entry_safe(rdy, tmp, &random_ready_list, list) {
640 struct module *owner = rdy->owner;
642 list_del_init(&rdy->list);
646 spin_unlock_irqrestore(&random_ready_list_lock, flags);
650 * Credit (or debit) the entropy store with n bits of entropy.
651 * Use credit_entropy_bits_safe() if the value comes from userspace
652 * or otherwise should be checked for extreme values.
654 static void credit_entropy_bits(struct entropy_store *r, int nbits)
656 int entropy_count, orig;
657 const int pool_size = r->poolinfo->poolfracbits;
658 int nfrac = nbits << ENTROPY_SHIFT;
664 entropy_count = orig = ACCESS_ONCE(r->entropy_count);
667 entropy_count += nfrac;
670 * Credit: we have to account for the possibility of
671 * overwriting already present entropy. Even in the
672 * ideal case of pure Shannon entropy, new contributions
673 * approach the full value asymptotically:
675 * entropy <- entropy + (pool_size - entropy) *
676 * (1 - exp(-add_entropy/pool_size))
678 * For add_entropy <= pool_size/2 then
679 * (1 - exp(-add_entropy/pool_size)) >=
680 * (add_entropy/pool_size)*0.7869...
681 * so we can approximate the exponential with
682 * 3/4*add_entropy/pool_size and still be on the
683 * safe side by adding at most pool_size/2 at a time.
685 * The use of pool_size-2 in the while statement is to
686 * prevent rounding artifacts from making the loop
687 * arbitrarily long; this limits the loop to log2(pool_size)*2
688 * turns no matter how large nbits is.
691 const int s = r->poolinfo->poolbitshift + ENTROPY_SHIFT + 2;
692 /* The +2 corresponds to the /4 in the denominator */
695 unsigned int anfrac = min(pnfrac, pool_size/2);
697 ((pool_size - entropy_count)*anfrac*3) >> s;
699 entropy_count += add;
701 } while (unlikely(entropy_count < pool_size-2 && pnfrac));
704 if (unlikely(entropy_count < 0)) {
705 pr_warn("random: negative entropy/overflow: pool %s count %d\n",
706 r->name, entropy_count);
709 } else if (entropy_count > pool_size)
710 entropy_count = pool_size;
711 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
714 r->entropy_total += nbits;
715 if (!r->initialized && r->entropy_total > 128) {
717 r->entropy_total = 0;
720 trace_credit_entropy_bits(r->name, nbits,
721 entropy_count >> ENTROPY_SHIFT,
722 r->entropy_total, _RET_IP_);
724 if (r == &input_pool) {
725 int entropy_bits = entropy_count >> ENTROPY_SHIFT;
727 if (crng_init < 2 && entropy_bits >= 128) {
728 crng_reseed(&primary_crng, r);
729 entropy_bits = r->entropy_count >> ENTROPY_SHIFT;
732 /* should we wake readers? */
733 if (entropy_bits >= random_read_wakeup_bits) {
734 wake_up_interruptible(&random_read_wait);
735 kill_fasync(&fasync, SIGIO, POLL_IN);
737 /* If the input pool is getting full, send some
738 * entropy to the blocking pool until it is 75% full.
740 if (entropy_bits > random_write_wakeup_bits &&
742 r->entropy_total >= 2*random_read_wakeup_bits) {
743 struct entropy_store *other = &blocking_pool;
745 if (other->entropy_count <=
746 3 * other->poolinfo->poolfracbits / 4) {
747 schedule_work(&other->push_work);
748 r->entropy_total = 0;
754 static int credit_entropy_bits_safe(struct entropy_store *r, int nbits)
756 const int nbits_max = r->poolinfo->poolwords * 32;
761 /* Cap the value to avoid overflows */
762 nbits = min(nbits, nbits_max);
764 credit_entropy_bits(r, nbits);
768 /*********************************************************************
770 * CRNG using CHACHA20
772 *********************************************************************/
774 #define CRNG_RESEED_INTERVAL (300*HZ)
776 static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
780 * Hack to deal with crazy userspace progams when they are all trying
781 * to access /dev/urandom in parallel. The programs are almost
782 * certainly doing something terribly wrong, but we'll work around
783 * their brain damage.
785 static struct crng_state **crng_node_pool __read_mostly;
788 static void crng_initialize(struct crng_state *crng)
793 memcpy(&crng->state[0], "expand 32-byte k", 16);
794 if (crng == &primary_crng)
795 _extract_entropy(&input_pool, &crng->state[4],
796 sizeof(__u32) * 12, 0);
798 get_random_bytes(&crng->state[4], sizeof(__u32) * 12);
799 for (i = 4; i < 16; i++) {
800 if (!arch_get_random_seed_long(&rv) &&
801 !arch_get_random_long(&rv))
802 rv = random_get_entropy();
803 crng->state[i] ^= rv;
805 crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
808 static int crng_fast_load(const char *cp, size_t len)
813 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
815 if (crng_init != 0) {
816 spin_unlock_irqrestore(&primary_crng.lock, flags);
819 p = (unsigned char *) &primary_crng.state[4];
820 while (len > 0 && crng_init_cnt < CRNG_INIT_CNT_THRESH) {
821 p[crng_init_cnt % CHACHA20_KEY_SIZE] ^= *cp;
822 cp++; crng_init_cnt++; len--;
824 if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
826 wake_up_interruptible(&crng_init_wait);
827 pr_notice("random: fast init done\n");
829 spin_unlock_irqrestore(&primary_crng.lock, flags);
834 static void do_numa_crng_init(struct work_struct *work)
837 struct crng_state *crng;
838 struct crng_state **pool;
840 pool = kcalloc(nr_node_ids, sizeof(*pool), GFP_KERNEL|__GFP_NOFAIL);
841 for_each_online_node(i) {
842 crng = kmalloc_node(sizeof(struct crng_state),
843 GFP_KERNEL | __GFP_NOFAIL, i);
844 spin_lock_init(&crng->lock);
845 crng_initialize(crng);
848 /* pairs with READ_ONCE() in select_crng() */
849 if (cmpxchg_release(&crng_node_pool, NULL, pool) != NULL) {
856 static DECLARE_WORK(numa_crng_init_work, do_numa_crng_init);
858 static void numa_crng_init(void)
860 schedule_work(&numa_crng_init_work);
863 static struct crng_state *select_crng(void)
865 struct crng_state **pool;
866 int nid = numa_node_id();
868 /* pairs with cmpxchg_release() in do_numa_crng_init() */
869 pool = READ_ONCE(crng_node_pool);
870 if (pool && pool[nid])
873 return &primary_crng;
876 static void numa_crng_init(void) {}
878 static struct crng_state *select_crng(void)
880 return &primary_crng;
884 static void crng_reseed(struct crng_state *crng, struct entropy_store *r)
889 __u8 block[CHACHA20_BLOCK_SIZE];
894 num = extract_entropy(r, &buf, 32, 16, 0);
898 _extract_crng(&primary_crng, buf.block);
899 _crng_backtrack_protect(&primary_crng, buf.block,
902 spin_lock_irqsave(&crng->lock, flags);
903 for (i = 0; i < 8; i++) {
905 if (!arch_get_random_seed_long(&rv) &&
906 !arch_get_random_long(&rv))
907 rv = random_get_entropy();
908 crng->state[i+4] ^= buf.key[i] ^ rv;
910 memzero_explicit(&buf, sizeof(buf));
911 WRITE_ONCE(crng->init_time, jiffies);
912 if (crng == &primary_crng && crng_init < 2) {
915 process_random_ready_list();
916 wake_up_interruptible(&crng_init_wait);
917 pr_notice("random: crng init done\n");
918 if (unseeded_warning.missed) {
919 pr_notice("random: %d get_random_xx warning(s) missed "
920 "due to ratelimiting\n",
921 unseeded_warning.missed);
922 unseeded_warning.missed = 0;
924 if (urandom_warning.missed) {
925 pr_notice("random: %d urandom warning(s) missed "
926 "due to ratelimiting\n",
927 urandom_warning.missed);
928 urandom_warning.missed = 0;
931 spin_unlock_irqrestore(&crng->lock, flags);
934 static inline void maybe_reseed_primary_crng(void)
937 time_after(jiffies, primary_crng.init_time + CRNG_RESEED_INTERVAL))
938 crng_reseed(&primary_crng, &input_pool);
941 static inline void crng_wait_ready(void)
943 wait_event_interruptible(crng_init_wait, crng_ready());
946 static void _extract_crng(struct crng_state *crng,
947 __u8 out[CHACHA20_BLOCK_SIZE])
949 unsigned long v, flags, init_time;
952 init_time = READ_ONCE(crng->init_time);
953 if (time_after(READ_ONCE(crng_global_init_time), init_time) ||
954 time_after(jiffies, init_time + CRNG_RESEED_INTERVAL))
955 crng_reseed(crng, crng == &primary_crng ?
958 spin_lock_irqsave(&crng->lock, flags);
959 if (arch_get_random_long(&v))
960 crng->state[14] ^= v;
961 chacha20_block(&crng->state[0], out);
962 if (crng->state[12] == 0)
964 spin_unlock_irqrestore(&crng->lock, flags);
967 static void extract_crng(__u8 out[CHACHA20_BLOCK_SIZE])
969 _extract_crng(select_crng(), out);
973 * Use the leftover bytes from the CRNG block output (if there is
974 * enough) to mutate the CRNG key to provide backtracking protection.
976 static void _crng_backtrack_protect(struct crng_state *crng,
977 __u8 tmp[CHACHA20_BLOCK_SIZE], int used)
983 used = round_up(used, sizeof(__u32));
984 if (used + CHACHA20_KEY_SIZE > CHACHA20_BLOCK_SIZE) {
988 spin_lock_irqsave(&crng->lock, flags);
989 s = (__u32 *) &tmp[used];
991 for (i=0; i < 8; i++)
993 spin_unlock_irqrestore(&crng->lock, flags);
996 static void crng_backtrack_protect(__u8 tmp[CHACHA20_BLOCK_SIZE], int used)
998 _crng_backtrack_protect(select_crng(), tmp, used);
1001 static ssize_t extract_crng_user(void __user *buf, size_t nbytes)
1003 ssize_t ret = 0, i = CHACHA20_BLOCK_SIZE;
1004 __u8 tmp[CHACHA20_BLOCK_SIZE];
1005 int large_request = (nbytes > 256);
1008 if (large_request && need_resched()) {
1009 if (signal_pending(current)) {
1018 i = min_t(int, nbytes, CHACHA20_BLOCK_SIZE);
1019 if (copy_to_user(buf, tmp, i)) {
1028 crng_backtrack_protect(tmp, i);
1030 /* Wipe data just written to memory */
1031 memzero_explicit(tmp, sizeof(tmp));
1037 /*********************************************************************
1039 * Entropy input management
1041 *********************************************************************/
1043 /* There is one of these per entropy source */
1044 struct timer_rand_state {
1046 long last_delta, last_delta2;
1047 unsigned dont_count_entropy:1;
1050 #define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
1053 * Add device- or boot-specific data to the input pool to help
1056 * None of this adds any entropy; it is meant to avoid the problem of
1057 * the entropy pool having similar initial state across largely
1058 * identical devices.
1060 void add_device_randomness(const void *buf, unsigned int size)
1062 unsigned long time = random_get_entropy() ^ jiffies;
1063 unsigned long flags;
1065 trace_add_device_randomness(size, _RET_IP_);
1066 spin_lock_irqsave(&input_pool.lock, flags);
1067 _mix_pool_bytes(&input_pool, buf, size);
1068 _mix_pool_bytes(&input_pool, &time, sizeof(time));
1069 spin_unlock_irqrestore(&input_pool.lock, flags);
1071 EXPORT_SYMBOL(add_device_randomness);
1073 static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
1076 * This function adds entropy to the entropy "pool" by using timing
1077 * delays. It uses the timer_rand_state structure to make an estimate
1078 * of how many bits of entropy this call has added to the pool.
1080 * The number "num" is also added to the pool - it should somehow describe
1081 * the type of event which just happened. This is currently 0-255 for
1082 * keyboard scan codes, and 256 upwards for interrupts.
1085 static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
1087 struct entropy_store *r;
1093 long delta, delta2, delta3;
1097 sample.jiffies = jiffies;
1098 sample.cycles = random_get_entropy();
1101 mix_pool_bytes(r, &sample, sizeof(sample));
1104 * Calculate number of bits of randomness we probably added.
1105 * We take into account the first, second and third-order deltas
1106 * in order to make our estimate.
1109 if (!state->dont_count_entropy) {
1110 delta = sample.jiffies - state->last_time;
1111 state->last_time = sample.jiffies;
1113 delta2 = delta - state->last_delta;
1114 state->last_delta = delta;
1116 delta3 = delta2 - state->last_delta2;
1117 state->last_delta2 = delta2;
1131 * delta is now minimum absolute delta.
1132 * Round down by 1 bit on general principles,
1133 * and limit entropy entimate to 12 bits.
1135 credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
1140 void add_input_randomness(unsigned int type, unsigned int code,
1143 static unsigned char last_value;
1145 /* ignore autorepeat and the like */
1146 if (value == last_value)
1150 add_timer_randomness(&input_timer_state,
1151 (type << 4) ^ code ^ (code >> 4) ^ value);
1152 trace_add_input_randomness(ENTROPY_BITS(&input_pool));
1154 EXPORT_SYMBOL_GPL(add_input_randomness);
1156 static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
1158 #ifdef ADD_INTERRUPT_BENCH
1159 static unsigned long avg_cycles, avg_deviation;
1161 #define AVG_SHIFT 8 /* Exponential average factor k=1/256 */
1162 #define FIXED_1_2 (1 << (AVG_SHIFT-1))
1164 static void add_interrupt_bench(cycles_t start)
1166 long delta = random_get_entropy() - start;
1168 /* Use a weighted moving average */
1169 delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
1170 avg_cycles += delta;
1171 /* And average deviation */
1172 delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
1173 avg_deviation += delta;
1176 #define add_interrupt_bench(x)
1179 static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
1181 __u32 *ptr = (__u32 *) regs;
1186 idx = READ_ONCE(f->reg_idx);
1187 if (idx >= sizeof(struct pt_regs) / sizeof(__u32))
1190 WRITE_ONCE(f->reg_idx, idx);
1194 void add_interrupt_randomness(int irq, int irq_flags)
1196 struct entropy_store *r;
1197 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1198 struct pt_regs *regs = get_irq_regs();
1199 unsigned long now = jiffies;
1200 cycles_t cycles = random_get_entropy();
1201 __u32 c_high, j_high;
1207 cycles = get_reg(fast_pool, regs);
1208 c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
1209 j_high = (sizeof(now) > 4) ? now >> 32 : 0;
1210 fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
1211 fast_pool->pool[1] ^= now ^ c_high;
1212 ip = regs ? instruction_pointer(regs) : _RET_IP_;
1213 fast_pool->pool[2] ^= ip;
1214 fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
1215 get_reg(fast_pool, regs);
1217 fast_mix(fast_pool);
1218 add_interrupt_bench(cycles);
1220 if (unlikely(crng_init == 0)) {
1221 if ((fast_pool->count >= 64) &&
1222 crng_fast_load((char *) fast_pool->pool,
1223 sizeof(fast_pool->pool))) {
1224 fast_pool->count = 0;
1225 fast_pool->last = now;
1230 if ((fast_pool->count < 64) &&
1231 !time_after(now, fast_pool->last + HZ))
1235 if (!spin_trylock(&r->lock))
1238 fast_pool->last = now;
1239 __mix_pool_bytes(r, &fast_pool->pool, sizeof(fast_pool->pool));
1242 * If we have architectural seed generator, produce a seed and
1243 * add it to the pool. For the sake of paranoia don't let the
1244 * architectural seed generator dominate the input from the
1247 if (arch_get_random_seed_long(&seed)) {
1248 __mix_pool_bytes(r, &seed, sizeof(seed));
1251 spin_unlock(&r->lock);
1253 fast_pool->count = 0;
1255 /* award one bit for the contents of the fast pool */
1256 credit_entropy_bits(r, credit + 1);
1258 EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1261 void add_disk_randomness(struct gendisk *disk)
1263 if (!disk || !disk->random)
1265 /* first major is 1, so we get >= 0x200 here */
1266 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1267 trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
1269 EXPORT_SYMBOL_GPL(add_disk_randomness);
1272 /*********************************************************************
1274 * Entropy extraction routines
1276 *********************************************************************/
1279 * This utility inline function is responsible for transferring entropy
1280 * from the primary pool to the secondary extraction pool. We make
1281 * sure we pull enough for a 'catastrophic reseed'.
1283 static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes);
1284 static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1287 r->entropy_count >= (nbytes << (ENTROPY_SHIFT + 3)) ||
1288 r->entropy_count > r->poolinfo->poolfracbits)
1291 if (r->limit == 0 && random_min_urandom_seed) {
1292 unsigned long now = jiffies;
1294 if (time_before(now,
1295 r->last_pulled + random_min_urandom_seed * HZ))
1297 r->last_pulled = now;
1300 _xfer_secondary_pool(r, nbytes);
1303 static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1305 __u32 tmp[OUTPUT_POOL_WORDS];
1307 /* For /dev/random's pool, always leave two wakeups' worth */
1308 int rsvd_bytes = r->limit ? 0 : random_read_wakeup_bits / 4;
1311 /* pull at least as much as a wakeup */
1312 bytes = max_t(int, bytes, random_read_wakeup_bits / 8);
1313 /* but never more than the buffer size */
1314 bytes = min_t(int, bytes, sizeof(tmp));
1316 trace_xfer_secondary_pool(r->name, bytes * 8, nbytes * 8,
1317 ENTROPY_BITS(r), ENTROPY_BITS(r->pull));
1318 bytes = extract_entropy(r->pull, tmp, bytes,
1319 random_read_wakeup_bits / 8, rsvd_bytes);
1320 mix_pool_bytes(r, tmp, bytes);
1321 credit_entropy_bits(r, bytes*8);
1325 * Used as a workqueue function so that when the input pool is getting
1326 * full, we can "spill over" some entropy to the output pools. That
1327 * way the output pools can store some of the excess entropy instead
1328 * of letting it go to waste.
1330 static void push_to_pool(struct work_struct *work)
1332 struct entropy_store *r = container_of(work, struct entropy_store,
1335 _xfer_secondary_pool(r, random_read_wakeup_bits/8);
1336 trace_push_to_pool(r->name, r->entropy_count >> ENTROPY_SHIFT,
1337 r->pull->entropy_count >> ENTROPY_SHIFT);
1341 * This function decides how many bytes to actually take from the
1342 * given pool, and also debits the entropy count accordingly.
1344 static size_t account(struct entropy_store *r, size_t nbytes, int min,
1347 int entropy_count, orig;
1348 size_t ibytes, nfrac;
1350 BUG_ON(r->entropy_count > r->poolinfo->poolfracbits);
1352 /* Can we pull enough? */
1354 entropy_count = orig = ACCESS_ONCE(r->entropy_count);
1356 /* If limited, never pull more than available */
1358 int have_bytes = entropy_count >> (ENTROPY_SHIFT + 3);
1360 if ((have_bytes -= reserved) < 0)
1362 ibytes = min_t(size_t, ibytes, have_bytes);
1367 if (unlikely(entropy_count < 0)) {
1368 pr_warn("random: negative entropy count: pool %s count %d\n",
1369 r->name, entropy_count);
1373 nfrac = ibytes << (ENTROPY_SHIFT + 3);
1374 if ((size_t) entropy_count > nfrac)
1375 entropy_count -= nfrac;
1379 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
1382 trace_debit_entropy(r->name, 8 * ibytes);
1384 (r->entropy_count >> ENTROPY_SHIFT) < random_write_wakeup_bits) {
1385 wake_up_interruptible(&random_write_wait);
1386 kill_fasync(&fasync, SIGIO, POLL_OUT);
1393 * This function does the actual extraction for extract_entropy and
1394 * extract_entropy_user.
1396 * Note: we assume that .poolwords is a multiple of 16 words.
1398 static void extract_buf(struct entropy_store *r, __u8 *out)
1403 unsigned long l[LONGS(20)];
1405 __u32 workspace[SHA_WORKSPACE_WORDS];
1406 unsigned long flags;
1409 * If we have an architectural hardware random number
1410 * generator, use it for SHA's initial vector
1413 for (i = 0; i < LONGS(20); i++) {
1415 if (!arch_get_random_long(&v))
1420 /* Generate a hash across the pool, 16 words (512 bits) at a time */
1421 spin_lock_irqsave(&r->lock, flags);
1422 for (i = 0; i < r->poolinfo->poolwords; i += 16)
1423 sha_transform(hash.w, (__u8 *)(r->pool + i), workspace);
1426 * We mix the hash back into the pool to prevent backtracking
1427 * attacks (where the attacker knows the state of the pool
1428 * plus the current outputs, and attempts to find previous
1429 * ouputs), unless the hash function can be inverted. By
1430 * mixing at least a SHA1 worth of hash data back, we make
1431 * brute-forcing the feedback as hard as brute-forcing the
1434 __mix_pool_bytes(r, hash.w, sizeof(hash.w));
1435 spin_unlock_irqrestore(&r->lock, flags);
1437 memzero_explicit(workspace, sizeof(workspace));
1440 * In case the hash function has some recognizable output
1441 * pattern, we fold it in half. Thus, we always feed back
1442 * twice as much data as we output.
1444 hash.w[0] ^= hash.w[3];
1445 hash.w[1] ^= hash.w[4];
1446 hash.w[2] ^= rol32(hash.w[2], 16);
1448 memcpy(out, &hash, EXTRACT_SIZE);
1449 memzero_explicit(&hash, sizeof(hash));
1452 static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
1453 size_t nbytes, int fips)
1456 __u8 tmp[EXTRACT_SIZE];
1457 unsigned long flags;
1460 extract_buf(r, tmp);
1463 spin_lock_irqsave(&r->lock, flags);
1464 if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
1465 panic("Hardware RNG duplicated output!\n");
1466 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1467 spin_unlock_irqrestore(&r->lock, flags);
1469 i = min_t(int, nbytes, EXTRACT_SIZE);
1470 memcpy(buf, tmp, i);
1476 /* Wipe data just returned from memory */
1477 memzero_explicit(tmp, sizeof(tmp));
1483 * This function extracts randomness from the "entropy pool", and
1484 * returns it in a buffer.
1486 * The min parameter specifies the minimum amount we can pull before
1487 * failing to avoid races that defeat catastrophic reseeding while the
1488 * reserved parameter indicates how much entropy we must leave in the
1489 * pool after each pull to avoid starving other readers.
1491 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1492 size_t nbytes, int min, int reserved)
1494 __u8 tmp[EXTRACT_SIZE];
1495 unsigned long flags;
1497 /* if last_data isn't primed, we need EXTRACT_SIZE extra bytes */
1499 spin_lock_irqsave(&r->lock, flags);
1500 if (!r->last_data_init) {
1501 r->last_data_init = 1;
1502 spin_unlock_irqrestore(&r->lock, flags);
1503 trace_extract_entropy(r->name, EXTRACT_SIZE,
1504 ENTROPY_BITS(r), _RET_IP_);
1505 xfer_secondary_pool(r, EXTRACT_SIZE);
1506 extract_buf(r, tmp);
1507 spin_lock_irqsave(&r->lock, flags);
1508 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1510 spin_unlock_irqrestore(&r->lock, flags);
1513 trace_extract_entropy(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1514 xfer_secondary_pool(r, nbytes);
1515 nbytes = account(r, nbytes, min, reserved);
1517 return _extract_entropy(r, buf, nbytes, fips_enabled);
1521 * This function extracts randomness from the "entropy pool", and
1522 * returns it in a userspace buffer.
1524 static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
1528 __u8 tmp[EXTRACT_SIZE];
1529 int large_request = (nbytes > 256);
1531 trace_extract_entropy_user(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1532 xfer_secondary_pool(r, nbytes);
1533 nbytes = account(r, nbytes, 0, 0);
1536 if (large_request && need_resched()) {
1537 if (signal_pending(current)) {
1545 extract_buf(r, tmp);
1546 i = min_t(int, nbytes, EXTRACT_SIZE);
1547 if (copy_to_user(buf, tmp, i)) {
1557 /* Wipe data just returned from memory */
1558 memzero_explicit(tmp, sizeof(tmp));
1564 * This function is the exported kernel interface. It returns some
1565 * number of good random numbers, suitable for key generation, seeding
1566 * TCP sequence numbers, etc. It does not rely on the hardware random
1567 * number generator. For random bytes direct from the hardware RNG
1568 * (when available), use get_random_bytes_arch().
1570 void get_random_bytes(void *buf, int nbytes)
1572 __u8 tmp[CHACHA20_BLOCK_SIZE];
1574 #if DEBUG_RANDOM_BOOT > 0
1576 printk(KERN_NOTICE "random: %pF get_random_bytes called "
1577 "with crng_init = %d\n", (void *) _RET_IP_, crng_init);
1579 trace_get_random_bytes(nbytes, _RET_IP_);
1581 while (nbytes >= CHACHA20_BLOCK_SIZE) {
1583 buf += CHACHA20_BLOCK_SIZE;
1584 nbytes -= CHACHA20_BLOCK_SIZE;
1589 memcpy(buf, tmp, nbytes);
1590 crng_backtrack_protect(tmp, nbytes);
1592 crng_backtrack_protect(tmp, CHACHA20_BLOCK_SIZE);
1593 memzero_explicit(tmp, sizeof(tmp));
1595 EXPORT_SYMBOL(get_random_bytes);
1598 * Add a callback function that will be invoked when the nonblocking
1599 * pool is initialised.
1601 * returns: 0 if callback is successfully added
1602 * -EALREADY if pool is already initialised (callback not called)
1603 * -ENOENT if module for callback is not alive
1605 int add_random_ready_callback(struct random_ready_callback *rdy)
1607 struct module *owner;
1608 unsigned long flags;
1609 int err = -EALREADY;
1615 if (!try_module_get(owner))
1618 spin_lock_irqsave(&random_ready_list_lock, flags);
1624 list_add(&rdy->list, &random_ready_list);
1628 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1634 EXPORT_SYMBOL(add_random_ready_callback);
1637 * Delete a previously registered readiness callback function.
1639 void del_random_ready_callback(struct random_ready_callback *rdy)
1641 unsigned long flags;
1642 struct module *owner = NULL;
1644 spin_lock_irqsave(&random_ready_list_lock, flags);
1645 if (!list_empty(&rdy->list)) {
1646 list_del_init(&rdy->list);
1649 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1653 EXPORT_SYMBOL(del_random_ready_callback);
1656 * This function will use the architecture-specific hardware random
1657 * number generator if it is available. The arch-specific hw RNG will
1658 * almost certainly be faster than what we can do in software, but it
1659 * is impossible to verify that it is implemented securely (as
1660 * opposed, to, say, the AES encryption of a sequence number using a
1661 * key known by the NSA). So it's useful if we need the speed, but
1662 * only if we're willing to trust the hardware manufacturer not to
1663 * have put in a back door.
1665 void get_random_bytes_arch(void *buf, int nbytes)
1669 trace_get_random_bytes_arch(nbytes, _RET_IP_);
1672 int chunk = min(nbytes, (int)sizeof(unsigned long));
1674 if (!arch_get_random_long(&v))
1677 memcpy(p, &v, chunk);
1683 get_random_bytes(p, nbytes);
1685 EXPORT_SYMBOL(get_random_bytes_arch);
1689 * init_std_data - initialize pool with system data
1691 * @r: pool to initialize
1693 * This function clears the pool's entropy count and mixes some system
1694 * data into the pool to prepare it for use. The pool is not cleared
1695 * as that can only decrease the entropy in the pool.
1697 static void init_std_data(struct entropy_store *r)
1700 ktime_t now = ktime_get_real();
1703 r->last_pulled = jiffies;
1704 mix_pool_bytes(r, &now, sizeof(now));
1705 for (i = r->poolinfo->poolbytes; i > 0; i -= sizeof(rv)) {
1706 if (!arch_get_random_seed_long(&rv) &&
1707 !arch_get_random_long(&rv))
1708 rv = random_get_entropy();
1709 mix_pool_bytes(r, &rv, sizeof(rv));
1711 mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1715 * Note that setup_arch() may call add_device_randomness()
1716 * long before we get here. This allows seeding of the pools
1717 * with some platform dependent data very early in the boot
1718 * process. But it limits our options here. We must use
1719 * statically allocated structures that already have all
1720 * initializations complete at compile time. We should also
1721 * take care not to overwrite the precious per platform data
1724 static int rand_initialize(void)
1726 init_std_data(&input_pool);
1727 init_std_data(&blocking_pool);
1728 crng_initialize(&primary_crng);
1729 crng_global_init_time = jiffies;
1730 if (ratelimit_disable) {
1731 urandom_warning.interval = 0;
1732 unseeded_warning.interval = 0;
1736 early_initcall(rand_initialize);
1739 void rand_initialize_disk(struct gendisk *disk)
1741 struct timer_rand_state *state;
1744 * If kzalloc returns null, we just won't use that entropy
1747 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1749 state->last_time = INITIAL_JIFFIES;
1750 disk->random = state;
1756 _random_read(int nonblock, char __user *buf, size_t nbytes)
1763 nbytes = min_t(size_t, nbytes, SEC_XFER_SIZE);
1765 n = extract_entropy_user(&blocking_pool, buf, nbytes);
1768 trace_random_read(n*8, (nbytes-n)*8,
1769 ENTROPY_BITS(&blocking_pool),
1770 ENTROPY_BITS(&input_pool));
1774 /* Pool is (near) empty. Maybe wait and retry. */
1778 wait_event_interruptible(random_read_wait,
1779 ENTROPY_BITS(&input_pool) >=
1780 random_read_wakeup_bits);
1781 if (signal_pending(current))
1782 return -ERESTARTSYS;
1787 random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1789 return _random_read(file->f_flags & O_NONBLOCK, buf, nbytes);
1793 urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1795 unsigned long flags;
1796 static int maxwarn = 10;
1799 if (!crng_ready() && maxwarn > 0) {
1801 if (__ratelimit(&urandom_warning))
1802 printk(KERN_NOTICE "random: %s: uninitialized "
1803 "urandom read (%zd bytes read)\n",
1804 current->comm, nbytes);
1805 spin_lock_irqsave(&primary_crng.lock, flags);
1807 spin_unlock_irqrestore(&primary_crng.lock, flags);
1809 nbytes = min_t(size_t, nbytes, INT_MAX >> (ENTROPY_SHIFT + 3));
1810 ret = extract_crng_user(buf, nbytes);
1811 trace_urandom_read(8 * nbytes, 0, ENTROPY_BITS(&input_pool));
1816 random_poll(struct file *file, poll_table * wait)
1820 poll_wait(file, &random_read_wait, wait);
1821 poll_wait(file, &random_write_wait, wait);
1823 if (ENTROPY_BITS(&input_pool) >= random_read_wakeup_bits)
1824 mask |= POLLIN | POLLRDNORM;
1825 if (ENTROPY_BITS(&input_pool) < random_write_wakeup_bits)
1826 mask |= POLLOUT | POLLWRNORM;
1831 write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1835 const char __user *p = buffer;
1840 bytes = min(count, sizeof(buf));
1841 if (copy_from_user(&buf, p, bytes))
1844 for (b = bytes ; b > 0 ; b -= sizeof(__u32), i++) {
1845 if (!arch_get_random_int(&t))
1853 mix_pool_bytes(r, buf, bytes);
1860 static ssize_t random_write(struct file *file, const char __user *buffer,
1861 size_t count, loff_t *ppos)
1865 ret = write_pool(&input_pool, buffer, count);
1869 return (ssize_t)count;
1872 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1874 int size, ent_count;
1875 int __user *p = (int __user *)arg;
1880 /* inherently racy, no point locking */
1881 ent_count = ENTROPY_BITS(&input_pool);
1882 if (put_user(ent_count, p))
1885 case RNDADDTOENTCNT:
1886 if (!capable(CAP_SYS_ADMIN))
1888 if (get_user(ent_count, p))
1890 return credit_entropy_bits_safe(&input_pool, ent_count);
1892 if (!capable(CAP_SYS_ADMIN))
1894 if (get_user(ent_count, p++))
1898 if (get_user(size, p++))
1900 retval = write_pool(&input_pool, (const char __user *)p,
1904 return credit_entropy_bits_safe(&input_pool, ent_count);
1908 * Clear the entropy pool counters. We no longer clear
1909 * the entropy pool, as that's silly.
1911 if (!capable(CAP_SYS_ADMIN))
1913 input_pool.entropy_count = 0;
1914 blocking_pool.entropy_count = 0;
1917 if (!capable(CAP_SYS_ADMIN))
1921 crng_reseed(&primary_crng, &input_pool);
1922 WRITE_ONCE(crng_global_init_time, jiffies - 1);
1929 static int random_fasync(int fd, struct file *filp, int on)
1931 return fasync_helper(fd, filp, on, &fasync);
1934 const struct file_operations random_fops = {
1935 .read = random_read,
1936 .write = random_write,
1937 .poll = random_poll,
1938 .unlocked_ioctl = random_ioctl,
1939 .fasync = random_fasync,
1940 .llseek = noop_llseek,
1943 const struct file_operations urandom_fops = {
1944 .read = urandom_read,
1945 .write = random_write,
1946 .unlocked_ioctl = random_ioctl,
1947 .fasync = random_fasync,
1948 .llseek = noop_llseek,
1951 SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count,
1952 unsigned int, flags)
1954 if (flags & ~(GRND_NONBLOCK|GRND_RANDOM))
1957 if (count > INT_MAX)
1960 if (flags & GRND_RANDOM)
1961 return _random_read(flags & GRND_NONBLOCK, buf, count);
1963 if (!crng_ready()) {
1964 if (flags & GRND_NONBLOCK)
1967 if (signal_pending(current))
1968 return -ERESTARTSYS;
1970 return urandom_read(NULL, buf, count, NULL);
1973 /********************************************************************
1977 ********************************************************************/
1979 #ifdef CONFIG_SYSCTL
1981 #include <linux/sysctl.h>
1983 static int min_read_thresh = 8, min_write_thresh;
1984 static int max_read_thresh = OUTPUT_POOL_WORDS * 32;
1985 static int max_write_thresh = INPUT_POOL_WORDS * 32;
1986 static char sysctl_bootid[16];
1989 * This function is used to return both the bootid UUID, and random
1990 * UUID. The difference is in whether table->data is NULL; if it is,
1991 * then a new UUID is generated and returned to the user.
1993 * If the user accesses this via the proc interface, the UUID will be
1994 * returned as an ASCII string in the standard UUID format; if via the
1995 * sysctl system call, as 16 bytes of binary data.
1997 static int proc_do_uuid(struct ctl_table *table, int write,
1998 void __user *buffer, size_t *lenp, loff_t *ppos)
2000 struct ctl_table fake_table;
2001 unsigned char buf[64], tmp_uuid[16], *uuid;
2006 generate_random_uuid(uuid);
2008 static DEFINE_SPINLOCK(bootid_spinlock);
2010 spin_lock(&bootid_spinlock);
2012 generate_random_uuid(uuid);
2013 spin_unlock(&bootid_spinlock);
2016 sprintf(buf, "%pU", uuid);
2018 fake_table.data = buf;
2019 fake_table.maxlen = sizeof(buf);
2021 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
2025 * Return entropy available scaled to integral bits
2027 static int proc_do_entropy(struct ctl_table *table, int write,
2028 void __user *buffer, size_t *lenp, loff_t *ppos)
2030 struct ctl_table fake_table;
2033 entropy_count = *(int *)table->data >> ENTROPY_SHIFT;
2035 fake_table.data = &entropy_count;
2036 fake_table.maxlen = sizeof(entropy_count);
2038 return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
2041 static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
2042 extern struct ctl_table random_table[];
2043 struct ctl_table random_table[] = {
2045 .procname = "poolsize",
2046 .data = &sysctl_poolsize,
2047 .maxlen = sizeof(int),
2049 .proc_handler = proc_dointvec,
2052 .procname = "entropy_avail",
2053 .maxlen = sizeof(int),
2055 .proc_handler = proc_do_entropy,
2056 .data = &input_pool.entropy_count,
2059 .procname = "read_wakeup_threshold",
2060 .data = &random_read_wakeup_bits,
2061 .maxlen = sizeof(int),
2063 .proc_handler = proc_dointvec_minmax,
2064 .extra1 = &min_read_thresh,
2065 .extra2 = &max_read_thresh,
2068 .procname = "write_wakeup_threshold",
2069 .data = &random_write_wakeup_bits,
2070 .maxlen = sizeof(int),
2072 .proc_handler = proc_dointvec_minmax,
2073 .extra1 = &min_write_thresh,
2074 .extra2 = &max_write_thresh,
2077 .procname = "urandom_min_reseed_secs",
2078 .data = &random_min_urandom_seed,
2079 .maxlen = sizeof(int),
2081 .proc_handler = proc_dointvec,
2084 .procname = "boot_id",
2085 .data = &sysctl_bootid,
2088 .proc_handler = proc_do_uuid,
2094 .proc_handler = proc_do_uuid,
2096 #ifdef ADD_INTERRUPT_BENCH
2098 .procname = "add_interrupt_avg_cycles",
2099 .data = &avg_cycles,
2100 .maxlen = sizeof(avg_cycles),
2102 .proc_handler = proc_doulongvec_minmax,
2105 .procname = "add_interrupt_avg_deviation",
2106 .data = &avg_deviation,
2107 .maxlen = sizeof(avg_deviation),
2109 .proc_handler = proc_doulongvec_minmax,
2114 #endif /* CONFIG_SYSCTL */
2116 struct batched_entropy {
2118 unsigned long entropy_long[CHACHA20_BLOCK_SIZE / sizeof(unsigned long)];
2119 unsigned int entropy_int[CHACHA20_BLOCK_SIZE / sizeof(unsigned int)];
2121 unsigned int position;
2125 * Get a random word for internal kernel use only. The quality of the random
2126 * number is good as /dev/urandom, but there is no backtrack protection, with
2127 * the goal of being quite fast and not depleting entropy.
2129 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_long);
2130 unsigned long get_random_long(void)
2133 struct batched_entropy *batch;
2135 batch = &get_cpu_var(batched_entropy_long);
2136 if (batch->position % ARRAY_SIZE(batch->entropy_long) == 0) {
2137 extract_crng((u8 *)batch->entropy_long);
2138 batch->position = 0;
2140 ret = batch->entropy_long[batch->position++];
2141 put_cpu_var(batched_entropy_long);
2144 EXPORT_SYMBOL(get_random_long);
2146 #if BITS_PER_LONG == 32
2147 unsigned int get_random_int(void)
2149 return get_random_long();
2152 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_int);
2153 unsigned int get_random_int(void)
2156 struct batched_entropy *batch;
2158 batch = &get_cpu_var(batched_entropy_int);
2159 if (batch->position % ARRAY_SIZE(batch->entropy_int) == 0) {
2160 extract_crng((u8 *)batch->entropy_int);
2161 batch->position = 0;
2163 ret = batch->entropy_int[batch->position++];
2164 put_cpu_var(batched_entropy_int);
2168 EXPORT_SYMBOL(get_random_int);
2171 * randomize_page - Generate a random, page aligned address
2172 * @start: The smallest acceptable address the caller will take.
2173 * @range: The size of the area, starting at @start, within which the
2174 * random address must fall.
2176 * If @start + @range would overflow, @range is capped.
2178 * NOTE: Historical use of randomize_range, which this replaces, presumed that
2179 * @start was already page aligned. We now align it regardless.
2181 * Return: A page aligned address within [start, start + range). On error,
2182 * @start is returned.
2185 randomize_page(unsigned long start, unsigned long range)
2187 if (!PAGE_ALIGNED(start)) {
2188 range -= PAGE_ALIGN(start) - start;
2189 start = PAGE_ALIGN(start);
2192 if (start > ULONG_MAX - range)
2193 range = ULONG_MAX - start;
2195 range >>= PAGE_SHIFT;
2200 return start + (get_random_long() % range << PAGE_SHIFT);
2203 /* Interface for in-kernel drivers of true hardware RNGs.
2204 * Those devices may produce endless random bits and will be throttled
2205 * when our pool is full.
2207 void add_hwgenerator_randomness(const char *buffer, size_t count,
2210 struct entropy_store *poolp = &input_pool;
2212 if (unlikely(crng_init == 0)) {
2213 crng_fast_load(buffer, count);
2217 /* Suspend writing if we're above the trickle threshold.
2218 * We'll be woken up again once below random_write_wakeup_thresh,
2219 * or when the calling thread is about to terminate.
2221 wait_event_interruptible(random_write_wait, kthread_should_stop() ||
2222 ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
2223 mix_pool_bytes(poolp, buffer, count);
2224 credit_entropy_bits(poolp, entropy);
2226 EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);