xref: /linux/drivers/char/random.c (revision 74ce1896c6c65b2f8cccbf59162d542988835835)
1 /*
2  * random.c -- A strong random number generator
3  *
4  * Copyright (C) 2017 Jason A. Donenfeld <Jason@zx2c4.com>. All
5  * Rights Reserved.
6  *
7  * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005
8  *
9  * Copyright Theodore Ts'o, 1994, 1995, 1996, 1997, 1998, 1999.  All
10  * rights reserved.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
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
23  *    written permission.
24  *
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.)
30  *
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
42  * DAMAGE.
43  */
44 
45 /*
46  * (now, with legal B.S. out of the way.....)
47  *
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.
54  *
55  * Theory of operation
56  * ===================
57  *
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.
68  *
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.
80  *
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.
92  *
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
99  * of purposes.
100  *
101  * Exported interfaces ---- output
102  * ===============================
103  *
104  * There are three exported interfaces; the first is one designed to
105  * be used from within the kernel:
106  *
107  * 	void get_random_bytes(void *buf, int nbytes);
108  *
109  * This interface will return the requested number of random bytes,
110  * and place it in the requested buffer.
111  *
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.
118  *
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.
124  *
125  * Exported interfaces ---- input
126  * ==============================
127  *
128  * The current exported interfaces for gathering environmental noise
129  * from the devices are:
130  *
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);
136  *
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).
144  *
145  * add_input_randomness() uses the input layer interrupt timing, as well as
146  * the event type information from the hardware.
147  *
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.
151  *
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.
157  *
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.
161  *
162  * Ensuring unpredictability at system startup
163  * ============================================
164  *
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
173  * sequence:
174  *
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
181  *	else
182  *		touch $random_seed
183  *	fi
184  *	chmod 600 $random_seed
185  *	dd if=/dev/urandom of=$random_seed count=1 bs=512
186  *
187  * and the following lines in an appropriate script which is run as
188  * the system is shutdown:
189  *
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
194  *	touch $random_seed
195  *	chmod 600 $random_seed
196  *	dd if=/dev/urandom of=$random_seed count=1 bs=512
197  *
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.
202  *
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
210  * the system.
211  *
212  * Configuring the /dev/random driver under Linux
213  * ==============================================
214  *
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:
219  *
220  * 	mknod /dev/random c 1 8
221  * 	mknod /dev/urandom c 1 9
222  *
223  * Acknowledgements:
224  * =================
225  *
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.
232  *
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.
235  *
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.
239  */
240 
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>
269 
270 #include <asm/processor.h>
271 #include <linux/uaccess.h>
272 #include <asm/irq.h>
273 #include <asm/irq_regs.h>
274 #include <asm/io.h>
275 
276 #define CREATE_TRACE_POINTS
277 #include <trace/events/random.h>
278 
279 /* #define ADD_INTERRUPT_BENCH */
280 
281 /*
282  * Configuration information
283  */
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
290 
291 
292 #define LONGS(x) (((x) + sizeof(unsigned long) - 1)/sizeof(unsigned long))
293 
294 /*
295  * To allow fractional bits to be tracked, the entropy_count field is
296  * denominated in units of 1/8th bits.
297  *
298  * 2*(ENTROPY_SHIFT + log2(poolbits)) must <= 31, or the multiply in
299  * credit_entropy_bits() needs to be 64 bits wide.
300  */
301 #define ENTROPY_SHIFT 3
302 #define ENTROPY_BITS(r) ((r)->entropy_count >> ENTROPY_SHIFT)
303 
304 /*
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.
307  */
308 static int random_read_wakeup_bits = 64;
309 
310 /*
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.
314  */
315 static int random_write_wakeup_bits = 28 * OUTPUT_POOL_WORDS;
316 
317 /*
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.
322  *
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)
330  *
331  * Thanks to Colin Plumb for suggesting this.
332  *
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).
349  *
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.
361  */
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 },
373 #if 0
374 	/* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1  -- 115 */
375 	{ S(2048),	1638,	1231,	819,	411,	1 },
376 
377 	/* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
378 	{ S(1024),	817,	615,	412,	204,	1 },
379 
380 	/* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
381 	{ S(1024),	819,	616,	410,	207,	2 },
382 
383 	/* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
384 	{ S(512),	411,	308,	208,	104,	1 },
385 
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 },
390 
391 	/* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
392 	{ S(256),	205,	155,	101,	52,	1 },
393 
394 	/* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
395 	{ S(128),	103,	78,	51,	27,	2 },
396 
397 	/* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
398 	{ S(64),	52,	39,	26,	14,	1 },
399 #endif
400 };
401 
402 /*
403  * Static global variables
404  */
405 static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
406 static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
407 static struct fasync_struct *fasync;
408 
409 static DEFINE_SPINLOCK(random_ready_list_lock);
410 static LIST_HEAD(random_ready_list);
411 
412 struct crng_state {
413 	__u32		state[16];
414 	unsigned long	init_time;
415 	spinlock_t	lock;
416 };
417 
418 struct crng_state primary_crng = {
419 	.lock = __SPIN_LOCK_UNLOCKED(primary_crng.lock),
420 };
421 
422 /*
423  * crng_init =  0 --> Uninitialized
424  *		1 --> Initialized
425  *		2 --> Initialized from input_pool
426  *
427  * crng_init is protected by primary_crng->lock, and only increases
428  * its value (from 0->1->2).
429  */
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);
439 static void _get_random_bytes(void *buf, int nbytes);
440 
441 /**********************************************************************
442  *
443  * OS independent entropy store.   Here are the functions which handle
444  * storing entropy in an entropy pool.
445  *
446  **********************************************************************/
447 
448 struct entropy_store;
449 struct entropy_store {
450 	/* read-only data: */
451 	const struct poolinfo *poolinfo;
452 	__u32 *pool;
453 	const char *name;
454 	struct entropy_store *pull;
455 	struct work_struct push_work;
456 
457 	/* read-write data: */
458 	unsigned long last_pulled;
459 	spinlock_t lock;
460 	unsigned short add_ptr;
461 	unsigned short input_rotate;
462 	int entropy_count;
463 	int entropy_total;
464 	unsigned int initialized:1;
465 	unsigned int last_data_init:1;
466 	__u8 last_data[EXTRACT_SIZE];
467 };
468 
469 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
470 			       size_t nbytes, int min, int rsvd);
471 static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
472 				size_t nbytes, int fips);
473 
474 static void crng_reseed(struct crng_state *crng, struct entropy_store *r);
475 static void push_to_pool(struct work_struct *work);
476 static __u32 input_pool_data[INPUT_POOL_WORDS] __latent_entropy;
477 static __u32 blocking_pool_data[OUTPUT_POOL_WORDS] __latent_entropy;
478 
479 static struct entropy_store input_pool = {
480 	.poolinfo = &poolinfo_table[0],
481 	.name = "input",
482 	.lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
483 	.pool = input_pool_data
484 };
485 
486 static struct entropy_store blocking_pool = {
487 	.poolinfo = &poolinfo_table[1],
488 	.name = "blocking",
489 	.pull = &input_pool,
490 	.lock = __SPIN_LOCK_UNLOCKED(blocking_pool.lock),
491 	.pool = blocking_pool_data,
492 	.push_work = __WORK_INITIALIZER(blocking_pool.push_work,
493 					push_to_pool),
494 };
495 
496 static __u32 const twist_table[8] = {
497 	0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
498 	0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
499 
500 /*
501  * This function adds bytes into the entropy "pool".  It does not
502  * update the entropy estimate.  The caller should call
503  * credit_entropy_bits if this is appropriate.
504  *
505  * The pool is stirred with a primitive polynomial of the appropriate
506  * degree, and then twisted.  We twist by three bits at a time because
507  * it's cheap to do so and helps slightly in the expected case where
508  * the entropy is concentrated in the low-order bits.
509  */
510 static void _mix_pool_bytes(struct entropy_store *r, const void *in,
511 			    int nbytes)
512 {
513 	unsigned long i, tap1, tap2, tap3, tap4, tap5;
514 	int input_rotate;
515 	int wordmask = r->poolinfo->poolwords - 1;
516 	const char *bytes = in;
517 	__u32 w;
518 
519 	tap1 = r->poolinfo->tap1;
520 	tap2 = r->poolinfo->tap2;
521 	tap3 = r->poolinfo->tap3;
522 	tap4 = r->poolinfo->tap4;
523 	tap5 = r->poolinfo->tap5;
524 
525 	input_rotate = r->input_rotate;
526 	i = r->add_ptr;
527 
528 	/* mix one byte at a time to simplify size handling and churn faster */
529 	while (nbytes--) {
530 		w = rol32(*bytes++, input_rotate);
531 		i = (i - 1) & wordmask;
532 
533 		/* XOR in the various taps */
534 		w ^= r->pool[i];
535 		w ^= r->pool[(i + tap1) & wordmask];
536 		w ^= r->pool[(i + tap2) & wordmask];
537 		w ^= r->pool[(i + tap3) & wordmask];
538 		w ^= r->pool[(i + tap4) & wordmask];
539 		w ^= r->pool[(i + tap5) & wordmask];
540 
541 		/* Mix the result back in with a twist */
542 		r->pool[i] = (w >> 3) ^ twist_table[w & 7];
543 
544 		/*
545 		 * Normally, we add 7 bits of rotation to the pool.
546 		 * At the beginning of the pool, add an extra 7 bits
547 		 * rotation, so that successive passes spread the
548 		 * input bits across the pool evenly.
549 		 */
550 		input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
551 	}
552 
553 	r->input_rotate = input_rotate;
554 	r->add_ptr = i;
555 }
556 
557 static void __mix_pool_bytes(struct entropy_store *r, const void *in,
558 			     int nbytes)
559 {
560 	trace_mix_pool_bytes_nolock(r->name, nbytes, _RET_IP_);
561 	_mix_pool_bytes(r, in, nbytes);
562 }
563 
564 static void mix_pool_bytes(struct entropy_store *r, const void *in,
565 			   int nbytes)
566 {
567 	unsigned long flags;
568 
569 	trace_mix_pool_bytes(r->name, nbytes, _RET_IP_);
570 	spin_lock_irqsave(&r->lock, flags);
571 	_mix_pool_bytes(r, in, nbytes);
572 	spin_unlock_irqrestore(&r->lock, flags);
573 }
574 
575 struct fast_pool {
576 	__u32		pool[4];
577 	unsigned long	last;
578 	unsigned short	reg_idx;
579 	unsigned char	count;
580 };
581 
582 /*
583  * This is a fast mixing routine used by the interrupt randomness
584  * collector.  It's hardcoded for an 128 bit pool and assumes that any
585  * locks that might be needed are taken by the caller.
586  */
587 static void fast_mix(struct fast_pool *f)
588 {
589 	__u32 a = f->pool[0],	b = f->pool[1];
590 	__u32 c = f->pool[2],	d = f->pool[3];
591 
592 	a += b;			c += d;
593 	b = rol32(b, 6);	d = rol32(d, 27);
594 	d ^= a;			b ^= c;
595 
596 	a += b;			c += d;
597 	b = rol32(b, 16);	d = rol32(d, 14);
598 	d ^= a;			b ^= c;
599 
600 	a += b;			c += d;
601 	b = rol32(b, 6);	d = rol32(d, 27);
602 	d ^= a;			b ^= c;
603 
604 	a += b;			c += d;
605 	b = rol32(b, 16);	d = rol32(d, 14);
606 	d ^= a;			b ^= c;
607 
608 	f->pool[0] = a;  f->pool[1] = b;
609 	f->pool[2] = c;  f->pool[3] = d;
610 	f->count++;
611 }
612 
613 static void process_random_ready_list(void)
614 {
615 	unsigned long flags;
616 	struct random_ready_callback *rdy, *tmp;
617 
618 	spin_lock_irqsave(&random_ready_list_lock, flags);
619 	list_for_each_entry_safe(rdy, tmp, &random_ready_list, list) {
620 		struct module *owner = rdy->owner;
621 
622 		list_del_init(&rdy->list);
623 		rdy->func(rdy);
624 		module_put(owner);
625 	}
626 	spin_unlock_irqrestore(&random_ready_list_lock, flags);
627 }
628 
629 /*
630  * Credit (or debit) the entropy store with n bits of entropy.
631  * Use credit_entropy_bits_safe() if the value comes from userspace
632  * or otherwise should be checked for extreme values.
633  */
634 static void credit_entropy_bits(struct entropy_store *r, int nbits)
635 {
636 	int entropy_count, orig;
637 	const int pool_size = r->poolinfo->poolfracbits;
638 	int nfrac = nbits << ENTROPY_SHIFT;
639 
640 	if (!nbits)
641 		return;
642 
643 retry:
644 	entropy_count = orig = ACCESS_ONCE(r->entropy_count);
645 	if (nfrac < 0) {
646 		/* Debit */
647 		entropy_count += nfrac;
648 	} else {
649 		/*
650 		 * Credit: we have to account for the possibility of
651 		 * overwriting already present entropy.	 Even in the
652 		 * ideal case of pure Shannon entropy, new contributions
653 		 * approach the full value asymptotically:
654 		 *
655 		 * entropy <- entropy + (pool_size - entropy) *
656 		 *	(1 - exp(-add_entropy/pool_size))
657 		 *
658 		 * For add_entropy <= pool_size/2 then
659 		 * (1 - exp(-add_entropy/pool_size)) >=
660 		 *    (add_entropy/pool_size)*0.7869...
661 		 * so we can approximate the exponential with
662 		 * 3/4*add_entropy/pool_size and still be on the
663 		 * safe side by adding at most pool_size/2 at a time.
664 		 *
665 		 * The use of pool_size-2 in the while statement is to
666 		 * prevent rounding artifacts from making the loop
667 		 * arbitrarily long; this limits the loop to log2(pool_size)*2
668 		 * turns no matter how large nbits is.
669 		 */
670 		int pnfrac = nfrac;
671 		const int s = r->poolinfo->poolbitshift + ENTROPY_SHIFT + 2;
672 		/* The +2 corresponds to the /4 in the denominator */
673 
674 		do {
675 			unsigned int anfrac = min(pnfrac, pool_size/2);
676 			unsigned int add =
677 				((pool_size - entropy_count)*anfrac*3) >> s;
678 
679 			entropy_count += add;
680 			pnfrac -= anfrac;
681 		} while (unlikely(entropy_count < pool_size-2 && pnfrac));
682 	}
683 
684 	if (unlikely(entropy_count < 0)) {
685 		pr_warn("random: negative entropy/overflow: pool %s count %d\n",
686 			r->name, entropy_count);
687 		WARN_ON(1);
688 		entropy_count = 0;
689 	} else if (entropy_count > pool_size)
690 		entropy_count = pool_size;
691 	if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
692 		goto retry;
693 
694 	r->entropy_total += nbits;
695 	if (!r->initialized && r->entropy_total > 128) {
696 		r->initialized = 1;
697 		r->entropy_total = 0;
698 	}
699 
700 	trace_credit_entropy_bits(r->name, nbits,
701 				  entropy_count >> ENTROPY_SHIFT,
702 				  r->entropy_total, _RET_IP_);
703 
704 	if (r == &input_pool) {
705 		int entropy_bits = entropy_count >> ENTROPY_SHIFT;
706 
707 		if (crng_init < 2 && entropy_bits >= 128) {
708 			crng_reseed(&primary_crng, r);
709 			entropy_bits = r->entropy_count >> ENTROPY_SHIFT;
710 		}
711 
712 		/* should we wake readers? */
713 		if (entropy_bits >= random_read_wakeup_bits) {
714 			wake_up_interruptible(&random_read_wait);
715 			kill_fasync(&fasync, SIGIO, POLL_IN);
716 		}
717 		/* If the input pool is getting full, send some
718 		 * entropy to the blocking pool until it is 75% full.
719 		 */
720 		if (entropy_bits > random_write_wakeup_bits &&
721 		    r->initialized &&
722 		    r->entropy_total >= 2*random_read_wakeup_bits) {
723 			struct entropy_store *other = &blocking_pool;
724 
725 			if (other->entropy_count <=
726 			    3 * other->poolinfo->poolfracbits / 4) {
727 				schedule_work(&other->push_work);
728 				r->entropy_total = 0;
729 			}
730 		}
731 	}
732 }
733 
734 static int credit_entropy_bits_safe(struct entropy_store *r, int nbits)
735 {
736 	const int nbits_max = (int)(~0U >> (ENTROPY_SHIFT + 1));
737 
738 	if (nbits < 0)
739 		return -EINVAL;
740 
741 	/* Cap the value to avoid overflows */
742 	nbits = min(nbits,  nbits_max);
743 
744 	credit_entropy_bits(r, nbits);
745 	return 0;
746 }
747 
748 /*********************************************************************
749  *
750  * CRNG using CHACHA20
751  *
752  *********************************************************************/
753 
754 #define CRNG_RESEED_INTERVAL (300*HZ)
755 
756 static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
757 
758 #ifdef CONFIG_NUMA
759 /*
760  * Hack to deal with crazy userspace progams when they are all trying
761  * to access /dev/urandom in parallel.  The programs are almost
762  * certainly doing something terribly wrong, but we'll work around
763  * their brain damage.
764  */
765 static struct crng_state **crng_node_pool __read_mostly;
766 #endif
767 
768 static void invalidate_batched_entropy(void);
769 
770 static void crng_initialize(struct crng_state *crng)
771 {
772 	int		i;
773 	unsigned long	rv;
774 
775 	memcpy(&crng->state[0], "expand 32-byte k", 16);
776 	if (crng == &primary_crng)
777 		_extract_entropy(&input_pool, &crng->state[4],
778 				 sizeof(__u32) * 12, 0);
779 	else
780 		_get_random_bytes(&crng->state[4], sizeof(__u32) * 12);
781 	for (i = 4; i < 16; i++) {
782 		if (!arch_get_random_seed_long(&rv) &&
783 		    !arch_get_random_long(&rv))
784 			rv = random_get_entropy();
785 		crng->state[i] ^= rv;
786 	}
787 	crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
788 }
789 
790 static int crng_fast_load(const char *cp, size_t len)
791 {
792 	unsigned long flags;
793 	char *p;
794 
795 	if (!spin_trylock_irqsave(&primary_crng.lock, flags))
796 		return 0;
797 	if (crng_ready()) {
798 		spin_unlock_irqrestore(&primary_crng.lock, flags);
799 		return 0;
800 	}
801 	p = (unsigned char *) &primary_crng.state[4];
802 	while (len > 0 && crng_init_cnt < CRNG_INIT_CNT_THRESH) {
803 		p[crng_init_cnt % CHACHA20_KEY_SIZE] ^= *cp;
804 		cp++; crng_init_cnt++; len--;
805 	}
806 	spin_unlock_irqrestore(&primary_crng.lock, flags);
807 	if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
808 		invalidate_batched_entropy();
809 		crng_init = 1;
810 		wake_up_interruptible(&crng_init_wait);
811 		pr_notice("random: fast init done\n");
812 	}
813 	return 1;
814 }
815 
816 static void crng_reseed(struct crng_state *crng, struct entropy_store *r)
817 {
818 	unsigned long	flags;
819 	int		i, num;
820 	union {
821 		__u8	block[CHACHA20_BLOCK_SIZE];
822 		__u32	key[8];
823 	} buf;
824 
825 	if (r) {
826 		num = extract_entropy(r, &buf, 32, 16, 0);
827 		if (num == 0)
828 			return;
829 	} else {
830 		_extract_crng(&primary_crng, buf.block);
831 		_crng_backtrack_protect(&primary_crng, buf.block,
832 					CHACHA20_KEY_SIZE);
833 	}
834 	spin_lock_irqsave(&primary_crng.lock, flags);
835 	for (i = 0; i < 8; i++) {
836 		unsigned long	rv;
837 		if (!arch_get_random_seed_long(&rv) &&
838 		    !arch_get_random_long(&rv))
839 			rv = random_get_entropy();
840 		crng->state[i+4] ^= buf.key[i] ^ rv;
841 	}
842 	memzero_explicit(&buf, sizeof(buf));
843 	crng->init_time = jiffies;
844 	spin_unlock_irqrestore(&primary_crng.lock, flags);
845 	if (crng == &primary_crng && crng_init < 2) {
846 		invalidate_batched_entropy();
847 		crng_init = 2;
848 		process_random_ready_list();
849 		wake_up_interruptible(&crng_init_wait);
850 		pr_notice("random: crng init done\n");
851 	}
852 }
853 
854 static void _extract_crng(struct crng_state *crng,
855 			  __u8 out[CHACHA20_BLOCK_SIZE])
856 {
857 	unsigned long v, flags;
858 
859 	if (crng_init > 1 &&
860 	    time_after(jiffies, crng->init_time + CRNG_RESEED_INTERVAL))
861 		crng_reseed(crng, crng == &primary_crng ? &input_pool : NULL);
862 	spin_lock_irqsave(&crng->lock, flags);
863 	if (arch_get_random_long(&v))
864 		crng->state[14] ^= v;
865 	chacha20_block(&crng->state[0], out);
866 	if (crng->state[12] == 0)
867 		crng->state[13]++;
868 	spin_unlock_irqrestore(&crng->lock, flags);
869 }
870 
871 static void extract_crng(__u8 out[CHACHA20_BLOCK_SIZE])
872 {
873 	struct crng_state *crng = NULL;
874 
875 #ifdef CONFIG_NUMA
876 	if (crng_node_pool)
877 		crng = crng_node_pool[numa_node_id()];
878 	if (crng == NULL)
879 #endif
880 		crng = &primary_crng;
881 	_extract_crng(crng, out);
882 }
883 
884 /*
885  * Use the leftover bytes from the CRNG block output (if there is
886  * enough) to mutate the CRNG key to provide backtracking protection.
887  */
888 static void _crng_backtrack_protect(struct crng_state *crng,
889 				    __u8 tmp[CHACHA20_BLOCK_SIZE], int used)
890 {
891 	unsigned long	flags;
892 	__u32		*s, *d;
893 	int		i;
894 
895 	used = round_up(used, sizeof(__u32));
896 	if (used + CHACHA20_KEY_SIZE > CHACHA20_BLOCK_SIZE) {
897 		extract_crng(tmp);
898 		used = 0;
899 	}
900 	spin_lock_irqsave(&crng->lock, flags);
901 	s = (__u32 *) &tmp[used];
902 	d = &crng->state[4];
903 	for (i=0; i < 8; i++)
904 		*d++ ^= *s++;
905 	spin_unlock_irqrestore(&crng->lock, flags);
906 }
907 
908 static void crng_backtrack_protect(__u8 tmp[CHACHA20_BLOCK_SIZE], int used)
909 {
910 	struct crng_state *crng = NULL;
911 
912 #ifdef CONFIG_NUMA
913 	if (crng_node_pool)
914 		crng = crng_node_pool[numa_node_id()];
915 	if (crng == NULL)
916 #endif
917 		crng = &primary_crng;
918 	_crng_backtrack_protect(crng, tmp, used);
919 }
920 
921 static ssize_t extract_crng_user(void __user *buf, size_t nbytes)
922 {
923 	ssize_t ret = 0, i = CHACHA20_BLOCK_SIZE;
924 	__u8 tmp[CHACHA20_BLOCK_SIZE];
925 	int large_request = (nbytes > 256);
926 
927 	while (nbytes) {
928 		if (large_request && need_resched()) {
929 			if (signal_pending(current)) {
930 				if (ret == 0)
931 					ret = -ERESTARTSYS;
932 				break;
933 			}
934 			schedule();
935 		}
936 
937 		extract_crng(tmp);
938 		i = min_t(int, nbytes, CHACHA20_BLOCK_SIZE);
939 		if (copy_to_user(buf, tmp, i)) {
940 			ret = -EFAULT;
941 			break;
942 		}
943 
944 		nbytes -= i;
945 		buf += i;
946 		ret += i;
947 	}
948 	crng_backtrack_protect(tmp, i);
949 
950 	/* Wipe data just written to memory */
951 	memzero_explicit(tmp, sizeof(tmp));
952 
953 	return ret;
954 }
955 
956 
957 /*********************************************************************
958  *
959  * Entropy input management
960  *
961  *********************************************************************/
962 
963 /* There is one of these per entropy source */
964 struct timer_rand_state {
965 	cycles_t last_time;
966 	long last_delta, last_delta2;
967 	unsigned dont_count_entropy:1;
968 };
969 
970 #define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
971 
972 /*
973  * Add device- or boot-specific data to the input pool to help
974  * initialize it.
975  *
976  * None of this adds any entropy; it is meant to avoid the problem of
977  * the entropy pool having similar initial state across largely
978  * identical devices.
979  */
980 void add_device_randomness(const void *buf, unsigned int size)
981 {
982 	unsigned long time = random_get_entropy() ^ jiffies;
983 	unsigned long flags;
984 
985 	if (!crng_ready()) {
986 		crng_fast_load(buf, size);
987 		return;
988 	}
989 
990 	trace_add_device_randomness(size, _RET_IP_);
991 	spin_lock_irqsave(&input_pool.lock, flags);
992 	_mix_pool_bytes(&input_pool, buf, size);
993 	_mix_pool_bytes(&input_pool, &time, sizeof(time));
994 	spin_unlock_irqrestore(&input_pool.lock, flags);
995 }
996 EXPORT_SYMBOL(add_device_randomness);
997 
998 static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
999 
1000 /*
1001  * This function adds entropy to the entropy "pool" by using timing
1002  * delays.  It uses the timer_rand_state structure to make an estimate
1003  * of how many bits of entropy this call has added to the pool.
1004  *
1005  * The number "num" is also added to the pool - it should somehow describe
1006  * the type of event which just happened.  This is currently 0-255 for
1007  * keyboard scan codes, and 256 upwards for interrupts.
1008  *
1009  */
1010 static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
1011 {
1012 	struct entropy_store	*r;
1013 	struct {
1014 		long jiffies;
1015 		unsigned cycles;
1016 		unsigned num;
1017 	} sample;
1018 	long delta, delta2, delta3;
1019 
1020 	preempt_disable();
1021 
1022 	sample.jiffies = jiffies;
1023 	sample.cycles = random_get_entropy();
1024 	sample.num = num;
1025 	r = &input_pool;
1026 	mix_pool_bytes(r, &sample, sizeof(sample));
1027 
1028 	/*
1029 	 * Calculate number of bits of randomness we probably added.
1030 	 * We take into account the first, second and third-order deltas
1031 	 * in order to make our estimate.
1032 	 */
1033 
1034 	if (!state->dont_count_entropy) {
1035 		delta = sample.jiffies - state->last_time;
1036 		state->last_time = sample.jiffies;
1037 
1038 		delta2 = delta - state->last_delta;
1039 		state->last_delta = delta;
1040 
1041 		delta3 = delta2 - state->last_delta2;
1042 		state->last_delta2 = delta2;
1043 
1044 		if (delta < 0)
1045 			delta = -delta;
1046 		if (delta2 < 0)
1047 			delta2 = -delta2;
1048 		if (delta3 < 0)
1049 			delta3 = -delta3;
1050 		if (delta > delta2)
1051 			delta = delta2;
1052 		if (delta > delta3)
1053 			delta = delta3;
1054 
1055 		/*
1056 		 * delta is now minimum absolute delta.
1057 		 * Round down by 1 bit on general principles,
1058 		 * and limit entropy entimate to 12 bits.
1059 		 */
1060 		credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
1061 	}
1062 	preempt_enable();
1063 }
1064 
1065 void add_input_randomness(unsigned int type, unsigned int code,
1066 				 unsigned int value)
1067 {
1068 	static unsigned char last_value;
1069 
1070 	/* ignore autorepeat and the like */
1071 	if (value == last_value)
1072 		return;
1073 
1074 	last_value = value;
1075 	add_timer_randomness(&input_timer_state,
1076 			     (type << 4) ^ code ^ (code >> 4) ^ value);
1077 	trace_add_input_randomness(ENTROPY_BITS(&input_pool));
1078 }
1079 EXPORT_SYMBOL_GPL(add_input_randomness);
1080 
1081 static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
1082 
1083 #ifdef ADD_INTERRUPT_BENCH
1084 static unsigned long avg_cycles, avg_deviation;
1085 
1086 #define AVG_SHIFT 8     /* Exponential average factor k=1/256 */
1087 #define FIXED_1_2 (1 << (AVG_SHIFT-1))
1088 
1089 static void add_interrupt_bench(cycles_t start)
1090 {
1091         long delta = random_get_entropy() - start;
1092 
1093         /* Use a weighted moving average */
1094         delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
1095         avg_cycles += delta;
1096         /* And average deviation */
1097         delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
1098         avg_deviation += delta;
1099 }
1100 #else
1101 #define add_interrupt_bench(x)
1102 #endif
1103 
1104 static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
1105 {
1106 	__u32 *ptr = (__u32 *) regs;
1107 	unsigned int idx;
1108 
1109 	if (regs == NULL)
1110 		return 0;
1111 	idx = READ_ONCE(f->reg_idx);
1112 	if (idx >= sizeof(struct pt_regs) / sizeof(__u32))
1113 		idx = 0;
1114 	ptr += idx++;
1115 	WRITE_ONCE(f->reg_idx, idx);
1116 	return *ptr;
1117 }
1118 
1119 void add_interrupt_randomness(int irq, int irq_flags)
1120 {
1121 	struct entropy_store	*r;
1122 	struct fast_pool	*fast_pool = this_cpu_ptr(&irq_randomness);
1123 	struct pt_regs		*regs = get_irq_regs();
1124 	unsigned long		now = jiffies;
1125 	cycles_t		cycles = random_get_entropy();
1126 	__u32			c_high, j_high;
1127 	__u64			ip;
1128 	unsigned long		seed;
1129 	int			credit = 0;
1130 
1131 	if (cycles == 0)
1132 		cycles = get_reg(fast_pool, regs);
1133 	c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
1134 	j_high = (sizeof(now) > 4) ? now >> 32 : 0;
1135 	fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
1136 	fast_pool->pool[1] ^= now ^ c_high;
1137 	ip = regs ? instruction_pointer(regs) : _RET_IP_;
1138 	fast_pool->pool[2] ^= ip;
1139 	fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
1140 		get_reg(fast_pool, regs);
1141 
1142 	fast_mix(fast_pool);
1143 	add_interrupt_bench(cycles);
1144 
1145 	if (!crng_ready()) {
1146 		if ((fast_pool->count >= 64) &&
1147 		    crng_fast_load((char *) fast_pool->pool,
1148 				   sizeof(fast_pool->pool))) {
1149 			fast_pool->count = 0;
1150 			fast_pool->last = now;
1151 		}
1152 		return;
1153 	}
1154 
1155 	if ((fast_pool->count < 64) &&
1156 	    !time_after(now, fast_pool->last + HZ))
1157 		return;
1158 
1159 	r = &input_pool;
1160 	if (!spin_trylock(&r->lock))
1161 		return;
1162 
1163 	fast_pool->last = now;
1164 	__mix_pool_bytes(r, &fast_pool->pool, sizeof(fast_pool->pool));
1165 
1166 	/*
1167 	 * If we have architectural seed generator, produce a seed and
1168 	 * add it to the pool.  For the sake of paranoia don't let the
1169 	 * architectural seed generator dominate the input from the
1170 	 * interrupt noise.
1171 	 */
1172 	if (arch_get_random_seed_long(&seed)) {
1173 		__mix_pool_bytes(r, &seed, sizeof(seed));
1174 		credit = 1;
1175 	}
1176 	spin_unlock(&r->lock);
1177 
1178 	fast_pool->count = 0;
1179 
1180 	/* award one bit for the contents of the fast pool */
1181 	credit_entropy_bits(r, credit + 1);
1182 }
1183 EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1184 
1185 #ifdef CONFIG_BLOCK
1186 void add_disk_randomness(struct gendisk *disk)
1187 {
1188 	if (!disk || !disk->random)
1189 		return;
1190 	/* first major is 1, so we get >= 0x200 here */
1191 	add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1192 	trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
1193 }
1194 EXPORT_SYMBOL_GPL(add_disk_randomness);
1195 #endif
1196 
1197 /*********************************************************************
1198  *
1199  * Entropy extraction routines
1200  *
1201  *********************************************************************/
1202 
1203 /*
1204  * This utility inline function is responsible for transferring entropy
1205  * from the primary pool to the secondary extraction pool. We make
1206  * sure we pull enough for a 'catastrophic reseed'.
1207  */
1208 static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes);
1209 static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1210 {
1211 	if (!r->pull ||
1212 	    r->entropy_count >= (nbytes << (ENTROPY_SHIFT + 3)) ||
1213 	    r->entropy_count > r->poolinfo->poolfracbits)
1214 		return;
1215 
1216 	_xfer_secondary_pool(r, nbytes);
1217 }
1218 
1219 static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1220 {
1221 	__u32	tmp[OUTPUT_POOL_WORDS];
1222 
1223 	int bytes = nbytes;
1224 
1225 	/* pull at least as much as a wakeup */
1226 	bytes = max_t(int, bytes, random_read_wakeup_bits / 8);
1227 	/* but never more than the buffer size */
1228 	bytes = min_t(int, bytes, sizeof(tmp));
1229 
1230 	trace_xfer_secondary_pool(r->name, bytes * 8, nbytes * 8,
1231 				  ENTROPY_BITS(r), ENTROPY_BITS(r->pull));
1232 	bytes = extract_entropy(r->pull, tmp, bytes,
1233 				random_read_wakeup_bits / 8, 0);
1234 	mix_pool_bytes(r, tmp, bytes);
1235 	credit_entropy_bits(r, bytes*8);
1236 }
1237 
1238 /*
1239  * Used as a workqueue function so that when the input pool is getting
1240  * full, we can "spill over" some entropy to the output pools.  That
1241  * way the output pools can store some of the excess entropy instead
1242  * of letting it go to waste.
1243  */
1244 static void push_to_pool(struct work_struct *work)
1245 {
1246 	struct entropy_store *r = container_of(work, struct entropy_store,
1247 					      push_work);
1248 	BUG_ON(!r);
1249 	_xfer_secondary_pool(r, random_read_wakeup_bits/8);
1250 	trace_push_to_pool(r->name, r->entropy_count >> ENTROPY_SHIFT,
1251 			   r->pull->entropy_count >> ENTROPY_SHIFT);
1252 }
1253 
1254 /*
1255  * This function decides how many bytes to actually take from the
1256  * given pool, and also debits the entropy count accordingly.
1257  */
1258 static size_t account(struct entropy_store *r, size_t nbytes, int min,
1259 		      int reserved)
1260 {
1261 	int entropy_count, orig, have_bytes;
1262 	size_t ibytes, nfrac;
1263 
1264 	BUG_ON(r->entropy_count > r->poolinfo->poolfracbits);
1265 
1266 	/* Can we pull enough? */
1267 retry:
1268 	entropy_count = orig = ACCESS_ONCE(r->entropy_count);
1269 	ibytes = nbytes;
1270 	/* never pull more than available */
1271 	have_bytes = entropy_count >> (ENTROPY_SHIFT + 3);
1272 
1273 	if ((have_bytes -= reserved) < 0)
1274 		have_bytes = 0;
1275 	ibytes = min_t(size_t, ibytes, have_bytes);
1276 	if (ibytes < min)
1277 		ibytes = 0;
1278 
1279 	if (unlikely(entropy_count < 0)) {
1280 		pr_warn("random: negative entropy count: pool %s count %d\n",
1281 			r->name, entropy_count);
1282 		WARN_ON(1);
1283 		entropy_count = 0;
1284 	}
1285 	nfrac = ibytes << (ENTROPY_SHIFT + 3);
1286 	if ((size_t) entropy_count > nfrac)
1287 		entropy_count -= nfrac;
1288 	else
1289 		entropy_count = 0;
1290 
1291 	if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
1292 		goto retry;
1293 
1294 	trace_debit_entropy(r->name, 8 * ibytes);
1295 	if (ibytes &&
1296 	    (r->entropy_count >> ENTROPY_SHIFT) < random_write_wakeup_bits) {
1297 		wake_up_interruptible(&random_write_wait);
1298 		kill_fasync(&fasync, SIGIO, POLL_OUT);
1299 	}
1300 
1301 	return ibytes;
1302 }
1303 
1304 /*
1305  * This function does the actual extraction for extract_entropy and
1306  * extract_entropy_user.
1307  *
1308  * Note: we assume that .poolwords is a multiple of 16 words.
1309  */
1310 static void extract_buf(struct entropy_store *r, __u8 *out)
1311 {
1312 	int i;
1313 	union {
1314 		__u32 w[5];
1315 		unsigned long l[LONGS(20)];
1316 	} hash;
1317 	__u32 workspace[SHA_WORKSPACE_WORDS];
1318 	unsigned long flags;
1319 
1320 	/*
1321 	 * If we have an architectural hardware random number
1322 	 * generator, use it for SHA's initial vector
1323 	 */
1324 	sha_init(hash.w);
1325 	for (i = 0; i < LONGS(20); i++) {
1326 		unsigned long v;
1327 		if (!arch_get_random_long(&v))
1328 			break;
1329 		hash.l[i] = v;
1330 	}
1331 
1332 	/* Generate a hash across the pool, 16 words (512 bits) at a time */
1333 	spin_lock_irqsave(&r->lock, flags);
1334 	for (i = 0; i < r->poolinfo->poolwords; i += 16)
1335 		sha_transform(hash.w, (__u8 *)(r->pool + i), workspace);
1336 
1337 	/*
1338 	 * We mix the hash back into the pool to prevent backtracking
1339 	 * attacks (where the attacker knows the state of the pool
1340 	 * plus the current outputs, and attempts to find previous
1341 	 * ouputs), unless the hash function can be inverted. By
1342 	 * mixing at least a SHA1 worth of hash data back, we make
1343 	 * brute-forcing the feedback as hard as brute-forcing the
1344 	 * hash.
1345 	 */
1346 	__mix_pool_bytes(r, hash.w, sizeof(hash.w));
1347 	spin_unlock_irqrestore(&r->lock, flags);
1348 
1349 	memzero_explicit(workspace, sizeof(workspace));
1350 
1351 	/*
1352 	 * In case the hash function has some recognizable output
1353 	 * pattern, we fold it in half. Thus, we always feed back
1354 	 * twice as much data as we output.
1355 	 */
1356 	hash.w[0] ^= hash.w[3];
1357 	hash.w[1] ^= hash.w[4];
1358 	hash.w[2] ^= rol32(hash.w[2], 16);
1359 
1360 	memcpy(out, &hash, EXTRACT_SIZE);
1361 	memzero_explicit(&hash, sizeof(hash));
1362 }
1363 
1364 static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
1365 				size_t nbytes, int fips)
1366 {
1367 	ssize_t ret = 0, i;
1368 	__u8 tmp[EXTRACT_SIZE];
1369 	unsigned long flags;
1370 
1371 	while (nbytes) {
1372 		extract_buf(r, tmp);
1373 
1374 		if (fips) {
1375 			spin_lock_irqsave(&r->lock, flags);
1376 			if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
1377 				panic("Hardware RNG duplicated output!\n");
1378 			memcpy(r->last_data, tmp, EXTRACT_SIZE);
1379 			spin_unlock_irqrestore(&r->lock, flags);
1380 		}
1381 		i = min_t(int, nbytes, EXTRACT_SIZE);
1382 		memcpy(buf, tmp, i);
1383 		nbytes -= i;
1384 		buf += i;
1385 		ret += i;
1386 	}
1387 
1388 	/* Wipe data just returned from memory */
1389 	memzero_explicit(tmp, sizeof(tmp));
1390 
1391 	return ret;
1392 }
1393 
1394 /*
1395  * This function extracts randomness from the "entropy pool", and
1396  * returns it in a buffer.
1397  *
1398  * The min parameter specifies the minimum amount we can pull before
1399  * failing to avoid races that defeat catastrophic reseeding while the
1400  * reserved parameter indicates how much entropy we must leave in the
1401  * pool after each pull to avoid starving other readers.
1402  */
1403 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1404 				 size_t nbytes, int min, int reserved)
1405 {
1406 	__u8 tmp[EXTRACT_SIZE];
1407 	unsigned long flags;
1408 
1409 	/* if last_data isn't primed, we need EXTRACT_SIZE extra bytes */
1410 	if (fips_enabled) {
1411 		spin_lock_irqsave(&r->lock, flags);
1412 		if (!r->last_data_init) {
1413 			r->last_data_init = 1;
1414 			spin_unlock_irqrestore(&r->lock, flags);
1415 			trace_extract_entropy(r->name, EXTRACT_SIZE,
1416 					      ENTROPY_BITS(r), _RET_IP_);
1417 			xfer_secondary_pool(r, EXTRACT_SIZE);
1418 			extract_buf(r, tmp);
1419 			spin_lock_irqsave(&r->lock, flags);
1420 			memcpy(r->last_data, tmp, EXTRACT_SIZE);
1421 		}
1422 		spin_unlock_irqrestore(&r->lock, flags);
1423 	}
1424 
1425 	trace_extract_entropy(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1426 	xfer_secondary_pool(r, nbytes);
1427 	nbytes = account(r, nbytes, min, reserved);
1428 
1429 	return _extract_entropy(r, buf, nbytes, fips_enabled);
1430 }
1431 
1432 /*
1433  * This function extracts randomness from the "entropy pool", and
1434  * returns it in a userspace buffer.
1435  */
1436 static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
1437 				    size_t nbytes)
1438 {
1439 	ssize_t ret = 0, i;
1440 	__u8 tmp[EXTRACT_SIZE];
1441 	int large_request = (nbytes > 256);
1442 
1443 	trace_extract_entropy_user(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1444 	xfer_secondary_pool(r, nbytes);
1445 	nbytes = account(r, nbytes, 0, 0);
1446 
1447 	while (nbytes) {
1448 		if (large_request && need_resched()) {
1449 			if (signal_pending(current)) {
1450 				if (ret == 0)
1451 					ret = -ERESTARTSYS;
1452 				break;
1453 			}
1454 			schedule();
1455 		}
1456 
1457 		extract_buf(r, tmp);
1458 		i = min_t(int, nbytes, EXTRACT_SIZE);
1459 		if (copy_to_user(buf, tmp, i)) {
1460 			ret = -EFAULT;
1461 			break;
1462 		}
1463 
1464 		nbytes -= i;
1465 		buf += i;
1466 		ret += i;
1467 	}
1468 
1469 	/* Wipe data just returned from memory */
1470 	memzero_explicit(tmp, sizeof(tmp));
1471 
1472 	return ret;
1473 }
1474 
1475 #define warn_unseeded_randomness(previous) \
1476 	_warn_unseeded_randomness(__func__, (void *) _RET_IP_, (previous))
1477 
1478 static void _warn_unseeded_randomness(const char *func_name, void *caller,
1479 				      void **previous)
1480 {
1481 #ifdef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1482 	const bool print_once = false;
1483 #else
1484 	static bool print_once __read_mostly;
1485 #endif
1486 
1487 	if (print_once ||
1488 	    crng_ready() ||
1489 	    (previous && (caller == READ_ONCE(*previous))))
1490 		return;
1491 	WRITE_ONCE(*previous, caller);
1492 #ifndef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1493 	print_once = true;
1494 #endif
1495 	pr_notice("random: %s called from %pS with crng_init=%d\n",
1496 		  func_name, caller, crng_init);
1497 }
1498 
1499 /*
1500  * This function is the exported kernel interface.  It returns some
1501  * number of good random numbers, suitable for key generation, seeding
1502  * TCP sequence numbers, etc.  It does not rely on the hardware random
1503  * number generator.  For random bytes direct from the hardware RNG
1504  * (when available), use get_random_bytes_arch(). In order to ensure
1505  * that the randomness provided by this function is okay, the function
1506  * wait_for_random_bytes() should be called and return 0 at least once
1507  * at any point prior.
1508  */
1509 static void _get_random_bytes(void *buf, int nbytes)
1510 {
1511 	__u8 tmp[CHACHA20_BLOCK_SIZE];
1512 
1513 	trace_get_random_bytes(nbytes, _RET_IP_);
1514 
1515 	while (nbytes >= CHACHA20_BLOCK_SIZE) {
1516 		extract_crng(buf);
1517 		buf += CHACHA20_BLOCK_SIZE;
1518 		nbytes -= CHACHA20_BLOCK_SIZE;
1519 	}
1520 
1521 	if (nbytes > 0) {
1522 		extract_crng(tmp);
1523 		memcpy(buf, tmp, nbytes);
1524 		crng_backtrack_protect(tmp, nbytes);
1525 	} else
1526 		crng_backtrack_protect(tmp, CHACHA20_BLOCK_SIZE);
1527 	memzero_explicit(tmp, sizeof(tmp));
1528 }
1529 
1530 void get_random_bytes(void *buf, int nbytes)
1531 {
1532 	static void *previous;
1533 
1534 	warn_unseeded_randomness(&previous);
1535 	_get_random_bytes(buf, nbytes);
1536 }
1537 EXPORT_SYMBOL(get_random_bytes);
1538 
1539 /*
1540  * Wait for the urandom pool to be seeded and thus guaranteed to supply
1541  * cryptographically secure random numbers. This applies to: the /dev/urandom
1542  * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
1543  * family of functions. Using any of these functions without first calling
1544  * this function forfeits the guarantee of security.
1545  *
1546  * Returns: 0 if the urandom pool has been seeded.
1547  *          -ERESTARTSYS if the function was interrupted by a signal.
1548  */
1549 int wait_for_random_bytes(void)
1550 {
1551 	if (likely(crng_ready()))
1552 		return 0;
1553 	return wait_event_interruptible(crng_init_wait, crng_ready());
1554 }
1555 EXPORT_SYMBOL(wait_for_random_bytes);
1556 
1557 /*
1558  * Add a callback function that will be invoked when the nonblocking
1559  * pool is initialised.
1560  *
1561  * returns: 0 if callback is successfully added
1562  *	    -EALREADY if pool is already initialised (callback not called)
1563  *	    -ENOENT if module for callback is not alive
1564  */
1565 int add_random_ready_callback(struct random_ready_callback *rdy)
1566 {
1567 	struct module *owner;
1568 	unsigned long flags;
1569 	int err = -EALREADY;
1570 
1571 	if (crng_ready())
1572 		return err;
1573 
1574 	owner = rdy->owner;
1575 	if (!try_module_get(owner))
1576 		return -ENOENT;
1577 
1578 	spin_lock_irqsave(&random_ready_list_lock, flags);
1579 	if (crng_ready())
1580 		goto out;
1581 
1582 	owner = NULL;
1583 
1584 	list_add(&rdy->list, &random_ready_list);
1585 	err = 0;
1586 
1587 out:
1588 	spin_unlock_irqrestore(&random_ready_list_lock, flags);
1589 
1590 	module_put(owner);
1591 
1592 	return err;
1593 }
1594 EXPORT_SYMBOL(add_random_ready_callback);
1595 
1596 /*
1597  * Delete a previously registered readiness callback function.
1598  */
1599 void del_random_ready_callback(struct random_ready_callback *rdy)
1600 {
1601 	unsigned long flags;
1602 	struct module *owner = NULL;
1603 
1604 	spin_lock_irqsave(&random_ready_list_lock, flags);
1605 	if (!list_empty(&rdy->list)) {
1606 		list_del_init(&rdy->list);
1607 		owner = rdy->owner;
1608 	}
1609 	spin_unlock_irqrestore(&random_ready_list_lock, flags);
1610 
1611 	module_put(owner);
1612 }
1613 EXPORT_SYMBOL(del_random_ready_callback);
1614 
1615 /*
1616  * This function will use the architecture-specific hardware random
1617  * number generator if it is available.  The arch-specific hw RNG will
1618  * almost certainly be faster than what we can do in software, but it
1619  * is impossible to verify that it is implemented securely (as
1620  * opposed, to, say, the AES encryption of a sequence number using a
1621  * key known by the NSA).  So it's useful if we need the speed, but
1622  * only if we're willing to trust the hardware manufacturer not to
1623  * have put in a back door.
1624  */
1625 void get_random_bytes_arch(void *buf, int nbytes)
1626 {
1627 	char *p = buf;
1628 
1629 	trace_get_random_bytes_arch(nbytes, _RET_IP_);
1630 	while (nbytes) {
1631 		unsigned long v;
1632 		int chunk = min(nbytes, (int)sizeof(unsigned long));
1633 
1634 		if (!arch_get_random_long(&v))
1635 			break;
1636 
1637 		memcpy(p, &v, chunk);
1638 		p += chunk;
1639 		nbytes -= chunk;
1640 	}
1641 
1642 	if (nbytes)
1643 		get_random_bytes(p, nbytes);
1644 }
1645 EXPORT_SYMBOL(get_random_bytes_arch);
1646 
1647 
1648 /*
1649  * init_std_data - initialize pool with system data
1650  *
1651  * @r: pool to initialize
1652  *
1653  * This function clears the pool's entropy count and mixes some system
1654  * data into the pool to prepare it for use. The pool is not cleared
1655  * as that can only decrease the entropy in the pool.
1656  */
1657 static void init_std_data(struct entropy_store *r)
1658 {
1659 	int i;
1660 	ktime_t now = ktime_get_real();
1661 	unsigned long rv;
1662 
1663 	r->last_pulled = jiffies;
1664 	mix_pool_bytes(r, &now, sizeof(now));
1665 	for (i = r->poolinfo->poolbytes; i > 0; i -= sizeof(rv)) {
1666 		if (!arch_get_random_seed_long(&rv) &&
1667 		    !arch_get_random_long(&rv))
1668 			rv = random_get_entropy();
1669 		mix_pool_bytes(r, &rv, sizeof(rv));
1670 	}
1671 	mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1672 }
1673 
1674 /*
1675  * Note that setup_arch() may call add_device_randomness()
1676  * long before we get here. This allows seeding of the pools
1677  * with some platform dependent data very early in the boot
1678  * process. But it limits our options here. We must use
1679  * statically allocated structures that already have all
1680  * initializations complete at compile time. We should also
1681  * take care not to overwrite the precious per platform data
1682  * we were given.
1683  */
1684 static int rand_initialize(void)
1685 {
1686 #ifdef CONFIG_NUMA
1687 	int i;
1688 	struct crng_state *crng;
1689 	struct crng_state **pool;
1690 #endif
1691 
1692 	init_std_data(&input_pool);
1693 	init_std_data(&blocking_pool);
1694 	crng_initialize(&primary_crng);
1695 
1696 #ifdef CONFIG_NUMA
1697 	pool = kcalloc(nr_node_ids, sizeof(*pool), GFP_KERNEL|__GFP_NOFAIL);
1698 	for_each_online_node(i) {
1699 		crng = kmalloc_node(sizeof(struct crng_state),
1700 				    GFP_KERNEL | __GFP_NOFAIL, i);
1701 		spin_lock_init(&crng->lock);
1702 		crng_initialize(crng);
1703 		pool[i] = crng;
1704 	}
1705 	mb();
1706 	crng_node_pool = pool;
1707 #endif
1708 	return 0;
1709 }
1710 early_initcall(rand_initialize);
1711 
1712 #ifdef CONFIG_BLOCK
1713 void rand_initialize_disk(struct gendisk *disk)
1714 {
1715 	struct timer_rand_state *state;
1716 
1717 	/*
1718 	 * If kzalloc returns null, we just won't use that entropy
1719 	 * source.
1720 	 */
1721 	state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1722 	if (state) {
1723 		state->last_time = INITIAL_JIFFIES;
1724 		disk->random = state;
1725 	}
1726 }
1727 #endif
1728 
1729 static ssize_t
1730 _random_read(int nonblock, char __user *buf, size_t nbytes)
1731 {
1732 	ssize_t n;
1733 
1734 	if (nbytes == 0)
1735 		return 0;
1736 
1737 	nbytes = min_t(size_t, nbytes, SEC_XFER_SIZE);
1738 	while (1) {
1739 		n = extract_entropy_user(&blocking_pool, buf, nbytes);
1740 		if (n < 0)
1741 			return n;
1742 		trace_random_read(n*8, (nbytes-n)*8,
1743 				  ENTROPY_BITS(&blocking_pool),
1744 				  ENTROPY_BITS(&input_pool));
1745 		if (n > 0)
1746 			return n;
1747 
1748 		/* Pool is (near) empty.  Maybe wait and retry. */
1749 		if (nonblock)
1750 			return -EAGAIN;
1751 
1752 		wait_event_interruptible(random_read_wait,
1753 			ENTROPY_BITS(&input_pool) >=
1754 			random_read_wakeup_bits);
1755 		if (signal_pending(current))
1756 			return -ERESTARTSYS;
1757 	}
1758 }
1759 
1760 static ssize_t
1761 random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1762 {
1763 	return _random_read(file->f_flags & O_NONBLOCK, buf, nbytes);
1764 }
1765 
1766 static ssize_t
1767 urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1768 {
1769 	unsigned long flags;
1770 	static int maxwarn = 10;
1771 	int ret;
1772 
1773 	if (!crng_ready() && maxwarn > 0) {
1774 		maxwarn--;
1775 		printk(KERN_NOTICE "random: %s: uninitialized urandom read "
1776 		       "(%zd bytes read)\n",
1777 		       current->comm, nbytes);
1778 		spin_lock_irqsave(&primary_crng.lock, flags);
1779 		crng_init_cnt = 0;
1780 		spin_unlock_irqrestore(&primary_crng.lock, flags);
1781 	}
1782 	nbytes = min_t(size_t, nbytes, INT_MAX >> (ENTROPY_SHIFT + 3));
1783 	ret = extract_crng_user(buf, nbytes);
1784 	trace_urandom_read(8 * nbytes, 0, ENTROPY_BITS(&input_pool));
1785 	return ret;
1786 }
1787 
1788 static unsigned int
1789 random_poll(struct file *file, poll_table * wait)
1790 {
1791 	unsigned int mask;
1792 
1793 	poll_wait(file, &random_read_wait, wait);
1794 	poll_wait(file, &random_write_wait, wait);
1795 	mask = 0;
1796 	if (ENTROPY_BITS(&input_pool) >= random_read_wakeup_bits)
1797 		mask |= POLLIN | POLLRDNORM;
1798 	if (ENTROPY_BITS(&input_pool) < random_write_wakeup_bits)
1799 		mask |= POLLOUT | POLLWRNORM;
1800 	return mask;
1801 }
1802 
1803 static int
1804 write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1805 {
1806 	size_t bytes;
1807 	__u32 buf[16];
1808 	const char __user *p = buffer;
1809 
1810 	while (count > 0) {
1811 		bytes = min(count, sizeof(buf));
1812 		if (copy_from_user(&buf, p, bytes))
1813 			return -EFAULT;
1814 
1815 		count -= bytes;
1816 		p += bytes;
1817 
1818 		mix_pool_bytes(r, buf, bytes);
1819 		cond_resched();
1820 	}
1821 
1822 	return 0;
1823 }
1824 
1825 static ssize_t random_write(struct file *file, const char __user *buffer,
1826 			    size_t count, loff_t *ppos)
1827 {
1828 	size_t ret;
1829 
1830 	ret = write_pool(&input_pool, buffer, count);
1831 	if (ret)
1832 		return ret;
1833 
1834 	return (ssize_t)count;
1835 }
1836 
1837 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1838 {
1839 	int size, ent_count;
1840 	int __user *p = (int __user *)arg;
1841 	int retval;
1842 
1843 	switch (cmd) {
1844 	case RNDGETENTCNT:
1845 		/* inherently racy, no point locking */
1846 		ent_count = ENTROPY_BITS(&input_pool);
1847 		if (put_user(ent_count, p))
1848 			return -EFAULT;
1849 		return 0;
1850 	case RNDADDTOENTCNT:
1851 		if (!capable(CAP_SYS_ADMIN))
1852 			return -EPERM;
1853 		if (get_user(ent_count, p))
1854 			return -EFAULT;
1855 		return credit_entropy_bits_safe(&input_pool, ent_count);
1856 	case RNDADDENTROPY:
1857 		if (!capable(CAP_SYS_ADMIN))
1858 			return -EPERM;
1859 		if (get_user(ent_count, p++))
1860 			return -EFAULT;
1861 		if (ent_count < 0)
1862 			return -EINVAL;
1863 		if (get_user(size, p++))
1864 			return -EFAULT;
1865 		retval = write_pool(&input_pool, (const char __user *)p,
1866 				    size);
1867 		if (retval < 0)
1868 			return retval;
1869 		return credit_entropy_bits_safe(&input_pool, ent_count);
1870 	case RNDZAPENTCNT:
1871 	case RNDCLEARPOOL:
1872 		/*
1873 		 * Clear the entropy pool counters. We no longer clear
1874 		 * the entropy pool, as that's silly.
1875 		 */
1876 		if (!capable(CAP_SYS_ADMIN))
1877 			return -EPERM;
1878 		input_pool.entropy_count = 0;
1879 		blocking_pool.entropy_count = 0;
1880 		return 0;
1881 	default:
1882 		return -EINVAL;
1883 	}
1884 }
1885 
1886 static int random_fasync(int fd, struct file *filp, int on)
1887 {
1888 	return fasync_helper(fd, filp, on, &fasync);
1889 }
1890 
1891 const struct file_operations random_fops = {
1892 	.read  = random_read,
1893 	.write = random_write,
1894 	.poll  = random_poll,
1895 	.unlocked_ioctl = random_ioctl,
1896 	.fasync = random_fasync,
1897 	.llseek = noop_llseek,
1898 };
1899 
1900 const struct file_operations urandom_fops = {
1901 	.read  = urandom_read,
1902 	.write = random_write,
1903 	.unlocked_ioctl = random_ioctl,
1904 	.fasync = random_fasync,
1905 	.llseek = noop_llseek,
1906 };
1907 
1908 SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count,
1909 		unsigned int, flags)
1910 {
1911 	int ret;
1912 
1913 	if (flags & ~(GRND_NONBLOCK|GRND_RANDOM))
1914 		return -EINVAL;
1915 
1916 	if (count > INT_MAX)
1917 		count = INT_MAX;
1918 
1919 	if (flags & GRND_RANDOM)
1920 		return _random_read(flags & GRND_NONBLOCK, buf, count);
1921 
1922 	if (!crng_ready()) {
1923 		if (flags & GRND_NONBLOCK)
1924 			return -EAGAIN;
1925 		ret = wait_for_random_bytes();
1926 		if (unlikely(ret))
1927 			return ret;
1928 	}
1929 	return urandom_read(NULL, buf, count, NULL);
1930 }
1931 
1932 /********************************************************************
1933  *
1934  * Sysctl interface
1935  *
1936  ********************************************************************/
1937 
1938 #ifdef CONFIG_SYSCTL
1939 
1940 #include <linux/sysctl.h>
1941 
1942 static int min_read_thresh = 8, min_write_thresh;
1943 static int max_read_thresh = OUTPUT_POOL_WORDS * 32;
1944 static int max_write_thresh = INPUT_POOL_WORDS * 32;
1945 static int random_min_urandom_seed = 60;
1946 static char sysctl_bootid[16];
1947 
1948 /*
1949  * This function is used to return both the bootid UUID, and random
1950  * UUID.  The difference is in whether table->data is NULL; if it is,
1951  * then a new UUID is generated and returned to the user.
1952  *
1953  * If the user accesses this via the proc interface, the UUID will be
1954  * returned as an ASCII string in the standard UUID format; if via the
1955  * sysctl system call, as 16 bytes of binary data.
1956  */
1957 static int proc_do_uuid(struct ctl_table *table, int write,
1958 			void __user *buffer, size_t *lenp, loff_t *ppos)
1959 {
1960 	struct ctl_table fake_table;
1961 	unsigned char buf[64], tmp_uuid[16], *uuid;
1962 
1963 	uuid = table->data;
1964 	if (!uuid) {
1965 		uuid = tmp_uuid;
1966 		generate_random_uuid(uuid);
1967 	} else {
1968 		static DEFINE_SPINLOCK(bootid_spinlock);
1969 
1970 		spin_lock(&bootid_spinlock);
1971 		if (!uuid[8])
1972 			generate_random_uuid(uuid);
1973 		spin_unlock(&bootid_spinlock);
1974 	}
1975 
1976 	sprintf(buf, "%pU", uuid);
1977 
1978 	fake_table.data = buf;
1979 	fake_table.maxlen = sizeof(buf);
1980 
1981 	return proc_dostring(&fake_table, write, buffer, lenp, ppos);
1982 }
1983 
1984 /*
1985  * Return entropy available scaled to integral bits
1986  */
1987 static int proc_do_entropy(struct ctl_table *table, int write,
1988 			   void __user *buffer, size_t *lenp, loff_t *ppos)
1989 {
1990 	struct ctl_table fake_table;
1991 	int entropy_count;
1992 
1993 	entropy_count = *(int *)table->data >> ENTROPY_SHIFT;
1994 
1995 	fake_table.data = &entropy_count;
1996 	fake_table.maxlen = sizeof(entropy_count);
1997 
1998 	return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
1999 }
2000 
2001 static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
2002 extern struct ctl_table random_table[];
2003 struct ctl_table random_table[] = {
2004 	{
2005 		.procname	= "poolsize",
2006 		.data		= &sysctl_poolsize,
2007 		.maxlen		= sizeof(int),
2008 		.mode		= 0444,
2009 		.proc_handler	= proc_dointvec,
2010 	},
2011 	{
2012 		.procname	= "entropy_avail",
2013 		.maxlen		= sizeof(int),
2014 		.mode		= 0444,
2015 		.proc_handler	= proc_do_entropy,
2016 		.data		= &input_pool.entropy_count,
2017 	},
2018 	{
2019 		.procname	= "read_wakeup_threshold",
2020 		.data		= &random_read_wakeup_bits,
2021 		.maxlen		= sizeof(int),
2022 		.mode		= 0644,
2023 		.proc_handler	= proc_dointvec_minmax,
2024 		.extra1		= &min_read_thresh,
2025 		.extra2		= &max_read_thresh,
2026 	},
2027 	{
2028 		.procname	= "write_wakeup_threshold",
2029 		.data		= &random_write_wakeup_bits,
2030 		.maxlen		= sizeof(int),
2031 		.mode		= 0644,
2032 		.proc_handler	= proc_dointvec_minmax,
2033 		.extra1		= &min_write_thresh,
2034 		.extra2		= &max_write_thresh,
2035 	},
2036 	{
2037 		.procname	= "urandom_min_reseed_secs",
2038 		.data		= &random_min_urandom_seed,
2039 		.maxlen		= sizeof(int),
2040 		.mode		= 0644,
2041 		.proc_handler	= proc_dointvec,
2042 	},
2043 	{
2044 		.procname	= "boot_id",
2045 		.data		= &sysctl_bootid,
2046 		.maxlen		= 16,
2047 		.mode		= 0444,
2048 		.proc_handler	= proc_do_uuid,
2049 	},
2050 	{
2051 		.procname	= "uuid",
2052 		.maxlen		= 16,
2053 		.mode		= 0444,
2054 		.proc_handler	= proc_do_uuid,
2055 	},
2056 #ifdef ADD_INTERRUPT_BENCH
2057 	{
2058 		.procname	= "add_interrupt_avg_cycles",
2059 		.data		= &avg_cycles,
2060 		.maxlen		= sizeof(avg_cycles),
2061 		.mode		= 0444,
2062 		.proc_handler	= proc_doulongvec_minmax,
2063 	},
2064 	{
2065 		.procname	= "add_interrupt_avg_deviation",
2066 		.data		= &avg_deviation,
2067 		.maxlen		= sizeof(avg_deviation),
2068 		.mode		= 0444,
2069 		.proc_handler	= proc_doulongvec_minmax,
2070 	},
2071 #endif
2072 	{ }
2073 };
2074 #endif 	/* CONFIG_SYSCTL */
2075 
2076 struct batched_entropy {
2077 	union {
2078 		u64 entropy_u64[CHACHA20_BLOCK_SIZE / sizeof(u64)];
2079 		u32 entropy_u32[CHACHA20_BLOCK_SIZE / sizeof(u32)];
2080 	};
2081 	unsigned int position;
2082 };
2083 static rwlock_t batched_entropy_reset_lock = __RW_LOCK_UNLOCKED(batched_entropy_reset_lock);
2084 
2085 /*
2086  * Get a random word for internal kernel use only. The quality of the random
2087  * number is either as good as RDRAND or as good as /dev/urandom, with the
2088  * goal of being quite fast and not depleting entropy. In order to ensure
2089  * that the randomness provided by this function is okay, the function
2090  * wait_for_random_bytes() should be called and return 0 at least once
2091  * at any point prior.
2092  */
2093 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64);
2094 u64 get_random_u64(void)
2095 {
2096 	u64 ret;
2097 	bool use_lock;
2098 	unsigned long flags = 0;
2099 	struct batched_entropy *batch;
2100 	static void *previous;
2101 
2102 #if BITS_PER_LONG == 64
2103 	if (arch_get_random_long((unsigned long *)&ret))
2104 		return ret;
2105 #else
2106 	if (arch_get_random_long((unsigned long *)&ret) &&
2107 	    arch_get_random_long((unsigned long *)&ret + 1))
2108 	    return ret;
2109 #endif
2110 
2111 	warn_unseeded_randomness(&previous);
2112 
2113 	use_lock = READ_ONCE(crng_init) < 2;
2114 	batch = &get_cpu_var(batched_entropy_u64);
2115 	if (use_lock)
2116 		read_lock_irqsave(&batched_entropy_reset_lock, flags);
2117 	if (batch->position % ARRAY_SIZE(batch->entropy_u64) == 0) {
2118 		extract_crng((u8 *)batch->entropy_u64);
2119 		batch->position = 0;
2120 	}
2121 	ret = batch->entropy_u64[batch->position++];
2122 	if (use_lock)
2123 		read_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2124 	put_cpu_var(batched_entropy_u64);
2125 	return ret;
2126 }
2127 EXPORT_SYMBOL(get_random_u64);
2128 
2129 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32);
2130 u32 get_random_u32(void)
2131 {
2132 	u32 ret;
2133 	bool use_lock;
2134 	unsigned long flags = 0;
2135 	struct batched_entropy *batch;
2136 	static void *previous;
2137 
2138 	if (arch_get_random_int(&ret))
2139 		return ret;
2140 
2141 	warn_unseeded_randomness(&previous);
2142 
2143 	use_lock = READ_ONCE(crng_init) < 2;
2144 	batch = &get_cpu_var(batched_entropy_u32);
2145 	if (use_lock)
2146 		read_lock_irqsave(&batched_entropy_reset_lock, flags);
2147 	if (batch->position % ARRAY_SIZE(batch->entropy_u32) == 0) {
2148 		extract_crng((u8 *)batch->entropy_u32);
2149 		batch->position = 0;
2150 	}
2151 	ret = batch->entropy_u32[batch->position++];
2152 	if (use_lock)
2153 		read_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2154 	put_cpu_var(batched_entropy_u32);
2155 	return ret;
2156 }
2157 EXPORT_SYMBOL(get_random_u32);
2158 
2159 /* It's important to invalidate all potential batched entropy that might
2160  * be stored before the crng is initialized, which we can do lazily by
2161  * simply resetting the counter to zero so that it's re-extracted on the
2162  * next usage. */
2163 static void invalidate_batched_entropy(void)
2164 {
2165 	int cpu;
2166 	unsigned long flags;
2167 
2168 	write_lock_irqsave(&batched_entropy_reset_lock, flags);
2169 	for_each_possible_cpu (cpu) {
2170 		per_cpu_ptr(&batched_entropy_u32, cpu)->position = 0;
2171 		per_cpu_ptr(&batched_entropy_u64, cpu)->position = 0;
2172 	}
2173 	write_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2174 }
2175 
2176 /**
2177  * randomize_page - Generate a random, page aligned address
2178  * @start:	The smallest acceptable address the caller will take.
2179  * @range:	The size of the area, starting at @start, within which the
2180  *		random address must fall.
2181  *
2182  * If @start + @range would overflow, @range is capped.
2183  *
2184  * NOTE: Historical use of randomize_range, which this replaces, presumed that
2185  * @start was already page aligned.  We now align it regardless.
2186  *
2187  * Return: A page aligned address within [start, start + range).  On error,
2188  * @start is returned.
2189  */
2190 unsigned long
2191 randomize_page(unsigned long start, unsigned long range)
2192 {
2193 	if (!PAGE_ALIGNED(start)) {
2194 		range -= PAGE_ALIGN(start) - start;
2195 		start = PAGE_ALIGN(start);
2196 	}
2197 
2198 	if (start > ULONG_MAX - range)
2199 		range = ULONG_MAX - start;
2200 
2201 	range >>= PAGE_SHIFT;
2202 
2203 	if (range == 0)
2204 		return start;
2205 
2206 	return start + (get_random_long() % range << PAGE_SHIFT);
2207 }
2208 
2209 /* Interface for in-kernel drivers of true hardware RNGs.
2210  * Those devices may produce endless random bits and will be throttled
2211  * when our pool is full.
2212  */
2213 void add_hwgenerator_randomness(const char *buffer, size_t count,
2214 				size_t entropy)
2215 {
2216 	struct entropy_store *poolp = &input_pool;
2217 
2218 	if (!crng_ready()) {
2219 		crng_fast_load(buffer, count);
2220 		return;
2221 	}
2222 
2223 	/* Suspend writing if we're above the trickle threshold.
2224 	 * We'll be woken up again once below random_write_wakeup_thresh,
2225 	 * or when the calling thread is about to terminate.
2226 	 */
2227 	wait_event_interruptible(random_write_wait, kthread_should_stop() ||
2228 			ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
2229 	mix_pool_bytes(poolp, buffer, count);
2230 	credit_entropy_bits(poolp, entropy);
2231 }
2232 EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
2233