xref: /openbmc/linux/drivers/char/random.c (revision a48c7709)
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/workqueue.h>
263 #include <linux/irq.h>
264 #include <linux/syscalls.h>
265 #include <linux/completion.h>
266 #include <linux/uuid.h>
267 #include <crypto/chacha20.h>
268 
269 #include <asm/processor.h>
270 #include <linux/uaccess.h>
271 #include <asm/irq.h>
272 #include <asm/irq_regs.h>
273 #include <asm/io.h>
274 
275 #define CREATE_TRACE_POINTS
276 #include <trace/events/random.h>
277 
278 /* #define ADD_INTERRUPT_BENCH */
279 
280 /*
281  * Configuration information
282  */
283 #define INPUT_POOL_SHIFT	12
284 #define INPUT_POOL_WORDS	(1 << (INPUT_POOL_SHIFT-5))
285 #define OUTPUT_POOL_SHIFT	10
286 #define OUTPUT_POOL_WORDS	(1 << (OUTPUT_POOL_SHIFT-5))
287 #define SEC_XFER_SIZE		512
288 #define EXTRACT_SIZE		10
289 
290 
291 #define LONGS(x) (((x) + sizeof(unsigned long) - 1)/sizeof(unsigned long))
292 
293 /*
294  * To allow fractional bits to be tracked, the entropy_count field is
295  * denominated in units of 1/8th bits.
296  *
297  * 2*(ENTROPY_SHIFT + log2(poolbits)) must <= 31, or the multiply in
298  * credit_entropy_bits() needs to be 64 bits wide.
299  */
300 #define ENTROPY_SHIFT 3
301 #define ENTROPY_BITS(r) ((r)->entropy_count >> ENTROPY_SHIFT)
302 
303 /*
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.
306  */
307 static int random_read_wakeup_bits = 64;
308 
309 /*
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.
313  */
314 static int random_write_wakeup_bits = 28 * OUTPUT_POOL_WORDS;
315 
316 /*
317  * Originally, we used a primitive polynomial of degree .poolwords
318  * over GF(2).  The taps for various sizes are defined below.  They
319  * were chosen to be evenly spaced except for the last tap, which is 1
320  * to get the twisting happening as fast as possible.
321  *
322  * For the purposes of better mixing, we use the CRC-32 polynomial as
323  * well to make a (modified) twisted Generalized Feedback Shift
324  * Register.  (See M. Matsumoto & Y. Kurita, 1992.  Twisted GFSR
325  * generators.  ACM Transactions on Modeling and Computer Simulation
326  * 2(3):179-194.  Also see M. Matsumoto & Y. Kurita, 1994.  Twisted
327  * GFSR generators II.  ACM Transactions on Modeling and Computer
328  * Simulation 4:254-266)
329  *
330  * Thanks to Colin Plumb for suggesting this.
331  *
332  * The mixing operation is much less sensitive than the output hash,
333  * where we use SHA-1.  All that we want of mixing operation is that
334  * it be a good non-cryptographic hash; i.e. it not produce collisions
335  * when fed "random" data of the sort we expect to see.  As long as
336  * the pool state differs for different inputs, we have preserved the
337  * input entropy and done a good job.  The fact that an intelligent
338  * attacker can construct inputs that will produce controlled
339  * alterations to the pool's state is not important because we don't
340  * consider such inputs to contribute any randomness.  The only
341  * property we need with respect to them is that the attacker can't
342  * increase his/her knowledge of the pool's state.  Since all
343  * additions are reversible (knowing the final state and the input,
344  * you can reconstruct the initial state), if an attacker has any
345  * uncertainty about the initial state, he/she can only shuffle that
346  * uncertainty about, but never cause any collisions (which would
347  * decrease the uncertainty).
348  *
349  * Our mixing functions were analyzed by Lacharme, Roeck, Strubel, and
350  * Videau in their paper, "The Linux Pseudorandom Number Generator
351  * Revisited" (see: http://eprint.iacr.org/2012/251.pdf).  In their
352  * paper, they point out that we are not using a true Twisted GFSR,
353  * since Matsumoto & Kurita used a trinomial feedback polynomial (that
354  * is, with only three taps, instead of the six that we are using).
355  * As a result, the resulting polynomial is neither primitive nor
356  * irreducible, and hence does not have a maximal period over
357  * GF(2**32).  They suggest a slight change to the generator
358  * polynomial which improves the resulting TGFSR polynomial to be
359  * irreducible, which we have made here.
360  */
361 static struct poolinfo {
362 	int poolbitshift, poolwords, poolbytes, poolbits, poolfracbits;
363 #define S(x) ilog2(x)+5, (x), (x)*4, (x)*32, (x) << (ENTROPY_SHIFT+5)
364 	int tap1, tap2, tap3, tap4, tap5;
365 } poolinfo_table[] = {
366 	/* was: x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 */
367 	/* x^128 + x^104 + x^76 + x^51 +x^25 + x + 1 */
368 	{ S(128),	104,	76,	51,	25,	1 },
369 	/* was: x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 */
370 	/* x^32 + x^26 + x^19 + x^14 + x^7 + x + 1 */
371 	{ S(32),	26,	19,	14,	7,	1 },
372 #if 0
373 	/* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1  -- 115 */
374 	{ S(2048),	1638,	1231,	819,	411,	1 },
375 
376 	/* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
377 	{ S(1024),	817,	615,	412,	204,	1 },
378 
379 	/* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
380 	{ S(1024),	819,	616,	410,	207,	2 },
381 
382 	/* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
383 	{ S(512),	411,	308,	208,	104,	1 },
384 
385 	/* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
386 	{ S(512),	409,	307,	206,	102,	2 },
387 	/* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
388 	{ S(512),	409,	309,	205,	103,	2 },
389 
390 	/* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
391 	{ S(256),	205,	155,	101,	52,	1 },
392 
393 	/* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
394 	{ S(128),	103,	78,	51,	27,	2 },
395 
396 	/* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
397 	{ S(64),	52,	39,	26,	14,	1 },
398 #endif
399 };
400 
401 /*
402  * Static global variables
403  */
404 static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
405 static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
406 static struct fasync_struct *fasync;
407 
408 static DEFINE_SPINLOCK(random_ready_list_lock);
409 static LIST_HEAD(random_ready_list);
410 
411 struct crng_state {
412 	__u32		state[16];
413 	unsigned long	init_time;
414 	spinlock_t	lock;
415 };
416 
417 struct crng_state primary_crng = {
418 	.lock = __SPIN_LOCK_UNLOCKED(primary_crng.lock),
419 };
420 
421 /*
422  * crng_init =  0 --> Uninitialized
423  *		1 --> Initialized
424  *		2 --> Initialized from input_pool
425  *
426  * crng_init is protected by primary_crng->lock, and only increases
427  * its value (from 0->1->2).
428  */
429 static int crng_init = 0;
430 #define crng_ready() (likely(crng_init > 0))
431 static int crng_init_cnt = 0;
432 #define CRNG_INIT_CNT_THRESH (2*CHACHA20_KEY_SIZE)
433 static void _extract_crng(struct crng_state *crng,
434 			  __u32 out[CHACHA20_BLOCK_WORDS]);
435 static void _crng_backtrack_protect(struct crng_state *crng,
436 				    __u32 tmp[CHACHA20_BLOCK_WORDS], int used);
437 static void process_random_ready_list(void);
438 static void _get_random_bytes(void *buf, int nbytes);
439 
440 /**********************************************************************
441  *
442  * OS independent entropy store.   Here are the functions which handle
443  * storing entropy in an entropy pool.
444  *
445  **********************************************************************/
446 
447 struct entropy_store;
448 struct entropy_store {
449 	/* read-only data: */
450 	const struct poolinfo *poolinfo;
451 	__u32 *pool;
452 	const char *name;
453 	struct entropy_store *pull;
454 	struct work_struct push_work;
455 
456 	/* read-write data: */
457 	unsigned long last_pulled;
458 	spinlock_t lock;
459 	unsigned short add_ptr;
460 	unsigned short input_rotate;
461 	int entropy_count;
462 	int entropy_total;
463 	unsigned int initialized:1;
464 	unsigned int last_data_init:1;
465 	__u8 last_data[EXTRACT_SIZE];
466 };
467 
468 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
469 			       size_t nbytes, int min, int rsvd);
470 static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
471 				size_t nbytes, int fips);
472 
473 static void crng_reseed(struct crng_state *crng, struct entropy_store *r);
474 static void push_to_pool(struct work_struct *work);
475 static __u32 input_pool_data[INPUT_POOL_WORDS] __latent_entropy;
476 static __u32 blocking_pool_data[OUTPUT_POOL_WORDS] __latent_entropy;
477 
478 static struct entropy_store input_pool = {
479 	.poolinfo = &poolinfo_table[0],
480 	.name = "input",
481 	.lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
482 	.pool = input_pool_data
483 };
484 
485 static struct entropy_store blocking_pool = {
486 	.poolinfo = &poolinfo_table[1],
487 	.name = "blocking",
488 	.pull = &input_pool,
489 	.lock = __SPIN_LOCK_UNLOCKED(blocking_pool.lock),
490 	.pool = blocking_pool_data,
491 	.push_work = __WORK_INITIALIZER(blocking_pool.push_work,
492 					push_to_pool),
493 };
494 
495 static __u32 const twist_table[8] = {
496 	0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
497 	0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
498 
499 /*
500  * This function adds bytes into the entropy "pool".  It does not
501  * update the entropy estimate.  The caller should call
502  * credit_entropy_bits if this is appropriate.
503  *
504  * The pool is stirred with a primitive polynomial of the appropriate
505  * degree, and then twisted.  We twist by three bits at a time because
506  * it's cheap to do so and helps slightly in the expected case where
507  * the entropy is concentrated in the low-order bits.
508  */
509 static void _mix_pool_bytes(struct entropy_store *r, const void *in,
510 			    int nbytes)
511 {
512 	unsigned long i, tap1, tap2, tap3, tap4, tap5;
513 	int input_rotate;
514 	int wordmask = r->poolinfo->poolwords - 1;
515 	const char *bytes = in;
516 	__u32 w;
517 
518 	tap1 = r->poolinfo->tap1;
519 	tap2 = r->poolinfo->tap2;
520 	tap3 = r->poolinfo->tap3;
521 	tap4 = r->poolinfo->tap4;
522 	tap5 = r->poolinfo->tap5;
523 
524 	input_rotate = r->input_rotate;
525 	i = r->add_ptr;
526 
527 	/* mix one byte at a time to simplify size handling and churn faster */
528 	while (nbytes--) {
529 		w = rol32(*bytes++, input_rotate);
530 		i = (i - 1) & wordmask;
531 
532 		/* XOR in the various taps */
533 		w ^= r->pool[i];
534 		w ^= r->pool[(i + tap1) & wordmask];
535 		w ^= r->pool[(i + tap2) & wordmask];
536 		w ^= r->pool[(i + tap3) & wordmask];
537 		w ^= r->pool[(i + tap4) & wordmask];
538 		w ^= r->pool[(i + tap5) & wordmask];
539 
540 		/* Mix the result back in with a twist */
541 		r->pool[i] = (w >> 3) ^ twist_table[w & 7];
542 
543 		/*
544 		 * Normally, we add 7 bits of rotation to the pool.
545 		 * At the beginning of the pool, add an extra 7 bits
546 		 * rotation, so that successive passes spread the
547 		 * input bits across the pool evenly.
548 		 */
549 		input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
550 	}
551 
552 	r->input_rotate = input_rotate;
553 	r->add_ptr = i;
554 }
555 
556 static void __mix_pool_bytes(struct entropy_store *r, const void *in,
557 			     int nbytes)
558 {
559 	trace_mix_pool_bytes_nolock(r->name, nbytes, _RET_IP_);
560 	_mix_pool_bytes(r, in, nbytes);
561 }
562 
563 static void mix_pool_bytes(struct entropy_store *r, const void *in,
564 			   int nbytes)
565 {
566 	unsigned long flags;
567 
568 	trace_mix_pool_bytes(r->name, nbytes, _RET_IP_);
569 	spin_lock_irqsave(&r->lock, flags);
570 	_mix_pool_bytes(r, in, nbytes);
571 	spin_unlock_irqrestore(&r->lock, flags);
572 }
573 
574 struct fast_pool {
575 	__u32		pool[4];
576 	unsigned long	last;
577 	unsigned short	reg_idx;
578 	unsigned char	count;
579 };
580 
581 /*
582  * This is a fast mixing routine used by the interrupt randomness
583  * collector.  It's hardcoded for an 128 bit pool and assumes that any
584  * locks that might be needed are taken by the caller.
585  */
586 static void fast_mix(struct fast_pool *f)
587 {
588 	__u32 a = f->pool[0],	b = f->pool[1];
589 	__u32 c = f->pool[2],	d = f->pool[3];
590 
591 	a += b;			c += d;
592 	b = rol32(b, 6);	d = rol32(d, 27);
593 	d ^= a;			b ^= c;
594 
595 	a += b;			c += d;
596 	b = rol32(b, 16);	d = rol32(d, 14);
597 	d ^= a;			b ^= c;
598 
599 	a += b;			c += d;
600 	b = rol32(b, 6);	d = rol32(d, 27);
601 	d ^= a;			b ^= c;
602 
603 	a += b;			c += d;
604 	b = rol32(b, 16);	d = rol32(d, 14);
605 	d ^= a;			b ^= c;
606 
607 	f->pool[0] = a;  f->pool[1] = b;
608 	f->pool[2] = c;  f->pool[3] = d;
609 	f->count++;
610 }
611 
612 static void process_random_ready_list(void)
613 {
614 	unsigned long flags;
615 	struct random_ready_callback *rdy, *tmp;
616 
617 	spin_lock_irqsave(&random_ready_list_lock, flags);
618 	list_for_each_entry_safe(rdy, tmp, &random_ready_list, list) {
619 		struct module *owner = rdy->owner;
620 
621 		list_del_init(&rdy->list);
622 		rdy->func(rdy);
623 		module_put(owner);
624 	}
625 	spin_unlock_irqrestore(&random_ready_list_lock, flags);
626 }
627 
628 /*
629  * Credit (or debit) the entropy store with n bits of entropy.
630  * Use credit_entropy_bits_safe() if the value comes from userspace
631  * or otherwise should be checked for extreme values.
632  */
633 static void credit_entropy_bits(struct entropy_store *r, int nbits)
634 {
635 	int entropy_count, orig;
636 	const int pool_size = r->poolinfo->poolfracbits;
637 	int nfrac = nbits << ENTROPY_SHIFT;
638 
639 	if (!nbits)
640 		return;
641 
642 retry:
643 	entropy_count = orig = READ_ONCE(r->entropy_count);
644 	if (nfrac < 0) {
645 		/* Debit */
646 		entropy_count += nfrac;
647 	} else {
648 		/*
649 		 * Credit: we have to account for the possibility of
650 		 * overwriting already present entropy.	 Even in the
651 		 * ideal case of pure Shannon entropy, new contributions
652 		 * approach the full value asymptotically:
653 		 *
654 		 * entropy <- entropy + (pool_size - entropy) *
655 		 *	(1 - exp(-add_entropy/pool_size))
656 		 *
657 		 * For add_entropy <= pool_size/2 then
658 		 * (1 - exp(-add_entropy/pool_size)) >=
659 		 *    (add_entropy/pool_size)*0.7869...
660 		 * so we can approximate the exponential with
661 		 * 3/4*add_entropy/pool_size and still be on the
662 		 * safe side by adding at most pool_size/2 at a time.
663 		 *
664 		 * The use of pool_size-2 in the while statement is to
665 		 * prevent rounding artifacts from making the loop
666 		 * arbitrarily long; this limits the loop to log2(pool_size)*2
667 		 * turns no matter how large nbits is.
668 		 */
669 		int pnfrac = nfrac;
670 		const int s = r->poolinfo->poolbitshift + ENTROPY_SHIFT + 2;
671 		/* The +2 corresponds to the /4 in the denominator */
672 
673 		do {
674 			unsigned int anfrac = min(pnfrac, pool_size/2);
675 			unsigned int add =
676 				((pool_size - entropy_count)*anfrac*3) >> s;
677 
678 			entropy_count += add;
679 			pnfrac -= anfrac;
680 		} while (unlikely(entropy_count < pool_size-2 && pnfrac));
681 	}
682 
683 	if (unlikely(entropy_count < 0)) {
684 		pr_warn("random: negative entropy/overflow: pool %s count %d\n",
685 			r->name, entropy_count);
686 		WARN_ON(1);
687 		entropy_count = 0;
688 	} else if (entropy_count > pool_size)
689 		entropy_count = pool_size;
690 	if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
691 		goto retry;
692 
693 	r->entropy_total += nbits;
694 	if (!r->initialized && r->entropy_total > 128) {
695 		r->initialized = 1;
696 		r->entropy_total = 0;
697 	}
698 
699 	trace_credit_entropy_bits(r->name, nbits,
700 				  entropy_count >> ENTROPY_SHIFT,
701 				  r->entropy_total, _RET_IP_);
702 
703 	if (r == &input_pool) {
704 		int entropy_bits = entropy_count >> ENTROPY_SHIFT;
705 
706 		if (crng_init < 2 && entropy_bits >= 128) {
707 			crng_reseed(&primary_crng, r);
708 			entropy_bits = r->entropy_count >> ENTROPY_SHIFT;
709 		}
710 
711 		/* should we wake readers? */
712 		if (entropy_bits >= random_read_wakeup_bits &&
713 		    wq_has_sleeper(&random_read_wait)) {
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 = r->poolinfo->poolwords * 32;
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 		__u32	block[CHACHA20_BLOCK_WORDS];
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 			  __u32 out[CHACHA20_BLOCK_WORDS])
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(__u32 out[CHACHA20_BLOCK_WORDS])
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 				    __u32 tmp[CHACHA20_BLOCK_WORDS], 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 = &tmp[used / sizeof(__u32)];
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(__u32 tmp[CHACHA20_BLOCK_WORDS], 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 	__u32 tmp[CHACHA20_BLOCK_WORDS];
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 };
968 
969 #define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
970 
971 /*
972  * Add device- or boot-specific data to the input pool to help
973  * initialize it.
974  *
975  * None of this adds any entropy; it is meant to avoid the problem of
976  * the entropy pool having similar initial state across largely
977  * identical devices.
978  */
979 void add_device_randomness(const void *buf, unsigned int size)
980 {
981 	unsigned long time = random_get_entropy() ^ jiffies;
982 	unsigned long flags;
983 
984 	if (!crng_ready()) {
985 		crng_fast_load(buf, size);
986 		return;
987 	}
988 
989 	trace_add_device_randomness(size, _RET_IP_);
990 	spin_lock_irqsave(&input_pool.lock, flags);
991 	_mix_pool_bytes(&input_pool, buf, size);
992 	_mix_pool_bytes(&input_pool, &time, sizeof(time));
993 	spin_unlock_irqrestore(&input_pool.lock, flags);
994 }
995 EXPORT_SYMBOL(add_device_randomness);
996 
997 static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
998 
999 /*
1000  * This function adds entropy to the entropy "pool" by using timing
1001  * delays.  It uses the timer_rand_state structure to make an estimate
1002  * of how many bits of entropy this call has added to the pool.
1003  *
1004  * The number "num" is also added to the pool - it should somehow describe
1005  * the type of event which just happened.  This is currently 0-255 for
1006  * keyboard scan codes, and 256 upwards for interrupts.
1007  *
1008  */
1009 static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
1010 {
1011 	struct entropy_store	*r;
1012 	struct {
1013 		long jiffies;
1014 		unsigned cycles;
1015 		unsigned num;
1016 	} sample;
1017 	long delta, delta2, delta3;
1018 
1019 	preempt_disable();
1020 
1021 	sample.jiffies = jiffies;
1022 	sample.cycles = random_get_entropy();
1023 	sample.num = num;
1024 	r = &input_pool;
1025 	mix_pool_bytes(r, &sample, sizeof(sample));
1026 
1027 	/*
1028 	 * Calculate number of bits of randomness we probably added.
1029 	 * We take into account the first, second and third-order deltas
1030 	 * in order to make our estimate.
1031 	 */
1032 	delta = sample.jiffies - state->last_time;
1033 	state->last_time = sample.jiffies;
1034 
1035 	delta2 = delta - state->last_delta;
1036 	state->last_delta = delta;
1037 
1038 	delta3 = delta2 - state->last_delta2;
1039 	state->last_delta2 = delta2;
1040 
1041 	if (delta < 0)
1042 		delta = -delta;
1043 	if (delta2 < 0)
1044 		delta2 = -delta2;
1045 	if (delta3 < 0)
1046 		delta3 = -delta3;
1047 	if (delta > delta2)
1048 		delta = delta2;
1049 	if (delta > delta3)
1050 		delta = delta3;
1051 
1052 	/*
1053 	 * delta is now minimum absolute delta.
1054 	 * Round down by 1 bit on general principles,
1055 	 * and limit entropy entimate to 12 bits.
1056 	 */
1057 	credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
1058 
1059 	preempt_enable();
1060 }
1061 
1062 void add_input_randomness(unsigned int type, unsigned int code,
1063 				 unsigned int value)
1064 {
1065 	static unsigned char last_value;
1066 
1067 	/* ignore autorepeat and the like */
1068 	if (value == last_value)
1069 		return;
1070 
1071 	last_value = value;
1072 	add_timer_randomness(&input_timer_state,
1073 			     (type << 4) ^ code ^ (code >> 4) ^ value);
1074 	trace_add_input_randomness(ENTROPY_BITS(&input_pool));
1075 }
1076 EXPORT_SYMBOL_GPL(add_input_randomness);
1077 
1078 static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
1079 
1080 #ifdef ADD_INTERRUPT_BENCH
1081 static unsigned long avg_cycles, avg_deviation;
1082 
1083 #define AVG_SHIFT 8     /* Exponential average factor k=1/256 */
1084 #define FIXED_1_2 (1 << (AVG_SHIFT-1))
1085 
1086 static void add_interrupt_bench(cycles_t start)
1087 {
1088         long delta = random_get_entropy() - start;
1089 
1090         /* Use a weighted moving average */
1091         delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
1092         avg_cycles += delta;
1093         /* And average deviation */
1094         delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
1095         avg_deviation += delta;
1096 }
1097 #else
1098 #define add_interrupt_bench(x)
1099 #endif
1100 
1101 static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
1102 {
1103 	__u32 *ptr = (__u32 *) regs;
1104 	unsigned int idx;
1105 
1106 	if (regs == NULL)
1107 		return 0;
1108 	idx = READ_ONCE(f->reg_idx);
1109 	if (idx >= sizeof(struct pt_regs) / sizeof(__u32))
1110 		idx = 0;
1111 	ptr += idx++;
1112 	WRITE_ONCE(f->reg_idx, idx);
1113 	return *ptr;
1114 }
1115 
1116 void add_interrupt_randomness(int irq, int irq_flags)
1117 {
1118 	struct entropy_store	*r;
1119 	struct fast_pool	*fast_pool = this_cpu_ptr(&irq_randomness);
1120 	struct pt_regs		*regs = get_irq_regs();
1121 	unsigned long		now = jiffies;
1122 	cycles_t		cycles = random_get_entropy();
1123 	__u32			c_high, j_high;
1124 	__u64			ip;
1125 	unsigned long		seed;
1126 	int			credit = 0;
1127 
1128 	if (cycles == 0)
1129 		cycles = get_reg(fast_pool, regs);
1130 	c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
1131 	j_high = (sizeof(now) > 4) ? now >> 32 : 0;
1132 	fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
1133 	fast_pool->pool[1] ^= now ^ c_high;
1134 	ip = regs ? instruction_pointer(regs) : _RET_IP_;
1135 	fast_pool->pool[2] ^= ip;
1136 	fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
1137 		get_reg(fast_pool, regs);
1138 
1139 	fast_mix(fast_pool);
1140 	add_interrupt_bench(cycles);
1141 
1142 	if (!crng_ready()) {
1143 		if ((fast_pool->count >= 64) &&
1144 		    crng_fast_load((char *) fast_pool->pool,
1145 				   sizeof(fast_pool->pool))) {
1146 			fast_pool->count = 0;
1147 			fast_pool->last = now;
1148 		}
1149 		return;
1150 	}
1151 
1152 	if ((fast_pool->count < 64) &&
1153 	    !time_after(now, fast_pool->last + HZ))
1154 		return;
1155 
1156 	r = &input_pool;
1157 	if (!spin_trylock(&r->lock))
1158 		return;
1159 
1160 	fast_pool->last = now;
1161 	__mix_pool_bytes(r, &fast_pool->pool, sizeof(fast_pool->pool));
1162 
1163 	/*
1164 	 * If we have architectural seed generator, produce a seed and
1165 	 * add it to the pool.  For the sake of paranoia don't let the
1166 	 * architectural seed generator dominate the input from the
1167 	 * interrupt noise.
1168 	 */
1169 	if (arch_get_random_seed_long(&seed)) {
1170 		__mix_pool_bytes(r, &seed, sizeof(seed));
1171 		credit = 1;
1172 	}
1173 	spin_unlock(&r->lock);
1174 
1175 	fast_pool->count = 0;
1176 
1177 	/* award one bit for the contents of the fast pool */
1178 	credit_entropy_bits(r, credit + 1);
1179 }
1180 EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1181 
1182 #ifdef CONFIG_BLOCK
1183 void add_disk_randomness(struct gendisk *disk)
1184 {
1185 	if (!disk || !disk->random)
1186 		return;
1187 	/* first major is 1, so we get >= 0x200 here */
1188 	add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1189 	trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
1190 }
1191 EXPORT_SYMBOL_GPL(add_disk_randomness);
1192 #endif
1193 
1194 /*********************************************************************
1195  *
1196  * Entropy extraction routines
1197  *
1198  *********************************************************************/
1199 
1200 /*
1201  * This utility inline function is responsible for transferring entropy
1202  * from the primary pool to the secondary extraction pool. We make
1203  * sure we pull enough for a 'catastrophic reseed'.
1204  */
1205 static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes);
1206 static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1207 {
1208 	if (!r->pull ||
1209 	    r->entropy_count >= (nbytes << (ENTROPY_SHIFT + 3)) ||
1210 	    r->entropy_count > r->poolinfo->poolfracbits)
1211 		return;
1212 
1213 	_xfer_secondary_pool(r, nbytes);
1214 }
1215 
1216 static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1217 {
1218 	__u32	tmp[OUTPUT_POOL_WORDS];
1219 
1220 	int bytes = nbytes;
1221 
1222 	/* pull at least as much as a wakeup */
1223 	bytes = max_t(int, bytes, random_read_wakeup_bits / 8);
1224 	/* but never more than the buffer size */
1225 	bytes = min_t(int, bytes, sizeof(tmp));
1226 
1227 	trace_xfer_secondary_pool(r->name, bytes * 8, nbytes * 8,
1228 				  ENTROPY_BITS(r), ENTROPY_BITS(r->pull));
1229 	bytes = extract_entropy(r->pull, tmp, bytes,
1230 				random_read_wakeup_bits / 8, 0);
1231 	mix_pool_bytes(r, tmp, bytes);
1232 	credit_entropy_bits(r, bytes*8);
1233 }
1234 
1235 /*
1236  * Used as a workqueue function so that when the input pool is getting
1237  * full, we can "spill over" some entropy to the output pools.  That
1238  * way the output pools can store some of the excess entropy instead
1239  * of letting it go to waste.
1240  */
1241 static void push_to_pool(struct work_struct *work)
1242 {
1243 	struct entropy_store *r = container_of(work, struct entropy_store,
1244 					      push_work);
1245 	BUG_ON(!r);
1246 	_xfer_secondary_pool(r, random_read_wakeup_bits/8);
1247 	trace_push_to_pool(r->name, r->entropy_count >> ENTROPY_SHIFT,
1248 			   r->pull->entropy_count >> ENTROPY_SHIFT);
1249 }
1250 
1251 /*
1252  * This function decides how many bytes to actually take from the
1253  * given pool, and also debits the entropy count accordingly.
1254  */
1255 static size_t account(struct entropy_store *r, size_t nbytes, int min,
1256 		      int reserved)
1257 {
1258 	int entropy_count, orig, have_bytes;
1259 	size_t ibytes, nfrac;
1260 
1261 	BUG_ON(r->entropy_count > r->poolinfo->poolfracbits);
1262 
1263 	/* Can we pull enough? */
1264 retry:
1265 	entropy_count = orig = READ_ONCE(r->entropy_count);
1266 	ibytes = nbytes;
1267 	/* never pull more than available */
1268 	have_bytes = entropy_count >> (ENTROPY_SHIFT + 3);
1269 
1270 	if ((have_bytes -= reserved) < 0)
1271 		have_bytes = 0;
1272 	ibytes = min_t(size_t, ibytes, have_bytes);
1273 	if (ibytes < min)
1274 		ibytes = 0;
1275 
1276 	if (unlikely(entropy_count < 0)) {
1277 		pr_warn("random: negative entropy count: pool %s count %d\n",
1278 			r->name, entropy_count);
1279 		WARN_ON(1);
1280 		entropy_count = 0;
1281 	}
1282 	nfrac = ibytes << (ENTROPY_SHIFT + 3);
1283 	if ((size_t) entropy_count > nfrac)
1284 		entropy_count -= nfrac;
1285 	else
1286 		entropy_count = 0;
1287 
1288 	if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
1289 		goto retry;
1290 
1291 	trace_debit_entropy(r->name, 8 * ibytes);
1292 	if (ibytes &&
1293 	    (r->entropy_count >> ENTROPY_SHIFT) < random_write_wakeup_bits) {
1294 		wake_up_interruptible(&random_write_wait);
1295 		kill_fasync(&fasync, SIGIO, POLL_OUT);
1296 	}
1297 
1298 	return ibytes;
1299 }
1300 
1301 /*
1302  * This function does the actual extraction for extract_entropy and
1303  * extract_entropy_user.
1304  *
1305  * Note: we assume that .poolwords is a multiple of 16 words.
1306  */
1307 static void extract_buf(struct entropy_store *r, __u8 *out)
1308 {
1309 	int i;
1310 	union {
1311 		__u32 w[5];
1312 		unsigned long l[LONGS(20)];
1313 	} hash;
1314 	__u32 workspace[SHA_WORKSPACE_WORDS];
1315 	unsigned long flags;
1316 
1317 	/*
1318 	 * If we have an architectural hardware random number
1319 	 * generator, use it for SHA's initial vector
1320 	 */
1321 	sha_init(hash.w);
1322 	for (i = 0; i < LONGS(20); i++) {
1323 		unsigned long v;
1324 		if (!arch_get_random_long(&v))
1325 			break;
1326 		hash.l[i] = v;
1327 	}
1328 
1329 	/* Generate a hash across the pool, 16 words (512 bits) at a time */
1330 	spin_lock_irqsave(&r->lock, flags);
1331 	for (i = 0; i < r->poolinfo->poolwords; i += 16)
1332 		sha_transform(hash.w, (__u8 *)(r->pool + i), workspace);
1333 
1334 	/*
1335 	 * We mix the hash back into the pool to prevent backtracking
1336 	 * attacks (where the attacker knows the state of the pool
1337 	 * plus the current outputs, and attempts to find previous
1338 	 * ouputs), unless the hash function can be inverted. By
1339 	 * mixing at least a SHA1 worth of hash data back, we make
1340 	 * brute-forcing the feedback as hard as brute-forcing the
1341 	 * hash.
1342 	 */
1343 	__mix_pool_bytes(r, hash.w, sizeof(hash.w));
1344 	spin_unlock_irqrestore(&r->lock, flags);
1345 
1346 	memzero_explicit(workspace, sizeof(workspace));
1347 
1348 	/*
1349 	 * In case the hash function has some recognizable output
1350 	 * pattern, we fold it in half. Thus, we always feed back
1351 	 * twice as much data as we output.
1352 	 */
1353 	hash.w[0] ^= hash.w[3];
1354 	hash.w[1] ^= hash.w[4];
1355 	hash.w[2] ^= rol32(hash.w[2], 16);
1356 
1357 	memcpy(out, &hash, EXTRACT_SIZE);
1358 	memzero_explicit(&hash, sizeof(hash));
1359 }
1360 
1361 static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
1362 				size_t nbytes, int fips)
1363 {
1364 	ssize_t ret = 0, i;
1365 	__u8 tmp[EXTRACT_SIZE];
1366 	unsigned long flags;
1367 
1368 	while (nbytes) {
1369 		extract_buf(r, tmp);
1370 
1371 		if (fips) {
1372 			spin_lock_irqsave(&r->lock, flags);
1373 			if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
1374 				panic("Hardware RNG duplicated output!\n");
1375 			memcpy(r->last_data, tmp, EXTRACT_SIZE);
1376 			spin_unlock_irqrestore(&r->lock, flags);
1377 		}
1378 		i = min_t(int, nbytes, EXTRACT_SIZE);
1379 		memcpy(buf, tmp, i);
1380 		nbytes -= i;
1381 		buf += i;
1382 		ret += i;
1383 	}
1384 
1385 	/* Wipe data just returned from memory */
1386 	memzero_explicit(tmp, sizeof(tmp));
1387 
1388 	return ret;
1389 }
1390 
1391 /*
1392  * This function extracts randomness from the "entropy pool", and
1393  * returns it in a buffer.
1394  *
1395  * The min parameter specifies the minimum amount we can pull before
1396  * failing to avoid races that defeat catastrophic reseeding while the
1397  * reserved parameter indicates how much entropy we must leave in the
1398  * pool after each pull to avoid starving other readers.
1399  */
1400 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1401 				 size_t nbytes, int min, int reserved)
1402 {
1403 	__u8 tmp[EXTRACT_SIZE];
1404 	unsigned long flags;
1405 
1406 	/* if last_data isn't primed, we need EXTRACT_SIZE extra bytes */
1407 	if (fips_enabled) {
1408 		spin_lock_irqsave(&r->lock, flags);
1409 		if (!r->last_data_init) {
1410 			r->last_data_init = 1;
1411 			spin_unlock_irqrestore(&r->lock, flags);
1412 			trace_extract_entropy(r->name, EXTRACT_SIZE,
1413 					      ENTROPY_BITS(r), _RET_IP_);
1414 			xfer_secondary_pool(r, EXTRACT_SIZE);
1415 			extract_buf(r, tmp);
1416 			spin_lock_irqsave(&r->lock, flags);
1417 			memcpy(r->last_data, tmp, EXTRACT_SIZE);
1418 		}
1419 		spin_unlock_irqrestore(&r->lock, flags);
1420 	}
1421 
1422 	trace_extract_entropy(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1423 	xfer_secondary_pool(r, nbytes);
1424 	nbytes = account(r, nbytes, min, reserved);
1425 
1426 	return _extract_entropy(r, buf, nbytes, fips_enabled);
1427 }
1428 
1429 /*
1430  * This function extracts randomness from the "entropy pool", and
1431  * returns it in a userspace buffer.
1432  */
1433 static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
1434 				    size_t nbytes)
1435 {
1436 	ssize_t ret = 0, i;
1437 	__u8 tmp[EXTRACT_SIZE];
1438 	int large_request = (nbytes > 256);
1439 
1440 	trace_extract_entropy_user(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1441 	xfer_secondary_pool(r, nbytes);
1442 	nbytes = account(r, nbytes, 0, 0);
1443 
1444 	while (nbytes) {
1445 		if (large_request && need_resched()) {
1446 			if (signal_pending(current)) {
1447 				if (ret == 0)
1448 					ret = -ERESTARTSYS;
1449 				break;
1450 			}
1451 			schedule();
1452 		}
1453 
1454 		extract_buf(r, tmp);
1455 		i = min_t(int, nbytes, EXTRACT_SIZE);
1456 		if (copy_to_user(buf, tmp, i)) {
1457 			ret = -EFAULT;
1458 			break;
1459 		}
1460 
1461 		nbytes -= i;
1462 		buf += i;
1463 		ret += i;
1464 	}
1465 
1466 	/* Wipe data just returned from memory */
1467 	memzero_explicit(tmp, sizeof(tmp));
1468 
1469 	return ret;
1470 }
1471 
1472 #define warn_unseeded_randomness(previous) \
1473 	_warn_unseeded_randomness(__func__, (void *) _RET_IP_, (previous))
1474 
1475 static void _warn_unseeded_randomness(const char *func_name, void *caller,
1476 				      void **previous)
1477 {
1478 #ifdef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1479 	const bool print_once = false;
1480 #else
1481 	static bool print_once __read_mostly;
1482 #endif
1483 
1484 	if (print_once ||
1485 	    crng_ready() ||
1486 	    (previous && (caller == READ_ONCE(*previous))))
1487 		return;
1488 	WRITE_ONCE(*previous, caller);
1489 #ifndef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1490 	print_once = true;
1491 #endif
1492 	pr_notice("random: %s called from %pS with crng_init=%d\n",
1493 		  func_name, caller, crng_init);
1494 }
1495 
1496 /*
1497  * This function is the exported kernel interface.  It returns some
1498  * number of good random numbers, suitable for key generation, seeding
1499  * TCP sequence numbers, etc.  It does not rely on the hardware random
1500  * number generator.  For random bytes direct from the hardware RNG
1501  * (when available), use get_random_bytes_arch(). In order to ensure
1502  * that the randomness provided by this function is okay, the function
1503  * wait_for_random_bytes() should be called and return 0 at least once
1504  * at any point prior.
1505  */
1506 static void _get_random_bytes(void *buf, int nbytes)
1507 {
1508 	__u32 tmp[CHACHA20_BLOCK_WORDS];
1509 
1510 	trace_get_random_bytes(nbytes, _RET_IP_);
1511 
1512 	while (nbytes >= CHACHA20_BLOCK_SIZE) {
1513 		extract_crng(buf);
1514 		buf += CHACHA20_BLOCK_SIZE;
1515 		nbytes -= CHACHA20_BLOCK_SIZE;
1516 	}
1517 
1518 	if (nbytes > 0) {
1519 		extract_crng(tmp);
1520 		memcpy(buf, tmp, nbytes);
1521 		crng_backtrack_protect(tmp, nbytes);
1522 	} else
1523 		crng_backtrack_protect(tmp, CHACHA20_BLOCK_SIZE);
1524 	memzero_explicit(tmp, sizeof(tmp));
1525 }
1526 
1527 void get_random_bytes(void *buf, int nbytes)
1528 {
1529 	static void *previous;
1530 
1531 	warn_unseeded_randomness(&previous);
1532 	_get_random_bytes(buf, nbytes);
1533 }
1534 EXPORT_SYMBOL(get_random_bytes);
1535 
1536 /*
1537  * Wait for the urandom pool to be seeded and thus guaranteed to supply
1538  * cryptographically secure random numbers. This applies to: the /dev/urandom
1539  * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
1540  * family of functions. Using any of these functions without first calling
1541  * this function forfeits the guarantee of security.
1542  *
1543  * Returns: 0 if the urandom pool has been seeded.
1544  *          -ERESTARTSYS if the function was interrupted by a signal.
1545  */
1546 int wait_for_random_bytes(void)
1547 {
1548 	if (likely(crng_ready()))
1549 		return 0;
1550 	return wait_event_interruptible(crng_init_wait, crng_ready());
1551 }
1552 EXPORT_SYMBOL(wait_for_random_bytes);
1553 
1554 /*
1555  * Add a callback function that will be invoked when the nonblocking
1556  * pool is initialised.
1557  *
1558  * returns: 0 if callback is successfully added
1559  *	    -EALREADY if pool is already initialised (callback not called)
1560  *	    -ENOENT if module for callback is not alive
1561  */
1562 int add_random_ready_callback(struct random_ready_callback *rdy)
1563 {
1564 	struct module *owner;
1565 	unsigned long flags;
1566 	int err = -EALREADY;
1567 
1568 	if (crng_ready())
1569 		return err;
1570 
1571 	owner = rdy->owner;
1572 	if (!try_module_get(owner))
1573 		return -ENOENT;
1574 
1575 	spin_lock_irqsave(&random_ready_list_lock, flags);
1576 	if (crng_ready())
1577 		goto out;
1578 
1579 	owner = NULL;
1580 
1581 	list_add(&rdy->list, &random_ready_list);
1582 	err = 0;
1583 
1584 out:
1585 	spin_unlock_irqrestore(&random_ready_list_lock, flags);
1586 
1587 	module_put(owner);
1588 
1589 	return err;
1590 }
1591 EXPORT_SYMBOL(add_random_ready_callback);
1592 
1593 /*
1594  * Delete a previously registered readiness callback function.
1595  */
1596 void del_random_ready_callback(struct random_ready_callback *rdy)
1597 {
1598 	unsigned long flags;
1599 	struct module *owner = NULL;
1600 
1601 	spin_lock_irqsave(&random_ready_list_lock, flags);
1602 	if (!list_empty(&rdy->list)) {
1603 		list_del_init(&rdy->list);
1604 		owner = rdy->owner;
1605 	}
1606 	spin_unlock_irqrestore(&random_ready_list_lock, flags);
1607 
1608 	module_put(owner);
1609 }
1610 EXPORT_SYMBOL(del_random_ready_callback);
1611 
1612 /*
1613  * This function will use the architecture-specific hardware random
1614  * number generator if it is available.  The arch-specific hw RNG will
1615  * almost certainly be faster than what we can do in software, but it
1616  * is impossible to verify that it is implemented securely (as
1617  * opposed, to, say, the AES encryption of a sequence number using a
1618  * key known by the NSA).  So it's useful if we need the speed, but
1619  * only if we're willing to trust the hardware manufacturer not to
1620  * have put in a back door.
1621  */
1622 void get_random_bytes_arch(void *buf, int nbytes)
1623 {
1624 	char *p = buf;
1625 
1626 	trace_get_random_bytes_arch(nbytes, _RET_IP_);
1627 	while (nbytes) {
1628 		unsigned long v;
1629 		int chunk = min(nbytes, (int)sizeof(unsigned long));
1630 
1631 		if (!arch_get_random_long(&v))
1632 			break;
1633 
1634 		memcpy(p, &v, chunk);
1635 		p += chunk;
1636 		nbytes -= chunk;
1637 	}
1638 
1639 	if (nbytes)
1640 		get_random_bytes(p, nbytes);
1641 }
1642 EXPORT_SYMBOL(get_random_bytes_arch);
1643 
1644 
1645 /*
1646  * init_std_data - initialize pool with system data
1647  *
1648  * @r: pool to initialize
1649  *
1650  * This function clears the pool's entropy count and mixes some system
1651  * data into the pool to prepare it for use. The pool is not cleared
1652  * as that can only decrease the entropy in the pool.
1653  */
1654 static void init_std_data(struct entropy_store *r)
1655 {
1656 	int i;
1657 	ktime_t now = ktime_get_real();
1658 	unsigned long rv;
1659 
1660 	r->last_pulled = jiffies;
1661 	mix_pool_bytes(r, &now, sizeof(now));
1662 	for (i = r->poolinfo->poolbytes; i > 0; i -= sizeof(rv)) {
1663 		if (!arch_get_random_seed_long(&rv) &&
1664 		    !arch_get_random_long(&rv))
1665 			rv = random_get_entropy();
1666 		mix_pool_bytes(r, &rv, sizeof(rv));
1667 	}
1668 	mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1669 }
1670 
1671 /*
1672  * Note that setup_arch() may call add_device_randomness()
1673  * long before we get here. This allows seeding of the pools
1674  * with some platform dependent data very early in the boot
1675  * process. But it limits our options here. We must use
1676  * statically allocated structures that already have all
1677  * initializations complete at compile time. We should also
1678  * take care not to overwrite the precious per platform data
1679  * we were given.
1680  */
1681 static int rand_initialize(void)
1682 {
1683 #ifdef CONFIG_NUMA
1684 	int i;
1685 	struct crng_state *crng;
1686 	struct crng_state **pool;
1687 #endif
1688 
1689 	init_std_data(&input_pool);
1690 	init_std_data(&blocking_pool);
1691 	crng_initialize(&primary_crng);
1692 
1693 #ifdef CONFIG_NUMA
1694 	pool = kcalloc(nr_node_ids, sizeof(*pool), GFP_KERNEL|__GFP_NOFAIL);
1695 	for_each_online_node(i) {
1696 		crng = kmalloc_node(sizeof(struct crng_state),
1697 				    GFP_KERNEL | __GFP_NOFAIL, i);
1698 		spin_lock_init(&crng->lock);
1699 		crng_initialize(crng);
1700 		pool[i] = crng;
1701 	}
1702 	mb();
1703 	crng_node_pool = pool;
1704 #endif
1705 	return 0;
1706 }
1707 early_initcall(rand_initialize);
1708 
1709 #ifdef CONFIG_BLOCK
1710 void rand_initialize_disk(struct gendisk *disk)
1711 {
1712 	struct timer_rand_state *state;
1713 
1714 	/*
1715 	 * If kzalloc returns null, we just won't use that entropy
1716 	 * source.
1717 	 */
1718 	state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1719 	if (state) {
1720 		state->last_time = INITIAL_JIFFIES;
1721 		disk->random = state;
1722 	}
1723 }
1724 #endif
1725 
1726 static ssize_t
1727 _random_read(int nonblock, char __user *buf, size_t nbytes)
1728 {
1729 	ssize_t n;
1730 
1731 	if (nbytes == 0)
1732 		return 0;
1733 
1734 	nbytes = min_t(size_t, nbytes, SEC_XFER_SIZE);
1735 	while (1) {
1736 		n = extract_entropy_user(&blocking_pool, buf, nbytes);
1737 		if (n < 0)
1738 			return n;
1739 		trace_random_read(n*8, (nbytes-n)*8,
1740 				  ENTROPY_BITS(&blocking_pool),
1741 				  ENTROPY_BITS(&input_pool));
1742 		if (n > 0)
1743 			return n;
1744 
1745 		/* Pool is (near) empty.  Maybe wait and retry. */
1746 		if (nonblock)
1747 			return -EAGAIN;
1748 
1749 		wait_event_interruptible(random_read_wait,
1750 			ENTROPY_BITS(&input_pool) >=
1751 			random_read_wakeup_bits);
1752 		if (signal_pending(current))
1753 			return -ERESTARTSYS;
1754 	}
1755 }
1756 
1757 static ssize_t
1758 random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1759 {
1760 	return _random_read(file->f_flags & O_NONBLOCK, buf, nbytes);
1761 }
1762 
1763 static ssize_t
1764 urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1765 {
1766 	unsigned long flags;
1767 	static int maxwarn = 10;
1768 	int ret;
1769 
1770 	if (!crng_ready() && maxwarn > 0) {
1771 		maxwarn--;
1772 		printk(KERN_NOTICE "random: %s: uninitialized urandom read "
1773 		       "(%zd bytes read)\n",
1774 		       current->comm, nbytes);
1775 		spin_lock_irqsave(&primary_crng.lock, flags);
1776 		crng_init_cnt = 0;
1777 		spin_unlock_irqrestore(&primary_crng.lock, flags);
1778 	}
1779 	nbytes = min_t(size_t, nbytes, INT_MAX >> (ENTROPY_SHIFT + 3));
1780 	ret = extract_crng_user(buf, nbytes);
1781 	trace_urandom_read(8 * nbytes, 0, ENTROPY_BITS(&input_pool));
1782 	return ret;
1783 }
1784 
1785 static __poll_t
1786 random_poll(struct file *file, poll_table * wait)
1787 {
1788 	__poll_t mask;
1789 
1790 	poll_wait(file, &random_read_wait, wait);
1791 	poll_wait(file, &random_write_wait, wait);
1792 	mask = 0;
1793 	if (ENTROPY_BITS(&input_pool) >= random_read_wakeup_bits)
1794 		mask |= EPOLLIN | EPOLLRDNORM;
1795 	if (ENTROPY_BITS(&input_pool) < random_write_wakeup_bits)
1796 		mask |= EPOLLOUT | EPOLLWRNORM;
1797 	return mask;
1798 }
1799 
1800 static int
1801 write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1802 {
1803 	size_t bytes;
1804 	__u32 buf[16];
1805 	const char __user *p = buffer;
1806 
1807 	while (count > 0) {
1808 		bytes = min(count, sizeof(buf));
1809 		if (copy_from_user(&buf, p, bytes))
1810 			return -EFAULT;
1811 
1812 		count -= bytes;
1813 		p += bytes;
1814 
1815 		mix_pool_bytes(r, buf, bytes);
1816 		cond_resched();
1817 	}
1818 
1819 	return 0;
1820 }
1821 
1822 static ssize_t random_write(struct file *file, const char __user *buffer,
1823 			    size_t count, loff_t *ppos)
1824 {
1825 	size_t ret;
1826 
1827 	ret = write_pool(&input_pool, buffer, count);
1828 	if (ret)
1829 		return ret;
1830 
1831 	return (ssize_t)count;
1832 }
1833 
1834 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1835 {
1836 	int size, ent_count;
1837 	int __user *p = (int __user *)arg;
1838 	int retval;
1839 
1840 	switch (cmd) {
1841 	case RNDGETENTCNT:
1842 		/* inherently racy, no point locking */
1843 		ent_count = ENTROPY_BITS(&input_pool);
1844 		if (put_user(ent_count, p))
1845 			return -EFAULT;
1846 		return 0;
1847 	case RNDADDTOENTCNT:
1848 		if (!capable(CAP_SYS_ADMIN))
1849 			return -EPERM;
1850 		if (get_user(ent_count, p))
1851 			return -EFAULT;
1852 		return credit_entropy_bits_safe(&input_pool, ent_count);
1853 	case RNDADDENTROPY:
1854 		if (!capable(CAP_SYS_ADMIN))
1855 			return -EPERM;
1856 		if (get_user(ent_count, p++))
1857 			return -EFAULT;
1858 		if (ent_count < 0)
1859 			return -EINVAL;
1860 		if (get_user(size, p++))
1861 			return -EFAULT;
1862 		retval = write_pool(&input_pool, (const char __user *)p,
1863 				    size);
1864 		if (retval < 0)
1865 			return retval;
1866 		return credit_entropy_bits_safe(&input_pool, ent_count);
1867 	case RNDZAPENTCNT:
1868 	case RNDCLEARPOOL:
1869 		/*
1870 		 * Clear the entropy pool counters. We no longer clear
1871 		 * the entropy pool, as that's silly.
1872 		 */
1873 		if (!capable(CAP_SYS_ADMIN))
1874 			return -EPERM;
1875 		input_pool.entropy_count = 0;
1876 		blocking_pool.entropy_count = 0;
1877 		return 0;
1878 	default:
1879 		return -EINVAL;
1880 	}
1881 }
1882 
1883 static int random_fasync(int fd, struct file *filp, int on)
1884 {
1885 	return fasync_helper(fd, filp, on, &fasync);
1886 }
1887 
1888 const struct file_operations random_fops = {
1889 	.read  = random_read,
1890 	.write = random_write,
1891 	.poll  = random_poll,
1892 	.unlocked_ioctl = random_ioctl,
1893 	.fasync = random_fasync,
1894 	.llseek = noop_llseek,
1895 };
1896 
1897 const struct file_operations urandom_fops = {
1898 	.read  = urandom_read,
1899 	.write = random_write,
1900 	.unlocked_ioctl = random_ioctl,
1901 	.fasync = random_fasync,
1902 	.llseek = noop_llseek,
1903 };
1904 
1905 SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count,
1906 		unsigned int, flags)
1907 {
1908 	int ret;
1909 
1910 	if (flags & ~(GRND_NONBLOCK|GRND_RANDOM))
1911 		return -EINVAL;
1912 
1913 	if (count > INT_MAX)
1914 		count = INT_MAX;
1915 
1916 	if (flags & GRND_RANDOM)
1917 		return _random_read(flags & GRND_NONBLOCK, buf, count);
1918 
1919 	if (!crng_ready()) {
1920 		if (flags & GRND_NONBLOCK)
1921 			return -EAGAIN;
1922 		ret = wait_for_random_bytes();
1923 		if (unlikely(ret))
1924 			return ret;
1925 	}
1926 	return urandom_read(NULL, buf, count, NULL);
1927 }
1928 
1929 /********************************************************************
1930  *
1931  * Sysctl interface
1932  *
1933  ********************************************************************/
1934 
1935 #ifdef CONFIG_SYSCTL
1936 
1937 #include <linux/sysctl.h>
1938 
1939 static int min_read_thresh = 8, min_write_thresh;
1940 static int max_read_thresh = OUTPUT_POOL_WORDS * 32;
1941 static int max_write_thresh = INPUT_POOL_WORDS * 32;
1942 static int random_min_urandom_seed = 60;
1943 static char sysctl_bootid[16];
1944 
1945 /*
1946  * This function is used to return both the bootid UUID, and random
1947  * UUID.  The difference is in whether table->data is NULL; if it is,
1948  * then a new UUID is generated and returned to the user.
1949  *
1950  * If the user accesses this via the proc interface, the UUID will be
1951  * returned as an ASCII string in the standard UUID format; if via the
1952  * sysctl system call, as 16 bytes of binary data.
1953  */
1954 static int proc_do_uuid(struct ctl_table *table, int write,
1955 			void __user *buffer, size_t *lenp, loff_t *ppos)
1956 {
1957 	struct ctl_table fake_table;
1958 	unsigned char buf[64], tmp_uuid[16], *uuid;
1959 
1960 	uuid = table->data;
1961 	if (!uuid) {
1962 		uuid = tmp_uuid;
1963 		generate_random_uuid(uuid);
1964 	} else {
1965 		static DEFINE_SPINLOCK(bootid_spinlock);
1966 
1967 		spin_lock(&bootid_spinlock);
1968 		if (!uuid[8])
1969 			generate_random_uuid(uuid);
1970 		spin_unlock(&bootid_spinlock);
1971 	}
1972 
1973 	sprintf(buf, "%pU", uuid);
1974 
1975 	fake_table.data = buf;
1976 	fake_table.maxlen = sizeof(buf);
1977 
1978 	return proc_dostring(&fake_table, write, buffer, lenp, ppos);
1979 }
1980 
1981 /*
1982  * Return entropy available scaled to integral bits
1983  */
1984 static int proc_do_entropy(struct ctl_table *table, int write,
1985 			   void __user *buffer, size_t *lenp, loff_t *ppos)
1986 {
1987 	struct ctl_table fake_table;
1988 	int entropy_count;
1989 
1990 	entropy_count = *(int *)table->data >> ENTROPY_SHIFT;
1991 
1992 	fake_table.data = &entropy_count;
1993 	fake_table.maxlen = sizeof(entropy_count);
1994 
1995 	return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
1996 }
1997 
1998 static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
1999 extern struct ctl_table random_table[];
2000 struct ctl_table random_table[] = {
2001 	{
2002 		.procname	= "poolsize",
2003 		.data		= &sysctl_poolsize,
2004 		.maxlen		= sizeof(int),
2005 		.mode		= 0444,
2006 		.proc_handler	= proc_dointvec,
2007 	},
2008 	{
2009 		.procname	= "entropy_avail",
2010 		.maxlen		= sizeof(int),
2011 		.mode		= 0444,
2012 		.proc_handler	= proc_do_entropy,
2013 		.data		= &input_pool.entropy_count,
2014 	},
2015 	{
2016 		.procname	= "read_wakeup_threshold",
2017 		.data		= &random_read_wakeup_bits,
2018 		.maxlen		= sizeof(int),
2019 		.mode		= 0644,
2020 		.proc_handler	= proc_dointvec_minmax,
2021 		.extra1		= &min_read_thresh,
2022 		.extra2		= &max_read_thresh,
2023 	},
2024 	{
2025 		.procname	= "write_wakeup_threshold",
2026 		.data		= &random_write_wakeup_bits,
2027 		.maxlen		= sizeof(int),
2028 		.mode		= 0644,
2029 		.proc_handler	= proc_dointvec_minmax,
2030 		.extra1		= &min_write_thresh,
2031 		.extra2		= &max_write_thresh,
2032 	},
2033 	{
2034 		.procname	= "urandom_min_reseed_secs",
2035 		.data		= &random_min_urandom_seed,
2036 		.maxlen		= sizeof(int),
2037 		.mode		= 0644,
2038 		.proc_handler	= proc_dointvec,
2039 	},
2040 	{
2041 		.procname	= "boot_id",
2042 		.data		= &sysctl_bootid,
2043 		.maxlen		= 16,
2044 		.mode		= 0444,
2045 		.proc_handler	= proc_do_uuid,
2046 	},
2047 	{
2048 		.procname	= "uuid",
2049 		.maxlen		= 16,
2050 		.mode		= 0444,
2051 		.proc_handler	= proc_do_uuid,
2052 	},
2053 #ifdef ADD_INTERRUPT_BENCH
2054 	{
2055 		.procname	= "add_interrupt_avg_cycles",
2056 		.data		= &avg_cycles,
2057 		.maxlen		= sizeof(avg_cycles),
2058 		.mode		= 0444,
2059 		.proc_handler	= proc_doulongvec_minmax,
2060 	},
2061 	{
2062 		.procname	= "add_interrupt_avg_deviation",
2063 		.data		= &avg_deviation,
2064 		.maxlen		= sizeof(avg_deviation),
2065 		.mode		= 0444,
2066 		.proc_handler	= proc_doulongvec_minmax,
2067 	},
2068 #endif
2069 	{ }
2070 };
2071 #endif 	/* CONFIG_SYSCTL */
2072 
2073 struct batched_entropy {
2074 	union {
2075 		u64 entropy_u64[CHACHA20_BLOCK_SIZE / sizeof(u64)];
2076 		u32 entropy_u32[CHACHA20_BLOCK_SIZE / sizeof(u32)];
2077 	};
2078 	unsigned int position;
2079 };
2080 static rwlock_t batched_entropy_reset_lock = __RW_LOCK_UNLOCKED(batched_entropy_reset_lock);
2081 
2082 /*
2083  * Get a random word for internal kernel use only. The quality of the random
2084  * number is either as good as RDRAND or as good as /dev/urandom, with the
2085  * goal of being quite fast and not depleting entropy. In order to ensure
2086  * that the randomness provided by this function is okay, the function
2087  * wait_for_random_bytes() should be called and return 0 at least once
2088  * at any point prior.
2089  */
2090 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64);
2091 u64 get_random_u64(void)
2092 {
2093 	u64 ret;
2094 	bool use_lock;
2095 	unsigned long flags = 0;
2096 	struct batched_entropy *batch;
2097 	static void *previous;
2098 
2099 #if BITS_PER_LONG == 64
2100 	if (arch_get_random_long((unsigned long *)&ret))
2101 		return ret;
2102 #else
2103 	if (arch_get_random_long((unsigned long *)&ret) &&
2104 	    arch_get_random_long((unsigned long *)&ret + 1))
2105 	    return ret;
2106 #endif
2107 
2108 	warn_unseeded_randomness(&previous);
2109 
2110 	use_lock = READ_ONCE(crng_init) < 2;
2111 	batch = &get_cpu_var(batched_entropy_u64);
2112 	if (use_lock)
2113 		read_lock_irqsave(&batched_entropy_reset_lock, flags);
2114 	if (batch->position % ARRAY_SIZE(batch->entropy_u64) == 0) {
2115 		extract_crng((__u32 *)batch->entropy_u64);
2116 		batch->position = 0;
2117 	}
2118 	ret = batch->entropy_u64[batch->position++];
2119 	if (use_lock)
2120 		read_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2121 	put_cpu_var(batched_entropy_u64);
2122 	return ret;
2123 }
2124 EXPORT_SYMBOL(get_random_u64);
2125 
2126 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32);
2127 u32 get_random_u32(void)
2128 {
2129 	u32 ret;
2130 	bool use_lock;
2131 	unsigned long flags = 0;
2132 	struct batched_entropy *batch;
2133 	static void *previous;
2134 
2135 	if (arch_get_random_int(&ret))
2136 		return ret;
2137 
2138 	warn_unseeded_randomness(&previous);
2139 
2140 	use_lock = READ_ONCE(crng_init) < 2;
2141 	batch = &get_cpu_var(batched_entropy_u32);
2142 	if (use_lock)
2143 		read_lock_irqsave(&batched_entropy_reset_lock, flags);
2144 	if (batch->position % ARRAY_SIZE(batch->entropy_u32) == 0) {
2145 		extract_crng(batch->entropy_u32);
2146 		batch->position = 0;
2147 	}
2148 	ret = batch->entropy_u32[batch->position++];
2149 	if (use_lock)
2150 		read_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2151 	put_cpu_var(batched_entropy_u32);
2152 	return ret;
2153 }
2154 EXPORT_SYMBOL(get_random_u32);
2155 
2156 /* It's important to invalidate all potential batched entropy that might
2157  * be stored before the crng is initialized, which we can do lazily by
2158  * simply resetting the counter to zero so that it's re-extracted on the
2159  * next usage. */
2160 static void invalidate_batched_entropy(void)
2161 {
2162 	int cpu;
2163 	unsigned long flags;
2164 
2165 	write_lock_irqsave(&batched_entropy_reset_lock, flags);
2166 	for_each_possible_cpu (cpu) {
2167 		per_cpu_ptr(&batched_entropy_u32, cpu)->position = 0;
2168 		per_cpu_ptr(&batched_entropy_u64, cpu)->position = 0;
2169 	}
2170 	write_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2171 }
2172 
2173 /**
2174  * randomize_page - Generate a random, page aligned address
2175  * @start:	The smallest acceptable address the caller will take.
2176  * @range:	The size of the area, starting at @start, within which the
2177  *		random address must fall.
2178  *
2179  * If @start + @range would overflow, @range is capped.
2180  *
2181  * NOTE: Historical use of randomize_range, which this replaces, presumed that
2182  * @start was already page aligned.  We now align it regardless.
2183  *
2184  * Return: A page aligned address within [start, start + range).  On error,
2185  * @start is returned.
2186  */
2187 unsigned long
2188 randomize_page(unsigned long start, unsigned long range)
2189 {
2190 	if (!PAGE_ALIGNED(start)) {
2191 		range -= PAGE_ALIGN(start) - start;
2192 		start = PAGE_ALIGN(start);
2193 	}
2194 
2195 	if (start > ULONG_MAX - range)
2196 		range = ULONG_MAX - start;
2197 
2198 	range >>= PAGE_SHIFT;
2199 
2200 	if (range == 0)
2201 		return start;
2202 
2203 	return start + (get_random_long() % range << PAGE_SHIFT);
2204 }
2205 
2206 /* Interface for in-kernel drivers of true hardware RNGs.
2207  * Those devices may produce endless random bits and will be throttled
2208  * when our pool is full.
2209  */
2210 void add_hwgenerator_randomness(const char *buffer, size_t count,
2211 				size_t entropy)
2212 {
2213 	struct entropy_store *poolp = &input_pool;
2214 
2215 	if (!crng_ready()) {
2216 		crng_fast_load(buffer, count);
2217 		return;
2218 	}
2219 
2220 	/* Suspend writing if we're above the trickle threshold.
2221 	 * We'll be woken up again once below random_write_wakeup_thresh,
2222 	 * or when the calling thread is about to terminate.
2223 	 */
2224 	wait_event_interruptible(random_write_wait, kthread_should_stop() ||
2225 			ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
2226 	mix_pool_bytes(poolp, buffer, count);
2227 	credit_entropy_bits(poolp, entropy);
2228 }
2229 EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
2230