[go: nahoru, domu]

1/*
2 * random.c -- A strong random number generator
3 *
4 * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005
5 *
6 * Copyright Theodore Ts'o, 1994, 1995, 1996, 1997, 1998, 1999.  All
7 * rights reserved.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, and the entire permission notice in its entirety,
14 *    including the disclaimer of warranties.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 *    notice, this list of conditions and the following disclaimer in the
17 *    documentation and/or other materials provided with the distribution.
18 * 3. The name of the author may not be used to endorse or promote
19 *    products derived from this software without specific prior
20 *    written permission.
21 *
22 * ALTERNATIVELY, this product may be distributed under the terms of
23 * the GNU General Public License, in which case the provisions of the GPL are
24 * required INSTEAD OF the above restrictions.  (This clause is
25 * necessary due to a potential bad interaction between the GPL and
26 * the restrictions contained in a BSD-style copyright.)
27 *
28 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
29 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
30 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ALL OF
31 * WHICH ARE HEREBY DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE
32 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
33 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT
34 * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
35 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
36 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
37 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
38 * USE OF THIS SOFTWARE, EVEN IF NOT ADVISED OF THE POSSIBILITY OF SUCH
39 * DAMAGE.
40 */
41
42/*
43 * (now, with legal B.S. out of the way.....)
44 *
45 * This routine gathers environmental noise from device drivers, etc.,
46 * and returns good random numbers, suitable for cryptographic use.
47 * Besides the obvious cryptographic uses, these numbers are also good
48 * for seeding TCP sequence numbers, and other places where it is
49 * desirable to have numbers which are not only random, but hard to
50 * predict by an attacker.
51 *
52 * Theory of operation
53 * ===================
54 *
55 * Computers are very predictable devices.  Hence it is extremely hard
56 * to produce truly random numbers on a computer --- as opposed to
57 * pseudo-random numbers, which can easily generated by using a
58 * algorithm.  Unfortunately, it is very easy for attackers to guess
59 * the sequence of pseudo-random number generators, and for some
60 * applications this is not acceptable.  So instead, we must try to
61 * gather "environmental noise" from the computer's environment, which
62 * must be hard for outside attackers to observe, and use that to
63 * generate random numbers.  In a Unix environment, this is best done
64 * from inside the kernel.
65 *
66 * Sources of randomness from the environment include inter-keyboard
67 * timings, inter-interrupt timings from some interrupts, and other
68 * events which are both (a) non-deterministic and (b) hard for an
69 * outside observer to measure.  Randomness from these sources are
70 * added to an "entropy pool", which is mixed using a CRC-like function.
71 * This is not cryptographically strong, but it is adequate assuming
72 * the randomness is not chosen maliciously, and it is fast enough that
73 * the overhead of doing it on every interrupt is very reasonable.
74 * As random bytes are mixed into the entropy pool, the routines keep
75 * an *estimate* of how many bits of randomness have been stored into
76 * the random number generator's internal state.
77 *
78 * When random bytes are desired, they are obtained by taking the SHA
79 * hash of the contents of the "entropy pool".  The SHA hash avoids
80 * exposing the internal state of the entropy pool.  It is believed to
81 * be computationally infeasible to derive any useful information
82 * about the input of SHA from its output.  Even if it is possible to
83 * analyze SHA in some clever way, as long as the amount of data
84 * returned from the generator is less than the inherent entropy in
85 * the pool, the output data is totally unpredictable.  For this
86 * reason, the routine decreases its internal estimate of how many
87 * bits of "true randomness" are contained in the entropy pool as it
88 * outputs random numbers.
89 *
90 * If this estimate goes to zero, the routine can still generate
91 * random numbers; however, an attacker may (at least in theory) be
92 * able to infer the future output of the generator from prior
93 * outputs.  This requires successful cryptanalysis of SHA, which is
94 * not believed to be feasible, but there is a remote possibility.
95 * Nonetheless, these numbers should be useful for the vast majority
96 * of purposes.
97 *
98 * Exported interfaces ---- output
99 * ===============================
100 *
101 * There are three exported interfaces; the first is one designed to
102 * be used from within the kernel:
103 *
104 * 	void get_random_bytes(void *buf, int nbytes);
105 *
106 * This interface will return the requested number of random bytes,
107 * and place it in the requested buffer.
108 *
109 * The two other interfaces are two character devices /dev/random and
110 * /dev/urandom.  /dev/random is suitable for use when very high
111 * quality randomness is desired (for example, for key generation or
112 * one-time pads), as it will only return a maximum of the number of
113 * bits of randomness (as estimated by the random number generator)
114 * contained in the entropy pool.
115 *
116 * The /dev/urandom device does not have this limit, and will return
117 * as many bytes as are requested.  As more and more random bytes are
118 * requested without giving time for the entropy pool to recharge,
119 * this will result in random numbers that are merely cryptographically
120 * strong.  For many applications, however, this is acceptable.
121 *
122 * Exported interfaces ---- input
123 * ==============================
124 *
125 * The current exported interfaces for gathering environmental noise
126 * from the devices are:
127 *
128 *	void add_device_randomness(const void *buf, unsigned int size);
129 * 	void add_input_randomness(unsigned int type, unsigned int code,
130 *                                unsigned int value);
131 *	void add_interrupt_randomness(int irq, int irq_flags);
132 * 	void add_disk_randomness(struct gendisk *disk);
133 *
134 * add_device_randomness() is for adding data to the random pool that
135 * is likely to differ between two devices (or possibly even per boot).
136 * This would be things like MAC addresses or serial numbers, or the
137 * read-out of the RTC. This does *not* add any actual entropy to the
138 * pool, but it initializes the pool to different values for devices
139 * that might otherwise be identical and have very little entropy
140 * available to them (particularly common in the embedded world).
141 *
142 * add_input_randomness() uses the input layer interrupt timing, as well as
143 * the event type information from the hardware.
144 *
145 * add_interrupt_randomness() uses the interrupt timing as random
146 * inputs to the entropy pool. Using the cycle counters and the irq source
147 * as inputs, it feeds the randomness roughly once a second.
148 *
149 * add_disk_randomness() uses what amounts to the seek time of block
150 * layer request events, on a per-disk_devt basis, as input to the
151 * entropy pool. Note that high-speed solid state drives with very low
152 * seek times do not make for good sources of entropy, as their seek
153 * times are usually fairly consistent.
154 *
155 * All of these routines try to estimate how many bits of randomness a
156 * particular randomness source.  They do this by keeping track of the
157 * first and second order deltas of the event timings.
158 *
159 * Ensuring unpredictability at system startup
160 * ============================================
161 *
162 * When any operating system starts up, it will go through a sequence
163 * of actions that are fairly predictable by an adversary, especially
164 * if the start-up does not involve interaction with a human operator.
165 * This reduces the actual number of bits of unpredictability in the
166 * entropy pool below the value in entropy_count.  In order to
167 * counteract this effect, it helps to carry information in the
168 * entropy pool across shut-downs and start-ups.  To do this, put the
169 * following lines an appropriate script which is run during the boot
170 * sequence:
171 *
172 *	echo "Initializing random number generator..."
173 *	random_seed=/var/run/random-seed
174 *	# Carry a random seed from start-up to start-up
175 *	# Load and then save the whole entropy pool
176 *	if [ -f $random_seed ]; then
177 *		cat $random_seed >/dev/urandom
178 *	else
179 *		touch $random_seed
180 *	fi
181 *	chmod 600 $random_seed
182 *	dd if=/dev/urandom of=$random_seed count=1 bs=512
183 *
184 * and the following lines in an appropriate script which is run as
185 * the system is shutdown:
186 *
187 *	# Carry a random seed from shut-down to start-up
188 *	# Save the whole entropy pool
189 *	echo "Saving random seed..."
190 *	random_seed=/var/run/random-seed
191 *	touch $random_seed
192 *	chmod 600 $random_seed
193 *	dd if=/dev/urandom of=$random_seed count=1 bs=512
194 *
195 * For example, on most modern systems using the System V init
196 * scripts, such code fragments would be found in
197 * /etc/rc.d/init.d/random.  On older Linux systems, the correct script
198 * location might be in /etc/rcb.d/rc.local or /etc/rc.d/rc.0.
199 *
200 * Effectively, these commands cause the contents of the entropy pool
201 * to be saved at shut-down time and reloaded into the entropy pool at
202 * start-up.  (The 'dd' in the addition to the bootup script is to
203 * make sure that /etc/random-seed is different for every start-up,
204 * even if the system crashes without executing rc.0.)  Even with
205 * complete knowledge of the start-up activities, predicting the state
206 * of the entropy pool requires knowledge of the previous history of
207 * the system.
208 *
209 * Configuring the /dev/random driver under Linux
210 * ==============================================
211 *
212 * The /dev/random driver under Linux uses minor numbers 8 and 9 of
213 * the /dev/mem major number (#1).  So if your system does not have
214 * /dev/random and /dev/urandom created already, they can be created
215 * by using the commands:
216 *
217 * 	mknod /dev/random c 1 8
218 * 	mknod /dev/urandom c 1 9
219 *
220 * Acknowledgements:
221 * =================
222 *
223 * Ideas for constructing this random number generator were derived
224 * from Pretty Good Privacy's random number generator, and from private
225 * discussions with Phil Karn.  Colin Plumb provided a faster random
226 * number generator, which speed up the mixing function of the entropy
227 * pool, taken from PGPfone.  Dale Worley has also contributed many
228 * useful ideas and suggestions to improve this driver.
229 *
230 * Any flaws in the design are solely my responsibility, and should
231 * not be attributed to the Phil, Colin, or any of authors of PGP.
232 *
233 * Further background information on this topic may be obtained from
234 * RFC 1750, "Randomness Recommendations for Security", by Donald
235 * Eastlake, Steve Crocker, and Jeff Schiller.
236 */
237
238#include <linux/utsname.h>
239#include <linux/module.h>
240#include <linux/kernel.h>
241#include <linux/major.h>
242#include <linux/string.h>
243#include <linux/fcntl.h>
244#include <linux/slab.h>
245#include <linux/random.h>
246#include <linux/poll.h>
247#include <linux/init.h>
248#include <linux/fs.h>
249#include <linux/genhd.h>
250#include <linux/interrupt.h>
251#include <linux/mm.h>
252#include <linux/spinlock.h>
253#include <linux/kthread.h>
254#include <linux/percpu.h>
255#include <linux/cryptohash.h>
256#include <linux/fips.h>
257#include <linux/ptrace.h>
258#include <linux/kmemcheck.h>
259#include <linux/workqueue.h>
260#include <linux/irq.h>
261#include <linux/syscalls.h>
262#include <linux/completion.h>
263
264#include <asm/processor.h>
265#include <asm/uaccess.h>
266#include <asm/irq.h>
267#include <asm/irq_regs.h>
268#include <asm/io.h>
269
270#define CREATE_TRACE_POINTS
271#include <trace/events/random.h>
272
273/* #define ADD_INTERRUPT_BENCH */
274
275/*
276 * Configuration information
277 */
278#define INPUT_POOL_SHIFT	12
279#define INPUT_POOL_WORDS	(1 << (INPUT_POOL_SHIFT-5))
280#define OUTPUT_POOL_SHIFT	10
281#define OUTPUT_POOL_WORDS	(1 << (OUTPUT_POOL_SHIFT-5))
282#define SEC_XFER_SIZE		512
283#define EXTRACT_SIZE		10
284
285#define DEBUG_RANDOM_BOOT 0
286
287#define LONGS(x) (((x) + sizeof(unsigned long) - 1)/sizeof(unsigned long))
288
289/*
290 * To allow fractional bits to be tracked, the entropy_count field is
291 * denominated in units of 1/8th bits.
292 *
293 * 2*(ENTROPY_SHIFT + log2(poolbits)) must <= 31, or the multiply in
294 * credit_entropy_bits() needs to be 64 bits wide.
295 */
296#define ENTROPY_SHIFT 3
297#define ENTROPY_BITS(r) ((r)->entropy_count >> ENTROPY_SHIFT)
298
299/*
300 * The minimum number of bits of entropy before we wake up a read on
301 * /dev/random.  Should be enough to do a significant reseed.
302 */
303static int random_read_wakeup_bits = 64;
304
305/*
306 * If the entropy count falls under this number of bits, then we
307 * should wake up processes which are selecting or polling on write
308 * access to /dev/random.
309 */
310static int random_write_wakeup_bits = 28 * OUTPUT_POOL_WORDS;
311
312/*
313 * The minimum number of seconds between urandom pool reseeding.  We
314 * do this to limit the amount of entropy that can be drained from the
315 * input pool even if there are heavy demands on /dev/urandom.
316 */
317static int random_min_urandom_seed = 60;
318
319/*
320 * Originally, we used a primitive polynomial of degree .poolwords
321 * over GF(2).  The taps for various sizes are defined below.  They
322 * were chosen to be evenly spaced except for the last tap, which is 1
323 * to get the twisting happening as fast as possible.
324 *
325 * For the purposes of better mixing, we use the CRC-32 polynomial as
326 * well to make a (modified) twisted Generalized Feedback Shift
327 * Register.  (See M. Matsumoto & Y. Kurita, 1992.  Twisted GFSR
328 * generators.  ACM Transactions on Modeling and Computer Simulation
329 * 2(3):179-194.  Also see M. Matsumoto & Y. Kurita, 1994.  Twisted
330 * GFSR generators II.  ACM Transactions on Modeling and Computer
331 * Simulation 4:254-266)
332 *
333 * Thanks to Colin Plumb for suggesting this.
334 *
335 * The mixing operation is much less sensitive than the output hash,
336 * where we use SHA-1.  All that we want of mixing operation is that
337 * it be a good non-cryptographic hash; i.e. it not produce collisions
338 * when fed "random" data of the sort we expect to see.  As long as
339 * the pool state differs for different inputs, we have preserved the
340 * input entropy and done a good job.  The fact that an intelligent
341 * attacker can construct inputs that will produce controlled
342 * alterations to the pool's state is not important because we don't
343 * consider such inputs to contribute any randomness.  The only
344 * property we need with respect to them is that the attacker can't
345 * increase his/her knowledge of the pool's state.  Since all
346 * additions are reversible (knowing the final state and the input,
347 * you can reconstruct the initial state), if an attacker has any
348 * uncertainty about the initial state, he/she can only shuffle that
349 * uncertainty about, but never cause any collisions (which would
350 * decrease the uncertainty).
351 *
352 * Our mixing functions were analyzed by Lacharme, Roeck, Strubel, and
353 * Videau in their paper, "The Linux Pseudorandom Number Generator
354 * Revisited" (see: http://eprint.iacr.org/2012/251.pdf).  In their
355 * paper, they point out that we are not using a true Twisted GFSR,
356 * since Matsumoto & Kurita used a trinomial feedback polynomial (that
357 * is, with only three taps, instead of the six that we are using).
358 * As a result, the resulting polynomial is neither primitive nor
359 * irreducible, and hence does not have a maximal period over
360 * GF(2**32).  They suggest a slight change to the generator
361 * polynomial which improves the resulting TGFSR polynomial to be
362 * irreducible, which we have made here.
363 */
364static struct poolinfo {
365	int poolbitshift, poolwords, poolbytes, poolbits, poolfracbits;
366#define S(x) ilog2(x)+5, (x), (x)*4, (x)*32, (x) << (ENTROPY_SHIFT+5)
367	int tap1, tap2, tap3, tap4, tap5;
368} poolinfo_table[] = {
369	/* was: x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 */
370	/* x^128 + x^104 + x^76 + x^51 +x^25 + x + 1 */
371	{ S(128),	104,	76,	51,	25,	1 },
372	/* was: x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 */
373	/* x^32 + x^26 + x^19 + x^14 + x^7 + x + 1 */
374	{ S(32),	26,	19,	14,	7,	1 },
375#if 0
376	/* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1  -- 115 */
377	{ S(2048),	1638,	1231,	819,	411,	1 },
378
379	/* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
380	{ S(1024),	817,	615,	412,	204,	1 },
381
382	/* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
383	{ S(1024),	819,	616,	410,	207,	2 },
384
385	/* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
386	{ S(512),	411,	308,	208,	104,	1 },
387
388	/* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
389	{ S(512),	409,	307,	206,	102,	2 },
390	/* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
391	{ S(512),	409,	309,	205,	103,	2 },
392
393	/* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
394	{ S(256),	205,	155,	101,	52,	1 },
395
396	/* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
397	{ S(128),	103,	78,	51,	27,	2 },
398
399	/* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
400	{ S(64),	52,	39,	26,	14,	1 },
401#endif
402};
403
404/*
405 * Static global variables
406 */
407static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
408static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
409static DECLARE_WAIT_QUEUE_HEAD(urandom_init_wait);
410static struct fasync_struct *fasync;
411
412/**********************************************************************
413 *
414 * OS independent entropy store.   Here are the functions which handle
415 * storing entropy in an entropy pool.
416 *
417 **********************************************************************/
418
419struct entropy_store;
420struct entropy_store {
421	/* read-only data: */
422	const struct poolinfo *poolinfo;
423	__u32 *pool;
424	const char *name;
425	struct entropy_store *pull;
426	struct work_struct push_work;
427
428	/* read-write data: */
429	unsigned long last_pulled;
430	spinlock_t lock;
431	unsigned short add_ptr;
432	unsigned short input_rotate;
433	int entropy_count;
434	int entropy_total;
435	unsigned int initialized:1;
436	unsigned int limit:1;
437	unsigned int last_data_init:1;
438	__u8 last_data[EXTRACT_SIZE];
439};
440
441static void push_to_pool(struct work_struct *work);
442static __u32 input_pool_data[INPUT_POOL_WORDS];
443static __u32 blocking_pool_data[OUTPUT_POOL_WORDS];
444static __u32 nonblocking_pool_data[OUTPUT_POOL_WORDS];
445
446static struct entropy_store input_pool = {
447	.poolinfo = &poolinfo_table[0],
448	.name = "input",
449	.limit = 1,
450	.lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
451	.pool = input_pool_data
452};
453
454static struct entropy_store blocking_pool = {
455	.poolinfo = &poolinfo_table[1],
456	.name = "blocking",
457	.limit = 1,
458	.pull = &input_pool,
459	.lock = __SPIN_LOCK_UNLOCKED(blocking_pool.lock),
460	.pool = blocking_pool_data,
461	.push_work = __WORK_INITIALIZER(blocking_pool.push_work,
462					push_to_pool),
463};
464
465static struct entropy_store nonblocking_pool = {
466	.poolinfo = &poolinfo_table[1],
467	.name = "nonblocking",
468	.pull = &input_pool,
469	.lock = __SPIN_LOCK_UNLOCKED(nonblocking_pool.lock),
470	.pool = nonblocking_pool_data,
471	.push_work = __WORK_INITIALIZER(nonblocking_pool.push_work,
472					push_to_pool),
473};
474
475static __u32 const twist_table[8] = {
476	0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
477	0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
478
479/*
480 * This function adds bytes into the entropy "pool".  It does not
481 * update the entropy estimate.  The caller should call
482 * credit_entropy_bits if this is appropriate.
483 *
484 * The pool is stirred with a primitive polynomial of the appropriate
485 * degree, and then twisted.  We twist by three bits at a time because
486 * it's cheap to do so and helps slightly in the expected case where
487 * the entropy is concentrated in the low-order bits.
488 */
489static void _mix_pool_bytes(struct entropy_store *r, const void *in,
490			    int nbytes)
491{
492	unsigned long i, tap1, tap2, tap3, tap4, tap5;
493	int input_rotate;
494	int wordmask = r->poolinfo->poolwords - 1;
495	const char *bytes = in;
496	__u32 w;
497
498	tap1 = r->poolinfo->tap1;
499	tap2 = r->poolinfo->tap2;
500	tap3 = r->poolinfo->tap3;
501	tap4 = r->poolinfo->tap4;
502	tap5 = r->poolinfo->tap5;
503
504	input_rotate = r->input_rotate;
505	i = r->add_ptr;
506
507	/* mix one byte at a time to simplify size handling and churn faster */
508	while (nbytes--) {
509		w = rol32(*bytes++, input_rotate);
510		i = (i - 1) & wordmask;
511
512		/* XOR in the various taps */
513		w ^= r->pool[i];
514		w ^= r->pool[(i + tap1) & wordmask];
515		w ^= r->pool[(i + tap2) & wordmask];
516		w ^= r->pool[(i + tap3) & wordmask];
517		w ^= r->pool[(i + tap4) & wordmask];
518		w ^= r->pool[(i + tap5) & wordmask];
519
520		/* Mix the result back in with a twist */
521		r->pool[i] = (w >> 3) ^ twist_table[w & 7];
522
523		/*
524		 * Normally, we add 7 bits of rotation to the pool.
525		 * At the beginning of the pool, add an extra 7 bits
526		 * rotation, so that successive passes spread the
527		 * input bits across the pool evenly.
528		 */
529		input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
530	}
531
532	r->input_rotate = input_rotate;
533	r->add_ptr = i;
534}
535
536static void __mix_pool_bytes(struct entropy_store *r, const void *in,
537			     int nbytes)
538{
539	trace_mix_pool_bytes_nolock(r->name, nbytes, _RET_IP_);
540	_mix_pool_bytes(r, in, nbytes);
541}
542
543static void mix_pool_bytes(struct entropy_store *r, const void *in,
544			   int nbytes)
545{
546	unsigned long flags;
547
548	trace_mix_pool_bytes(r->name, nbytes, _RET_IP_);
549	spin_lock_irqsave(&r->lock, flags);
550	_mix_pool_bytes(r, in, nbytes);
551	spin_unlock_irqrestore(&r->lock, flags);
552}
553
554struct fast_pool {
555	__u32		pool[4];
556	unsigned long	last;
557	unsigned short	reg_idx;
558	unsigned char	count;
559};
560
561/*
562 * This is a fast mixing routine used by the interrupt randomness
563 * collector.  It's hardcoded for an 128 bit pool and assumes that any
564 * locks that might be needed are taken by the caller.
565 */
566static void fast_mix(struct fast_pool *f)
567{
568	__u32 a = f->pool[0],	b = f->pool[1];
569	__u32 c = f->pool[2],	d = f->pool[3];
570
571	a += b;			c += d;
572	b = rol32(a, 6);	d = rol32(c, 27);
573	d ^= a;			b ^= c;
574
575	a += b;			c += d;
576	b = rol32(a, 16);	d = rol32(c, 14);
577	d ^= a;			b ^= c;
578
579	a += b;			c += d;
580	b = rol32(a, 6);	d = rol32(c, 27);
581	d ^= a;			b ^= c;
582
583	a += b;			c += d;
584	b = rol32(a, 16);	d = rol32(c, 14);
585	d ^= a;			b ^= c;
586
587	f->pool[0] = a;  f->pool[1] = b;
588	f->pool[2] = c;  f->pool[3] = d;
589	f->count++;
590}
591
592/*
593 * Credit (or debit) the entropy store with n bits of entropy.
594 * Use credit_entropy_bits_safe() if the value comes from userspace
595 * or otherwise should be checked for extreme values.
596 */
597static void credit_entropy_bits(struct entropy_store *r, int nbits)
598{
599	int entropy_count, orig;
600	const int pool_size = r->poolinfo->poolfracbits;
601	int nfrac = nbits << ENTROPY_SHIFT;
602
603	if (!nbits)
604		return;
605
606retry:
607	entropy_count = orig = ACCESS_ONCE(r->entropy_count);
608	if (nfrac < 0) {
609		/* Debit */
610		entropy_count += nfrac;
611	} else {
612		/*
613		 * Credit: we have to account for the possibility of
614		 * overwriting already present entropy.	 Even in the
615		 * ideal case of pure Shannon entropy, new contributions
616		 * approach the full value asymptotically:
617		 *
618		 * entropy <- entropy + (pool_size - entropy) *
619		 *	(1 - exp(-add_entropy/pool_size))
620		 *
621		 * For add_entropy <= pool_size/2 then
622		 * (1 - exp(-add_entropy/pool_size)) >=
623		 *    (add_entropy/pool_size)*0.7869...
624		 * so we can approximate the exponential with
625		 * 3/4*add_entropy/pool_size and still be on the
626		 * safe side by adding at most pool_size/2 at a time.
627		 *
628		 * The use of pool_size-2 in the while statement is to
629		 * prevent rounding artifacts from making the loop
630		 * arbitrarily long; this limits the loop to log2(pool_size)*2
631		 * turns no matter how large nbits is.
632		 */
633		int pnfrac = nfrac;
634		const int s = r->poolinfo->poolbitshift + ENTROPY_SHIFT + 2;
635		/* The +2 corresponds to the /4 in the denominator */
636
637		do {
638			unsigned int anfrac = min(pnfrac, pool_size/2);
639			unsigned int add =
640				((pool_size - entropy_count)*anfrac*3) >> s;
641
642			entropy_count += add;
643			pnfrac -= anfrac;
644		} while (unlikely(entropy_count < pool_size-2 && pnfrac));
645	}
646
647	if (unlikely(entropy_count < 0)) {
648		pr_warn("random: negative entropy/overflow: pool %s count %d\n",
649			r->name, entropy_count);
650		WARN_ON(1);
651		entropy_count = 0;
652	} else if (entropy_count > pool_size)
653		entropy_count = pool_size;
654	if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
655		goto retry;
656
657	r->entropy_total += nbits;
658	if (!r->initialized && r->entropy_total > 128) {
659		r->initialized = 1;
660		r->entropy_total = 0;
661		if (r == &nonblocking_pool) {
662			prandom_reseed_late();
663			wake_up_interruptible(&urandom_init_wait);
664			pr_notice("random: %s pool is initialized\n", r->name);
665		}
666	}
667
668	trace_credit_entropy_bits(r->name, nbits,
669				  entropy_count >> ENTROPY_SHIFT,
670				  r->entropy_total, _RET_IP_);
671
672	if (r == &input_pool) {
673		int entropy_bits = entropy_count >> ENTROPY_SHIFT;
674
675		/* should we wake readers? */
676		if (entropy_bits >= random_read_wakeup_bits) {
677			wake_up_interruptible(&random_read_wait);
678			kill_fasync(&fasync, SIGIO, POLL_IN);
679		}
680		/* If the input pool is getting full, send some
681		 * entropy to the two output pools, flipping back and
682		 * forth between them, until the output pools are 75%
683		 * full.
684		 */
685		if (entropy_bits > random_write_wakeup_bits &&
686		    r->initialized &&
687		    r->entropy_total >= 2*random_read_wakeup_bits) {
688			static struct entropy_store *last = &blocking_pool;
689			struct entropy_store *other = &blocking_pool;
690
691			if (last == &blocking_pool)
692				other = &nonblocking_pool;
693			if (other->entropy_count <=
694			    3 * other->poolinfo->poolfracbits / 4)
695				last = other;
696			if (last->entropy_count <=
697			    3 * last->poolinfo->poolfracbits / 4) {
698				schedule_work(&last->push_work);
699				r->entropy_total = 0;
700			}
701		}
702	}
703}
704
705static void credit_entropy_bits_safe(struct entropy_store *r, int nbits)
706{
707	const int nbits_max = (int)(~0U >> (ENTROPY_SHIFT + 1));
708
709	/* Cap the value to avoid overflows */
710	nbits = min(nbits,  nbits_max);
711	nbits = max(nbits, -nbits_max);
712
713	credit_entropy_bits(r, nbits);
714}
715
716/*********************************************************************
717 *
718 * Entropy input management
719 *
720 *********************************************************************/
721
722/* There is one of these per entropy source */
723struct timer_rand_state {
724	cycles_t last_time;
725	long last_delta, last_delta2;
726	unsigned dont_count_entropy:1;
727};
728
729#define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
730
731/*
732 * Add device- or boot-specific data to the input and nonblocking
733 * pools to help initialize them to unique values.
734 *
735 * None of this adds any entropy, it is meant to avoid the
736 * problem of the nonblocking pool having similar initial state
737 * across largely identical devices.
738 */
739void add_device_randomness(const void *buf, unsigned int size)
740{
741	unsigned long time = random_get_entropy() ^ jiffies;
742	unsigned long flags;
743
744	trace_add_device_randomness(size, _RET_IP_);
745	spin_lock_irqsave(&input_pool.lock, flags);
746	_mix_pool_bytes(&input_pool, buf, size);
747	_mix_pool_bytes(&input_pool, &time, sizeof(time));
748	spin_unlock_irqrestore(&input_pool.lock, flags);
749
750	spin_lock_irqsave(&nonblocking_pool.lock, flags);
751	_mix_pool_bytes(&nonblocking_pool, buf, size);
752	_mix_pool_bytes(&nonblocking_pool, &time, sizeof(time));
753	spin_unlock_irqrestore(&nonblocking_pool.lock, flags);
754}
755EXPORT_SYMBOL(add_device_randomness);
756
757static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
758
759/*
760 * This function adds entropy to the entropy "pool" by using timing
761 * delays.  It uses the timer_rand_state structure to make an estimate
762 * of how many bits of entropy this call has added to the pool.
763 *
764 * The number "num" is also added to the pool - it should somehow describe
765 * the type of event which just happened.  This is currently 0-255 for
766 * keyboard scan codes, and 256 upwards for interrupts.
767 *
768 */
769static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
770{
771	struct entropy_store	*r;
772	struct {
773		long jiffies;
774		unsigned cycles;
775		unsigned num;
776	} sample;
777	long delta, delta2, delta3;
778
779	preempt_disable();
780
781	sample.jiffies = jiffies;
782	sample.cycles = random_get_entropy();
783	sample.num = num;
784	r = nonblocking_pool.initialized ? &input_pool : &nonblocking_pool;
785	mix_pool_bytes(r, &sample, sizeof(sample));
786
787	/*
788	 * Calculate number of bits of randomness we probably added.
789	 * We take into account the first, second and third-order deltas
790	 * in order to make our estimate.
791	 */
792
793	if (!state->dont_count_entropy) {
794		delta = sample.jiffies - state->last_time;
795		state->last_time = sample.jiffies;
796
797		delta2 = delta - state->last_delta;
798		state->last_delta = delta;
799
800		delta3 = delta2 - state->last_delta2;
801		state->last_delta2 = delta2;
802
803		if (delta < 0)
804			delta = -delta;
805		if (delta2 < 0)
806			delta2 = -delta2;
807		if (delta3 < 0)
808			delta3 = -delta3;
809		if (delta > delta2)
810			delta = delta2;
811		if (delta > delta3)
812			delta = delta3;
813
814		/*
815		 * delta is now minimum absolute delta.
816		 * Round down by 1 bit on general principles,
817		 * and limit entropy entimate to 12 bits.
818		 */
819		credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
820	}
821	preempt_enable();
822}
823
824void add_input_randomness(unsigned int type, unsigned int code,
825				 unsigned int value)
826{
827	static unsigned char last_value;
828
829	/* ignore autorepeat and the like */
830	if (value == last_value)
831		return;
832
833	last_value = value;
834	add_timer_randomness(&input_timer_state,
835			     (type << 4) ^ code ^ (code >> 4) ^ value);
836	trace_add_input_randomness(ENTROPY_BITS(&input_pool));
837}
838EXPORT_SYMBOL_GPL(add_input_randomness);
839
840static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
841
842#ifdef ADD_INTERRUPT_BENCH
843static unsigned long avg_cycles, avg_deviation;
844
845#define AVG_SHIFT 8     /* Exponential average factor k=1/256 */
846#define FIXED_1_2 (1 << (AVG_SHIFT-1))
847
848static void add_interrupt_bench(cycles_t start)
849{
850        long delta = random_get_entropy() - start;
851
852        /* Use a weighted moving average */
853        delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
854        avg_cycles += delta;
855        /* And average deviation */
856        delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
857        avg_deviation += delta;
858}
859#else
860#define add_interrupt_bench(x)
861#endif
862
863static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
864{
865	__u32 *ptr = (__u32 *) regs;
866
867	if (regs == NULL)
868		return 0;
869	if (f->reg_idx >= sizeof(struct pt_regs) / sizeof(__u32))
870		f->reg_idx = 0;
871	return *(ptr + f->reg_idx++);
872}
873
874void add_interrupt_randomness(int irq, int irq_flags)
875{
876	struct entropy_store	*r;
877	struct fast_pool	*fast_pool = this_cpu_ptr(&irq_randomness);
878	struct pt_regs		*regs = get_irq_regs();
879	unsigned long		now = jiffies;
880	cycles_t		cycles = random_get_entropy();
881	__u32			c_high, j_high;
882	__u64			ip;
883	unsigned long		seed;
884	int			credit = 0;
885
886	if (cycles == 0)
887		cycles = get_reg(fast_pool, regs);
888	c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
889	j_high = (sizeof(now) > 4) ? now >> 32 : 0;
890	fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
891	fast_pool->pool[1] ^= now ^ c_high;
892	ip = regs ? instruction_pointer(regs) : _RET_IP_;
893	fast_pool->pool[2] ^= ip;
894	fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
895		get_reg(fast_pool, regs);
896
897	fast_mix(fast_pool);
898	add_interrupt_bench(cycles);
899
900	if ((fast_pool->count < 64) &&
901	    !time_after(now, fast_pool->last + HZ))
902		return;
903
904	r = nonblocking_pool.initialized ? &input_pool : &nonblocking_pool;
905	if (!spin_trylock(&r->lock))
906		return;
907
908	fast_pool->last = now;
909	__mix_pool_bytes(r, &fast_pool->pool, sizeof(fast_pool->pool));
910
911	/*
912	 * If we have architectural seed generator, produce a seed and
913	 * add it to the pool.  For the sake of paranoia don't let the
914	 * architectural seed generator dominate the input from the
915	 * interrupt noise.
916	 */
917	if (arch_get_random_seed_long(&seed)) {
918		__mix_pool_bytes(r, &seed, sizeof(seed));
919		credit = 1;
920	}
921	spin_unlock(&r->lock);
922
923	fast_pool->count = 0;
924
925	/* award one bit for the contents of the fast pool */
926	credit_entropy_bits(r, credit + 1);
927}
928
929#ifdef CONFIG_BLOCK
930void add_disk_randomness(struct gendisk *disk)
931{
932	if (!disk || !disk->random)
933		return;
934	/* first major is 1, so we get >= 0x200 here */
935	add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
936	trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
937}
938EXPORT_SYMBOL_GPL(add_disk_randomness);
939#endif
940
941/*********************************************************************
942 *
943 * Entropy extraction routines
944 *
945 *********************************************************************/
946
947static ssize_t extract_entropy(struct entropy_store *r, void *buf,
948			       size_t nbytes, int min, int rsvd);
949
950/*
951 * This utility inline function is responsible for transferring entropy
952 * from the primary pool to the secondary extraction pool. We make
953 * sure we pull enough for a 'catastrophic reseed'.
954 */
955static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes);
956static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
957{
958	if (!r->pull ||
959	    r->entropy_count >= (nbytes << (ENTROPY_SHIFT + 3)) ||
960	    r->entropy_count > r->poolinfo->poolfracbits)
961		return;
962
963	if (r->limit == 0 && random_min_urandom_seed) {
964		unsigned long now = jiffies;
965
966		if (time_before(now,
967				r->last_pulled + random_min_urandom_seed * HZ))
968			return;
969		r->last_pulled = now;
970	}
971
972	_xfer_secondary_pool(r, nbytes);
973}
974
975static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
976{
977	__u32	tmp[OUTPUT_POOL_WORDS];
978
979	/* For /dev/random's pool, always leave two wakeups' worth */
980	int rsvd_bytes = r->limit ? 0 : random_read_wakeup_bits / 4;
981	int bytes = nbytes;
982
983	/* pull at least as much as a wakeup */
984	bytes = max_t(int, bytes, random_read_wakeup_bits / 8);
985	/* but never more than the buffer size */
986	bytes = min_t(int, bytes, sizeof(tmp));
987
988	trace_xfer_secondary_pool(r->name, bytes * 8, nbytes * 8,
989				  ENTROPY_BITS(r), ENTROPY_BITS(r->pull));
990	bytes = extract_entropy(r->pull, tmp, bytes,
991				random_read_wakeup_bits / 8, rsvd_bytes);
992	mix_pool_bytes(r, tmp, bytes);
993	credit_entropy_bits(r, bytes*8);
994}
995
996/*
997 * Used as a workqueue function so that when the input pool is getting
998 * full, we can "spill over" some entropy to the output pools.  That
999 * way the output pools can store some of the excess entropy instead
1000 * of letting it go to waste.
1001 */
1002static void push_to_pool(struct work_struct *work)
1003{
1004	struct entropy_store *r = container_of(work, struct entropy_store,
1005					      push_work);
1006	BUG_ON(!r);
1007	_xfer_secondary_pool(r, random_read_wakeup_bits/8);
1008	trace_push_to_pool(r->name, r->entropy_count >> ENTROPY_SHIFT,
1009			   r->pull->entropy_count >> ENTROPY_SHIFT);
1010}
1011
1012/*
1013 * This function decides how many bytes to actually take from the
1014 * given pool, and also debits the entropy count accordingly.
1015 */
1016static size_t account(struct entropy_store *r, size_t nbytes, int min,
1017		      int reserved)
1018{
1019	int entropy_count, orig;
1020	size_t ibytes, nfrac;
1021
1022	BUG_ON(r->entropy_count > r->poolinfo->poolfracbits);
1023
1024	/* Can we pull enough? */
1025retry:
1026	entropy_count = orig = ACCESS_ONCE(r->entropy_count);
1027	ibytes = nbytes;
1028	/* If limited, never pull more than available */
1029	if (r->limit) {
1030		int have_bytes = entropy_count >> (ENTROPY_SHIFT + 3);
1031
1032		if ((have_bytes -= reserved) < 0)
1033			have_bytes = 0;
1034		ibytes = min_t(size_t, ibytes, have_bytes);
1035	}
1036	if (ibytes < min)
1037		ibytes = 0;
1038
1039	if (unlikely(entropy_count < 0)) {
1040		pr_warn("random: negative entropy count: pool %s count %d\n",
1041			r->name, entropy_count);
1042		WARN_ON(1);
1043		entropy_count = 0;
1044	}
1045	nfrac = ibytes << (ENTROPY_SHIFT + 3);
1046	if ((size_t) entropy_count > nfrac)
1047		entropy_count -= nfrac;
1048	else
1049		entropy_count = 0;
1050
1051	if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
1052		goto retry;
1053
1054	trace_debit_entropy(r->name, 8 * ibytes);
1055	if (ibytes &&
1056	    (r->entropy_count >> ENTROPY_SHIFT) < random_write_wakeup_bits) {
1057		wake_up_interruptible(&random_write_wait);
1058		kill_fasync(&fasync, SIGIO, POLL_OUT);
1059	}
1060
1061	return ibytes;
1062}
1063
1064/*
1065 * This function does the actual extraction for extract_entropy and
1066 * extract_entropy_user.
1067 *
1068 * Note: we assume that .poolwords is a multiple of 16 words.
1069 */
1070static void extract_buf(struct entropy_store *r, __u8 *out)
1071{
1072	int i;
1073	union {
1074		__u32 w[5];
1075		unsigned long l[LONGS(20)];
1076	} hash;
1077	__u32 workspace[SHA_WORKSPACE_WORDS];
1078	unsigned long flags;
1079
1080	/*
1081	 * If we have an architectural hardware random number
1082	 * generator, use it for SHA's initial vector
1083	 */
1084	sha_init(hash.w);
1085	for (i = 0; i < LONGS(20); i++) {
1086		unsigned long v;
1087		if (!arch_get_random_long(&v))
1088			break;
1089		hash.l[i] = v;
1090	}
1091
1092	/* Generate a hash across the pool, 16 words (512 bits) at a time */
1093	spin_lock_irqsave(&r->lock, flags);
1094	for (i = 0; i < r->poolinfo->poolwords; i += 16)
1095		sha_transform(hash.w, (__u8 *)(r->pool + i), workspace);
1096
1097	/*
1098	 * We mix the hash back into the pool to prevent backtracking
1099	 * attacks (where the attacker knows the state of the pool
1100	 * plus the current outputs, and attempts to find previous
1101	 * ouputs), unless the hash function can be inverted. By
1102	 * mixing at least a SHA1 worth of hash data back, we make
1103	 * brute-forcing the feedback as hard as brute-forcing the
1104	 * hash.
1105	 */
1106	__mix_pool_bytes(r, hash.w, sizeof(hash.w));
1107	spin_unlock_irqrestore(&r->lock, flags);
1108
1109	memzero_explicit(workspace, sizeof(workspace));
1110
1111	/*
1112	 * In case the hash function has some recognizable output
1113	 * pattern, we fold it in half. Thus, we always feed back
1114	 * twice as much data as we output.
1115	 */
1116	hash.w[0] ^= hash.w[3];
1117	hash.w[1] ^= hash.w[4];
1118	hash.w[2] ^= rol32(hash.w[2], 16);
1119
1120	memcpy(out, &hash, EXTRACT_SIZE);
1121	memzero_explicit(&hash, sizeof(hash));
1122}
1123
1124/*
1125 * This function extracts randomness from the "entropy pool", and
1126 * returns it in a buffer.
1127 *
1128 * The min parameter specifies the minimum amount we can pull before
1129 * failing to avoid races that defeat catastrophic reseeding while the
1130 * reserved parameter indicates how much entropy we must leave in the
1131 * pool after each pull to avoid starving other readers.
1132 */
1133static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1134				 size_t nbytes, int min, int reserved)
1135{
1136	ssize_t ret = 0, i;
1137	__u8 tmp[EXTRACT_SIZE];
1138	unsigned long flags;
1139
1140	/* if last_data isn't primed, we need EXTRACT_SIZE extra bytes */
1141	if (fips_enabled) {
1142		spin_lock_irqsave(&r->lock, flags);
1143		if (!r->last_data_init) {
1144			r->last_data_init = 1;
1145			spin_unlock_irqrestore(&r->lock, flags);
1146			trace_extract_entropy(r->name, EXTRACT_SIZE,
1147					      ENTROPY_BITS(r), _RET_IP_);
1148			xfer_secondary_pool(r, EXTRACT_SIZE);
1149			extract_buf(r, tmp);
1150			spin_lock_irqsave(&r->lock, flags);
1151			memcpy(r->last_data, tmp, EXTRACT_SIZE);
1152		}
1153		spin_unlock_irqrestore(&r->lock, flags);
1154	}
1155
1156	trace_extract_entropy(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1157	xfer_secondary_pool(r, nbytes);
1158	nbytes = account(r, nbytes, min, reserved);
1159
1160	while (nbytes) {
1161		extract_buf(r, tmp);
1162
1163		if (fips_enabled) {
1164			spin_lock_irqsave(&r->lock, flags);
1165			if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
1166				panic("Hardware RNG duplicated output!\n");
1167			memcpy(r->last_data, tmp, EXTRACT_SIZE);
1168			spin_unlock_irqrestore(&r->lock, flags);
1169		}
1170		i = min_t(int, nbytes, EXTRACT_SIZE);
1171		memcpy(buf, tmp, i);
1172		nbytes -= i;
1173		buf += i;
1174		ret += i;
1175	}
1176
1177	/* Wipe data just returned from memory */
1178	memzero_explicit(tmp, sizeof(tmp));
1179
1180	return ret;
1181}
1182
1183/*
1184 * This function extracts randomness from the "entropy pool", and
1185 * returns it in a userspace buffer.
1186 */
1187static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
1188				    size_t nbytes)
1189{
1190	ssize_t ret = 0, i;
1191	__u8 tmp[EXTRACT_SIZE];
1192	int large_request = (nbytes > 256);
1193
1194	trace_extract_entropy_user(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1195	xfer_secondary_pool(r, nbytes);
1196	nbytes = account(r, nbytes, 0, 0);
1197
1198	while (nbytes) {
1199		if (large_request && need_resched()) {
1200			if (signal_pending(current)) {
1201				if (ret == 0)
1202					ret = -ERESTARTSYS;
1203				break;
1204			}
1205			schedule();
1206		}
1207
1208		extract_buf(r, tmp);
1209		i = min_t(int, nbytes, EXTRACT_SIZE);
1210		if (copy_to_user(buf, tmp, i)) {
1211			ret = -EFAULT;
1212			break;
1213		}
1214
1215		nbytes -= i;
1216		buf += i;
1217		ret += i;
1218	}
1219
1220	/* Wipe data just returned from memory */
1221	memzero_explicit(tmp, sizeof(tmp));
1222
1223	return ret;
1224}
1225
1226/*
1227 * This function is the exported kernel interface.  It returns some
1228 * number of good random numbers, suitable for key generation, seeding
1229 * TCP sequence numbers, etc.  It does not rely on the hardware random
1230 * number generator.  For random bytes direct from the hardware RNG
1231 * (when available), use get_random_bytes_arch().
1232 */
1233void get_random_bytes(void *buf, int nbytes)
1234{
1235#if DEBUG_RANDOM_BOOT > 0
1236	if (unlikely(nonblocking_pool.initialized == 0))
1237		printk(KERN_NOTICE "random: %pF get_random_bytes called "
1238		       "with %d bits of entropy available\n",
1239		       (void *) _RET_IP_,
1240		       nonblocking_pool.entropy_total);
1241#endif
1242	trace_get_random_bytes(nbytes, _RET_IP_);
1243	extract_entropy(&nonblocking_pool, buf, nbytes, 0, 0);
1244}
1245EXPORT_SYMBOL(get_random_bytes);
1246
1247/*
1248 * This function will use the architecture-specific hardware random
1249 * number generator if it is available.  The arch-specific hw RNG will
1250 * almost certainly be faster than what we can do in software, but it
1251 * is impossible to verify that it is implemented securely (as
1252 * opposed, to, say, the AES encryption of a sequence number using a
1253 * key known by the NSA).  So it's useful if we need the speed, but
1254 * only if we're willing to trust the hardware manufacturer not to
1255 * have put in a back door.
1256 */
1257void get_random_bytes_arch(void *buf, int nbytes)
1258{
1259	char *p = buf;
1260
1261	trace_get_random_bytes_arch(nbytes, _RET_IP_);
1262	while (nbytes) {
1263		unsigned long v;
1264		int chunk = min(nbytes, (int)sizeof(unsigned long));
1265
1266		if (!arch_get_random_long(&v))
1267			break;
1268
1269		memcpy(p, &v, chunk);
1270		p += chunk;
1271		nbytes -= chunk;
1272	}
1273
1274	if (nbytes)
1275		extract_entropy(&nonblocking_pool, p, nbytes, 0, 0);
1276}
1277EXPORT_SYMBOL(get_random_bytes_arch);
1278
1279
1280/*
1281 * init_std_data - initialize pool with system data
1282 *
1283 * @r: pool to initialize
1284 *
1285 * This function clears the pool's entropy count and mixes some system
1286 * data into the pool to prepare it for use. The pool is not cleared
1287 * as that can only decrease the entropy in the pool.
1288 */
1289static void init_std_data(struct entropy_store *r)
1290{
1291	int i;
1292	ktime_t now = ktime_get_real();
1293	unsigned long rv;
1294
1295	r->last_pulled = jiffies;
1296	mix_pool_bytes(r, &now, sizeof(now));
1297	for (i = r->poolinfo->poolbytes; i > 0; i -= sizeof(rv)) {
1298		if (!arch_get_random_seed_long(&rv) &&
1299		    !arch_get_random_long(&rv))
1300			rv = random_get_entropy();
1301		mix_pool_bytes(r, &rv, sizeof(rv));
1302	}
1303	mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1304}
1305
1306/*
1307 * Note that setup_arch() may call add_device_randomness()
1308 * long before we get here. This allows seeding of the pools
1309 * with some platform dependent data very early in the boot
1310 * process. But it limits our options here. We must use
1311 * statically allocated structures that already have all
1312 * initializations complete at compile time. We should also
1313 * take care not to overwrite the precious per platform data
1314 * we were given.
1315 */
1316static int rand_initialize(void)
1317{
1318	init_std_data(&input_pool);
1319	init_std_data(&blocking_pool);
1320	init_std_data(&nonblocking_pool);
1321	return 0;
1322}
1323early_initcall(rand_initialize);
1324
1325#ifdef CONFIG_BLOCK
1326void rand_initialize_disk(struct gendisk *disk)
1327{
1328	struct timer_rand_state *state;
1329
1330	/*
1331	 * If kzalloc returns null, we just won't use that entropy
1332	 * source.
1333	 */
1334	state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1335	if (state) {
1336		state->last_time = INITIAL_JIFFIES;
1337		disk->random = state;
1338	}
1339}
1340#endif
1341
1342static ssize_t
1343_random_read(int nonblock, char __user *buf, size_t nbytes)
1344{
1345	ssize_t n;
1346
1347	if (nbytes == 0)
1348		return 0;
1349
1350	nbytes = min_t(size_t, nbytes, SEC_XFER_SIZE);
1351	while (1) {
1352		n = extract_entropy_user(&blocking_pool, buf, nbytes);
1353		if (n < 0)
1354			return n;
1355		trace_random_read(n*8, (nbytes-n)*8,
1356				  ENTROPY_BITS(&blocking_pool),
1357				  ENTROPY_BITS(&input_pool));
1358		if (n > 0)
1359			return n;
1360
1361		/* Pool is (near) empty.  Maybe wait and retry. */
1362		if (nonblock)
1363			return -EAGAIN;
1364
1365		wait_event_interruptible(random_read_wait,
1366			ENTROPY_BITS(&input_pool) >=
1367			random_read_wakeup_bits);
1368		if (signal_pending(current))
1369			return -ERESTARTSYS;
1370	}
1371}
1372
1373static ssize_t
1374random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1375{
1376	return _random_read(file->f_flags & O_NONBLOCK, buf, nbytes);
1377}
1378
1379static ssize_t
1380urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1381{
1382	int ret;
1383
1384	if (unlikely(nonblocking_pool.initialized == 0))
1385		printk_once(KERN_NOTICE "random: %s urandom read "
1386			    "with %d bits of entropy available\n",
1387			    current->comm, nonblocking_pool.entropy_total);
1388
1389	nbytes = min_t(size_t, nbytes, INT_MAX >> (ENTROPY_SHIFT + 3));
1390	ret = extract_entropy_user(&nonblocking_pool, buf, nbytes);
1391
1392	trace_urandom_read(8 * nbytes, ENTROPY_BITS(&nonblocking_pool),
1393			   ENTROPY_BITS(&input_pool));
1394	return ret;
1395}
1396
1397static unsigned int
1398random_poll(struct file *file, poll_table * wait)
1399{
1400	unsigned int mask;
1401
1402	poll_wait(file, &random_read_wait, wait);
1403	poll_wait(file, &random_write_wait, wait);
1404	mask = 0;
1405	if (ENTROPY_BITS(&input_pool) >= random_read_wakeup_bits)
1406		mask |= POLLIN | POLLRDNORM;
1407	if (ENTROPY_BITS(&input_pool) < random_write_wakeup_bits)
1408		mask |= POLLOUT | POLLWRNORM;
1409	return mask;
1410}
1411
1412static int
1413write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1414{
1415	size_t bytes;
1416	__u32 buf[16];
1417	const char __user *p = buffer;
1418
1419	while (count > 0) {
1420		bytes = min(count, sizeof(buf));
1421		if (copy_from_user(&buf, p, bytes))
1422			return -EFAULT;
1423
1424		count -= bytes;
1425		p += bytes;
1426
1427		mix_pool_bytes(r, buf, bytes);
1428		cond_resched();
1429	}
1430
1431	return 0;
1432}
1433
1434static ssize_t random_write(struct file *file, const char __user *buffer,
1435			    size_t count, loff_t *ppos)
1436{
1437	size_t ret;
1438
1439	ret = write_pool(&blocking_pool, buffer, count);
1440	if (ret)
1441		return ret;
1442	ret = write_pool(&nonblocking_pool, buffer, count);
1443	if (ret)
1444		return ret;
1445
1446	return (ssize_t)count;
1447}
1448
1449static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1450{
1451	int size, ent_count;
1452	int __user *p = (int __user *)arg;
1453	int retval;
1454
1455	switch (cmd) {
1456	case RNDGETENTCNT:
1457		/* inherently racy, no point locking */
1458		ent_count = ENTROPY_BITS(&input_pool);
1459		if (put_user(ent_count, p))
1460			return -EFAULT;
1461		return 0;
1462	case RNDADDTOENTCNT:
1463		if (!capable(CAP_SYS_ADMIN))
1464			return -EPERM;
1465		if (get_user(ent_count, p))
1466			return -EFAULT;
1467		credit_entropy_bits_safe(&input_pool, ent_count);
1468		return 0;
1469	case RNDADDENTROPY:
1470		if (!capable(CAP_SYS_ADMIN))
1471			return -EPERM;
1472		if (get_user(ent_count, p++))
1473			return -EFAULT;
1474		if (ent_count < 0)
1475			return -EINVAL;
1476		if (get_user(size, p++))
1477			return -EFAULT;
1478		retval = write_pool(&input_pool, (const char __user *)p,
1479				    size);
1480		if (retval < 0)
1481			return retval;
1482		credit_entropy_bits_safe(&input_pool, ent_count);
1483		return 0;
1484	case RNDZAPENTCNT:
1485	case RNDCLEARPOOL:
1486		/*
1487		 * Clear the entropy pool counters. We no longer clear
1488		 * the entropy pool, as that's silly.
1489		 */
1490		if (!capable(CAP_SYS_ADMIN))
1491			return -EPERM;
1492		input_pool.entropy_count = 0;
1493		nonblocking_pool.entropy_count = 0;
1494		blocking_pool.entropy_count = 0;
1495		return 0;
1496	default:
1497		return -EINVAL;
1498	}
1499}
1500
1501static int random_fasync(int fd, struct file *filp, int on)
1502{
1503	return fasync_helper(fd, filp, on, &fasync);
1504}
1505
1506const struct file_operations random_fops = {
1507	.read  = random_read,
1508	.write = random_write,
1509	.poll  = random_poll,
1510	.unlocked_ioctl = random_ioctl,
1511	.fasync = random_fasync,
1512	.llseek = noop_llseek,
1513};
1514
1515const struct file_operations urandom_fops = {
1516	.read  = urandom_read,
1517	.write = random_write,
1518	.unlocked_ioctl = random_ioctl,
1519	.fasync = random_fasync,
1520	.llseek = noop_llseek,
1521};
1522
1523SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count,
1524		unsigned int, flags)
1525{
1526	if (flags & ~(GRND_NONBLOCK|GRND_RANDOM))
1527		return -EINVAL;
1528
1529	if (count > INT_MAX)
1530		count = INT_MAX;
1531
1532	if (flags & GRND_RANDOM)
1533		return _random_read(flags & GRND_NONBLOCK, buf, count);
1534
1535	if (unlikely(nonblocking_pool.initialized == 0)) {
1536		if (flags & GRND_NONBLOCK)
1537			return -EAGAIN;
1538		wait_event_interruptible(urandom_init_wait,
1539					 nonblocking_pool.initialized);
1540		if (signal_pending(current))
1541			return -ERESTARTSYS;
1542	}
1543	return urandom_read(NULL, buf, count, NULL);
1544}
1545
1546/***************************************************************
1547 * Random UUID interface
1548 *
1549 * Used here for a Boot ID, but can be useful for other kernel
1550 * drivers.
1551 ***************************************************************/
1552
1553/*
1554 * Generate random UUID
1555 */
1556void generate_random_uuid(unsigned char uuid_out[16])
1557{
1558	get_random_bytes(uuid_out, 16);
1559	/* Set UUID version to 4 --- truly random generation */
1560	uuid_out[6] = (uuid_out[6] & 0x0F) | 0x40;
1561	/* Set the UUID variant to DCE */
1562	uuid_out[8] = (uuid_out[8] & 0x3F) | 0x80;
1563}
1564EXPORT_SYMBOL(generate_random_uuid);
1565
1566/********************************************************************
1567 *
1568 * Sysctl interface
1569 *
1570 ********************************************************************/
1571
1572#ifdef CONFIG_SYSCTL
1573
1574#include <linux/sysctl.h>
1575
1576static int min_read_thresh = 8, min_write_thresh;
1577static int max_read_thresh = OUTPUT_POOL_WORDS * 32;
1578static int max_write_thresh = INPUT_POOL_WORDS * 32;
1579static char sysctl_bootid[16];
1580
1581/*
1582 * This function is used to return both the bootid UUID, and random
1583 * UUID.  The difference is in whether table->data is NULL; if it is,
1584 * then a new UUID is generated and returned to the user.
1585 *
1586 * If the user accesses this via the proc interface, the UUID will be
1587 * returned as an ASCII string in the standard UUID format; if via the
1588 * sysctl system call, as 16 bytes of binary data.
1589 */
1590static int proc_do_uuid(struct ctl_table *table, int write,
1591			void __user *buffer, size_t *lenp, loff_t *ppos)
1592{
1593	struct ctl_table fake_table;
1594	unsigned char buf[64], tmp_uuid[16], *uuid;
1595
1596	uuid = table->data;
1597	if (!uuid) {
1598		uuid = tmp_uuid;
1599		generate_random_uuid(uuid);
1600	} else {
1601		static DEFINE_SPINLOCK(bootid_spinlock);
1602
1603		spin_lock(&bootid_spinlock);
1604		if (!uuid[8])
1605			generate_random_uuid(uuid);
1606		spin_unlock(&bootid_spinlock);
1607	}
1608
1609	sprintf(buf, "%pU", uuid);
1610
1611	fake_table.data = buf;
1612	fake_table.maxlen = sizeof(buf);
1613
1614	return proc_dostring(&fake_table, write, buffer, lenp, ppos);
1615}
1616
1617/*
1618 * Return entropy available scaled to integral bits
1619 */
1620static int proc_do_entropy(struct ctl_table *table, int write,
1621			   void __user *buffer, size_t *lenp, loff_t *ppos)
1622{
1623	struct ctl_table fake_table;
1624	int entropy_count;
1625
1626	entropy_count = *(int *)table->data >> ENTROPY_SHIFT;
1627
1628	fake_table.data = &entropy_count;
1629	fake_table.maxlen = sizeof(entropy_count);
1630
1631	return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
1632}
1633
1634static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
1635extern struct ctl_table random_table[];
1636struct ctl_table random_table[] = {
1637	{
1638		.procname	= "poolsize",
1639		.data		= &sysctl_poolsize,
1640		.maxlen		= sizeof(int),
1641		.mode		= 0444,
1642		.proc_handler	= proc_dointvec,
1643	},
1644	{
1645		.procname	= "entropy_avail",
1646		.maxlen		= sizeof(int),
1647		.mode		= 0444,
1648		.proc_handler	= proc_do_entropy,
1649		.data		= &input_pool.entropy_count,
1650	},
1651	{
1652		.procname	= "read_wakeup_threshold",
1653		.data		= &random_read_wakeup_bits,
1654		.maxlen		= sizeof(int),
1655		.mode		= 0644,
1656		.proc_handler	= proc_dointvec_minmax,
1657		.extra1		= &min_read_thresh,
1658		.extra2		= &max_read_thresh,
1659	},
1660	{
1661		.procname	= "write_wakeup_threshold",
1662		.data		= &random_write_wakeup_bits,
1663		.maxlen		= sizeof(int),
1664		.mode		= 0644,
1665		.proc_handler	= proc_dointvec_minmax,
1666		.extra1		= &min_write_thresh,
1667		.extra2		= &max_write_thresh,
1668	},
1669	{
1670		.procname	= "urandom_min_reseed_secs",
1671		.data		= &random_min_urandom_seed,
1672		.maxlen		= sizeof(int),
1673		.mode		= 0644,
1674		.proc_handler	= proc_dointvec,
1675	},
1676	{
1677		.procname	= "boot_id",
1678		.data		= &sysctl_bootid,
1679		.maxlen		= 16,
1680		.mode		= 0444,
1681		.proc_handler	= proc_do_uuid,
1682	},
1683	{
1684		.procname	= "uuid",
1685		.maxlen		= 16,
1686		.mode		= 0444,
1687		.proc_handler	= proc_do_uuid,
1688	},
1689#ifdef ADD_INTERRUPT_BENCH
1690	{
1691		.procname	= "add_interrupt_avg_cycles",
1692		.data		= &avg_cycles,
1693		.maxlen		= sizeof(avg_cycles),
1694		.mode		= 0444,
1695		.proc_handler	= proc_doulongvec_minmax,
1696	},
1697	{
1698		.procname	= "add_interrupt_avg_deviation",
1699		.data		= &avg_deviation,
1700		.maxlen		= sizeof(avg_deviation),
1701		.mode		= 0444,
1702		.proc_handler	= proc_doulongvec_minmax,
1703	},
1704#endif
1705	{ }
1706};
1707#endif 	/* CONFIG_SYSCTL */
1708
1709static u32 random_int_secret[MD5_MESSAGE_BYTES / 4] ____cacheline_aligned;
1710
1711int random_int_secret_init(void)
1712{
1713	get_random_bytes(random_int_secret, sizeof(random_int_secret));
1714	return 0;
1715}
1716
1717/*
1718 * Get a random word for internal kernel use only. Similar to urandom but
1719 * with the goal of minimal entropy pool depletion. As a result, the random
1720 * value is not cryptographically secure but for several uses the cost of
1721 * depleting entropy is too high
1722 */
1723static DEFINE_PER_CPU(__u32 [MD5_DIGEST_WORDS], get_random_int_hash);
1724unsigned int get_random_int(void)
1725{
1726	__u32 *hash;
1727	unsigned int ret;
1728
1729	if (arch_get_random_int(&ret))
1730		return ret;
1731
1732	hash = get_cpu_var(get_random_int_hash);
1733
1734	hash[0] += current->pid + jiffies + random_get_entropy();
1735	md5_transform(hash, random_int_secret);
1736	ret = hash[0];
1737	put_cpu_var(get_random_int_hash);
1738
1739	return ret;
1740}
1741EXPORT_SYMBOL(get_random_int);
1742
1743/*
1744 * randomize_range() returns a start address such that
1745 *
1746 *    [...... <range> .....]
1747 *  start                  end
1748 *
1749 * a <range> with size "len" starting at the return value is inside in the
1750 * area defined by [start, end], but is otherwise randomized.
1751 */
1752unsigned long
1753randomize_range(unsigned long start, unsigned long end, unsigned long len)
1754{
1755	unsigned long range = end - len - start;
1756
1757	if (end <= start + len)
1758		return 0;
1759	return PAGE_ALIGN(get_random_int() % range + start);
1760}
1761
1762/* Interface for in-kernel drivers of true hardware RNGs.
1763 * Those devices may produce endless random bits and will be throttled
1764 * when our pool is full.
1765 */
1766void add_hwgenerator_randomness(const char *buffer, size_t count,
1767				size_t entropy)
1768{
1769	struct entropy_store *poolp = &input_pool;
1770
1771	/* Suspend writing if we're above the trickle threshold.
1772	 * We'll be woken up again once below random_write_wakeup_thresh,
1773	 * or when the calling thread is about to terminate.
1774	 */
1775	wait_event_interruptible(random_write_wait, kthread_should_stop() ||
1776			ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
1777	mix_pool_bytes(poolp, buffer, count);
1778	credit_entropy_bits(poolp, entropy);
1779}
1780EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
1781