Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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 */
307static 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 */
314static 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 */
361static 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 */
404static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
405static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
406static struct fasync_struct *fasync;
407
408static DEFINE_SPINLOCK(random_ready_list_lock);
409static LIST_HEAD(random_ready_list);
410
411struct crng_state {
412 __u32 state[16];
413 unsigned long init_time;
414 spinlock_t lock;
415};
416
417struct 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 */
429static int crng_init = 0;
430#define crng_ready() (likely(crng_init > 1))
431static int crng_init_cnt = 0;
432static unsigned long crng_global_init_time = 0;
433#define CRNG_INIT_CNT_THRESH (2*CHACHA20_KEY_SIZE)
434static void _extract_crng(struct crng_state *crng,
435 __u32 out[CHACHA20_BLOCK_WORDS]);
436static void _crng_backtrack_protect(struct crng_state *crng,
437 __u32 tmp[CHACHA20_BLOCK_WORDS], int used);
438static void process_random_ready_list(void);
439static void _get_random_bytes(void *buf, int nbytes);
440
441/**********************************************************************
442 *
443 * OS independent entropy store. Here are the functions which handle
444 * storing entropy in an entropy pool.
445 *
446 **********************************************************************/
447
448struct entropy_store;
449struct entropy_store {
450 /* read-only data: */
451 const struct poolinfo *poolinfo;
452 __u32 *pool;
453 const char *name;
454 struct entropy_store *pull;
455 struct work_struct push_work;
456
457 /* read-write data: */
458 unsigned long last_pulled;
459 spinlock_t lock;
460 unsigned short add_ptr;
461 unsigned short input_rotate;
462 int entropy_count;
463 int entropy_total;
464 unsigned int initialized:1;
465 unsigned int last_data_init:1;
466 __u8 last_data[EXTRACT_SIZE];
467};
468
469static ssize_t extract_entropy(struct entropy_store *r, void *buf,
470 size_t nbytes, int min, int rsvd);
471static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
472 size_t nbytes, int fips);
473
474static void crng_reseed(struct crng_state *crng, struct entropy_store *r);
475static void push_to_pool(struct work_struct *work);
476static __u32 input_pool_data[INPUT_POOL_WORDS] __latent_entropy;
477static __u32 blocking_pool_data[OUTPUT_POOL_WORDS] __latent_entropy;
478
479static struct entropy_store input_pool = {
480 .poolinfo = &poolinfo_table[0],
481 .name = "input",
482 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
483 .pool = input_pool_data
484};
485
486static struct entropy_store blocking_pool = {
487 .poolinfo = &poolinfo_table[1],
488 .name = "blocking",
489 .pull = &input_pool,
490 .lock = __SPIN_LOCK_UNLOCKED(blocking_pool.lock),
491 .pool = blocking_pool_data,
492 .push_work = __WORK_INITIALIZER(blocking_pool.push_work,
493 push_to_pool),
494};
495
496static __u32 const twist_table[8] = {
497 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
498 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
499
500/*
501 * This function adds bytes into the entropy "pool". It does not
502 * update the entropy estimate. The caller should call
503 * credit_entropy_bits if this is appropriate.
504 *
505 * The pool is stirred with a primitive polynomial of the appropriate
506 * degree, and then twisted. We twist by three bits at a time because
507 * it's cheap to do so and helps slightly in the expected case where
508 * the entropy is concentrated in the low-order bits.
509 */
510static void _mix_pool_bytes(struct entropy_store *r, const void *in,
511 int nbytes)
512{
513 unsigned long i, tap1, tap2, tap3, tap4, tap5;
514 int input_rotate;
515 int wordmask = r->poolinfo->poolwords - 1;
516 const char *bytes = in;
517 __u32 w;
518
519 tap1 = r->poolinfo->tap1;
520 tap2 = r->poolinfo->tap2;
521 tap3 = r->poolinfo->tap3;
522 tap4 = r->poolinfo->tap4;
523 tap5 = r->poolinfo->tap5;
524
525 input_rotate = r->input_rotate;
526 i = r->add_ptr;
527
528 /* mix one byte at a time to simplify size handling and churn faster */
529 while (nbytes--) {
530 w = rol32(*bytes++, input_rotate);
531 i = (i - 1) & wordmask;
532
533 /* XOR in the various taps */
534 w ^= r->pool[i];
535 w ^= r->pool[(i + tap1) & wordmask];
536 w ^= r->pool[(i + tap2) & wordmask];
537 w ^= r->pool[(i + tap3) & wordmask];
538 w ^= r->pool[(i + tap4) & wordmask];
539 w ^= r->pool[(i + tap5) & wordmask];
540
541 /* Mix the result back in with a twist */
542 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
543
544 /*
545 * Normally, we add 7 bits of rotation to the pool.
546 * At the beginning of the pool, add an extra 7 bits
547 * rotation, so that successive passes spread the
548 * input bits across the pool evenly.
549 */
550 input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
551 }
552
553 r->input_rotate = input_rotate;
554 r->add_ptr = i;
555}
556
557static void __mix_pool_bytes(struct entropy_store *r, const void *in,
558 int nbytes)
559{
560 trace_mix_pool_bytes_nolock(r->name, nbytes, _RET_IP_);
561 _mix_pool_bytes(r, in, nbytes);
562}
563
564static void mix_pool_bytes(struct entropy_store *r, const void *in,
565 int nbytes)
566{
567 unsigned long flags;
568
569 trace_mix_pool_bytes(r->name, nbytes, _RET_IP_);
570 spin_lock_irqsave(&r->lock, flags);
571 _mix_pool_bytes(r, in, nbytes);
572 spin_unlock_irqrestore(&r->lock, flags);
573}
574
575struct fast_pool {
576 __u32 pool[4];
577 unsigned long last;
578 unsigned short reg_idx;
579 unsigned char count;
580};
581
582/*
583 * This is a fast mixing routine used by the interrupt randomness
584 * collector. It's hardcoded for an 128 bit pool and assumes that any
585 * locks that might be needed are taken by the caller.
586 */
587static void fast_mix(struct fast_pool *f)
588{
589 __u32 a = f->pool[0], b = f->pool[1];
590 __u32 c = f->pool[2], d = f->pool[3];
591
592 a += b; c += d;
593 b = rol32(b, 6); d = rol32(d, 27);
594 d ^= a; b ^= c;
595
596 a += b; c += d;
597 b = rol32(b, 16); d = rol32(d, 14);
598 d ^= a; b ^= c;
599
600 a += b; c += d;
601 b = rol32(b, 6); d = rol32(d, 27);
602 d ^= a; b ^= c;
603
604 a += b; c += d;
605 b = rol32(b, 16); d = rol32(d, 14);
606 d ^= a; b ^= c;
607
608 f->pool[0] = a; f->pool[1] = b;
609 f->pool[2] = c; f->pool[3] = d;
610 f->count++;
611}
612
613static void process_random_ready_list(void)
614{
615 unsigned long flags;
616 struct random_ready_callback *rdy, *tmp;
617
618 spin_lock_irqsave(&random_ready_list_lock, flags);
619 list_for_each_entry_safe(rdy, tmp, &random_ready_list, list) {
620 struct module *owner = rdy->owner;
621
622 list_del_init(&rdy->list);
623 rdy->func(rdy);
624 module_put(owner);
625 }
626 spin_unlock_irqrestore(&random_ready_list_lock, flags);
627}
628
629/*
630 * Credit (or debit) the entropy store with n bits of entropy.
631 * Use credit_entropy_bits_safe() if the value comes from userspace
632 * or otherwise should be checked for extreme values.
633 */
634static void credit_entropy_bits(struct entropy_store *r, int nbits)
635{
636 int entropy_count, orig;
637 const int pool_size = r->poolinfo->poolfracbits;
638 int nfrac = nbits << ENTROPY_SHIFT;
639
640 if (!nbits)
641 return;
642
643retry:
644 entropy_count = orig = READ_ONCE(r->entropy_count);
645 if (nfrac < 0) {
646 /* Debit */
647 entropy_count += nfrac;
648 } else {
649 /*
650 * Credit: we have to account for the possibility of
651 * overwriting already present entropy. Even in the
652 * ideal case of pure Shannon entropy, new contributions
653 * approach the full value asymptotically:
654 *
655 * entropy <- entropy + (pool_size - entropy) *
656 * (1 - exp(-add_entropy/pool_size))
657 *
658 * For add_entropy <= pool_size/2 then
659 * (1 - exp(-add_entropy/pool_size)) >=
660 * (add_entropy/pool_size)*0.7869...
661 * so we can approximate the exponential with
662 * 3/4*add_entropy/pool_size and still be on the
663 * safe side by adding at most pool_size/2 at a time.
664 *
665 * The use of pool_size-2 in the while statement is to
666 * prevent rounding artifacts from making the loop
667 * arbitrarily long; this limits the loop to log2(pool_size)*2
668 * turns no matter how large nbits is.
669 */
670 int pnfrac = nfrac;
671 const int s = r->poolinfo->poolbitshift + ENTROPY_SHIFT + 2;
672 /* The +2 corresponds to the /4 in the denominator */
673
674 do {
675 unsigned int anfrac = min(pnfrac, pool_size/2);
676 unsigned int add =
677 ((pool_size - entropy_count)*anfrac*3) >> s;
678
679 entropy_count += add;
680 pnfrac -= anfrac;
681 } while (unlikely(entropy_count < pool_size-2 && pnfrac));
682 }
683
684 if (unlikely(entropy_count < 0)) {
685 pr_warn("random: negative entropy/overflow: pool %s count %d\n",
686 r->name, entropy_count);
687 WARN_ON(1);
688 entropy_count = 0;
689 } else if (entropy_count > pool_size)
690 entropy_count = pool_size;
691 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
692 goto retry;
693
694 r->entropy_total += nbits;
695 if (!r->initialized && r->entropy_total > 128) {
696 r->initialized = 1;
697 r->entropy_total = 0;
698 }
699
700 trace_credit_entropy_bits(r->name, nbits,
701 entropy_count >> ENTROPY_SHIFT,
702 r->entropy_total, _RET_IP_);
703
704 if (r == &input_pool) {
705 int entropy_bits = entropy_count >> ENTROPY_SHIFT;
706
707 if (crng_init < 2 && entropy_bits >= 128) {
708 crng_reseed(&primary_crng, r);
709 entropy_bits = r->entropy_count >> ENTROPY_SHIFT;
710 }
711
712 /* should we wake readers? */
713 if (entropy_bits >= random_read_wakeup_bits &&
714 wq_has_sleeper(&random_read_wait)) {
715 wake_up_interruptible(&random_read_wait);
716 kill_fasync(&fasync, SIGIO, POLL_IN);
717 }
718 /* If the input pool is getting full, send some
719 * entropy to the blocking pool until it is 75% full.
720 */
721 if (entropy_bits > random_write_wakeup_bits &&
722 r->initialized &&
723 r->entropy_total >= 2*random_read_wakeup_bits) {
724 struct entropy_store *other = &blocking_pool;
725
726 if (other->entropy_count <=
727 3 * other->poolinfo->poolfracbits / 4) {
728 schedule_work(&other->push_work);
729 r->entropy_total = 0;
730 }
731 }
732 }
733}
734
735static int credit_entropy_bits_safe(struct entropy_store *r, int nbits)
736{
737 const int nbits_max = r->poolinfo->poolwords * 32;
738
739 if (nbits < 0)
740 return -EINVAL;
741
742 /* Cap the value to avoid overflows */
743 nbits = min(nbits, nbits_max);
744
745 credit_entropy_bits(r, nbits);
746 return 0;
747}
748
749/*********************************************************************
750 *
751 * CRNG using CHACHA20
752 *
753 *********************************************************************/
754
755#define CRNG_RESEED_INTERVAL (300*HZ)
756
757static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
758
759#ifdef CONFIG_NUMA
760/*
761 * Hack to deal with crazy userspace progams when they are all trying
762 * to access /dev/urandom in parallel. The programs are almost
763 * certainly doing something terribly wrong, but we'll work around
764 * their brain damage.
765 */
766static struct crng_state **crng_node_pool __read_mostly;
767#endif
768
769static void invalidate_batched_entropy(void);
770
771static void crng_initialize(struct crng_state *crng)
772{
773 int i;
774 unsigned long rv;
775
776 memcpy(&crng->state[0], "expand 32-byte k", 16);
777 if (crng == &primary_crng)
778 _extract_entropy(&input_pool, &crng->state[4],
779 sizeof(__u32) * 12, 0);
780 else
781 _get_random_bytes(&crng->state[4], sizeof(__u32) * 12);
782 for (i = 4; i < 16; i++) {
783 if (!arch_get_random_seed_long(&rv) &&
784 !arch_get_random_long(&rv))
785 rv = random_get_entropy();
786 crng->state[i] ^= rv;
787 }
788 crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
789}
790
791#ifdef CONFIG_NUMA
792static void numa_crng_init(void)
793{
794 int i;
795 struct crng_state *crng;
796 struct crng_state **pool;
797
798 pool = kcalloc(nr_node_ids, sizeof(*pool), GFP_KERNEL|__GFP_NOFAIL);
799 for_each_online_node(i) {
800 crng = kmalloc_node(sizeof(struct crng_state),
801 GFP_KERNEL | __GFP_NOFAIL, i);
802 spin_lock_init(&crng->lock);
803 crng_initialize(crng);
804 pool[i] = crng;
805 }
806 mb();
807 if (cmpxchg(&crng_node_pool, NULL, pool)) {
808 for_each_node(i)
809 kfree(pool[i]);
810 kfree(pool);
811 }
812}
813#else
814static void numa_crng_init(void) {}
815#endif
816
817/*
818 * crng_fast_load() can be called by code in the interrupt service
819 * path. So we can't afford to dilly-dally.
820 */
821static int crng_fast_load(const char *cp, size_t len)
822{
823 unsigned long flags;
824 char *p;
825
826 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
827 return 0;
828 if (crng_init != 0) {
829 spin_unlock_irqrestore(&primary_crng.lock, flags);
830 return 0;
831 }
832 p = (unsigned char *) &primary_crng.state[4];
833 while (len > 0 && crng_init_cnt < CRNG_INIT_CNT_THRESH) {
834 p[crng_init_cnt % CHACHA20_KEY_SIZE] ^= *cp;
835 cp++; crng_init_cnt++; len--;
836 }
837 spin_unlock_irqrestore(&primary_crng.lock, flags);
838 if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
839 invalidate_batched_entropy();
840 crng_init = 1;
841 wake_up_interruptible(&crng_init_wait);
842 pr_notice("random: fast init done\n");
843 }
844 return 1;
845}
846
847/*
848 * crng_slow_load() is called by add_device_randomness, which has two
849 * attributes. (1) We can't trust the buffer passed to it is
850 * guaranteed to be unpredictable (so it might not have any entropy at
851 * all), and (2) it doesn't have the performance constraints of
852 * crng_fast_load().
853 *
854 * So we do something more comprehensive which is guaranteed to touch
855 * all of the primary_crng's state, and which uses a LFSR with a
856 * period of 255 as part of the mixing algorithm. Finally, we do
857 * *not* advance crng_init_cnt since buffer we may get may be something
858 * like a fixed DMI table (for example), which might very well be
859 * unique to the machine, but is otherwise unvarying.
860 */
861static int crng_slow_load(const char *cp, size_t len)
862{
863 unsigned long flags;
864 static unsigned char lfsr = 1;
865 unsigned char tmp;
866 unsigned i, max = CHACHA20_KEY_SIZE;
867 const char * src_buf = cp;
868 char * dest_buf = (char *) &primary_crng.state[4];
869
870 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
871 return 0;
872 if (crng_init != 0) {
873 spin_unlock_irqrestore(&primary_crng.lock, flags);
874 return 0;
875 }
876 if (len > max)
877 max = len;
878
879 for (i = 0; i < max ; i++) {
880 tmp = lfsr;
881 lfsr >>= 1;
882 if (tmp & 1)
883 lfsr ^= 0xE1;
884 tmp = dest_buf[i % CHACHA20_KEY_SIZE];
885 dest_buf[i % CHACHA20_KEY_SIZE] ^= src_buf[i % len] ^ lfsr;
886 lfsr += (tmp << 3) | (tmp >> 5);
887 }
888 spin_unlock_irqrestore(&primary_crng.lock, flags);
889 return 1;
890}
891
892static void crng_reseed(struct crng_state *crng, struct entropy_store *r)
893{
894 unsigned long flags;
895 int i, num;
896 union {
897 __u32 block[CHACHA20_BLOCK_WORDS];
898 __u32 key[8];
899 } buf;
900
901 if (r) {
902 num = extract_entropy(r, &buf, 32, 16, 0);
903 if (num == 0)
904 return;
905 } else {
906 _extract_crng(&primary_crng, buf.block);
907 _crng_backtrack_protect(&primary_crng, buf.block,
908 CHACHA20_KEY_SIZE);
909 }
910 spin_lock_irqsave(&crng->lock, flags);
911 for (i = 0; i < 8; i++) {
912 unsigned long rv;
913 if (!arch_get_random_seed_long(&rv) &&
914 !arch_get_random_long(&rv))
915 rv = random_get_entropy();
916 crng->state[i+4] ^= buf.key[i] ^ rv;
917 }
918 memzero_explicit(&buf, sizeof(buf));
919 crng->init_time = jiffies;
920 spin_unlock_irqrestore(&crng->lock, flags);
921 if (crng == &primary_crng && crng_init < 2) {
922 invalidate_batched_entropy();
923 numa_crng_init();
924 crng_init = 2;
925 process_random_ready_list();
926 wake_up_interruptible(&crng_init_wait);
927 pr_notice("random: crng init done\n");
928 }
929}
930
931static void _extract_crng(struct crng_state *crng,
932 __u32 out[CHACHA20_BLOCK_WORDS])
933{
934 unsigned long v, flags;
935
936 if (crng_ready() &&
937 (time_after(crng_global_init_time, crng->init_time) ||
938 time_after(jiffies, crng->init_time + CRNG_RESEED_INTERVAL)))
939 crng_reseed(crng, crng == &primary_crng ? &input_pool : NULL);
940 spin_lock_irqsave(&crng->lock, flags);
941 if (arch_get_random_long(&v))
942 crng->state[14] ^= v;
943 chacha20_block(&crng->state[0], out);
944 if (crng->state[12] == 0)
945 crng->state[13]++;
946 spin_unlock_irqrestore(&crng->lock, flags);
947}
948
949static void extract_crng(__u32 out[CHACHA20_BLOCK_WORDS])
950{
951 struct crng_state *crng = NULL;
952
953#ifdef CONFIG_NUMA
954 if (crng_node_pool)
955 crng = crng_node_pool[numa_node_id()];
956 if (crng == NULL)
957#endif
958 crng = &primary_crng;
959 _extract_crng(crng, out);
960}
961
962/*
963 * Use the leftover bytes from the CRNG block output (if there is
964 * enough) to mutate the CRNG key to provide backtracking protection.
965 */
966static void _crng_backtrack_protect(struct crng_state *crng,
967 __u32 tmp[CHACHA20_BLOCK_WORDS], int used)
968{
969 unsigned long flags;
970 __u32 *s, *d;
971 int i;
972
973 used = round_up(used, sizeof(__u32));
974 if (used + CHACHA20_KEY_SIZE > CHACHA20_BLOCK_SIZE) {
975 extract_crng(tmp);
976 used = 0;
977 }
978 spin_lock_irqsave(&crng->lock, flags);
979 s = &tmp[used / sizeof(__u32)];
980 d = &crng->state[4];
981 for (i=0; i < 8; i++)
982 *d++ ^= *s++;
983 spin_unlock_irqrestore(&crng->lock, flags);
984}
985
986static void crng_backtrack_protect(__u32 tmp[CHACHA20_BLOCK_WORDS], int used)
987{
988 struct crng_state *crng = NULL;
989
990#ifdef CONFIG_NUMA
991 if (crng_node_pool)
992 crng = crng_node_pool[numa_node_id()];
993 if (crng == NULL)
994#endif
995 crng = &primary_crng;
996 _crng_backtrack_protect(crng, tmp, used);
997}
998
999static ssize_t extract_crng_user(void __user *buf, size_t nbytes)
1000{
1001 ssize_t ret = 0, i = CHACHA20_BLOCK_SIZE;
1002 __u32 tmp[CHACHA20_BLOCK_WORDS];
1003 int large_request = (nbytes > 256);
1004
1005 while (nbytes) {
1006 if (large_request && need_resched()) {
1007 if (signal_pending(current)) {
1008 if (ret == 0)
1009 ret = -ERESTARTSYS;
1010 break;
1011 }
1012 schedule();
1013 }
1014
1015 extract_crng(tmp);
1016 i = min_t(int, nbytes, CHACHA20_BLOCK_SIZE);
1017 if (copy_to_user(buf, tmp, i)) {
1018 ret = -EFAULT;
1019 break;
1020 }
1021
1022 nbytes -= i;
1023 buf += i;
1024 ret += i;
1025 }
1026 crng_backtrack_protect(tmp, i);
1027
1028 /* Wipe data just written to memory */
1029 memzero_explicit(tmp, sizeof(tmp));
1030
1031 return ret;
1032}
1033
1034
1035/*********************************************************************
1036 *
1037 * Entropy input management
1038 *
1039 *********************************************************************/
1040
1041/* There is one of these per entropy source */
1042struct timer_rand_state {
1043 cycles_t last_time;
1044 long last_delta, last_delta2;
1045};
1046
1047#define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
1048
1049/*
1050 * Add device- or boot-specific data to the input pool to help
1051 * initialize it.
1052 *
1053 * None of this adds any entropy; it is meant to avoid the problem of
1054 * the entropy pool having similar initial state across largely
1055 * identical devices.
1056 */
1057void add_device_randomness(const void *buf, unsigned int size)
1058{
1059 unsigned long time = random_get_entropy() ^ jiffies;
1060 unsigned long flags;
1061
1062 if (!crng_ready() && size)
1063 crng_slow_load(buf, size);
1064
1065 trace_add_device_randomness(size, _RET_IP_);
1066 spin_lock_irqsave(&input_pool.lock, flags);
1067 _mix_pool_bytes(&input_pool, buf, size);
1068 _mix_pool_bytes(&input_pool, &time, sizeof(time));
1069 spin_unlock_irqrestore(&input_pool.lock, flags);
1070}
1071EXPORT_SYMBOL(add_device_randomness);
1072
1073static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
1074
1075/*
1076 * This function adds entropy to the entropy "pool" by using timing
1077 * delays. It uses the timer_rand_state structure to make an estimate
1078 * of how many bits of entropy this call has added to the pool.
1079 *
1080 * The number "num" is also added to the pool - it should somehow describe
1081 * the type of event which just happened. This is currently 0-255 for
1082 * keyboard scan codes, and 256 upwards for interrupts.
1083 *
1084 */
1085static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
1086{
1087 struct entropy_store *r;
1088 struct {
1089 long jiffies;
1090 unsigned cycles;
1091 unsigned num;
1092 } sample;
1093 long delta, delta2, delta3;
1094
1095 preempt_disable();
1096
1097 sample.jiffies = jiffies;
1098 sample.cycles = random_get_entropy();
1099 sample.num = num;
1100 r = &input_pool;
1101 mix_pool_bytes(r, &sample, sizeof(sample));
1102
1103 /*
1104 * Calculate number of bits of randomness we probably added.
1105 * We take into account the first, second and third-order deltas
1106 * in order to make our estimate.
1107 */
1108 delta = sample.jiffies - state->last_time;
1109 state->last_time = sample.jiffies;
1110
1111 delta2 = delta - state->last_delta;
1112 state->last_delta = delta;
1113
1114 delta3 = delta2 - state->last_delta2;
1115 state->last_delta2 = delta2;
1116
1117 if (delta < 0)
1118 delta = -delta;
1119 if (delta2 < 0)
1120 delta2 = -delta2;
1121 if (delta3 < 0)
1122 delta3 = -delta3;
1123 if (delta > delta2)
1124 delta = delta2;
1125 if (delta > delta3)
1126 delta = delta3;
1127
1128 /*
1129 * delta is now minimum absolute delta.
1130 * Round down by 1 bit on general principles,
1131 * and limit entropy entimate to 12 bits.
1132 */
1133 credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
1134
1135 preempt_enable();
1136}
1137
1138void add_input_randomness(unsigned int type, unsigned int code,
1139 unsigned int value)
1140{
1141 static unsigned char last_value;
1142
1143 /* ignore autorepeat and the like */
1144 if (value == last_value)
1145 return;
1146
1147 last_value = value;
1148 add_timer_randomness(&input_timer_state,
1149 (type << 4) ^ code ^ (code >> 4) ^ value);
1150 trace_add_input_randomness(ENTROPY_BITS(&input_pool));
1151}
1152EXPORT_SYMBOL_GPL(add_input_randomness);
1153
1154static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
1155
1156#ifdef ADD_INTERRUPT_BENCH
1157static unsigned long avg_cycles, avg_deviation;
1158
1159#define AVG_SHIFT 8 /* Exponential average factor k=1/256 */
1160#define FIXED_1_2 (1 << (AVG_SHIFT-1))
1161
1162static void add_interrupt_bench(cycles_t start)
1163{
1164 long delta = random_get_entropy() - start;
1165
1166 /* Use a weighted moving average */
1167 delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
1168 avg_cycles += delta;
1169 /* And average deviation */
1170 delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
1171 avg_deviation += delta;
1172}
1173#else
1174#define add_interrupt_bench(x)
1175#endif
1176
1177static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
1178{
1179 __u32 *ptr = (__u32 *) regs;
1180 unsigned int idx;
1181
1182 if (regs == NULL)
1183 return 0;
1184 idx = READ_ONCE(f->reg_idx);
1185 if (idx >= sizeof(struct pt_regs) / sizeof(__u32))
1186 idx = 0;
1187 ptr += idx++;
1188 WRITE_ONCE(f->reg_idx, idx);
1189 return *ptr;
1190}
1191
1192void add_interrupt_randomness(int irq, int irq_flags)
1193{
1194 struct entropy_store *r;
1195 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1196 struct pt_regs *regs = get_irq_regs();
1197 unsigned long now = jiffies;
1198 cycles_t cycles = random_get_entropy();
1199 __u32 c_high, j_high;
1200 __u64 ip;
1201 unsigned long seed;
1202 int credit = 0;
1203
1204 if (cycles == 0)
1205 cycles = get_reg(fast_pool, regs);
1206 c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
1207 j_high = (sizeof(now) > 4) ? now >> 32 : 0;
1208 fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
1209 fast_pool->pool[1] ^= now ^ c_high;
1210 ip = regs ? instruction_pointer(regs) : _RET_IP_;
1211 fast_pool->pool[2] ^= ip;
1212 fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
1213 get_reg(fast_pool, regs);
1214
1215 fast_mix(fast_pool);
1216 add_interrupt_bench(cycles);
1217
1218 if (unlikely(crng_init == 0)) {
1219 if ((fast_pool->count >= 64) &&
1220 crng_fast_load((char *) fast_pool->pool,
1221 sizeof(fast_pool->pool))) {
1222 fast_pool->count = 0;
1223 fast_pool->last = now;
1224 }
1225 return;
1226 }
1227
1228 if ((fast_pool->count < 64) &&
1229 !time_after(now, fast_pool->last + HZ))
1230 return;
1231
1232 r = &input_pool;
1233 if (!spin_trylock(&r->lock))
1234 return;
1235
1236 fast_pool->last = now;
1237 __mix_pool_bytes(r, &fast_pool->pool, sizeof(fast_pool->pool));
1238
1239 /*
1240 * If we have architectural seed generator, produce a seed and
1241 * add it to the pool. For the sake of paranoia don't let the
1242 * architectural seed generator dominate the input from the
1243 * interrupt noise.
1244 */
1245 if (arch_get_random_seed_long(&seed)) {
1246 __mix_pool_bytes(r, &seed, sizeof(seed));
1247 credit = 1;
1248 }
1249 spin_unlock(&r->lock);
1250
1251 fast_pool->count = 0;
1252
1253 /* award one bit for the contents of the fast pool */
1254 credit_entropy_bits(r, credit + 1);
1255}
1256EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1257
1258#ifdef CONFIG_BLOCK
1259void add_disk_randomness(struct gendisk *disk)
1260{
1261 if (!disk || !disk->random)
1262 return;
1263 /* first major is 1, so we get >= 0x200 here */
1264 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1265 trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
1266}
1267EXPORT_SYMBOL_GPL(add_disk_randomness);
1268#endif
1269
1270/*********************************************************************
1271 *
1272 * Entropy extraction routines
1273 *
1274 *********************************************************************/
1275
1276/*
1277 * This utility inline function is responsible for transferring entropy
1278 * from the primary pool to the secondary extraction pool. We make
1279 * sure we pull enough for a 'catastrophic reseed'.
1280 */
1281static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes);
1282static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1283{
1284 if (!r->pull ||
1285 r->entropy_count >= (nbytes << (ENTROPY_SHIFT + 3)) ||
1286 r->entropy_count > r->poolinfo->poolfracbits)
1287 return;
1288
1289 _xfer_secondary_pool(r, nbytes);
1290}
1291
1292static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1293{
1294 __u32 tmp[OUTPUT_POOL_WORDS];
1295
1296 int bytes = nbytes;
1297
1298 /* pull at least as much as a wakeup */
1299 bytes = max_t(int, bytes, random_read_wakeup_bits / 8);
1300 /* but never more than the buffer size */
1301 bytes = min_t(int, bytes, sizeof(tmp));
1302
1303 trace_xfer_secondary_pool(r->name, bytes * 8, nbytes * 8,
1304 ENTROPY_BITS(r), ENTROPY_BITS(r->pull));
1305 bytes = extract_entropy(r->pull, tmp, bytes,
1306 random_read_wakeup_bits / 8, 0);
1307 mix_pool_bytes(r, tmp, bytes);
1308 credit_entropy_bits(r, bytes*8);
1309}
1310
1311/*
1312 * Used as a workqueue function so that when the input pool is getting
1313 * full, we can "spill over" some entropy to the output pools. That
1314 * way the output pools can store some of the excess entropy instead
1315 * of letting it go to waste.
1316 */
1317static void push_to_pool(struct work_struct *work)
1318{
1319 struct entropy_store *r = container_of(work, struct entropy_store,
1320 push_work);
1321 BUG_ON(!r);
1322 _xfer_secondary_pool(r, random_read_wakeup_bits/8);
1323 trace_push_to_pool(r->name, r->entropy_count >> ENTROPY_SHIFT,
1324 r->pull->entropy_count >> ENTROPY_SHIFT);
1325}
1326
1327/*
1328 * This function decides how many bytes to actually take from the
1329 * given pool, and also debits the entropy count accordingly.
1330 */
1331static size_t account(struct entropy_store *r, size_t nbytes, int min,
1332 int reserved)
1333{
1334 int entropy_count, orig, have_bytes;
1335 size_t ibytes, nfrac;
1336
1337 BUG_ON(r->entropy_count > r->poolinfo->poolfracbits);
1338
1339 /* Can we pull enough? */
1340retry:
1341 entropy_count = orig = READ_ONCE(r->entropy_count);
1342 ibytes = nbytes;
1343 /* never pull more than available */
1344 have_bytes = entropy_count >> (ENTROPY_SHIFT + 3);
1345
1346 if ((have_bytes -= reserved) < 0)
1347 have_bytes = 0;
1348 ibytes = min_t(size_t, ibytes, have_bytes);
1349 if (ibytes < min)
1350 ibytes = 0;
1351
1352 if (unlikely(entropy_count < 0)) {
1353 pr_warn("random: negative entropy count: pool %s count %d\n",
1354 r->name, entropy_count);
1355 WARN_ON(1);
1356 entropy_count = 0;
1357 }
1358 nfrac = ibytes << (ENTROPY_SHIFT + 3);
1359 if ((size_t) entropy_count > nfrac)
1360 entropy_count -= nfrac;
1361 else
1362 entropy_count = 0;
1363
1364 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
1365 goto retry;
1366
1367 trace_debit_entropy(r->name, 8 * ibytes);
1368 if (ibytes &&
1369 (r->entropy_count >> ENTROPY_SHIFT) < random_write_wakeup_bits) {
1370 wake_up_interruptible(&random_write_wait);
1371 kill_fasync(&fasync, SIGIO, POLL_OUT);
1372 }
1373
1374 return ibytes;
1375}
1376
1377/*
1378 * This function does the actual extraction for extract_entropy and
1379 * extract_entropy_user.
1380 *
1381 * Note: we assume that .poolwords is a multiple of 16 words.
1382 */
1383static void extract_buf(struct entropy_store *r, __u8 *out)
1384{
1385 int i;
1386 union {
1387 __u32 w[5];
1388 unsigned long l[LONGS(20)];
1389 } hash;
1390 __u32 workspace[SHA_WORKSPACE_WORDS];
1391 unsigned long flags;
1392
1393 /*
1394 * If we have an architectural hardware random number
1395 * generator, use it for SHA's initial vector
1396 */
1397 sha_init(hash.w);
1398 for (i = 0; i < LONGS(20); i++) {
1399 unsigned long v;
1400 if (!arch_get_random_long(&v))
1401 break;
1402 hash.l[i] = v;
1403 }
1404
1405 /* Generate a hash across the pool, 16 words (512 bits) at a time */
1406 spin_lock_irqsave(&r->lock, flags);
1407 for (i = 0; i < r->poolinfo->poolwords; i += 16)
1408 sha_transform(hash.w, (__u8 *)(r->pool + i), workspace);
1409
1410 /*
1411 * We mix the hash back into the pool to prevent backtracking
1412 * attacks (where the attacker knows the state of the pool
1413 * plus the current outputs, and attempts to find previous
1414 * ouputs), unless the hash function can be inverted. By
1415 * mixing at least a SHA1 worth of hash data back, we make
1416 * brute-forcing the feedback as hard as brute-forcing the
1417 * hash.
1418 */
1419 __mix_pool_bytes(r, hash.w, sizeof(hash.w));
1420 spin_unlock_irqrestore(&r->lock, flags);
1421
1422 memzero_explicit(workspace, sizeof(workspace));
1423
1424 /*
1425 * In case the hash function has some recognizable output
1426 * pattern, we fold it in half. Thus, we always feed back
1427 * twice as much data as we output.
1428 */
1429 hash.w[0] ^= hash.w[3];
1430 hash.w[1] ^= hash.w[4];
1431 hash.w[2] ^= rol32(hash.w[2], 16);
1432
1433 memcpy(out, &hash, EXTRACT_SIZE);
1434 memzero_explicit(&hash, sizeof(hash));
1435}
1436
1437static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
1438 size_t nbytes, int fips)
1439{
1440 ssize_t ret = 0, i;
1441 __u8 tmp[EXTRACT_SIZE];
1442 unsigned long flags;
1443
1444 while (nbytes) {
1445 extract_buf(r, tmp);
1446
1447 if (fips) {
1448 spin_lock_irqsave(&r->lock, flags);
1449 if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
1450 panic("Hardware RNG duplicated output!\n");
1451 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1452 spin_unlock_irqrestore(&r->lock, flags);
1453 }
1454 i = min_t(int, nbytes, EXTRACT_SIZE);
1455 memcpy(buf, tmp, i);
1456 nbytes -= i;
1457 buf += i;
1458 ret += i;
1459 }
1460
1461 /* Wipe data just returned from memory */
1462 memzero_explicit(tmp, sizeof(tmp));
1463
1464 return ret;
1465}
1466
1467/*
1468 * This function extracts randomness from the "entropy pool", and
1469 * returns it in a buffer.
1470 *
1471 * The min parameter specifies the minimum amount we can pull before
1472 * failing to avoid races that defeat catastrophic reseeding while the
1473 * reserved parameter indicates how much entropy we must leave in the
1474 * pool after each pull to avoid starving other readers.
1475 */
1476static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1477 size_t nbytes, int min, int reserved)
1478{
1479 __u8 tmp[EXTRACT_SIZE];
1480 unsigned long flags;
1481
1482 /* if last_data isn't primed, we need EXTRACT_SIZE extra bytes */
1483 if (fips_enabled) {
1484 spin_lock_irqsave(&r->lock, flags);
1485 if (!r->last_data_init) {
1486 r->last_data_init = 1;
1487 spin_unlock_irqrestore(&r->lock, flags);
1488 trace_extract_entropy(r->name, EXTRACT_SIZE,
1489 ENTROPY_BITS(r), _RET_IP_);
1490 xfer_secondary_pool(r, EXTRACT_SIZE);
1491 extract_buf(r, tmp);
1492 spin_lock_irqsave(&r->lock, flags);
1493 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1494 }
1495 spin_unlock_irqrestore(&r->lock, flags);
1496 }
1497
1498 trace_extract_entropy(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1499 xfer_secondary_pool(r, nbytes);
1500 nbytes = account(r, nbytes, min, reserved);
1501
1502 return _extract_entropy(r, buf, nbytes, fips_enabled);
1503}
1504
1505/*
1506 * This function extracts randomness from the "entropy pool", and
1507 * returns it in a userspace buffer.
1508 */
1509static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
1510 size_t nbytes)
1511{
1512 ssize_t ret = 0, i;
1513 __u8 tmp[EXTRACT_SIZE];
1514 int large_request = (nbytes > 256);
1515
1516 trace_extract_entropy_user(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1517 xfer_secondary_pool(r, nbytes);
1518 nbytes = account(r, nbytes, 0, 0);
1519
1520 while (nbytes) {
1521 if (large_request && need_resched()) {
1522 if (signal_pending(current)) {
1523 if (ret == 0)
1524 ret = -ERESTARTSYS;
1525 break;
1526 }
1527 schedule();
1528 }
1529
1530 extract_buf(r, tmp);
1531 i = min_t(int, nbytes, EXTRACT_SIZE);
1532 if (copy_to_user(buf, tmp, i)) {
1533 ret = -EFAULT;
1534 break;
1535 }
1536
1537 nbytes -= i;
1538 buf += i;
1539 ret += i;
1540 }
1541
1542 /* Wipe data just returned from memory */
1543 memzero_explicit(tmp, sizeof(tmp));
1544
1545 return ret;
1546}
1547
1548#define warn_unseeded_randomness(previous) \
1549 _warn_unseeded_randomness(__func__, (void *) _RET_IP_, (previous))
1550
1551static void _warn_unseeded_randomness(const char *func_name, void *caller,
1552 void **previous)
1553{
1554#ifdef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1555 const bool print_once = false;
1556#else
1557 static bool print_once __read_mostly;
1558#endif
1559
1560 if (print_once ||
1561 crng_ready() ||
1562 (previous && (caller == READ_ONCE(*previous))))
1563 return;
1564 WRITE_ONCE(*previous, caller);
1565#ifndef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1566 print_once = true;
1567#endif
1568 pr_notice("random: %s called from %pS with crng_init=%d\n",
1569 func_name, caller, crng_init);
1570}
1571
1572/*
1573 * This function is the exported kernel interface. It returns some
1574 * number of good random numbers, suitable for key generation, seeding
1575 * TCP sequence numbers, etc. It does not rely on the hardware random
1576 * number generator. For random bytes direct from the hardware RNG
1577 * (when available), use get_random_bytes_arch(). In order to ensure
1578 * that the randomness provided by this function is okay, the function
1579 * wait_for_random_bytes() should be called and return 0 at least once
1580 * at any point prior.
1581 */
1582static void _get_random_bytes(void *buf, int nbytes)
1583{
1584 __u32 tmp[CHACHA20_BLOCK_WORDS];
1585
1586 trace_get_random_bytes(nbytes, _RET_IP_);
1587
1588 while (nbytes >= CHACHA20_BLOCK_SIZE) {
1589 extract_crng(buf);
1590 buf += CHACHA20_BLOCK_SIZE;
1591 nbytes -= CHACHA20_BLOCK_SIZE;
1592 }
1593
1594 if (nbytes > 0) {
1595 extract_crng(tmp);
1596 memcpy(buf, tmp, nbytes);
1597 crng_backtrack_protect(tmp, nbytes);
1598 } else
1599 crng_backtrack_protect(tmp, CHACHA20_BLOCK_SIZE);
1600 memzero_explicit(tmp, sizeof(tmp));
1601}
1602
1603void get_random_bytes(void *buf, int nbytes)
1604{
1605 static void *previous;
1606
1607 warn_unseeded_randomness(&previous);
1608 _get_random_bytes(buf, nbytes);
1609}
1610EXPORT_SYMBOL(get_random_bytes);
1611
1612/*
1613 * Wait for the urandom pool to be seeded and thus guaranteed to supply
1614 * cryptographically secure random numbers. This applies to: the /dev/urandom
1615 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
1616 * family of functions. Using any of these functions without first calling
1617 * this function forfeits the guarantee of security.
1618 *
1619 * Returns: 0 if the urandom pool has been seeded.
1620 * -ERESTARTSYS if the function was interrupted by a signal.
1621 */
1622int wait_for_random_bytes(void)
1623{
1624 if (likely(crng_ready()))
1625 return 0;
1626 return wait_event_interruptible(crng_init_wait, crng_ready());
1627}
1628EXPORT_SYMBOL(wait_for_random_bytes);
1629
1630/*
1631 * Add a callback function that will be invoked when the nonblocking
1632 * pool is initialised.
1633 *
1634 * returns: 0 if callback is successfully added
1635 * -EALREADY if pool is already initialised (callback not called)
1636 * -ENOENT if module for callback is not alive
1637 */
1638int add_random_ready_callback(struct random_ready_callback *rdy)
1639{
1640 struct module *owner;
1641 unsigned long flags;
1642 int err = -EALREADY;
1643
1644 if (crng_ready())
1645 return err;
1646
1647 owner = rdy->owner;
1648 if (!try_module_get(owner))
1649 return -ENOENT;
1650
1651 spin_lock_irqsave(&random_ready_list_lock, flags);
1652 if (crng_ready())
1653 goto out;
1654
1655 owner = NULL;
1656
1657 list_add(&rdy->list, &random_ready_list);
1658 err = 0;
1659
1660out:
1661 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1662
1663 module_put(owner);
1664
1665 return err;
1666}
1667EXPORT_SYMBOL(add_random_ready_callback);
1668
1669/*
1670 * Delete a previously registered readiness callback function.
1671 */
1672void del_random_ready_callback(struct random_ready_callback *rdy)
1673{
1674 unsigned long flags;
1675 struct module *owner = NULL;
1676
1677 spin_lock_irqsave(&random_ready_list_lock, flags);
1678 if (!list_empty(&rdy->list)) {
1679 list_del_init(&rdy->list);
1680 owner = rdy->owner;
1681 }
1682 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1683
1684 module_put(owner);
1685}
1686EXPORT_SYMBOL(del_random_ready_callback);
1687
1688/*
1689 * This function will use the architecture-specific hardware random
1690 * number generator if it is available. The arch-specific hw RNG will
1691 * almost certainly be faster than what we can do in software, but it
1692 * is impossible to verify that it is implemented securely (as
1693 * opposed, to, say, the AES encryption of a sequence number using a
1694 * key known by the NSA). So it's useful if we need the speed, but
1695 * only if we're willing to trust the hardware manufacturer not to
1696 * have put in a back door.
1697 */
1698void get_random_bytes_arch(void *buf, int nbytes)
1699{
1700 char *p = buf;
1701
1702 trace_get_random_bytes_arch(nbytes, _RET_IP_);
1703 while (nbytes) {
1704 unsigned long v;
1705 int chunk = min(nbytes, (int)sizeof(unsigned long));
1706
1707 if (!arch_get_random_long(&v))
1708 break;
1709
1710 memcpy(p, &v, chunk);
1711 p += chunk;
1712 nbytes -= chunk;
1713 }
1714
1715 if (nbytes)
1716 get_random_bytes(p, nbytes);
1717}
1718EXPORT_SYMBOL(get_random_bytes_arch);
1719
1720
1721/*
1722 * init_std_data - initialize pool with system data
1723 *
1724 * @r: pool to initialize
1725 *
1726 * This function clears the pool's entropy count and mixes some system
1727 * data into the pool to prepare it for use. The pool is not cleared
1728 * as that can only decrease the entropy in the pool.
1729 */
1730static void init_std_data(struct entropy_store *r)
1731{
1732 int i;
1733 ktime_t now = ktime_get_real();
1734 unsigned long rv;
1735
1736 r->last_pulled = jiffies;
1737 mix_pool_bytes(r, &now, sizeof(now));
1738 for (i = r->poolinfo->poolbytes; i > 0; i -= sizeof(rv)) {
1739 if (!arch_get_random_seed_long(&rv) &&
1740 !arch_get_random_long(&rv))
1741 rv = random_get_entropy();
1742 mix_pool_bytes(r, &rv, sizeof(rv));
1743 }
1744 mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1745}
1746
1747/*
1748 * Note that setup_arch() may call add_device_randomness()
1749 * long before we get here. This allows seeding of the pools
1750 * with some platform dependent data very early in the boot
1751 * process. But it limits our options here. We must use
1752 * statically allocated structures that already have all
1753 * initializations complete at compile time. We should also
1754 * take care not to overwrite the precious per platform data
1755 * we were given.
1756 */
1757static int rand_initialize(void)
1758{
1759 init_std_data(&input_pool);
1760 init_std_data(&blocking_pool);
1761 crng_initialize(&primary_crng);
1762 crng_global_init_time = jiffies;
1763 return 0;
1764}
1765early_initcall(rand_initialize);
1766
1767#ifdef CONFIG_BLOCK
1768void rand_initialize_disk(struct gendisk *disk)
1769{
1770 struct timer_rand_state *state;
1771
1772 /*
1773 * If kzalloc returns null, we just won't use that entropy
1774 * source.
1775 */
1776 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1777 if (state) {
1778 state->last_time = INITIAL_JIFFIES;
1779 disk->random = state;
1780 }
1781}
1782#endif
1783
1784static ssize_t
1785_random_read(int nonblock, char __user *buf, size_t nbytes)
1786{
1787 ssize_t n;
1788
1789 if (nbytes == 0)
1790 return 0;
1791
1792 nbytes = min_t(size_t, nbytes, SEC_XFER_SIZE);
1793 while (1) {
1794 n = extract_entropy_user(&blocking_pool, buf, nbytes);
1795 if (n < 0)
1796 return n;
1797 trace_random_read(n*8, (nbytes-n)*8,
1798 ENTROPY_BITS(&blocking_pool),
1799 ENTROPY_BITS(&input_pool));
1800 if (n > 0)
1801 return n;
1802
1803 /* Pool is (near) empty. Maybe wait and retry. */
1804 if (nonblock)
1805 return -EAGAIN;
1806
1807 wait_event_interruptible(random_read_wait,
1808 ENTROPY_BITS(&input_pool) >=
1809 random_read_wakeup_bits);
1810 if (signal_pending(current))
1811 return -ERESTARTSYS;
1812 }
1813}
1814
1815static ssize_t
1816random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1817{
1818 return _random_read(file->f_flags & O_NONBLOCK, buf, nbytes);
1819}
1820
1821static ssize_t
1822urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1823{
1824 unsigned long flags;
1825 static int maxwarn = 10;
1826 int ret;
1827
1828 if (!crng_ready() && maxwarn > 0) {
1829 maxwarn--;
1830 printk(KERN_NOTICE "random: %s: uninitialized urandom read "
1831 "(%zd bytes read)\n",
1832 current->comm, nbytes);
1833 spin_lock_irqsave(&primary_crng.lock, flags);
1834 crng_init_cnt = 0;
1835 spin_unlock_irqrestore(&primary_crng.lock, flags);
1836 }
1837 nbytes = min_t(size_t, nbytes, INT_MAX >> (ENTROPY_SHIFT + 3));
1838 ret = extract_crng_user(buf, nbytes);
1839 trace_urandom_read(8 * nbytes, 0, ENTROPY_BITS(&input_pool));
1840 return ret;
1841}
1842
1843static __poll_t
1844random_poll(struct file *file, poll_table * wait)
1845{
1846 __poll_t mask;
1847
1848 poll_wait(file, &random_read_wait, wait);
1849 poll_wait(file, &random_write_wait, wait);
1850 mask = 0;
1851 if (ENTROPY_BITS(&input_pool) >= random_read_wakeup_bits)
1852 mask |= EPOLLIN | EPOLLRDNORM;
1853 if (ENTROPY_BITS(&input_pool) < random_write_wakeup_bits)
1854 mask |= EPOLLOUT | EPOLLWRNORM;
1855 return mask;
1856}
1857
1858static int
1859write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1860{
1861 size_t bytes;
1862 __u32 buf[16];
1863 const char __user *p = buffer;
1864
1865 while (count > 0) {
1866 bytes = min(count, sizeof(buf));
1867 if (copy_from_user(&buf, p, bytes))
1868 return -EFAULT;
1869
1870 count -= bytes;
1871 p += bytes;
1872
1873 mix_pool_bytes(r, buf, bytes);
1874 cond_resched();
1875 }
1876
1877 return 0;
1878}
1879
1880static ssize_t random_write(struct file *file, const char __user *buffer,
1881 size_t count, loff_t *ppos)
1882{
1883 size_t ret;
1884
1885 ret = write_pool(&input_pool, buffer, count);
1886 if (ret)
1887 return ret;
1888
1889 return (ssize_t)count;
1890}
1891
1892static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1893{
1894 int size, ent_count;
1895 int __user *p = (int __user *)arg;
1896 int retval;
1897
1898 switch (cmd) {
1899 case RNDGETENTCNT:
1900 /* inherently racy, no point locking */
1901 ent_count = ENTROPY_BITS(&input_pool);
1902 if (put_user(ent_count, p))
1903 return -EFAULT;
1904 return 0;
1905 case RNDADDTOENTCNT:
1906 if (!capable(CAP_SYS_ADMIN))
1907 return -EPERM;
1908 if (get_user(ent_count, p))
1909 return -EFAULT;
1910 return credit_entropy_bits_safe(&input_pool, ent_count);
1911 case RNDADDENTROPY:
1912 if (!capable(CAP_SYS_ADMIN))
1913 return -EPERM;
1914 if (get_user(ent_count, p++))
1915 return -EFAULT;
1916 if (ent_count < 0)
1917 return -EINVAL;
1918 if (get_user(size, p++))
1919 return -EFAULT;
1920 retval = write_pool(&input_pool, (const char __user *)p,
1921 size);
1922 if (retval < 0)
1923 return retval;
1924 return credit_entropy_bits_safe(&input_pool, ent_count);
1925 case RNDZAPENTCNT:
1926 case RNDCLEARPOOL:
1927 /*
1928 * Clear the entropy pool counters. We no longer clear
1929 * the entropy pool, as that's silly.
1930 */
1931 if (!capable(CAP_SYS_ADMIN))
1932 return -EPERM;
1933 input_pool.entropy_count = 0;
1934 blocking_pool.entropy_count = 0;
1935 return 0;
1936 case RNDRESEEDCRNG:
1937 if (!capable(CAP_SYS_ADMIN))
1938 return -EPERM;
1939 if (crng_init < 2)
1940 return -ENODATA;
1941 crng_reseed(&primary_crng, NULL);
1942 crng_global_init_time = jiffies - 1;
1943 return 0;
1944 default:
1945 return -EINVAL;
1946 }
1947}
1948
1949static int random_fasync(int fd, struct file *filp, int on)
1950{
1951 return fasync_helper(fd, filp, on, &fasync);
1952}
1953
1954const struct file_operations random_fops = {
1955 .read = random_read,
1956 .write = random_write,
1957 .poll = random_poll,
1958 .unlocked_ioctl = random_ioctl,
1959 .fasync = random_fasync,
1960 .llseek = noop_llseek,
1961};
1962
1963const struct file_operations urandom_fops = {
1964 .read = urandom_read,
1965 .write = random_write,
1966 .unlocked_ioctl = random_ioctl,
1967 .fasync = random_fasync,
1968 .llseek = noop_llseek,
1969};
1970
1971SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count,
1972 unsigned int, flags)
1973{
1974 int ret;
1975
1976 if (flags & ~(GRND_NONBLOCK|GRND_RANDOM))
1977 return -EINVAL;
1978
1979 if (count > INT_MAX)
1980 count = INT_MAX;
1981
1982 if (flags & GRND_RANDOM)
1983 return _random_read(flags & GRND_NONBLOCK, buf, count);
1984
1985 if (!crng_ready()) {
1986 if (flags & GRND_NONBLOCK)
1987 return -EAGAIN;
1988 ret = wait_for_random_bytes();
1989 if (unlikely(ret))
1990 return ret;
1991 }
1992 return urandom_read(NULL, buf, count, NULL);
1993}
1994
1995/********************************************************************
1996 *
1997 * Sysctl interface
1998 *
1999 ********************************************************************/
2000
2001#ifdef CONFIG_SYSCTL
2002
2003#include <linux/sysctl.h>
2004
2005static int min_read_thresh = 8, min_write_thresh;
2006static int max_read_thresh = OUTPUT_POOL_WORDS * 32;
2007static int max_write_thresh = INPUT_POOL_WORDS * 32;
2008static int random_min_urandom_seed = 60;
2009static char sysctl_bootid[16];
2010
2011/*
2012 * This function is used to return both the bootid UUID, and random
2013 * UUID. The difference is in whether table->data is NULL; if it is,
2014 * then a new UUID is generated and returned to the user.
2015 *
2016 * If the user accesses this via the proc interface, the UUID will be
2017 * returned as an ASCII string in the standard UUID format; if via the
2018 * sysctl system call, as 16 bytes of binary data.
2019 */
2020static int proc_do_uuid(struct ctl_table *table, int write,
2021 void __user *buffer, size_t *lenp, loff_t *ppos)
2022{
2023 struct ctl_table fake_table;
2024 unsigned char buf[64], tmp_uuid[16], *uuid;
2025
2026 uuid = table->data;
2027 if (!uuid) {
2028 uuid = tmp_uuid;
2029 generate_random_uuid(uuid);
2030 } else {
2031 static DEFINE_SPINLOCK(bootid_spinlock);
2032
2033 spin_lock(&bootid_spinlock);
2034 if (!uuid[8])
2035 generate_random_uuid(uuid);
2036 spin_unlock(&bootid_spinlock);
2037 }
2038
2039 sprintf(buf, "%pU", uuid);
2040
2041 fake_table.data = buf;
2042 fake_table.maxlen = sizeof(buf);
2043
2044 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
2045}
2046
2047/*
2048 * Return entropy available scaled to integral bits
2049 */
2050static int proc_do_entropy(struct ctl_table *table, int write,
2051 void __user *buffer, size_t *lenp, loff_t *ppos)
2052{
2053 struct ctl_table fake_table;
2054 int entropy_count;
2055
2056 entropy_count = *(int *)table->data >> ENTROPY_SHIFT;
2057
2058 fake_table.data = &entropy_count;
2059 fake_table.maxlen = sizeof(entropy_count);
2060
2061 return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
2062}
2063
2064static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
2065extern struct ctl_table random_table[];
2066struct ctl_table random_table[] = {
2067 {
2068 .procname = "poolsize",
2069 .data = &sysctl_poolsize,
2070 .maxlen = sizeof(int),
2071 .mode = 0444,
2072 .proc_handler = proc_dointvec,
2073 },
2074 {
2075 .procname = "entropy_avail",
2076 .maxlen = sizeof(int),
2077 .mode = 0444,
2078 .proc_handler = proc_do_entropy,
2079 .data = &input_pool.entropy_count,
2080 },
2081 {
2082 .procname = "read_wakeup_threshold",
2083 .data = &random_read_wakeup_bits,
2084 .maxlen = sizeof(int),
2085 .mode = 0644,
2086 .proc_handler = proc_dointvec_minmax,
2087 .extra1 = &min_read_thresh,
2088 .extra2 = &max_read_thresh,
2089 },
2090 {
2091 .procname = "write_wakeup_threshold",
2092 .data = &random_write_wakeup_bits,
2093 .maxlen = sizeof(int),
2094 .mode = 0644,
2095 .proc_handler = proc_dointvec_minmax,
2096 .extra1 = &min_write_thresh,
2097 .extra2 = &max_write_thresh,
2098 },
2099 {
2100 .procname = "urandom_min_reseed_secs",
2101 .data = &random_min_urandom_seed,
2102 .maxlen = sizeof(int),
2103 .mode = 0644,
2104 .proc_handler = proc_dointvec,
2105 },
2106 {
2107 .procname = "boot_id",
2108 .data = &sysctl_bootid,
2109 .maxlen = 16,
2110 .mode = 0444,
2111 .proc_handler = proc_do_uuid,
2112 },
2113 {
2114 .procname = "uuid",
2115 .maxlen = 16,
2116 .mode = 0444,
2117 .proc_handler = proc_do_uuid,
2118 },
2119#ifdef ADD_INTERRUPT_BENCH
2120 {
2121 .procname = "add_interrupt_avg_cycles",
2122 .data = &avg_cycles,
2123 .maxlen = sizeof(avg_cycles),
2124 .mode = 0444,
2125 .proc_handler = proc_doulongvec_minmax,
2126 },
2127 {
2128 .procname = "add_interrupt_avg_deviation",
2129 .data = &avg_deviation,
2130 .maxlen = sizeof(avg_deviation),
2131 .mode = 0444,
2132 .proc_handler = proc_doulongvec_minmax,
2133 },
2134#endif
2135 { }
2136};
2137#endif /* CONFIG_SYSCTL */
2138
2139struct batched_entropy {
2140 union {
2141 u64 entropy_u64[CHACHA20_BLOCK_SIZE / sizeof(u64)];
2142 u32 entropy_u32[CHACHA20_BLOCK_SIZE / sizeof(u32)];
2143 };
2144 unsigned int position;
2145};
2146static rwlock_t batched_entropy_reset_lock = __RW_LOCK_UNLOCKED(batched_entropy_reset_lock);
2147
2148/*
2149 * Get a random word for internal kernel use only. The quality of the random
2150 * number is either as good as RDRAND or as good as /dev/urandom, with the
2151 * goal of being quite fast and not depleting entropy. In order to ensure
2152 * that the randomness provided by this function is okay, the function
2153 * wait_for_random_bytes() should be called and return 0 at least once
2154 * at any point prior.
2155 */
2156static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64);
2157u64 get_random_u64(void)
2158{
2159 u64 ret;
2160 bool use_lock;
2161 unsigned long flags = 0;
2162 struct batched_entropy *batch;
2163 static void *previous;
2164
2165#if BITS_PER_LONG == 64
2166 if (arch_get_random_long((unsigned long *)&ret))
2167 return ret;
2168#else
2169 if (arch_get_random_long((unsigned long *)&ret) &&
2170 arch_get_random_long((unsigned long *)&ret + 1))
2171 return ret;
2172#endif
2173
2174 warn_unseeded_randomness(&previous);
2175
2176 use_lock = READ_ONCE(crng_init) < 2;
2177 batch = &get_cpu_var(batched_entropy_u64);
2178 if (use_lock)
2179 read_lock_irqsave(&batched_entropy_reset_lock, flags);
2180 if (batch->position % ARRAY_SIZE(batch->entropy_u64) == 0) {
2181 extract_crng((__u32 *)batch->entropy_u64);
2182 batch->position = 0;
2183 }
2184 ret = batch->entropy_u64[batch->position++];
2185 if (use_lock)
2186 read_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2187 put_cpu_var(batched_entropy_u64);
2188 return ret;
2189}
2190EXPORT_SYMBOL(get_random_u64);
2191
2192static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32);
2193u32 get_random_u32(void)
2194{
2195 u32 ret;
2196 bool use_lock;
2197 unsigned long flags = 0;
2198 struct batched_entropy *batch;
2199 static void *previous;
2200
2201 if (arch_get_random_int(&ret))
2202 return ret;
2203
2204 warn_unseeded_randomness(&previous);
2205
2206 use_lock = READ_ONCE(crng_init) < 2;
2207 batch = &get_cpu_var(batched_entropy_u32);
2208 if (use_lock)
2209 read_lock_irqsave(&batched_entropy_reset_lock, flags);
2210 if (batch->position % ARRAY_SIZE(batch->entropy_u32) == 0) {
2211 extract_crng(batch->entropy_u32);
2212 batch->position = 0;
2213 }
2214 ret = batch->entropy_u32[batch->position++];
2215 if (use_lock)
2216 read_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2217 put_cpu_var(batched_entropy_u32);
2218 return ret;
2219}
2220EXPORT_SYMBOL(get_random_u32);
2221
2222/* It's important to invalidate all potential batched entropy that might
2223 * be stored before the crng is initialized, which we can do lazily by
2224 * simply resetting the counter to zero so that it's re-extracted on the
2225 * next usage. */
2226static void invalidate_batched_entropy(void)
2227{
2228 int cpu;
2229 unsigned long flags;
2230
2231 write_lock_irqsave(&batched_entropy_reset_lock, flags);
2232 for_each_possible_cpu (cpu) {
2233 per_cpu_ptr(&batched_entropy_u32, cpu)->position = 0;
2234 per_cpu_ptr(&batched_entropy_u64, cpu)->position = 0;
2235 }
2236 write_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2237}
2238
2239/**
2240 * randomize_page - Generate a random, page aligned address
2241 * @start: The smallest acceptable address the caller will take.
2242 * @range: The size of the area, starting at @start, within which the
2243 * random address must fall.
2244 *
2245 * If @start + @range would overflow, @range is capped.
2246 *
2247 * NOTE: Historical use of randomize_range, which this replaces, presumed that
2248 * @start was already page aligned. We now align it regardless.
2249 *
2250 * Return: A page aligned address within [start, start + range). On error,
2251 * @start is returned.
2252 */
2253unsigned long
2254randomize_page(unsigned long start, unsigned long range)
2255{
2256 if (!PAGE_ALIGNED(start)) {
2257 range -= PAGE_ALIGN(start) - start;
2258 start = PAGE_ALIGN(start);
2259 }
2260
2261 if (start > ULONG_MAX - range)
2262 range = ULONG_MAX - start;
2263
2264 range >>= PAGE_SHIFT;
2265
2266 if (range == 0)
2267 return start;
2268
2269 return start + (get_random_long() % range << PAGE_SHIFT);
2270}
2271
2272/* Interface for in-kernel drivers of true hardware RNGs.
2273 * Those devices may produce endless random bits and will be throttled
2274 * when our pool is full.
2275 */
2276void add_hwgenerator_randomness(const char *buffer, size_t count,
2277 size_t entropy)
2278{
2279 struct entropy_store *poolp = &input_pool;
2280
2281 if (unlikely(crng_init == 0)) {
2282 crng_fast_load(buffer, count);
2283 return;
2284 }
2285
2286 /* Suspend writing if we're above the trickle threshold.
2287 * We'll be woken up again once below random_write_wakeup_thresh,
2288 * or when the calling thread is about to terminate.
2289 */
2290 wait_event_interruptible(random_write_wait, kthread_should_stop() ||
2291 ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
2292 mix_pool_bytes(poolp, buffer, count);
2293 credit_entropy_bits(poolp, entropy);
2294}
2295EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);