blob: b202f66fc383b4320f87ab726e671b0ba08da60f [file] [log] [blame]
rjw1f884582022-01-06 17:20:42 +08001/*
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/ratelimit.h>
265#include <linux/syscalls.h>
266#include <linux/completion.h>
267#include <linux/uuid.h>
268#include <crypto/chacha20.h>
269
270#include <asm/processor.h>
271#include <linux/uaccess.h>
272#include <asm/irq.h>
273#include <asm/irq_regs.h>
274#include <asm/io.h>
275
276#define CREATE_TRACE_POINTS
277#include <trace/events/random.h>
278
279/* #define ADD_INTERRUPT_BENCH */
280
281/*
282 * Configuration information
283 */
284#define INPUT_POOL_SHIFT 12
285#define INPUT_POOL_WORDS (1 << (INPUT_POOL_SHIFT-5))
286#define OUTPUT_POOL_SHIFT 10
287#define OUTPUT_POOL_WORDS (1 << (OUTPUT_POOL_SHIFT-5))
288#define SEC_XFER_SIZE 512
289#define EXTRACT_SIZE 10
290
291
292#define LONGS(x) (((x) + sizeof(unsigned long) - 1)/sizeof(unsigned long))
293
294/*
295 * To allow fractional bits to be tracked, the entropy_count field is
296 * denominated in units of 1/8th bits.
297 *
298 * 2*(ENTROPY_SHIFT + log2(poolbits)) must <= 31, or the multiply in
299 * credit_entropy_bits() needs to be 64 bits wide.
300 */
301#define ENTROPY_SHIFT 3
302#define ENTROPY_BITS(r) ((r)->entropy_count >> ENTROPY_SHIFT)
303
304/*
305 * The minimum number of bits of entropy before we wake up a read on
306 * /dev/random. Should be enough to do a significant reseed.
307 */
308static int random_read_wakeup_bits = 64;
309
310/*
311 * If the entropy count falls under this number of bits, then we
312 * should wake up processes which are selecting or polling on write
313 * access to /dev/random.
314 */
315static int random_write_wakeup_bits = 28 * OUTPUT_POOL_WORDS;
316
317/*
318 * Originally, we used a primitive polynomial of degree .poolwords
319 * over GF(2). The taps for various sizes are defined below. They
320 * were chosen to be evenly spaced except for the last tap, which is 1
321 * to get the twisting happening as fast as possible.
322 *
323 * For the purposes of better mixing, we use the CRC-32 polynomial as
324 * well to make a (modified) twisted Generalized Feedback Shift
325 * Register. (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR
326 * generators. ACM Transactions on Modeling and Computer Simulation
327 * 2(3):179-194. Also see M. Matsumoto & Y. Kurita, 1994. Twisted
328 * GFSR generators II. ACM Transactions on Modeling and Computer
329 * Simulation 4:254-266)
330 *
331 * Thanks to Colin Plumb for suggesting this.
332 *
333 * The mixing operation is much less sensitive than the output hash,
334 * where we use SHA-1. All that we want of mixing operation is that
335 * it be a good non-cryptographic hash; i.e. it not produce collisions
336 * when fed "random" data of the sort we expect to see. As long as
337 * the pool state differs for different inputs, we have preserved the
338 * input entropy and done a good job. The fact that an intelligent
339 * attacker can construct inputs that will produce controlled
340 * alterations to the pool's state is not important because we don't
341 * consider such inputs to contribute any randomness. The only
342 * property we need with respect to them is that the attacker can't
343 * increase his/her knowledge of the pool's state. Since all
344 * additions are reversible (knowing the final state and the input,
345 * you can reconstruct the initial state), if an attacker has any
346 * uncertainty about the initial state, he/she can only shuffle that
347 * uncertainty about, but never cause any collisions (which would
348 * decrease the uncertainty).
349 *
350 * Our mixing functions were analyzed by Lacharme, Roeck, Strubel, and
351 * Videau in their paper, "The Linux Pseudorandom Number Generator
352 * Revisited" (see: http://eprint.iacr.org/2012/251.pdf). In their
353 * paper, they point out that we are not using a true Twisted GFSR,
354 * since Matsumoto & Kurita used a trinomial feedback polynomial (that
355 * is, with only three taps, instead of the six that we are using).
356 * As a result, the resulting polynomial is neither primitive nor
357 * irreducible, and hence does not have a maximal period over
358 * GF(2**32). They suggest a slight change to the generator
359 * polynomial which improves the resulting TGFSR polynomial to be
360 * irreducible, which we have made here.
361 */
362static struct poolinfo {
363 int poolbitshift, poolwords, poolbytes, poolbits, poolfracbits;
364#define S(x) ilog2(x)+5, (x), (x)*4, (x)*32, (x) << (ENTROPY_SHIFT+5)
365 int tap1, tap2, tap3, tap4, tap5;
366} poolinfo_table[] = {
367 /* was: x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 */
368 /* x^128 + x^104 + x^76 + x^51 +x^25 + x + 1 */
369 { S(128), 104, 76, 51, 25, 1 },
370 /* was: x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 */
371 /* x^32 + x^26 + x^19 + x^14 + x^7 + x + 1 */
372 { S(32), 26, 19, 14, 7, 1 },
373#if 0
374 /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1 -- 115 */
375 { S(2048), 1638, 1231, 819, 411, 1 },
376
377 /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
378 { S(1024), 817, 615, 412, 204, 1 },
379
380 /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
381 { S(1024), 819, 616, 410, 207, 2 },
382
383 /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
384 { S(512), 411, 308, 208, 104, 1 },
385
386 /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
387 { S(512), 409, 307, 206, 102, 2 },
388 /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
389 { S(512), 409, 309, 205, 103, 2 },
390
391 /* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
392 { S(256), 205, 155, 101, 52, 1 },
393
394 /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
395 { S(128), 103, 78, 51, 27, 2 },
396
397 /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
398 { S(64), 52, 39, 26, 14, 1 },
399#endif
400};
401
402/*
403 * Static global variables
404 */
405static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
406static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
407static struct fasync_struct *fasync;
408
409static DEFINE_SPINLOCK(random_ready_list_lock);
410static LIST_HEAD(random_ready_list);
411
412struct crng_state {
413 __u32 state[16];
414 unsigned long init_time;
415 spinlock_t lock;
416};
417
418struct crng_state primary_crng = {
419 .lock = __SPIN_LOCK_UNLOCKED(primary_crng.lock),
420};
421
422/*
423 * crng_init = 0 --> Uninitialized
424 * 1 --> Initialized
425 * 2 --> Initialized from input_pool
426 *
427 * crng_init is protected by primary_crng->lock, and only increases
428 * its value (from 0->1->2).
429 */
430static int crng_init = 0;
431#define crng_ready() (likely(crng_init > 1))
432static int crng_init_cnt = 0;
433static unsigned long crng_global_init_time = 0;
434#define CRNG_INIT_CNT_THRESH (2*CHACHA20_KEY_SIZE)
435static void _extract_crng(struct crng_state *crng,
436 __u8 out[CHACHA20_BLOCK_SIZE]);
437static void _crng_backtrack_protect(struct crng_state *crng,
438 __u8 tmp[CHACHA20_BLOCK_SIZE], int used);
439static void process_random_ready_list(void);
440static void _get_random_bytes(void *buf, int nbytes);
441
442static struct ratelimit_state unseeded_warning =
443 RATELIMIT_STATE_INIT("warn_unseeded_randomness", HZ, 3);
444static struct ratelimit_state urandom_warning =
445 RATELIMIT_STATE_INIT("warn_urandom_randomness", HZ, 3);
446
447static int ratelimit_disable __read_mostly;
448
449module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
450MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
451
452/**********************************************************************
453 *
454 * OS independent entropy store. Here are the functions which handle
455 * storing entropy in an entropy pool.
456 *
457 **********************************************************************/
458
459struct entropy_store;
460struct entropy_store {
461 /* read-only data: */
462 const struct poolinfo *poolinfo;
463 __u32 *pool;
464 const char *name;
465 struct entropy_store *pull;
466 struct work_struct push_work;
467
468 /* read-write data: */
469 unsigned long last_pulled;
470 spinlock_t lock;
471 unsigned short add_ptr;
472 unsigned short input_rotate;
473 int entropy_count;
474 int entropy_total;
475 unsigned int initialized:1;
476 unsigned int last_data_init:1;
477 __u8 last_data[EXTRACT_SIZE];
478};
479
480static ssize_t extract_entropy(struct entropy_store *r, void *buf,
481 size_t nbytes, int min, int rsvd);
482static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
483 size_t nbytes, int fips);
484
485static void crng_reseed(struct crng_state *crng, struct entropy_store *r);
486static void push_to_pool(struct work_struct *work);
487static __u32 input_pool_data[INPUT_POOL_WORDS] __latent_entropy;
488static __u32 blocking_pool_data[OUTPUT_POOL_WORDS] __latent_entropy;
489
490static struct entropy_store input_pool = {
491 .poolinfo = &poolinfo_table[0],
492 .name = "input",
493 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
494 .pool = input_pool_data
495};
496
497static struct entropy_store blocking_pool = {
498 .poolinfo = &poolinfo_table[1],
499 .name = "blocking",
500 .pull = &input_pool,
501 .lock = __SPIN_LOCK_UNLOCKED(blocking_pool.lock),
502 .pool = blocking_pool_data,
503 .push_work = __WORK_INITIALIZER(blocking_pool.push_work,
504 push_to_pool),
505};
506
507static __u32 const twist_table[8] = {
508 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
509 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
510
511/*
512 * This function adds bytes into the entropy "pool". It does not
513 * update the entropy estimate. The caller should call
514 * credit_entropy_bits if this is appropriate.
515 *
516 * The pool is stirred with a primitive polynomial of the appropriate
517 * degree, and then twisted. We twist by three bits at a time because
518 * it's cheap to do so and helps slightly in the expected case where
519 * the entropy is concentrated in the low-order bits.
520 */
521static void _mix_pool_bytes(struct entropy_store *r, const void *in,
522 int nbytes)
523{
524 unsigned long i, tap1, tap2, tap3, tap4, tap5;
525 int input_rotate;
526 int wordmask = r->poolinfo->poolwords - 1;
527 const char *bytes = in;
528 __u32 w;
529
530 tap1 = r->poolinfo->tap1;
531 tap2 = r->poolinfo->tap2;
532 tap3 = r->poolinfo->tap3;
533 tap4 = r->poolinfo->tap4;
534 tap5 = r->poolinfo->tap5;
535
536 input_rotate = r->input_rotate;
537 i = r->add_ptr;
538
539 /* mix one byte at a time to simplify size handling and churn faster */
540 while (nbytes--) {
541 w = rol32(*bytes++, input_rotate);
542 i = (i - 1) & wordmask;
543
544 /* XOR in the various taps */
545 w ^= r->pool[i];
546 w ^= r->pool[(i + tap1) & wordmask];
547 w ^= r->pool[(i + tap2) & wordmask];
548 w ^= r->pool[(i + tap3) & wordmask];
549 w ^= r->pool[(i + tap4) & wordmask];
550 w ^= r->pool[(i + tap5) & wordmask];
551
552 /* Mix the result back in with a twist */
553 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
554
555 /*
556 * Normally, we add 7 bits of rotation to the pool.
557 * At the beginning of the pool, add an extra 7 bits
558 * rotation, so that successive passes spread the
559 * input bits across the pool evenly.
560 */
561 input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
562 }
563
564 r->input_rotate = input_rotate;
565 r->add_ptr = i;
566}
567
568static void __mix_pool_bytes(struct entropy_store *r, const void *in,
569 int nbytes)
570{
571 trace_mix_pool_bytes_nolock(r->name, nbytes, _RET_IP_);
572 _mix_pool_bytes(r, in, nbytes);
573}
574
575static void mix_pool_bytes(struct entropy_store *r, const void *in,
576 int nbytes)
577{
578 unsigned long flags;
579
580 trace_mix_pool_bytes(r->name, nbytes, _RET_IP_);
581 spin_lock_irqsave(&r->lock, flags);
582 _mix_pool_bytes(r, in, nbytes);
583 spin_unlock_irqrestore(&r->lock, flags);
584}
585
586struct fast_pool {
587 __u32 pool[4];
588 unsigned long last;
589 unsigned short reg_idx;
590 unsigned char count;
591};
592
593/*
594 * This is a fast mixing routine used by the interrupt randomness
595 * collector. It's hardcoded for an 128 bit pool and assumes that any
596 * locks that might be needed are taken by the caller.
597 */
598static void fast_mix(struct fast_pool *f)
599{
600 __u32 a = f->pool[0], b = f->pool[1];
601 __u32 c = f->pool[2], d = f->pool[3];
602
603 a += b; c += d;
604 b = rol32(b, 6); d = rol32(d, 27);
605 d ^= a; b ^= c;
606
607 a += b; c += d;
608 b = rol32(b, 16); d = rol32(d, 14);
609 d ^= a; b ^= c;
610
611 a += b; c += d;
612 b = rol32(b, 6); d = rol32(d, 27);
613 d ^= a; b ^= c;
614
615 a += b; c += d;
616 b = rol32(b, 16); d = rol32(d, 14);
617 d ^= a; b ^= c;
618
619 f->pool[0] = a; f->pool[1] = b;
620 f->pool[2] = c; f->pool[3] = d;
621 f->count++;
622}
623
624static void process_random_ready_list(void)
625{
626 unsigned long flags;
627 struct random_ready_callback *rdy, *tmp;
628
629 spin_lock_irqsave(&random_ready_list_lock, flags);
630 list_for_each_entry_safe(rdy, tmp, &random_ready_list, list) {
631 struct module *owner = rdy->owner;
632
633 list_del_init(&rdy->list);
634 rdy->func(rdy);
635 module_put(owner);
636 }
637 spin_unlock_irqrestore(&random_ready_list_lock, flags);
638}
639
640/*
641 * Credit (or debit) the entropy store with n bits of entropy.
642 * Use credit_entropy_bits_safe() if the value comes from userspace
643 * or otherwise should be checked for extreme values.
644 */
645static void credit_entropy_bits(struct entropy_store *r, int nbits)
646{
647 int entropy_count, orig;
648 const int pool_size = r->poolinfo->poolfracbits;
649 int nfrac = nbits << ENTROPY_SHIFT;
650
651 if (!nbits)
652 return;
653
654retry:
655 entropy_count = orig = ACCESS_ONCE(r->entropy_count);
656 if (nfrac < 0) {
657 /* Debit */
658 entropy_count += nfrac;
659 } else {
660 /*
661 * Credit: we have to account for the possibility of
662 * overwriting already present entropy. Even in the
663 * ideal case of pure Shannon entropy, new contributions
664 * approach the full value asymptotically:
665 *
666 * entropy <- entropy + (pool_size - entropy) *
667 * (1 - exp(-add_entropy/pool_size))
668 *
669 * For add_entropy <= pool_size/2 then
670 * (1 - exp(-add_entropy/pool_size)) >=
671 * (add_entropy/pool_size)*0.7869...
672 * so we can approximate the exponential with
673 * 3/4*add_entropy/pool_size and still be on the
674 * safe side by adding at most pool_size/2 at a time.
675 *
676 * The use of pool_size-2 in the while statement is to
677 * prevent rounding artifacts from making the loop
678 * arbitrarily long; this limits the loop to log2(pool_size)*2
679 * turns no matter how large nbits is.
680 */
681 int pnfrac = nfrac;
682 const int s = r->poolinfo->poolbitshift + ENTROPY_SHIFT + 2;
683 /* The +2 corresponds to the /4 in the denominator */
684
685 do {
686 unsigned int anfrac = min(pnfrac, pool_size/2);
687 unsigned int add =
688 ((pool_size - entropy_count)*anfrac*3) >> s;
689
690 entropy_count += add;
691 pnfrac -= anfrac;
692 } while (unlikely(entropy_count < pool_size-2 && pnfrac));
693 }
694
695 if (unlikely(entropy_count < 0)) {
696 pr_warn("random: negative entropy/overflow: pool %s count %d\n",
697 r->name, entropy_count);
698 WARN_ON(1);
699 entropy_count = 0;
700 } else if (entropy_count > pool_size)
701 entropy_count = pool_size;
702 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
703 goto retry;
704
705 r->entropy_total += nbits;
706 if (!r->initialized && r->entropy_total > 128) {
707 r->initialized = 1;
708 r->entropy_total = 0;
709 }
710
711 trace_credit_entropy_bits(r->name, nbits,
712 entropy_count >> ENTROPY_SHIFT,
713 r->entropy_total, _RET_IP_);
714
715 if (r == &input_pool) {
716 int entropy_bits = entropy_count >> ENTROPY_SHIFT;
717
718 if (crng_init < 2 && entropy_bits >= 128) {
719 crng_reseed(&primary_crng, r);
720 entropy_bits = r->entropy_count >> ENTROPY_SHIFT;
721 }
722
723 /* should we wake readers? */
724 if (entropy_bits >= random_read_wakeup_bits) {
725 wake_up_interruptible(&random_read_wait);
726 kill_fasync(&fasync, SIGIO, POLL_IN);
727 }
728 /* If the input pool is getting full, send some
729 * entropy to the blocking pool until it is 75% full.
730 */
731 if (entropy_bits > random_write_wakeup_bits &&
732 r->initialized &&
733 r->entropy_total >= 2*random_read_wakeup_bits) {
734 struct entropy_store *other = &blocking_pool;
735
736 if (other->entropy_count <=
737 3 * other->poolinfo->poolfracbits / 4) {
738 schedule_work(&other->push_work);
739 r->entropy_total = 0;
740 }
741 }
742 }
743}
744
745static int credit_entropy_bits_safe(struct entropy_store *r, int nbits)
746{
747 const int nbits_max = r->poolinfo->poolwords * 32;
748
749 if (nbits < 0)
750 return -EINVAL;
751
752 /* Cap the value to avoid overflows */
753 nbits = min(nbits, nbits_max);
754
755 credit_entropy_bits(r, nbits);
756 return 0;
757}
758
759/*********************************************************************
760 *
761 * CRNG using CHACHA20
762 *
763 *********************************************************************/
764
765#define CRNG_RESEED_INTERVAL (300*HZ)
766
767static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
768
769#ifdef CONFIG_NUMA
770/*
771 * Hack to deal with crazy userspace progams when they are all trying
772 * to access /dev/urandom in parallel. The programs are almost
773 * certainly doing something terribly wrong, but we'll work around
774 * their brain damage.
775 */
776static struct crng_state **crng_node_pool __read_mostly;
777#endif
778
779static void invalidate_batched_entropy(void);
780
781static void crng_initialize(struct crng_state *crng)
782{
783 int i;
784 unsigned long rv;
785
786 memcpy(&crng->state[0], "expand 32-byte k", 16);
787 if (crng == &primary_crng)
788 _extract_entropy(&input_pool, &crng->state[4],
789 sizeof(__u32) * 12, 0);
790 else
791 _get_random_bytes(&crng->state[4], sizeof(__u32) * 12);
792 for (i = 4; i < 16; i++) {
793 if (!arch_get_random_seed_long(&rv) &&
794 !arch_get_random_long(&rv))
795 rv = random_get_entropy();
796 crng->state[i] ^= rv;
797 }
798 crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
799}
800
801#ifdef CONFIG_NUMA
802static void do_numa_crng_init(struct work_struct *work)
803{
804 int i;
805 struct crng_state *crng;
806 struct crng_state **pool;
807
808 pool = kcalloc(nr_node_ids, sizeof(*pool), GFP_KERNEL|__GFP_NOFAIL);
809 for_each_online_node(i) {
810 crng = kmalloc_node(sizeof(struct crng_state),
811 GFP_KERNEL | __GFP_NOFAIL, i);
812 spin_lock_init(&crng->lock);
813 crng_initialize(crng);
814 pool[i] = crng;
815 }
816 mb();
817 if (cmpxchg(&crng_node_pool, NULL, pool)) {
818 for_each_node(i)
819 kfree(pool[i]);
820 kfree(pool);
821 }
822}
823
824static DECLARE_WORK(numa_crng_init_work, do_numa_crng_init);
825
826static void numa_crng_init(void)
827{
828 schedule_work(&numa_crng_init_work);
829}
830#else
831static void numa_crng_init(void) {}
832#endif
833
834/*
835 * crng_fast_load() can be called by code in the interrupt service
836 * path. So we can't afford to dilly-dally.
837 */
838static int crng_fast_load(const char *cp, size_t len)
839{
840 unsigned long flags;
841 char *p;
842
843 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
844 return 0;
845 if (crng_init != 0) {
846 spin_unlock_irqrestore(&primary_crng.lock, flags);
847 return 0;
848 }
849 p = (unsigned char *) &primary_crng.state[4];
850 while (len > 0 && crng_init_cnt < CRNG_INIT_CNT_THRESH) {
851 p[crng_init_cnt % CHACHA20_KEY_SIZE] ^= *cp;
852 cp++; crng_init_cnt++; len--;
853 }
854 spin_unlock_irqrestore(&primary_crng.lock, flags);
855 if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
856 invalidate_batched_entropy();
857 crng_init = 1;
858 wake_up_interruptible(&crng_init_wait);
859 pr_notice("random: fast init done\n");
860 }
861 return 1;
862}
863
864/*
865 * crng_slow_load() is called by add_device_randomness, which has two
866 * attributes. (1) We can't trust the buffer passed to it is
867 * guaranteed to be unpredictable (so it might not have any entropy at
868 * all), and (2) it doesn't have the performance constraints of
869 * crng_fast_load().
870 *
871 * So we do something more comprehensive which is guaranteed to touch
872 * all of the primary_crng's state, and which uses a LFSR with a
873 * period of 255 as part of the mixing algorithm. Finally, we do
874 * *not* advance crng_init_cnt since buffer we may get may be something
875 * like a fixed DMI table (for example), which might very well be
876 * unique to the machine, but is otherwise unvarying.
877 */
878static int crng_slow_load(const char *cp, size_t len)
879{
880 unsigned long flags;
881 static unsigned char lfsr = 1;
882 unsigned char tmp;
883 unsigned i, max = CHACHA20_KEY_SIZE;
884 const char * src_buf = cp;
885 char * dest_buf = (char *) &primary_crng.state[4];
886
887 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
888 return 0;
889 if (crng_init != 0) {
890 spin_unlock_irqrestore(&primary_crng.lock, flags);
891 return 0;
892 }
893 if (len > max)
894 max = len;
895
896 for (i = 0; i < max ; i++) {
897 tmp = lfsr;
898 lfsr >>= 1;
899 if (tmp & 1)
900 lfsr ^= 0xE1;
901 tmp = dest_buf[i % CHACHA20_KEY_SIZE];
902 dest_buf[i % CHACHA20_KEY_SIZE] ^= src_buf[i % len] ^ lfsr;
903 lfsr += (tmp << 3) | (tmp >> 5);
904 }
905 spin_unlock_irqrestore(&primary_crng.lock, flags);
906 return 1;
907}
908
909static void crng_reseed(struct crng_state *crng, struct entropy_store *r)
910{
911 unsigned long flags;
912 int i, num;
913 union {
914 __u8 block[CHACHA20_BLOCK_SIZE];
915 __u32 key[8];
916 } buf;
917
918 if (r) {
919 num = extract_entropy(r, &buf, 32, 16, 0);
920 if (num == 0)
921 return;
922 } else {
923 _extract_crng(&primary_crng, buf.block);
924 _crng_backtrack_protect(&primary_crng, buf.block,
925 CHACHA20_KEY_SIZE);
926 }
927 spin_lock_irqsave(&crng->lock, flags);
928 for (i = 0; i < 8; i++) {
929 unsigned long rv;
930 if (!arch_get_random_seed_long(&rv) &&
931 !arch_get_random_long(&rv))
932 rv = random_get_entropy();
933 crng->state[i+4] ^= buf.key[i] ^ rv;
934 }
935 memzero_explicit(&buf, sizeof(buf));
936 crng->init_time = jiffies;
937 spin_unlock_irqrestore(&crng->lock, flags);
938 if (crng == &primary_crng && crng_init < 2) {
939 invalidate_batched_entropy();
940 numa_crng_init();
941 crng_init = 2;
942 process_random_ready_list();
943 wake_up_interruptible(&crng_init_wait);
944 pr_notice("random: crng init done\n");
945 if (unseeded_warning.missed) {
946 pr_notice("random: %d get_random_xx warning(s) missed "
947 "due to ratelimiting\n",
948 unseeded_warning.missed);
949 unseeded_warning.missed = 0;
950 }
951 if (urandom_warning.missed) {
952 pr_notice("random: %d urandom warning(s) missed "
953 "due to ratelimiting\n",
954 urandom_warning.missed);
955 urandom_warning.missed = 0;
956 }
957 }
958}
959
960static void _extract_crng(struct crng_state *crng,
961 __u8 out[CHACHA20_BLOCK_SIZE])
962{
963 unsigned long v, flags;
964
965 if (crng_ready() &&
966 (time_after(crng_global_init_time, crng->init_time) ||
967 time_after(jiffies, crng->init_time + CRNG_RESEED_INTERVAL)))
968 crng_reseed(crng, crng == &primary_crng ? &input_pool : NULL);
969 spin_lock_irqsave(&crng->lock, flags);
970 if (arch_get_random_long(&v))
971 crng->state[14] ^= v;
972 chacha20_block(&crng->state[0], out);
973 if (crng->state[12] == 0)
974 crng->state[13]++;
975 spin_unlock_irqrestore(&crng->lock, flags);
976}
977
978static void extract_crng(__u8 out[CHACHA20_BLOCK_SIZE])
979{
980 struct crng_state *crng = NULL;
981
982#ifdef CONFIG_NUMA
983 if (crng_node_pool)
984 crng = crng_node_pool[numa_node_id()];
985 if (crng == NULL)
986#endif
987 crng = &primary_crng;
988 _extract_crng(crng, out);
989}
990
991/*
992 * Use the leftover bytes from the CRNG block output (if there is
993 * enough) to mutate the CRNG key to provide backtracking protection.
994 */
995static void _crng_backtrack_protect(struct crng_state *crng,
996 __u8 tmp[CHACHA20_BLOCK_SIZE], int used)
997{
998 unsigned long flags;
999 __u32 *s, *d;
1000 int i;
1001
1002 used = round_up(used, sizeof(__u32));
1003 if (used + CHACHA20_KEY_SIZE > CHACHA20_BLOCK_SIZE) {
1004 extract_crng(tmp);
1005 used = 0;
1006 }
1007 spin_lock_irqsave(&crng->lock, flags);
1008 s = (__u32 *) &tmp[used];
1009 d = &crng->state[4];
1010 for (i=0; i < 8; i++)
1011 *d++ ^= *s++;
1012 spin_unlock_irqrestore(&crng->lock, flags);
1013}
1014
1015static void crng_backtrack_protect(__u8 tmp[CHACHA20_BLOCK_SIZE], int used)
1016{
1017 struct crng_state *crng = NULL;
1018
1019#ifdef CONFIG_NUMA
1020 if (crng_node_pool)
1021 crng = crng_node_pool[numa_node_id()];
1022 if (crng == NULL)
1023#endif
1024 crng = &primary_crng;
1025 _crng_backtrack_protect(crng, tmp, used);
1026}
1027
1028static ssize_t extract_crng_user(void __user *buf, size_t nbytes)
1029{
1030 ssize_t ret = 0, i = CHACHA20_BLOCK_SIZE;
1031 __u8 tmp[CHACHA20_BLOCK_SIZE];
1032 int large_request = (nbytes > 256);
1033
1034 while (nbytes) {
1035 if (large_request && need_resched()) {
1036 if (signal_pending(current)) {
1037 if (ret == 0)
1038 ret = -ERESTARTSYS;
1039 break;
1040 }
1041 schedule();
1042 }
1043
1044 extract_crng(tmp);
1045 i = min_t(int, nbytes, CHACHA20_BLOCK_SIZE);
1046 if (copy_to_user(buf, tmp, i)) {
1047 ret = -EFAULT;
1048 break;
1049 }
1050
1051 nbytes -= i;
1052 buf += i;
1053 ret += i;
1054 }
1055 crng_backtrack_protect(tmp, i);
1056
1057 /* Wipe data just written to memory */
1058 memzero_explicit(tmp, sizeof(tmp));
1059
1060 return ret;
1061}
1062
1063
1064/*********************************************************************
1065 *
1066 * Entropy input management
1067 *
1068 *********************************************************************/
1069
1070/* There is one of these per entropy source */
1071struct timer_rand_state {
1072 cycles_t last_time;
1073 long last_delta, last_delta2;
1074 unsigned dont_count_entropy:1;
1075};
1076
1077#define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
1078
1079/*
1080 * Add device- or boot-specific data to the input pool to help
1081 * initialize it.
1082 *
1083 * None of this adds any entropy; it is meant to avoid the problem of
1084 * the entropy pool having similar initial state across largely
1085 * identical devices.
1086 */
1087void add_device_randomness(const void *buf, unsigned int size)
1088{
1089 unsigned long time = random_get_entropy() ^ jiffies;
1090 unsigned long flags;
1091
1092 if (!crng_ready() && size)
1093 crng_slow_load(buf, size);
1094
1095 trace_add_device_randomness(size, _RET_IP_);
1096 spin_lock_irqsave(&input_pool.lock, flags);
1097 _mix_pool_bytes(&input_pool, buf, size);
1098 _mix_pool_bytes(&input_pool, &time, sizeof(time));
1099 spin_unlock_irqrestore(&input_pool.lock, flags);
1100}
1101EXPORT_SYMBOL(add_device_randomness);
1102
1103static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
1104
1105/*
1106 * This function adds entropy to the entropy "pool" by using timing
1107 * delays. It uses the timer_rand_state structure to make an estimate
1108 * of how many bits of entropy this call has added to the pool.
1109 *
1110 * The number "num" is also added to the pool - it should somehow describe
1111 * the type of event which just happened. This is currently 0-255 for
1112 * keyboard scan codes, and 256 upwards for interrupts.
1113 *
1114 */
1115static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
1116{
1117 struct entropy_store *r;
1118 struct {
1119 long jiffies;
1120 unsigned cycles;
1121 unsigned num;
1122 } sample;
1123 long delta, delta2, delta3;
1124
1125 preempt_disable();
1126
1127 sample.jiffies = jiffies;
1128 sample.cycles = random_get_entropy();
1129 sample.num = num;
1130 r = &input_pool;
1131 mix_pool_bytes(r, &sample, sizeof(sample));
1132
1133 /*
1134 * Calculate number of bits of randomness we probably added.
1135 * We take into account the first, second and third-order deltas
1136 * in order to make our estimate.
1137 */
1138
1139 if (!state->dont_count_entropy) {
1140 delta = sample.jiffies - state->last_time;
1141 state->last_time = sample.jiffies;
1142
1143 delta2 = delta - state->last_delta;
1144 state->last_delta = delta;
1145
1146 delta3 = delta2 - state->last_delta2;
1147 state->last_delta2 = delta2;
1148
1149 if (delta < 0)
1150 delta = -delta;
1151 if (delta2 < 0)
1152 delta2 = -delta2;
1153 if (delta3 < 0)
1154 delta3 = -delta3;
1155 if (delta > delta2)
1156 delta = delta2;
1157 if (delta > delta3)
1158 delta = delta3;
1159
1160 /*
1161 * delta is now minimum absolute delta.
1162 * Round down by 1 bit on general principles,
1163 * and limit entropy entimate to 12 bits.
1164 */
1165 credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
1166 }
1167 preempt_enable();
1168}
1169
1170void add_input_randomness(unsigned int type, unsigned int code,
1171 unsigned int value)
1172{
1173 static unsigned char last_value;
1174
1175 /* ignore autorepeat and the like */
1176 if (value == last_value)
1177 return;
1178
1179 last_value = value;
1180 add_timer_randomness(&input_timer_state,
1181 (type << 4) ^ code ^ (code >> 4) ^ value);
1182 trace_add_input_randomness(ENTROPY_BITS(&input_pool));
1183}
1184EXPORT_SYMBOL_GPL(add_input_randomness);
1185
1186static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
1187
1188#ifdef ADD_INTERRUPT_BENCH
1189static unsigned long avg_cycles, avg_deviation;
1190
1191#define AVG_SHIFT 8 /* Exponential average factor k=1/256 */
1192#define FIXED_1_2 (1 << (AVG_SHIFT-1))
1193
1194static void add_interrupt_bench(cycles_t start)
1195{
1196 long delta = random_get_entropy() - start;
1197
1198 /* Use a weighted moving average */
1199 delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
1200 avg_cycles += delta;
1201 /* And average deviation */
1202 delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
1203 avg_deviation += delta;
1204}
1205#else
1206#define add_interrupt_bench(x)
1207#endif
1208
1209static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
1210{
1211 __u32 *ptr = (__u32 *) regs;
1212 unsigned int idx;
1213
1214 if (regs == NULL)
1215 return 0;
1216 idx = READ_ONCE(f->reg_idx);
1217 if (idx >= sizeof(struct pt_regs) / sizeof(__u32))
1218 idx = 0;
1219 ptr += idx++;
1220 WRITE_ONCE(f->reg_idx, idx);
1221 return *ptr;
1222}
1223
1224void add_interrupt_randomness(int irq, int irq_flags)
1225{
1226 struct entropy_store *r;
1227 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1228 struct pt_regs *regs = get_irq_regs();
1229 unsigned long now = jiffies;
1230 cycles_t cycles = random_get_entropy();
1231 __u32 c_high, j_high;
1232 __u64 ip;
1233 unsigned long seed;
1234 int credit = 0;
1235
1236 if (cycles == 0)
1237 cycles = get_reg(fast_pool, regs);
1238 c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
1239 j_high = (sizeof(now) > 4) ? now >> 32 : 0;
1240 fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
1241 fast_pool->pool[1] ^= now ^ c_high;
1242 ip = regs ? instruction_pointer(regs) : _RET_IP_;
1243 fast_pool->pool[2] ^= ip;
1244 fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
1245 get_reg(fast_pool, regs);
1246
1247 fast_mix(fast_pool);
1248 add_interrupt_bench(cycles);
1249 this_cpu_add(net_rand_state.s1, fast_pool->pool[cycles & 3]);
1250
1251 if (unlikely(crng_init == 0)) {
1252 if ((fast_pool->count >= 64) &&
1253 crng_fast_load((char *) fast_pool->pool,
1254 sizeof(fast_pool->pool))) {
1255 fast_pool->count = 0;
1256 fast_pool->last = now;
1257 }
1258 return;
1259 }
1260
1261 if ((fast_pool->count < 64) &&
1262 !time_after(now, fast_pool->last + HZ))
1263 return;
1264
1265 r = &input_pool;
1266 if (!spin_trylock(&r->lock))
1267 return;
1268
1269 fast_pool->last = now;
1270 __mix_pool_bytes(r, &fast_pool->pool, sizeof(fast_pool->pool));
1271
1272 /*
1273 * If we have architectural seed generator, produce a seed and
1274 * add it to the pool. For the sake of paranoia don't let the
1275 * architectural seed generator dominate the input from the
1276 * interrupt noise.
1277 */
1278 if (arch_get_random_seed_long(&seed)) {
1279 __mix_pool_bytes(r, &seed, sizeof(seed));
1280 credit = 1;
1281 }
1282 spin_unlock(&r->lock);
1283
1284 fast_pool->count = 0;
1285
1286 /* award one bit for the contents of the fast pool */
1287 credit_entropy_bits(r, credit + 1);
1288}
1289EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1290
1291#ifdef CONFIG_BLOCK
1292void add_disk_randomness(struct gendisk *disk)
1293{
1294 if (!disk || !disk->random)
1295 return;
1296 /* first major is 1, so we get >= 0x200 here */
1297 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1298 trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
1299}
1300EXPORT_SYMBOL_GPL(add_disk_randomness);
1301#endif
1302
1303/*********************************************************************
1304 *
1305 * Entropy extraction routines
1306 *
1307 *********************************************************************/
1308
1309/*
1310 * This utility inline function is responsible for transferring entropy
1311 * from the primary pool to the secondary extraction pool. We make
1312 * sure we pull enough for a 'catastrophic reseed'.
1313 */
1314static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes);
1315static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1316{
1317 if (!r->pull ||
1318 r->entropy_count >= (nbytes << (ENTROPY_SHIFT + 3)) ||
1319 r->entropy_count > r->poolinfo->poolfracbits)
1320 return;
1321
1322 _xfer_secondary_pool(r, nbytes);
1323}
1324
1325static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1326{
1327 __u32 tmp[OUTPUT_POOL_WORDS];
1328
1329 int bytes = nbytes;
1330
1331 /* pull at least as much as a wakeup */
1332 bytes = max_t(int, bytes, random_read_wakeup_bits / 8);
1333 /* but never more than the buffer size */
1334 bytes = min_t(int, bytes, sizeof(tmp));
1335
1336 trace_xfer_secondary_pool(r->name, bytes * 8, nbytes * 8,
1337 ENTROPY_BITS(r), ENTROPY_BITS(r->pull));
1338 bytes = extract_entropy(r->pull, tmp, bytes,
1339 random_read_wakeup_bits / 8, 0);
1340 mix_pool_bytes(r, tmp, bytes);
1341 credit_entropy_bits(r, bytes*8);
1342}
1343
1344/*
1345 * Used as a workqueue function so that when the input pool is getting
1346 * full, we can "spill over" some entropy to the output pools. That
1347 * way the output pools can store some of the excess entropy instead
1348 * of letting it go to waste.
1349 */
1350static void push_to_pool(struct work_struct *work)
1351{
1352 struct entropy_store *r = container_of(work, struct entropy_store,
1353 push_work);
1354 BUG_ON(!r);
1355 _xfer_secondary_pool(r, random_read_wakeup_bits/8);
1356 trace_push_to_pool(r->name, r->entropy_count >> ENTROPY_SHIFT,
1357 r->pull->entropy_count >> ENTROPY_SHIFT);
1358}
1359
1360/*
1361 * This function decides how many bytes to actually take from the
1362 * given pool, and also debits the entropy count accordingly.
1363 */
1364static size_t account(struct entropy_store *r, size_t nbytes, int min,
1365 int reserved)
1366{
1367 int entropy_count, orig, have_bytes;
1368 size_t ibytes, nfrac;
1369
1370 BUG_ON(r->entropy_count > r->poolinfo->poolfracbits);
1371
1372 /* Can we pull enough? */
1373retry:
1374 entropy_count = orig = ACCESS_ONCE(r->entropy_count);
1375 ibytes = nbytes;
1376 /* never pull more than available */
1377 have_bytes = entropy_count >> (ENTROPY_SHIFT + 3);
1378
1379 if ((have_bytes -= reserved) < 0)
1380 have_bytes = 0;
1381 ibytes = min_t(size_t, ibytes, have_bytes);
1382 if (ibytes < min)
1383 ibytes = 0;
1384
1385 if (unlikely(entropy_count < 0)) {
1386 pr_warn("random: negative entropy count: pool %s count %d\n",
1387 r->name, entropy_count);
1388 WARN_ON(1);
1389 entropy_count = 0;
1390 }
1391 nfrac = ibytes << (ENTROPY_SHIFT + 3);
1392 if ((size_t) entropy_count > nfrac)
1393 entropy_count -= nfrac;
1394 else
1395 entropy_count = 0;
1396
1397 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
1398 goto retry;
1399
1400 trace_debit_entropy(r->name, 8 * ibytes);
1401 if (ibytes &&
1402 (r->entropy_count >> ENTROPY_SHIFT) < random_write_wakeup_bits) {
1403 wake_up_interruptible(&random_write_wait);
1404 kill_fasync(&fasync, SIGIO, POLL_OUT);
1405 }
1406
1407 return ibytes;
1408}
1409
1410/*
1411 * This function does the actual extraction for extract_entropy and
1412 * extract_entropy_user.
1413 *
1414 * Note: we assume that .poolwords is a multiple of 16 words.
1415 */
1416static void extract_buf(struct entropy_store *r, __u8 *out)
1417{
1418 int i;
1419 union {
1420 __u32 w[5];
1421 unsigned long l[LONGS(20)];
1422 } hash;
1423 __u32 workspace[SHA_WORKSPACE_WORDS];
1424 unsigned long flags;
1425
1426 /*
1427 * If we have an architectural hardware random number
1428 * generator, use it for SHA's initial vector
1429 */
1430 sha_init(hash.w);
1431 for (i = 0; i < LONGS(20); i++) {
1432 unsigned long v;
1433 if (!arch_get_random_long(&v))
1434 break;
1435 hash.l[i] = v;
1436 }
1437
1438 /* Generate a hash across the pool, 16 words (512 bits) at a time */
1439 spin_lock_irqsave(&r->lock, flags);
1440 for (i = 0; i < r->poolinfo->poolwords; i += 16)
1441 sha_transform(hash.w, (__u8 *)(r->pool + i), workspace);
1442
1443 /*
1444 * We mix the hash back into the pool to prevent backtracking
1445 * attacks (where the attacker knows the state of the pool
1446 * plus the current outputs, and attempts to find previous
1447 * ouputs), unless the hash function can be inverted. By
1448 * mixing at least a SHA1 worth of hash data back, we make
1449 * brute-forcing the feedback as hard as brute-forcing the
1450 * hash.
1451 */
1452 __mix_pool_bytes(r, hash.w, sizeof(hash.w));
1453 spin_unlock_irqrestore(&r->lock, flags);
1454
1455 memzero_explicit(workspace, sizeof(workspace));
1456
1457 /*
1458 * In case the hash function has some recognizable output
1459 * pattern, we fold it in half. Thus, we always feed back
1460 * twice as much data as we output.
1461 */
1462 hash.w[0] ^= hash.w[3];
1463 hash.w[1] ^= hash.w[4];
1464 hash.w[2] ^= rol32(hash.w[2], 16);
1465
1466 memcpy(out, &hash, EXTRACT_SIZE);
1467 memzero_explicit(&hash, sizeof(hash));
1468}
1469
1470static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
1471 size_t nbytes, int fips)
1472{
1473 ssize_t ret = 0, i;
1474 __u8 tmp[EXTRACT_SIZE];
1475 unsigned long flags;
1476
1477 while (nbytes) {
1478 extract_buf(r, tmp);
1479
1480 if (fips) {
1481 spin_lock_irqsave(&r->lock, flags);
1482 if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
1483 panic("Hardware RNG duplicated output!\n");
1484 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1485 spin_unlock_irqrestore(&r->lock, flags);
1486 }
1487 i = min_t(int, nbytes, EXTRACT_SIZE);
1488 memcpy(buf, tmp, i);
1489 nbytes -= i;
1490 buf += i;
1491 ret += i;
1492 }
1493
1494 /* Wipe data just returned from memory */
1495 memzero_explicit(tmp, sizeof(tmp));
1496
1497 return ret;
1498}
1499
1500/*
1501 * This function extracts randomness from the "entropy pool", and
1502 * returns it in a buffer.
1503 *
1504 * The min parameter specifies the minimum amount we can pull before
1505 * failing to avoid races that defeat catastrophic reseeding while the
1506 * reserved parameter indicates how much entropy we must leave in the
1507 * pool after each pull to avoid starving other readers.
1508 */
1509static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1510 size_t nbytes, int min, int reserved)
1511{
1512 __u8 tmp[EXTRACT_SIZE];
1513 unsigned long flags;
1514
1515 /* if last_data isn't primed, we need EXTRACT_SIZE extra bytes */
1516 if (fips_enabled) {
1517 spin_lock_irqsave(&r->lock, flags);
1518 if (!r->last_data_init) {
1519 r->last_data_init = 1;
1520 spin_unlock_irqrestore(&r->lock, flags);
1521 trace_extract_entropy(r->name, EXTRACT_SIZE,
1522 ENTROPY_BITS(r), _RET_IP_);
1523 xfer_secondary_pool(r, EXTRACT_SIZE);
1524 extract_buf(r, tmp);
1525 spin_lock_irqsave(&r->lock, flags);
1526 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1527 }
1528 spin_unlock_irqrestore(&r->lock, flags);
1529 }
1530
1531 trace_extract_entropy(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1532 xfer_secondary_pool(r, nbytes);
1533 nbytes = account(r, nbytes, min, reserved);
1534
1535 return _extract_entropy(r, buf, nbytes, fips_enabled);
1536}
1537
1538/*
1539 * This function extracts randomness from the "entropy pool", and
1540 * returns it in a userspace buffer.
1541 */
1542static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
1543 size_t nbytes)
1544{
1545 ssize_t ret = 0, i;
1546 __u8 tmp[EXTRACT_SIZE];
1547 int large_request = (nbytes > 256);
1548
1549 trace_extract_entropy_user(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1550 xfer_secondary_pool(r, nbytes);
1551 nbytes = account(r, nbytes, 0, 0);
1552
1553 while (nbytes) {
1554 if (large_request && need_resched()) {
1555 if (signal_pending(current)) {
1556 if (ret == 0)
1557 ret = -ERESTARTSYS;
1558 break;
1559 }
1560 schedule();
1561 }
1562
1563 extract_buf(r, tmp);
1564 i = min_t(int, nbytes, EXTRACT_SIZE);
1565 if (copy_to_user(buf, tmp, i)) {
1566 ret = -EFAULT;
1567 break;
1568 }
1569
1570 nbytes -= i;
1571 buf += i;
1572 ret += i;
1573 }
1574
1575 /* Wipe data just returned from memory */
1576 memzero_explicit(tmp, sizeof(tmp));
1577
1578 return ret;
1579}
1580
1581#define warn_unseeded_randomness(previous) \
1582 _warn_unseeded_randomness(__func__, (void *) _RET_IP_, (previous))
1583
1584static void _warn_unseeded_randomness(const char *func_name, void *caller,
1585 void **previous)
1586{
1587#ifdef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1588 const bool print_once = false;
1589#else
1590 static bool print_once __read_mostly;
1591#endif
1592
1593 if (print_once ||
1594 crng_ready() ||
1595 (previous && (caller == READ_ONCE(*previous))))
1596 return;
1597 WRITE_ONCE(*previous, caller);
1598#ifndef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1599 print_once = true;
1600#endif
1601 if (__ratelimit(&unseeded_warning))
1602 pr_notice("random: %s called from %pS with crng_init=%d\n",
1603 func_name, caller, crng_init);
1604}
1605
1606/*
1607 * This function is the exported kernel interface. It returns some
1608 * number of good random numbers, suitable for key generation, seeding
1609 * TCP sequence numbers, etc. It does not rely on the hardware random
1610 * number generator. For random bytes direct from the hardware RNG
1611 * (when available), use get_random_bytes_arch(). In order to ensure
1612 * that the randomness provided by this function is okay, the function
1613 * wait_for_random_bytes() should be called and return 0 at least once
1614 * at any point prior.
1615 */
1616static void _get_random_bytes(void *buf, int nbytes)
1617{
1618 __u8 tmp[CHACHA20_BLOCK_SIZE];
1619
1620 trace_get_random_bytes(nbytes, _RET_IP_);
1621
1622 while (nbytes >= CHACHA20_BLOCK_SIZE) {
1623 extract_crng(buf);
1624 buf += CHACHA20_BLOCK_SIZE;
1625 nbytes -= CHACHA20_BLOCK_SIZE;
1626 }
1627
1628 if (nbytes > 0) {
1629 extract_crng(tmp);
1630 memcpy(buf, tmp, nbytes);
1631 crng_backtrack_protect(tmp, nbytes);
1632 } else
1633 crng_backtrack_protect(tmp, CHACHA20_BLOCK_SIZE);
1634 memzero_explicit(tmp, sizeof(tmp));
1635}
1636
1637void get_random_bytes(void *buf, int nbytes)
1638{
1639 static void *previous;
1640
1641 warn_unseeded_randomness(&previous);
1642 _get_random_bytes(buf, nbytes);
1643}
1644EXPORT_SYMBOL(get_random_bytes);
1645
1646/*
1647 * Wait for the urandom pool to be seeded and thus guaranteed to supply
1648 * cryptographically secure random numbers. This applies to: the /dev/urandom
1649 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
1650 * family of functions. Using any of these functions without first calling
1651 * this function forfeits the guarantee of security.
1652 *
1653 * Returns: 0 if the urandom pool has been seeded.
1654 * -ERESTARTSYS if the function was interrupted by a signal.
1655 */
1656int wait_for_random_bytes(void)
1657{
1658 if (likely(crng_ready()))
1659 return 0;
1660 return wait_event_interruptible(crng_init_wait, crng_ready());
1661}
1662EXPORT_SYMBOL(wait_for_random_bytes);
1663
1664/*
1665 * Add a callback function that will be invoked when the nonblocking
1666 * pool is initialised.
1667 *
1668 * returns: 0 if callback is successfully added
1669 * -EALREADY if pool is already initialised (callback not called)
1670 * -ENOENT if module for callback is not alive
1671 */
1672int add_random_ready_callback(struct random_ready_callback *rdy)
1673{
1674 struct module *owner;
1675 unsigned long flags;
1676 int err = -EALREADY;
1677
1678 if (crng_ready())
1679 return err;
1680
1681 owner = rdy->owner;
1682 if (!try_module_get(owner))
1683 return -ENOENT;
1684
1685 spin_lock_irqsave(&random_ready_list_lock, flags);
1686 if (crng_ready())
1687 goto out;
1688
1689 owner = NULL;
1690
1691 list_add(&rdy->list, &random_ready_list);
1692 err = 0;
1693
1694out:
1695 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1696
1697 module_put(owner);
1698
1699 return err;
1700}
1701EXPORT_SYMBOL(add_random_ready_callback);
1702
1703/*
1704 * Delete a previously registered readiness callback function.
1705 */
1706void del_random_ready_callback(struct random_ready_callback *rdy)
1707{
1708 unsigned long flags;
1709 struct module *owner = NULL;
1710
1711 spin_lock_irqsave(&random_ready_list_lock, flags);
1712 if (!list_empty(&rdy->list)) {
1713 list_del_init(&rdy->list);
1714 owner = rdy->owner;
1715 }
1716 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1717
1718 module_put(owner);
1719}
1720EXPORT_SYMBOL(del_random_ready_callback);
1721
1722/*
1723 * This function will use the architecture-specific hardware random
1724 * number generator if it is available. The arch-specific hw RNG will
1725 * almost certainly be faster than what we can do in software, but it
1726 * is impossible to verify that it is implemented securely (as
1727 * opposed, to, say, the AES encryption of a sequence number using a
1728 * key known by the NSA). So it's useful if we need the speed, but
1729 * only if we're willing to trust the hardware manufacturer not to
1730 * have put in a back door.
1731 */
1732void get_random_bytes_arch(void *buf, int nbytes)
1733{
1734 char *p = buf;
1735
1736 trace_get_random_bytes_arch(nbytes, _RET_IP_);
1737 while (nbytes) {
1738 unsigned long v;
1739 int chunk = min(nbytes, (int)sizeof(unsigned long));
1740
1741 if (!arch_get_random_long(&v))
1742 break;
1743
1744 memcpy(p, &v, chunk);
1745 p += chunk;
1746 nbytes -= chunk;
1747 }
1748
1749 if (nbytes)
1750 get_random_bytes(p, nbytes);
1751}
1752EXPORT_SYMBOL(get_random_bytes_arch);
1753
1754
1755/*
1756 * init_std_data - initialize pool with system data
1757 *
1758 * @r: pool to initialize
1759 *
1760 * This function clears the pool's entropy count and mixes some system
1761 * data into the pool to prepare it for use. The pool is not cleared
1762 * as that can only decrease the entropy in the pool.
1763 */
1764static void init_std_data(struct entropy_store *r)
1765{
1766 int i;
1767 ktime_t now = ktime_get_real();
1768 unsigned long rv;
1769
1770 r->last_pulled = jiffies;
1771 mix_pool_bytes(r, &now, sizeof(now));
1772 for (i = r->poolinfo->poolbytes; i > 0; i -= sizeof(rv)) {
1773 if (!arch_get_random_seed_long(&rv) &&
1774 !arch_get_random_long(&rv))
1775 rv = random_get_entropy();
1776 mix_pool_bytes(r, &rv, sizeof(rv));
1777 }
1778 mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1779}
1780
1781/*
1782 * Note that setup_arch() may call add_device_randomness()
1783 * long before we get here. This allows seeding of the pools
1784 * with some platform dependent data very early in the boot
1785 * process. But it limits our options here. We must use
1786 * statically allocated structures that already have all
1787 * initializations complete at compile time. We should also
1788 * take care not to overwrite the precious per platform data
1789 * we were given.
1790 */
1791static int rand_initialize(void)
1792{
1793 init_std_data(&input_pool);
1794 init_std_data(&blocking_pool);
1795 crng_initialize(&primary_crng);
1796 crng_global_init_time = jiffies;
1797 if (ratelimit_disable) {
1798 urandom_warning.interval = 0;
1799 unseeded_warning.interval = 0;
1800 }
1801 return 0;
1802}
1803early_initcall(rand_initialize);
1804
1805#ifdef CONFIG_BLOCK
1806void rand_initialize_disk(struct gendisk *disk)
1807{
1808 struct timer_rand_state *state;
1809
1810 /*
1811 * If kzalloc returns null, we just won't use that entropy
1812 * source.
1813 */
1814 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1815 if (state) {
1816 state->last_time = INITIAL_JIFFIES;
1817 disk->random = state;
1818 }
1819}
1820#endif
1821
1822static ssize_t
1823_random_read(int nonblock, char __user *buf, size_t nbytes)
1824{
1825 ssize_t n;
1826
1827 if (nbytes == 0)
1828 return 0;
1829
1830 nbytes = min_t(size_t, nbytes, SEC_XFER_SIZE);
1831 while (1) {
1832 n = extract_entropy_user(&blocking_pool, buf, nbytes);
1833 if (n < 0)
1834 return n;
1835 trace_random_read(n*8, (nbytes-n)*8,
1836 ENTROPY_BITS(&blocking_pool),
1837 ENTROPY_BITS(&input_pool));
1838 if (n > 0)
1839 return n;
1840
1841 /* Pool is (near) empty. Maybe wait and retry. */
1842 if (nonblock)
1843 return -EAGAIN;
1844
1845 wait_event_interruptible(random_read_wait,
1846 ENTROPY_BITS(&input_pool) >=
1847 random_read_wakeup_bits);
1848 if (signal_pending(current))
1849 return -ERESTARTSYS;
1850 }
1851}
1852
1853static ssize_t
1854random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1855{
1856 return _random_read(file->f_flags & O_NONBLOCK, buf, nbytes);
1857}
1858
1859static ssize_t
1860urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1861{
1862 unsigned long flags;
1863 static int maxwarn = 10;
1864 int ret;
1865
1866 if (!crng_ready() && maxwarn > 0) {
1867 maxwarn--;
1868 if (__ratelimit(&urandom_warning))
1869 printk(KERN_NOTICE "random: %s: uninitialized "
1870 "urandom read (%zd bytes read)\n",
1871 current->comm, nbytes);
1872 spin_lock_irqsave(&primary_crng.lock, flags);
1873 crng_init_cnt = 0;
1874 spin_unlock_irqrestore(&primary_crng.lock, flags);
1875 }
1876 nbytes = min_t(size_t, nbytes, INT_MAX >> (ENTROPY_SHIFT + 3));
1877 ret = extract_crng_user(buf, nbytes);
1878 trace_urandom_read(8 * nbytes, 0, ENTROPY_BITS(&input_pool));
1879 return ret;
1880}
1881
1882static unsigned int
1883random_poll(struct file *file, poll_table * wait)
1884{
1885 unsigned int mask;
1886
1887 poll_wait(file, &random_read_wait, wait);
1888 poll_wait(file, &random_write_wait, wait);
1889 mask = 0;
1890 if (ENTROPY_BITS(&input_pool) >= random_read_wakeup_bits)
1891 mask |= POLLIN | POLLRDNORM;
1892 if (ENTROPY_BITS(&input_pool) < random_write_wakeup_bits)
1893 mask |= POLLOUT | POLLWRNORM;
1894 return mask;
1895}
1896
1897static int
1898write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1899{
1900 size_t bytes;
1901 __u32 t, buf[16];
1902 const char __user *p = buffer;
1903
1904 while (count > 0) {
1905 int b, i = 0;
1906
1907 bytes = min(count, sizeof(buf));
1908 if (copy_from_user(&buf, p, bytes))
1909 return -EFAULT;
1910
1911 for (b = bytes ; b > 0 ; b -= sizeof(__u32), i++) {
1912 if (!arch_get_random_int(&t))
1913 break;
1914 buf[i] ^= t;
1915 }
1916
1917 count -= bytes;
1918 p += bytes;
1919
1920 mix_pool_bytes(r, buf, bytes);
1921 cond_resched();
1922 }
1923
1924 return 0;
1925}
1926
1927static ssize_t random_write(struct file *file, const char __user *buffer,
1928 size_t count, loff_t *ppos)
1929{
1930 size_t ret;
1931
1932 ret = write_pool(&input_pool, buffer, count);
1933 if (ret)
1934 return ret;
1935
1936 return (ssize_t)count;
1937}
1938
1939static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1940{
1941 int size, ent_count;
1942 int __user *p = (int __user *)arg;
1943 int retval;
1944
1945 switch (cmd) {
1946 case RNDGETENTCNT:
1947 /* inherently racy, no point locking */
1948 ent_count = ENTROPY_BITS(&input_pool);
1949 if (put_user(ent_count, p))
1950 return -EFAULT;
1951 return 0;
1952 case RNDADDTOENTCNT:
1953 if (!capable(CAP_SYS_ADMIN))
1954 return -EPERM;
1955 if (get_user(ent_count, p))
1956 return -EFAULT;
1957 return credit_entropy_bits_safe(&input_pool, ent_count);
1958 case RNDADDENTROPY:
1959 if (!capable(CAP_SYS_ADMIN))
1960 return -EPERM;
1961 if (get_user(ent_count, p++))
1962 return -EFAULT;
1963 if (ent_count < 0)
1964 return -EINVAL;
1965 if (get_user(size, p++))
1966 return -EFAULT;
1967 retval = write_pool(&input_pool, (const char __user *)p,
1968 size);
1969 if (retval < 0)
1970 return retval;
1971 return credit_entropy_bits_safe(&input_pool, ent_count);
1972 case RNDZAPENTCNT:
1973 case RNDCLEARPOOL:
1974 /*
1975 * Clear the entropy pool counters. We no longer clear
1976 * the entropy pool, as that's silly.
1977 */
1978 if (!capable(CAP_SYS_ADMIN))
1979 return -EPERM;
1980 input_pool.entropy_count = 0;
1981 blocking_pool.entropy_count = 0;
1982 return 0;
1983 case RNDRESEEDCRNG:
1984 if (!capable(CAP_SYS_ADMIN))
1985 return -EPERM;
1986 if (crng_init < 2)
1987 return -ENODATA;
1988 crng_reseed(&primary_crng, NULL);
1989 crng_global_init_time = jiffies - 1;
1990 return 0;
1991 default:
1992 return -EINVAL;
1993 }
1994}
1995
1996static int random_fasync(int fd, struct file *filp, int on)
1997{
1998 return fasync_helper(fd, filp, on, &fasync);
1999}
2000
2001const struct file_operations random_fops = {
2002 .read = random_read,
2003 .write = random_write,
2004 .poll = random_poll,
2005 .unlocked_ioctl = random_ioctl,
2006 .fasync = random_fasync,
2007 .llseek = noop_llseek,
2008};
2009
2010const struct file_operations urandom_fops = {
2011 .read = urandom_read,
2012 .write = random_write,
2013 .unlocked_ioctl = random_ioctl,
2014 .fasync = random_fasync,
2015 .llseek = noop_llseek,
2016};
2017
2018SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count,
2019 unsigned int, flags)
2020{
2021 int ret;
2022
2023 if (flags & ~(GRND_NONBLOCK|GRND_RANDOM))
2024 return -EINVAL;
2025
2026 if (count > INT_MAX)
2027 count = INT_MAX;
2028
2029 if (flags & GRND_RANDOM)
2030 return _random_read(flags & GRND_NONBLOCK, buf, count);
2031
2032 if (!crng_ready()) {
2033 if (flags & GRND_NONBLOCK)
2034 return -EAGAIN;
2035 ret = wait_for_random_bytes();
2036 if (unlikely(ret))
2037 return ret;
2038 }
2039 return urandom_read(NULL, buf, count, NULL);
2040}
2041
2042/********************************************************************
2043 *
2044 * Sysctl interface
2045 *
2046 ********************************************************************/
2047
2048#ifdef CONFIG_SYSCTL
2049
2050#include <linux/sysctl.h>
2051
2052static int min_read_thresh = 8, min_write_thresh;
2053static int max_read_thresh = OUTPUT_POOL_WORDS * 32;
2054static int max_write_thresh = INPUT_POOL_WORDS * 32;
2055static int random_min_urandom_seed = 60;
2056static char sysctl_bootid[16];
2057
2058/*
2059 * This function is used to return both the bootid UUID, and random
2060 * UUID. The difference is in whether table->data is NULL; if it is,
2061 * then a new UUID is generated and returned to the user.
2062 *
2063 * If the user accesses this via the proc interface, the UUID will be
2064 * returned as an ASCII string in the standard UUID format; if via the
2065 * sysctl system call, as 16 bytes of binary data.
2066 */
2067static int proc_do_uuid(struct ctl_table *table, int write,
2068 void __user *buffer, size_t *lenp, loff_t *ppos)
2069{
2070 struct ctl_table fake_table;
2071 unsigned char buf[64], tmp_uuid[16], *uuid;
2072
2073 uuid = table->data;
2074 if (!uuid) {
2075 uuid = tmp_uuid;
2076 generate_random_uuid(uuid);
2077 } else {
2078 static DEFINE_SPINLOCK(bootid_spinlock);
2079
2080 spin_lock(&bootid_spinlock);
2081 if (!uuid[8])
2082 generate_random_uuid(uuid);
2083 spin_unlock(&bootid_spinlock);
2084 }
2085
2086 sprintf(buf, "%pU", uuid);
2087
2088 fake_table.data = buf;
2089 fake_table.maxlen = sizeof(buf);
2090
2091 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
2092}
2093
2094/*
2095 * Return entropy available scaled to integral bits
2096 */
2097static int proc_do_entropy(struct ctl_table *table, int write,
2098 void __user *buffer, size_t *lenp, loff_t *ppos)
2099{
2100 struct ctl_table fake_table;
2101 int entropy_count;
2102
2103 entropy_count = *(int *)table->data >> ENTROPY_SHIFT;
2104
2105 fake_table.data = &entropy_count;
2106 fake_table.maxlen = sizeof(entropy_count);
2107
2108 return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
2109}
2110
2111static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
2112extern struct ctl_table random_table[];
2113struct ctl_table random_table[] = {
2114 {
2115 .procname = "poolsize",
2116 .data = &sysctl_poolsize,
2117 .maxlen = sizeof(int),
2118 .mode = 0444,
2119 .proc_handler = proc_dointvec,
2120 },
2121 {
2122 .procname = "entropy_avail",
2123 .maxlen = sizeof(int),
2124 .mode = 0444,
2125 .proc_handler = proc_do_entropy,
2126 .data = &input_pool.entropy_count,
2127 },
2128 {
2129 .procname = "read_wakeup_threshold",
2130 .data = &random_read_wakeup_bits,
2131 .maxlen = sizeof(int),
2132 .mode = 0644,
2133 .proc_handler = proc_dointvec_minmax,
2134 .extra1 = &min_read_thresh,
2135 .extra2 = &max_read_thresh,
2136 },
2137 {
2138 .procname = "write_wakeup_threshold",
2139 .data = &random_write_wakeup_bits,
2140 .maxlen = sizeof(int),
2141 .mode = 0644,
2142 .proc_handler = proc_dointvec_minmax,
2143 .extra1 = &min_write_thresh,
2144 .extra2 = &max_write_thresh,
2145 },
2146 {
2147 .procname = "urandom_min_reseed_secs",
2148 .data = &random_min_urandom_seed,
2149 .maxlen = sizeof(int),
2150 .mode = 0644,
2151 .proc_handler = proc_dointvec,
2152 },
2153 {
2154 .procname = "boot_id",
2155 .data = &sysctl_bootid,
2156 .maxlen = 16,
2157 .mode = 0444,
2158 .proc_handler = proc_do_uuid,
2159 },
2160 {
2161 .procname = "uuid",
2162 .maxlen = 16,
2163 .mode = 0444,
2164 .proc_handler = proc_do_uuid,
2165 },
2166#ifdef ADD_INTERRUPT_BENCH
2167 {
2168 .procname = "add_interrupt_avg_cycles",
2169 .data = &avg_cycles,
2170 .maxlen = sizeof(avg_cycles),
2171 .mode = 0444,
2172 .proc_handler = proc_doulongvec_minmax,
2173 },
2174 {
2175 .procname = "add_interrupt_avg_deviation",
2176 .data = &avg_deviation,
2177 .maxlen = sizeof(avg_deviation),
2178 .mode = 0444,
2179 .proc_handler = proc_doulongvec_minmax,
2180 },
2181#endif
2182 { }
2183};
2184#endif /* CONFIG_SYSCTL */
2185
2186struct batched_entropy {
2187 union {
2188 u64 entropy_u64[CHACHA20_BLOCK_SIZE / sizeof(u64)];
2189 u32 entropy_u32[CHACHA20_BLOCK_SIZE / sizeof(u32)];
2190 };
2191 unsigned int position;
2192 spinlock_t batch_lock;
2193};
2194
2195/*
2196 * Get a random word for internal kernel use only. The quality of the random
2197 * number is good as /dev/urandom, but there is no backtrack protection, with
2198 * the goal of being quite fast and not depleting entropy. In order to ensure
2199 * that the randomness provided by this function is okay, the function
2200 * wait_for_random_bytes() should be called and return 0 at least once at any
2201 * point prior.
2202 */
2203static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64) = {
2204 .batch_lock = __SPIN_LOCK_UNLOCKED(batched_entropy_u64.lock),
2205};
2206
2207u64 get_random_u64(void)
2208{
2209 u64 ret;
2210 unsigned long flags;
2211 struct batched_entropy *batch;
2212 static void *previous;
2213
2214 warn_unseeded_randomness(&previous);
2215
2216 batch = raw_cpu_ptr(&batched_entropy_u64);
2217 spin_lock_irqsave(&batch->batch_lock, flags);
2218 if (batch->position % ARRAY_SIZE(batch->entropy_u64) == 0) {
2219 extract_crng((u8 *)batch->entropy_u64);
2220 batch->position = 0;
2221 }
2222 ret = batch->entropy_u64[batch->position++];
2223 spin_unlock_irqrestore(&batch->batch_lock, flags);
2224 return ret;
2225}
2226EXPORT_SYMBOL(get_random_u64);
2227
2228static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32) = {
2229 .batch_lock = __SPIN_LOCK_UNLOCKED(batched_entropy_u32.lock),
2230};
2231u32 get_random_u32(void)
2232{
2233 u32 ret;
2234 unsigned long flags;
2235 struct batched_entropy *batch;
2236 static void *previous;
2237
2238 warn_unseeded_randomness(&previous);
2239
2240 batch = raw_cpu_ptr(&batched_entropy_u32);
2241 spin_lock_irqsave(&batch->batch_lock, flags);
2242 if (batch->position % ARRAY_SIZE(batch->entropy_u32) == 0) {
2243 extract_crng((u8 *)batch->entropy_u32);
2244 batch->position = 0;
2245 }
2246 ret = batch->entropy_u32[batch->position++];
2247 spin_unlock_irqrestore(&batch->batch_lock, flags);
2248 return ret;
2249}
2250EXPORT_SYMBOL(get_random_u32);
2251
2252/* It's important to invalidate all potential batched entropy that might
2253 * be stored before the crng is initialized, which we can do lazily by
2254 * simply resetting the counter to zero so that it's re-extracted on the
2255 * next usage. */
2256static void invalidate_batched_entropy(void)
2257{
2258 int cpu;
2259 unsigned long flags;
2260
2261 for_each_possible_cpu (cpu) {
2262 struct batched_entropy *batched_entropy;
2263
2264 batched_entropy = per_cpu_ptr(&batched_entropy_u32, cpu);
2265 spin_lock_irqsave(&batched_entropy->batch_lock, flags);
2266 batched_entropy->position = 0;
2267 spin_unlock(&batched_entropy->batch_lock);
2268
2269 batched_entropy = per_cpu_ptr(&batched_entropy_u64, cpu);
2270 spin_lock(&batched_entropy->batch_lock);
2271 batched_entropy->position = 0;
2272 spin_unlock_irqrestore(&batched_entropy->batch_lock, flags);
2273 }
2274}
2275
2276/**
2277 * randomize_page - Generate a random, page aligned address
2278 * @start: The smallest acceptable address the caller will take.
2279 * @range: The size of the area, starting at @start, within which the
2280 * random address must fall.
2281 *
2282 * If @start + @range would overflow, @range is capped.
2283 *
2284 * NOTE: Historical use of randomize_range, which this replaces, presumed that
2285 * @start was already page aligned. We now align it regardless.
2286 *
2287 * Return: A page aligned address within [start, start + range). On error,
2288 * @start is returned.
2289 */
2290unsigned long
2291randomize_page(unsigned long start, unsigned long range)
2292{
2293 if (!PAGE_ALIGNED(start)) {
2294 range -= PAGE_ALIGN(start) - start;
2295 start = PAGE_ALIGN(start);
2296 }
2297
2298 if (start > ULONG_MAX - range)
2299 range = ULONG_MAX - start;
2300
2301 range >>= PAGE_SHIFT;
2302
2303 if (range == 0)
2304 return start;
2305
2306 return start + (get_random_long() % range << PAGE_SHIFT);
2307}
2308
2309/* Interface for in-kernel drivers of true hardware RNGs.
2310 * Those devices may produce endless random bits and will be throttled
2311 * when our pool is full.
2312 */
2313void add_hwgenerator_randomness(const char *buffer, size_t count,
2314 size_t entropy)
2315{
2316 struct entropy_store *poolp = &input_pool;
2317
2318 if (unlikely(crng_init == 0)) {
2319 crng_fast_load(buffer, count);
2320 return;
2321 }
2322
2323 /* Suspend writing if we're above the trickle threshold.
2324 * We'll be woken up again once below random_write_wakeup_thresh,
2325 * or when the calling thread is about to terminate.
2326 */
2327 wait_event_interruptible(random_write_wait, kthread_should_stop() ||
2328 ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
2329 mix_pool_bytes(poolp, buffer, count);
2330 credit_entropy_bits(poolp, entropy);
2331}
2332EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);