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 four exported interfaces; two for use within the kernel,
105 * and two or use from userspace.
106 *
107 * Exported interfaces ---- userspace output
108 * -----------------------------------------
109 *
110 * The userspace interfaces are two character devices /dev/random and
111 * /dev/urandom. /dev/random is suitable for use when very high
112 * quality randomness is desired (for example, for key generation or
113 * one-time pads), as it will only return a maximum of the number of
114 * bits of randomness (as estimated by the random number generator)
115 * contained in the entropy pool.
116 *
117 * The /dev/urandom device does not have this limit, and will return
118 * as many bytes as are requested. As more and more random bytes are
119 * requested without giving time for the entropy pool to recharge,
120 * this will result in random numbers that are merely cryptographically
121 * strong. For many applications, however, this is acceptable.
122 *
123 * Exported interfaces ---- kernel output
124 * --------------------------------------
125 *
126 * The primary kernel interface is
127 *
128 * void get_random_bytes(void *buf, int nbytes);
129 *
130 * This interface will return the requested number of random bytes,
131 * and place it in the requested buffer. This is equivalent to a
132 * read from /dev/urandom.
133 *
134 * For less critical applications, there are the functions:
135 *
136 * u32 get_random_u32()
137 * u64 get_random_u64()
138 * unsigned int get_random_int()
139 * unsigned long get_random_long()
140 *
141 * These are produced by a cryptographic RNG seeded from get_random_bytes,
142 * and so do not deplete the entropy pool as much. These are recommended
143 * for most in-kernel operations *if the result is going to be stored in
144 * the kernel*.
145 *
146 * Specifically, the get_random_int() family do not attempt to do
147 * "anti-backtracking". If you capture the state of the kernel (e.g.
148 * by snapshotting the VM), you can figure out previous get_random_int()
149 * return values. But if the value is stored in the kernel anyway,
150 * this is not a problem.
151 *
152 * It *is* safe to expose get_random_int() output to attackers (e.g. as
153 * network cookies); given outputs 1..n, it's not feasible to predict
154 * outputs 0 or n+1. The only concern is an attacker who breaks into
155 * the kernel later; the get_random_int() engine is not reseeded as
156 * often as the get_random_bytes() one.
157 *
158 * get_random_bytes() is needed for keys that need to stay secret after
159 * they are erased from the kernel. For example, any key that will
160 * be wrapped and stored encrypted. And session encryption keys: we'd
161 * like to know that after the session is closed and the keys erased,
162 * the plaintext is unrecoverable to someone who recorded the ciphertext.
163 *
164 * But for network ports/cookies, stack canaries, PRNG seeds, address
165 * space layout randomization, session *authentication* keys, or other
166 * applications where the sensitive data is stored in the kernel in
167 * plaintext for as long as it's sensitive, the get_random_int() family
168 * is just fine.
169 *
170 * Consider ASLR. We want to keep the address space secret from an
171 * outside attacker while the process is running, but once the address
172 * space is torn down, it's of no use to an attacker any more. And it's
173 * stored in kernel data structures as long as it's alive, so worrying
174 * about an attacker's ability to extrapolate it from the get_random_int()
175 * CRNG is silly.
176 *
177 * Even some cryptographic keys are safe to generate with get_random_int().
178 * In particular, keys for SipHash are generally fine. Here, knowledge
179 * of the key authorizes you to do something to a kernel object (inject
180 * packets to a network connection, or flood a hash table), and the
181 * key is stored with the object being protected. Once it goes away,
182 * we no longer care if anyone knows the key.
183 *
184 * prandom_u32()
185 * -------------
186 *
187 * For even weaker applications, see the pseudorandom generator
188 * prandom_u32(), prandom_max(), and prandom_bytes(). If the random
189 * numbers aren't security-critical at all, these are *far* cheaper.
190 * Useful for self-tests, random error simulation, randomized backoffs,
191 * and any other application where you trust that nobody is trying to
192 * maliciously mess with you by guessing the "random" numbers.
193 *
194 * Exported interfaces ---- input
195 * ==============================
196 *
197 * The current exported interfaces for gathering environmental noise
198 * from the devices are:
199 *
200 * void add_device_randomness(const void *buf, unsigned int size);
201 * void add_input_randomness(unsigned int type, unsigned int code,
202 * unsigned int value);
203 * void add_interrupt_randomness(int irq, int irq_flags);
204 * void add_disk_randomness(struct gendisk *disk);
205 *
206 * add_device_randomness() is for adding data to the random pool that
207 * is likely to differ between two devices (or possibly even per boot).
208 * This would be things like MAC addresses or serial numbers, or the
209 * read-out of the RTC. This does *not* add any actual entropy to the
210 * pool, but it initializes the pool to different values for devices
211 * that might otherwise be identical and have very little entropy
212 * available to them (particularly common in the embedded world).
213 *
214 * add_input_randomness() uses the input layer interrupt timing, as well as
215 * the event type information from the hardware.
216 *
217 * add_interrupt_randomness() uses the interrupt timing as random
218 * inputs to the entropy pool. Using the cycle counters and the irq source
219 * as inputs, it feeds the randomness roughly once a second.
220 *
221 * add_disk_randomness() uses what amounts to the seek time of block
222 * layer request events, on a per-disk_devt basis, as input to the
223 * entropy pool. Note that high-speed solid state drives with very low
224 * seek times do not make for good sources of entropy, as their seek
225 * times are usually fairly consistent.
226 *
227 * All of these routines try to estimate how many bits of randomness a
228 * particular randomness source. They do this by keeping track of the
229 * first and second order deltas of the event timings.
230 *
231 * Ensuring unpredictability at system startup
232 * ============================================
233 *
234 * When any operating system starts up, it will go through a sequence
235 * of actions that are fairly predictable by an adversary, especially
236 * if the start-up does not involve interaction with a human operator.
237 * This reduces the actual number of bits of unpredictability in the
238 * entropy pool below the value in entropy_count. In order to
239 * counteract this effect, it helps to carry information in the
240 * entropy pool across shut-downs and start-ups. To do this, put the
241 * following lines an appropriate script which is run during the boot
242 * sequence:
243 *
244 * echo "Initializing random number generator..."
245 * random_seed=/var/run/random-seed
246 * # Carry a random seed from start-up to start-up
247 * # Load and then save the whole entropy pool
248 * if [ -f $random_seed ]; then
249 * cat $random_seed >/dev/urandom
250 * else
251 * touch $random_seed
252 * fi
253 * chmod 600 $random_seed
254 * dd if=/dev/urandom of=$random_seed count=1 bs=512
255 *
256 * and the following lines in an appropriate script which is run as
257 * the system is shutdown:
258 *
259 * # Carry a random seed from shut-down to start-up
260 * # Save the whole entropy pool
261 * echo "Saving random seed..."
262 * random_seed=/var/run/random-seed
263 * touch $random_seed
264 * chmod 600 $random_seed
265 * dd if=/dev/urandom of=$random_seed count=1 bs=512
266 *
267 * For example, on most modern systems using the System V init
268 * scripts, such code fragments would be found in
269 * /etc/rc.d/init.d/random. On older Linux systems, the correct script
270 * location might be in /etc/rcb.d/rc.local or /etc/rc.d/rc.0.
271 *
272 * Effectively, these commands cause the contents of the entropy pool
273 * to be saved at shut-down time and reloaded into the entropy pool at
274 * start-up. (The 'dd' in the addition to the bootup script is to
275 * make sure that /etc/random-seed is different for every start-up,
276 * even if the system crashes without executing rc.0.) Even with
277 * complete knowledge of the start-up activities, predicting the state
278 * of the entropy pool requires knowledge of the previous history of
279 * the system.
280 *
281 * Configuring the /dev/random driver under Linux
282 * ==============================================
283 *
284 * The /dev/random driver under Linux uses minor numbers 8 and 9 of
285 * the /dev/mem major number (#1). So if your system does not have
286 * /dev/random and /dev/urandom created already, they can be created
287 * by using the commands:
288 *
289 * mknod /dev/random c 1 8
290 * mknod /dev/urandom c 1 9
291 *
292 * Acknowledgements:
293 * =================
294 *
295 * Ideas for constructing this random number generator were derived
296 * from Pretty Good Privacy's random number generator, and from private
297 * discussions with Phil Karn. Colin Plumb provided a faster random
298 * number generator, which speed up the mixing function of the entropy
299 * pool, taken from PGPfone. Dale Worley has also contributed many
300 * useful ideas and suggestions to improve this driver.
301 *
302 * Any flaws in the design are solely my responsibility, and should
303 * not be attributed to the Phil, Colin, or any of authors of PGP.
304 *
305 * Further background information on this topic may be obtained from
306 * RFC 1750, "Randomness Recommendations for Security", by Donald
307 * Eastlake, Steve Crocker, and Jeff Schiller.
308 */
309
310 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
311
312 #include <linux/utsname.h>
313 #include <linux/module.h>
314 #include <linux/kernel.h>
315 #include <linux/major.h>
316 #include <linux/string.h>
317 #include <linux/fcntl.h>
318 #include <linux/slab.h>
319 #include <linux/random.h>
320 #include <linux/poll.h>
321 #include <linux/init.h>
322 #include <linux/fs.h>
323 #include <linux/genhd.h>
324 #include <linux/interrupt.h>
325 #include <linux/mm.h>
326 #include <linux/nodemask.h>
327 #include <linux/spinlock.h>
328 #include <linux/kthread.h>
329 #include <linux/percpu.h>
330 #include <linux/fips.h>
331 #include <linux/ptrace.h>
332 #include <linux/workqueue.h>
333 #include <linux/irq.h>
334 #include <linux/ratelimit.h>
335 #include <linux/syscalls.h>
336 #include <linux/completion.h>
337 #include <linux/uuid.h>
338 #include <crypto/chacha.h>
339 #include <crypto/sha.h>
340
341 #include <asm/processor.h>
342 #include <linux/uaccess.h>
343 #include <asm/irq.h>
344 #include <asm/irq_regs.h>
345 #include <asm/io.h>
346
347 #define CREATE_TRACE_POINTS
348 #include <trace/events/random.h>
349
350 /* #define ADD_INTERRUPT_BENCH */
351
352 /*
353 * Configuration information
354 */
355 #define INPUT_POOL_SHIFT 12
356 #define INPUT_POOL_WORDS (1 << (INPUT_POOL_SHIFT-5))
357 #define OUTPUT_POOL_SHIFT 10
358 #define OUTPUT_POOL_WORDS (1 << (OUTPUT_POOL_SHIFT-5))
359 #define EXTRACT_SIZE 10
360
361
362 #define LONGS(x) (((x) + sizeof(unsigned long) - 1)/sizeof(unsigned long))
363
364 /*
365 * To allow fractional bits to be tracked, the entropy_count field is
366 * denominated in units of 1/8th bits.
367 *
368 * 2*(ENTROPY_SHIFT + poolbitshift) must <= 31, or the multiply in
369 * credit_entropy_bits() needs to be 64 bits wide.
370 */
371 #define ENTROPY_SHIFT 3
372 #define ENTROPY_BITS(r) ((r)->entropy_count >> ENTROPY_SHIFT)
373
374 /*
375 * If the entropy count falls under this number of bits, then we
376 * should wake up processes which are selecting or polling on write
377 * access to /dev/random.
378 */
379 static int random_write_wakeup_bits = 28 * OUTPUT_POOL_WORDS;
380
381 /*
382 * Originally, we used a primitive polynomial of degree .poolwords
383 * over GF(2). The taps for various sizes are defined below. They
384 * were chosen to be evenly spaced except for the last tap, which is 1
385 * to get the twisting happening as fast as possible.
386 *
387 * For the purposes of better mixing, we use the CRC-32 polynomial as
388 * well to make a (modified) twisted Generalized Feedback Shift
389 * Register. (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR
390 * generators. ACM Transactions on Modeling and Computer Simulation
391 * 2(3):179-194. Also see M. Matsumoto & Y. Kurita, 1994. Twisted
392 * GFSR generators II. ACM Transactions on Modeling and Computer
393 * Simulation 4:254-266)
394 *
395 * Thanks to Colin Plumb for suggesting this.
396 *
397 * The mixing operation is much less sensitive than the output hash,
398 * where we use SHA-1. All that we want of mixing operation is that
399 * it be a good non-cryptographic hash; i.e. it not produce collisions
400 * when fed "random" data of the sort we expect to see. As long as
401 * the pool state differs for different inputs, we have preserved the
402 * input entropy and done a good job. The fact that an intelligent
403 * attacker can construct inputs that will produce controlled
404 * alterations to the pool's state is not important because we don't
405 * consider such inputs to contribute any randomness. The only
406 * property we need with respect to them is that the attacker can't
407 * increase his/her knowledge of the pool's state. Since all
408 * additions are reversible (knowing the final state and the input,
409 * you can reconstruct the initial state), if an attacker has any
410 * uncertainty about the initial state, he/she can only shuffle that
411 * uncertainty about, but never cause any collisions (which would
412 * decrease the uncertainty).
413 *
414 * Our mixing functions were analyzed by Lacharme, Roeck, Strubel, and
415 * Videau in their paper, "The Linux Pseudorandom Number Generator
416 * Revisited" (see: http://eprint.iacr.org/2012/251.pdf). In their
417 * paper, they point out that we are not using a true Twisted GFSR,
418 * since Matsumoto & Kurita used a trinomial feedback polynomial (that
419 * is, with only three taps, instead of the six that we are using).
420 * As a result, the resulting polynomial is neither primitive nor
421 * irreducible, and hence does not have a maximal period over
422 * GF(2**32). They suggest a slight change to the generator
423 * polynomial which improves the resulting TGFSR polynomial to be
424 * irreducible, which we have made here.
425 */
426 static const struct poolinfo {
427 int poolbitshift, poolwords, poolbytes, poolfracbits;
428 #define S(x) ilog2(x)+5, (x), (x)*4, (x) << (ENTROPY_SHIFT+5)
429 int tap1, tap2, tap3, tap4, tap5;
430 } poolinfo_table[] = {
431 /* was: x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 */
432 /* x^128 + x^104 + x^76 + x^51 +x^25 + x + 1 */
433 { S(128), 104, 76, 51, 25, 1 },
434 };
435
436 /*
437 * Static global variables
438 */
439 static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
440 static struct fasync_struct *fasync;
441
442 static DEFINE_SPINLOCK(random_ready_list_lock);
443 static LIST_HEAD(random_ready_list);
444
445 struct crng_state {
446 __u32 state[16];
447 unsigned long init_time;
448 spinlock_t lock;
449 };
450
451 static struct crng_state primary_crng = {
452 .lock = __SPIN_LOCK_UNLOCKED(primary_crng.lock),
453 };
454
455 /*
456 * crng_init = 0 --> Uninitialized
457 * 1 --> Initialized
458 * 2 --> Initialized from input_pool
459 *
460 * crng_init is protected by primary_crng->lock, and only increases
461 * its value (from 0->1->2).
462 */
463 static int crng_init = 0;
464 static bool crng_need_final_init = false;
465 #define crng_ready() (likely(crng_init > 1))
466 static int crng_init_cnt = 0;
467 static unsigned long crng_global_init_time = 0;
468 #define CRNG_INIT_CNT_THRESH (2*CHACHA_KEY_SIZE)
469 static void _extract_crng(struct crng_state *crng, __u8 out[CHACHA_BLOCK_SIZE]);
470 static void _crng_backtrack_protect(struct crng_state *crng,
471 __u8 tmp[CHACHA_BLOCK_SIZE], int used);
472 static void process_random_ready_list(void);
473 static void _get_random_bytes(void *buf, int nbytes);
474
475 static struct ratelimit_state unseeded_warning =
476 RATELIMIT_STATE_INIT("warn_unseeded_randomness", HZ, 3);
477 static struct ratelimit_state urandom_warning =
478 RATELIMIT_STATE_INIT("warn_urandom_randomness", HZ, 3);
479
480 static int ratelimit_disable __read_mostly;
481
482 module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
483 MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
484
485 /**********************************************************************
486 *
487 * OS independent entropy store. Here are the functions which handle
488 * storing entropy in an entropy pool.
489 *
490 **********************************************************************/
491
492 struct entropy_store;
493 struct entropy_store {
494 /* read-only data: */
495 const struct poolinfo *poolinfo;
496 __u32 *pool;
497 const char *name;
498
499 /* read-write data: */
500 spinlock_t lock;
501 unsigned short add_ptr;
502 unsigned short input_rotate;
503 int entropy_count;
504 unsigned int initialized:1;
505 unsigned int last_data_init:1;
506 __u8 last_data[EXTRACT_SIZE];
507 };
508
509 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
510 size_t nbytes, int min, int rsvd);
511 static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
512 size_t nbytes, int fips);
513
514 static void crng_reseed(struct crng_state *crng, struct entropy_store *r);
515 static __u32 input_pool_data[INPUT_POOL_WORDS] __latent_entropy;
516
517 static struct entropy_store input_pool = {
518 .poolinfo = &poolinfo_table[0],
519 .name = "input",
520 .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
521 .pool = input_pool_data
522 };
523
524 static __u32 const twist_table[8] = {
525 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
526 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
527
528 /*
529 * This function adds bytes into the entropy "pool". It does not
530 * update the entropy estimate. The caller should call
531 * credit_entropy_bits if this is appropriate.
532 *
533 * The pool is stirred with a primitive polynomial of the appropriate
534 * degree, and then twisted. We twist by three bits at a time because
535 * it's cheap to do so and helps slightly in the expected case where
536 * the entropy is concentrated in the low-order bits.
537 */
_mix_pool_bytes(struct entropy_store * r,const void * in,int nbytes)538 static void _mix_pool_bytes(struct entropy_store *r, const void *in,
539 int nbytes)
540 {
541 unsigned long i, tap1, tap2, tap3, tap4, tap5;
542 int input_rotate;
543 int wordmask = r->poolinfo->poolwords - 1;
544 const char *bytes = in;
545 __u32 w;
546
547 tap1 = r->poolinfo->tap1;
548 tap2 = r->poolinfo->tap2;
549 tap3 = r->poolinfo->tap3;
550 tap4 = r->poolinfo->tap4;
551 tap5 = r->poolinfo->tap5;
552
553 input_rotate = r->input_rotate;
554 i = r->add_ptr;
555
556 /* mix one byte at a time to simplify size handling and churn faster */
557 while (nbytes--) {
558 w = rol32(*bytes++, input_rotate);
559 i = (i - 1) & wordmask;
560
561 /* XOR in the various taps */
562 w ^= r->pool[i];
563 w ^= r->pool[(i + tap1) & wordmask];
564 w ^= r->pool[(i + tap2) & wordmask];
565 w ^= r->pool[(i + tap3) & wordmask];
566 w ^= r->pool[(i + tap4) & wordmask];
567 w ^= r->pool[(i + tap5) & wordmask];
568
569 /* Mix the result back in with a twist */
570 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
571
572 /*
573 * Normally, we add 7 bits of rotation to the pool.
574 * At the beginning of the pool, add an extra 7 bits
575 * rotation, so that successive passes spread the
576 * input bits across the pool evenly.
577 */
578 input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
579 }
580
581 r->input_rotate = input_rotate;
582 r->add_ptr = i;
583 }
584
__mix_pool_bytes(struct entropy_store * r,const void * in,int nbytes)585 static void __mix_pool_bytes(struct entropy_store *r, const void *in,
586 int nbytes)
587 {
588 trace_mix_pool_bytes_nolock(r->name, nbytes, _RET_IP_);
589 _mix_pool_bytes(r, in, nbytes);
590 }
591
mix_pool_bytes(struct entropy_store * r,const void * in,int nbytes)592 static void mix_pool_bytes(struct entropy_store *r, const void *in,
593 int nbytes)
594 {
595 unsigned long flags;
596
597 trace_mix_pool_bytes(r->name, nbytes, _RET_IP_);
598 spin_lock_irqsave(&r->lock, flags);
599 _mix_pool_bytes(r, in, nbytes);
600 spin_unlock_irqrestore(&r->lock, flags);
601 }
602
603 struct fast_pool {
604 __u32 pool[4];
605 unsigned long last;
606 unsigned short reg_idx;
607 unsigned char count;
608 };
609
610 /*
611 * This is a fast mixing routine used by the interrupt randomness
612 * collector. It's hardcoded for an 128 bit pool and assumes that any
613 * locks that might be needed are taken by the caller.
614 */
fast_mix(struct fast_pool * f)615 static void fast_mix(struct fast_pool *f)
616 {
617 __u32 a = f->pool[0], b = f->pool[1];
618 __u32 c = f->pool[2], d = f->pool[3];
619
620 a += b; c += d;
621 b = rol32(b, 6); d = rol32(d, 27);
622 d ^= a; b ^= c;
623
624 a += b; c += d;
625 b = rol32(b, 16); d = rol32(d, 14);
626 d ^= a; b ^= c;
627
628 a += b; c += d;
629 b = rol32(b, 6); d = rol32(d, 27);
630 d ^= a; b ^= c;
631
632 a += b; c += d;
633 b = rol32(b, 16); d = rol32(d, 14);
634 d ^= a; b ^= c;
635
636 f->pool[0] = a; f->pool[1] = b;
637 f->pool[2] = c; f->pool[3] = d;
638 f->count++;
639 }
640
process_random_ready_list(void)641 static void process_random_ready_list(void)
642 {
643 unsigned long flags;
644 struct random_ready_callback *rdy, *tmp;
645
646 spin_lock_irqsave(&random_ready_list_lock, flags);
647 list_for_each_entry_safe(rdy, tmp, &random_ready_list, list) {
648 struct module *owner = rdy->owner;
649
650 list_del_init(&rdy->list);
651 rdy->func(rdy);
652 module_put(owner);
653 }
654 spin_unlock_irqrestore(&random_ready_list_lock, flags);
655 }
656
657 /*
658 * Credit (or debit) the entropy store with n bits of entropy.
659 * Use credit_entropy_bits_safe() if the value comes from userspace
660 * or otherwise should be checked for extreme values.
661 */
credit_entropy_bits(struct entropy_store * r,int nbits)662 static void credit_entropy_bits(struct entropy_store *r, int nbits)
663 {
664 int entropy_count, orig, has_initialized = 0;
665 const int pool_size = r->poolinfo->poolfracbits;
666 int nfrac = nbits << ENTROPY_SHIFT;
667
668 if (!nbits)
669 return;
670
671 retry:
672 entropy_count = orig = READ_ONCE(r->entropy_count);
673 if (nfrac < 0) {
674 /* Debit */
675 entropy_count += nfrac;
676 } else {
677 /*
678 * Credit: we have to account for the possibility of
679 * overwriting already present entropy. Even in the
680 * ideal case of pure Shannon entropy, new contributions
681 * approach the full value asymptotically:
682 *
683 * entropy <- entropy + (pool_size - entropy) *
684 * (1 - exp(-add_entropy/pool_size))
685 *
686 * For add_entropy <= pool_size/2 then
687 * (1 - exp(-add_entropy/pool_size)) >=
688 * (add_entropy/pool_size)*0.7869...
689 * so we can approximate the exponential with
690 * 3/4*add_entropy/pool_size and still be on the
691 * safe side by adding at most pool_size/2 at a time.
692 *
693 * The use of pool_size-2 in the while statement is to
694 * prevent rounding artifacts from making the loop
695 * arbitrarily long; this limits the loop to log2(pool_size)*2
696 * turns no matter how large nbits is.
697 */
698 int pnfrac = nfrac;
699 const int s = r->poolinfo->poolbitshift + ENTROPY_SHIFT + 2;
700 /* The +2 corresponds to the /4 in the denominator */
701
702 do {
703 unsigned int anfrac = min(pnfrac, pool_size/2);
704 unsigned int add =
705 ((pool_size - entropy_count)*anfrac*3) >> s;
706
707 entropy_count += add;
708 pnfrac -= anfrac;
709 } while (unlikely(entropy_count < pool_size-2 && pnfrac));
710 }
711
712 if (WARN_ON(entropy_count < 0)) {
713 pr_warn("negative entropy/overflow: pool %s count %d\n",
714 r->name, entropy_count);
715 entropy_count = 0;
716 } else if (entropy_count > pool_size)
717 entropy_count = pool_size;
718 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
719 goto retry;
720
721 if (has_initialized) {
722 r->initialized = 1;
723 kill_fasync(&fasync, SIGIO, POLL_IN);
724 }
725
726 trace_credit_entropy_bits(r->name, nbits,
727 entropy_count >> ENTROPY_SHIFT, _RET_IP_);
728
729 if (r == &input_pool) {
730 int entropy_bits = entropy_count >> ENTROPY_SHIFT;
731
732 if (crng_init < 2) {
733 if (entropy_bits < 128)
734 return;
735 crng_reseed(&primary_crng, r);
736 entropy_bits = ENTROPY_BITS(r);
737 }
738 }
739 }
740
credit_entropy_bits_safe(struct entropy_store * r,int nbits)741 static int credit_entropy_bits_safe(struct entropy_store *r, int nbits)
742 {
743 const int nbits_max = r->poolinfo->poolwords * 32;
744
745 if (nbits < 0)
746 return -EINVAL;
747
748 /* Cap the value to avoid overflows */
749 nbits = min(nbits, nbits_max);
750
751 credit_entropy_bits(r, nbits);
752 return 0;
753 }
754
755 /*********************************************************************
756 *
757 * CRNG using CHACHA20
758 *
759 *********************************************************************/
760
761 #define CRNG_RESEED_INTERVAL (300*HZ)
762
763 static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
764
765 #ifdef CONFIG_NUMA
766 /*
767 * Hack to deal with crazy userspace progams when they are all trying
768 * to access /dev/urandom in parallel. The programs are almost
769 * certainly doing something terribly wrong, but we'll work around
770 * their brain damage.
771 */
772 static struct crng_state **crng_node_pool __read_mostly;
773 #endif
774
775 static void invalidate_batched_entropy(void);
776 static void numa_crng_init(void);
777
778 static bool trust_cpu __ro_after_init = IS_ENABLED(CONFIG_RANDOM_TRUST_CPU);
parse_trust_cpu(char * arg)779 static int __init parse_trust_cpu(char *arg)
780 {
781 return kstrtobool(arg, &trust_cpu);
782 }
783 early_param("random.trust_cpu", parse_trust_cpu);
784
crng_init_try_arch(struct crng_state * crng)785 static bool crng_init_try_arch(struct crng_state *crng)
786 {
787 int i;
788 bool arch_init = true;
789 unsigned long rv;
790
791 for (i = 4; i < 16; i++) {
792 if (!arch_get_random_seed_long(&rv) &&
793 !arch_get_random_long(&rv)) {
794 rv = random_get_entropy();
795 arch_init = false;
796 }
797 crng->state[i] ^= rv;
798 }
799
800 return arch_init;
801 }
802
crng_init_try_arch_early(struct crng_state * crng)803 static bool __init crng_init_try_arch_early(struct crng_state *crng)
804 {
805 int i;
806 bool arch_init = true;
807 unsigned long rv;
808
809 for (i = 4; i < 16; i++) {
810 if (!arch_get_random_seed_long_early(&rv) &&
811 !arch_get_random_long_early(&rv)) {
812 rv = random_get_entropy();
813 arch_init = false;
814 }
815 crng->state[i] ^= rv;
816 }
817
818 return arch_init;
819 }
820
crng_initialize_secondary(struct crng_state * crng)821 static void __maybe_unused crng_initialize_secondary(struct crng_state *crng)
822 {
823 chacha_init_consts(crng->state);
824 _get_random_bytes(&crng->state[4], sizeof(__u32) * 12);
825 crng_init_try_arch(crng);
826 crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
827 }
828
crng_initialize_primary(struct crng_state * crng)829 static void __init crng_initialize_primary(struct crng_state *crng)
830 {
831 chacha_init_consts(crng->state);
832 _extract_entropy(&input_pool, &crng->state[4], sizeof(__u32) * 12, 0);
833 if (crng_init_try_arch_early(crng) && trust_cpu) {
834 invalidate_batched_entropy();
835 numa_crng_init();
836 crng_init = 2;
837 pr_notice("crng done (trusting CPU's manufacturer)\n");
838 }
839 crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
840 }
841
crng_finalize_init(struct crng_state * crng)842 static void crng_finalize_init(struct crng_state *crng)
843 {
844 if (crng != &primary_crng || crng_init >= 2)
845 return;
846 if (!system_wq) {
847 /* We can't call numa_crng_init until we have workqueues,
848 * so mark this for processing later. */
849 crng_need_final_init = true;
850 return;
851 }
852
853 invalidate_batched_entropy();
854 numa_crng_init();
855 crng_init = 2;
856 process_random_ready_list();
857 wake_up_interruptible(&crng_init_wait);
858 kill_fasync(&fasync, SIGIO, POLL_IN);
859 pr_notice("crng init done\n");
860 if (unseeded_warning.missed) {
861 pr_notice("%d get_random_xx warning(s) missed due to ratelimiting\n",
862 unseeded_warning.missed);
863 unseeded_warning.missed = 0;
864 }
865 if (urandom_warning.missed) {
866 pr_notice("%d urandom warning(s) missed due to ratelimiting\n",
867 urandom_warning.missed);
868 urandom_warning.missed = 0;
869 }
870 }
871
872 #ifdef CONFIG_NUMA
do_numa_crng_init(struct work_struct * work)873 static void do_numa_crng_init(struct work_struct *work)
874 {
875 int i;
876 struct crng_state *crng;
877 struct crng_state **pool;
878
879 pool = kcalloc(nr_node_ids, sizeof(*pool), GFP_KERNEL|__GFP_NOFAIL);
880 for_each_online_node(i) {
881 crng = kmalloc_node(sizeof(struct crng_state),
882 GFP_KERNEL | __GFP_NOFAIL, i);
883 spin_lock_init(&crng->lock);
884 crng_initialize_secondary(crng);
885 pool[i] = crng;
886 }
887 /* pairs with READ_ONCE() in select_crng() */
888 if (cmpxchg_release(&crng_node_pool, NULL, pool) != NULL) {
889 for_each_node(i)
890 kfree(pool[i]);
891 kfree(pool);
892 }
893 }
894
895 static DECLARE_WORK(numa_crng_init_work, do_numa_crng_init);
896
numa_crng_init(void)897 static void numa_crng_init(void)
898 {
899 schedule_work(&numa_crng_init_work);
900 }
901
select_crng(void)902 static struct crng_state *select_crng(void)
903 {
904 struct crng_state **pool;
905 int nid = numa_node_id();
906
907 /* pairs with cmpxchg_release() in do_numa_crng_init() */
908 pool = READ_ONCE(crng_node_pool);
909 if (pool && pool[nid])
910 return pool[nid];
911
912 return &primary_crng;
913 }
914 #else
numa_crng_init(void)915 static void numa_crng_init(void) {}
916
select_crng(void)917 static struct crng_state *select_crng(void)
918 {
919 return &primary_crng;
920 }
921 #endif
922
923 /*
924 * crng_fast_load() can be called by code in the interrupt service
925 * path. So we can't afford to dilly-dally. Returns the number of
926 * bytes processed from cp.
927 */
crng_fast_load(const char * cp,size_t len)928 static size_t crng_fast_load(const char *cp, size_t len)
929 {
930 unsigned long flags;
931 char *p;
932 size_t ret = 0;
933
934 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
935 return 0;
936 if (crng_init != 0) {
937 spin_unlock_irqrestore(&primary_crng.lock, flags);
938 return 0;
939 }
940 p = (unsigned char *) &primary_crng.state[4];
941 while (len > 0 && crng_init_cnt < CRNG_INIT_CNT_THRESH) {
942 p[crng_init_cnt % CHACHA_KEY_SIZE] ^= *cp;
943 cp++; crng_init_cnt++; len--; ret++;
944 }
945 spin_unlock_irqrestore(&primary_crng.lock, flags);
946 if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
947 invalidate_batched_entropy();
948 crng_init = 1;
949 pr_notice("fast init done\n");
950 }
951 return ret;
952 }
953
954 /*
955 * crng_slow_load() is called by add_device_randomness, which has two
956 * attributes. (1) We can't trust the buffer passed to it is
957 * guaranteed to be unpredictable (so it might not have any entropy at
958 * all), and (2) it doesn't have the performance constraints of
959 * crng_fast_load().
960 *
961 * So we do something more comprehensive which is guaranteed to touch
962 * all of the primary_crng's state, and which uses a LFSR with a
963 * period of 255 as part of the mixing algorithm. Finally, we do
964 * *not* advance crng_init_cnt since buffer we may get may be something
965 * like a fixed DMI table (for example), which might very well be
966 * unique to the machine, but is otherwise unvarying.
967 */
crng_slow_load(const char * cp,size_t len)968 static int crng_slow_load(const char *cp, size_t len)
969 {
970 unsigned long flags;
971 static unsigned char lfsr = 1;
972 unsigned char tmp;
973 unsigned i, max = CHACHA_KEY_SIZE;
974 const char * src_buf = cp;
975 char * dest_buf = (char *) &primary_crng.state[4];
976
977 if (!spin_trylock_irqsave(&primary_crng.lock, flags))
978 return 0;
979 if (crng_init != 0) {
980 spin_unlock_irqrestore(&primary_crng.lock, flags);
981 return 0;
982 }
983 if (len > max)
984 max = len;
985
986 for (i = 0; i < max ; i++) {
987 tmp = lfsr;
988 lfsr >>= 1;
989 if (tmp & 1)
990 lfsr ^= 0xE1;
991 tmp = dest_buf[i % CHACHA_KEY_SIZE];
992 dest_buf[i % CHACHA_KEY_SIZE] ^= src_buf[i % len] ^ lfsr;
993 lfsr += (tmp << 3) | (tmp >> 5);
994 }
995 spin_unlock_irqrestore(&primary_crng.lock, flags);
996 return 1;
997 }
998
crng_reseed(struct crng_state * crng,struct entropy_store * r)999 static void crng_reseed(struct crng_state *crng, struct entropy_store *r)
1000 {
1001 unsigned long flags;
1002 int i, num;
1003 union {
1004 __u8 block[CHACHA_BLOCK_SIZE];
1005 __u32 key[8];
1006 } buf;
1007
1008 if (r) {
1009 num = extract_entropy(r, &buf, 32, 16, 0);
1010 if (num == 0)
1011 return;
1012 } else {
1013 _extract_crng(&primary_crng, buf.block);
1014 _crng_backtrack_protect(&primary_crng, buf.block,
1015 CHACHA_KEY_SIZE);
1016 }
1017 spin_lock_irqsave(&crng->lock, flags);
1018 for (i = 0; i < 8; i++) {
1019 unsigned long rv;
1020 if (!arch_get_random_seed_long(&rv) &&
1021 !arch_get_random_long(&rv))
1022 rv = random_get_entropy();
1023 crng->state[i+4] ^= buf.key[i] ^ rv;
1024 }
1025 memzero_explicit(&buf, sizeof(buf));
1026 WRITE_ONCE(crng->init_time, jiffies);
1027 spin_unlock_irqrestore(&crng->lock, flags);
1028 crng_finalize_init(crng);
1029 }
1030
_extract_crng(struct crng_state * crng,__u8 out[CHACHA_BLOCK_SIZE])1031 static void _extract_crng(struct crng_state *crng,
1032 __u8 out[CHACHA_BLOCK_SIZE])
1033 {
1034 unsigned long v, flags, init_time;
1035
1036 if (crng_ready()) {
1037 init_time = READ_ONCE(crng->init_time);
1038 if (time_after(READ_ONCE(crng_global_init_time), init_time) ||
1039 time_after(jiffies, init_time + CRNG_RESEED_INTERVAL))
1040 crng_reseed(crng, crng == &primary_crng ?
1041 &input_pool : NULL);
1042 }
1043 spin_lock_irqsave(&crng->lock, flags);
1044 if (arch_get_random_long(&v))
1045 crng->state[14] ^= v;
1046 chacha20_block(&crng->state[0], out);
1047 if (crng->state[12] == 0)
1048 crng->state[13]++;
1049 spin_unlock_irqrestore(&crng->lock, flags);
1050 }
1051
extract_crng(__u8 out[CHACHA_BLOCK_SIZE])1052 static void extract_crng(__u8 out[CHACHA_BLOCK_SIZE])
1053 {
1054 _extract_crng(select_crng(), out);
1055 }
1056
1057 /*
1058 * Use the leftover bytes from the CRNG block output (if there is
1059 * enough) to mutate the CRNG key to provide backtracking protection.
1060 */
_crng_backtrack_protect(struct crng_state * crng,__u8 tmp[CHACHA_BLOCK_SIZE],int used)1061 static void _crng_backtrack_protect(struct crng_state *crng,
1062 __u8 tmp[CHACHA_BLOCK_SIZE], int used)
1063 {
1064 unsigned long flags;
1065 __u32 *s, *d;
1066 int i;
1067
1068 used = round_up(used, sizeof(__u32));
1069 if (used + CHACHA_KEY_SIZE > CHACHA_BLOCK_SIZE) {
1070 extract_crng(tmp);
1071 used = 0;
1072 }
1073 spin_lock_irqsave(&crng->lock, flags);
1074 s = (__u32 *) &tmp[used];
1075 d = &crng->state[4];
1076 for (i=0; i < 8; i++)
1077 *d++ ^= *s++;
1078 spin_unlock_irqrestore(&crng->lock, flags);
1079 }
1080
crng_backtrack_protect(__u8 tmp[CHACHA_BLOCK_SIZE],int used)1081 static void crng_backtrack_protect(__u8 tmp[CHACHA_BLOCK_SIZE], int used)
1082 {
1083 _crng_backtrack_protect(select_crng(), tmp, used);
1084 }
1085
extract_crng_user(void __user * buf,size_t nbytes)1086 static ssize_t extract_crng_user(void __user *buf, size_t nbytes)
1087 {
1088 ssize_t ret = 0, i = CHACHA_BLOCK_SIZE;
1089 __u8 tmp[CHACHA_BLOCK_SIZE] __aligned(4);
1090 int large_request = (nbytes > 256);
1091
1092 while (nbytes) {
1093 if (large_request && need_resched()) {
1094 if (signal_pending(current)) {
1095 if (ret == 0)
1096 ret = -ERESTARTSYS;
1097 break;
1098 }
1099 schedule();
1100 }
1101
1102 extract_crng(tmp);
1103 i = min_t(int, nbytes, CHACHA_BLOCK_SIZE);
1104 if (copy_to_user(buf, tmp, i)) {
1105 ret = -EFAULT;
1106 break;
1107 }
1108
1109 nbytes -= i;
1110 buf += i;
1111 ret += i;
1112 }
1113 crng_backtrack_protect(tmp, i);
1114
1115 /* Wipe data just written to memory */
1116 memzero_explicit(tmp, sizeof(tmp));
1117
1118 return ret;
1119 }
1120
1121
1122 /*********************************************************************
1123 *
1124 * Entropy input management
1125 *
1126 *********************************************************************/
1127
1128 /* There is one of these per entropy source */
1129 struct timer_rand_state {
1130 cycles_t last_time;
1131 long last_delta, last_delta2;
1132 };
1133
1134 #define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
1135
1136 /*
1137 * Add device- or boot-specific data to the input pool to help
1138 * initialize it.
1139 *
1140 * None of this adds any entropy; it is meant to avoid the problem of
1141 * the entropy pool having similar initial state across largely
1142 * identical devices.
1143 */
add_device_randomness(const void * buf,unsigned int size)1144 void add_device_randomness(const void *buf, unsigned int size)
1145 {
1146 unsigned long time = random_get_entropy() ^ jiffies;
1147 unsigned long flags;
1148
1149 if (!crng_ready() && size)
1150 crng_slow_load(buf, size);
1151
1152 trace_add_device_randomness(size, _RET_IP_);
1153 spin_lock_irqsave(&input_pool.lock, flags);
1154 _mix_pool_bytes(&input_pool, buf, size);
1155 _mix_pool_bytes(&input_pool, &time, sizeof(time));
1156 spin_unlock_irqrestore(&input_pool.lock, flags);
1157 }
1158 EXPORT_SYMBOL(add_device_randomness);
1159
1160 static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
1161
1162 /*
1163 * This function adds entropy to the entropy "pool" by using timing
1164 * delays. It uses the timer_rand_state structure to make an estimate
1165 * of how many bits of entropy this call has added to the pool.
1166 *
1167 * The number "num" is also added to the pool - it should somehow describe
1168 * the type of event which just happened. This is currently 0-255 for
1169 * keyboard scan codes, and 256 upwards for interrupts.
1170 *
1171 */
add_timer_randomness(struct timer_rand_state * state,unsigned num)1172 static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
1173 {
1174 struct entropy_store *r;
1175 struct {
1176 long jiffies;
1177 unsigned cycles;
1178 unsigned num;
1179 } sample;
1180 long delta, delta2, delta3;
1181
1182 sample.jiffies = jiffies;
1183 sample.cycles = random_get_entropy();
1184 sample.num = num;
1185 r = &input_pool;
1186 mix_pool_bytes(r, &sample, sizeof(sample));
1187
1188 /*
1189 * Calculate number of bits of randomness we probably added.
1190 * We take into account the first, second and third-order deltas
1191 * in order to make our estimate.
1192 */
1193 delta = sample.jiffies - READ_ONCE(state->last_time);
1194 WRITE_ONCE(state->last_time, sample.jiffies);
1195
1196 delta2 = delta - READ_ONCE(state->last_delta);
1197 WRITE_ONCE(state->last_delta, delta);
1198
1199 delta3 = delta2 - READ_ONCE(state->last_delta2);
1200 WRITE_ONCE(state->last_delta2, delta2);
1201
1202 if (delta < 0)
1203 delta = -delta;
1204 if (delta2 < 0)
1205 delta2 = -delta2;
1206 if (delta3 < 0)
1207 delta3 = -delta3;
1208 if (delta > delta2)
1209 delta = delta2;
1210 if (delta > delta3)
1211 delta = delta3;
1212
1213 /*
1214 * delta is now minimum absolute delta.
1215 * Round down by 1 bit on general principles,
1216 * and limit entropy estimate to 12 bits.
1217 */
1218 credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
1219 }
1220
add_input_randomness(unsigned int type,unsigned int code,unsigned int value)1221 void add_input_randomness(unsigned int type, unsigned int code,
1222 unsigned int value)
1223 {
1224 static unsigned char last_value;
1225
1226 /* ignore autorepeat and the like */
1227 if (value == last_value)
1228 return;
1229
1230 last_value = value;
1231 add_timer_randomness(&input_timer_state,
1232 (type << 4) ^ code ^ (code >> 4) ^ value);
1233 trace_add_input_randomness(ENTROPY_BITS(&input_pool));
1234 }
1235 EXPORT_SYMBOL_GPL(add_input_randomness);
1236
1237 static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
1238
1239 #ifdef ADD_INTERRUPT_BENCH
1240 static unsigned long avg_cycles, avg_deviation;
1241
1242 #define AVG_SHIFT 8 /* Exponential average factor k=1/256 */
1243 #define FIXED_1_2 (1 << (AVG_SHIFT-1))
1244
add_interrupt_bench(cycles_t start)1245 static void add_interrupt_bench(cycles_t start)
1246 {
1247 long delta = random_get_entropy() - start;
1248
1249 /* Use a weighted moving average */
1250 delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
1251 avg_cycles += delta;
1252 /* And average deviation */
1253 delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
1254 avg_deviation += delta;
1255 }
1256 #else
1257 #define add_interrupt_bench(x)
1258 #endif
1259
get_reg(struct fast_pool * f,struct pt_regs * regs)1260 static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
1261 {
1262 __u32 *ptr = (__u32 *) regs;
1263 unsigned int idx;
1264
1265 if (regs == NULL)
1266 return 0;
1267 idx = READ_ONCE(f->reg_idx);
1268 if (idx >= sizeof(struct pt_regs) / sizeof(__u32))
1269 idx = 0;
1270 ptr += idx++;
1271 WRITE_ONCE(f->reg_idx, idx);
1272 return *ptr;
1273 }
1274
add_interrupt_randomness(int irq,int irq_flags)1275 void add_interrupt_randomness(int irq, int irq_flags)
1276 {
1277 struct entropy_store *r;
1278 struct fast_pool *fast_pool = this_cpu_ptr(&irq_randomness);
1279 struct pt_regs *regs = get_irq_regs();
1280 unsigned long now = jiffies;
1281 cycles_t cycles = random_get_entropy();
1282 __u32 c_high, j_high;
1283 __u64 ip;
1284 unsigned long seed;
1285 int credit = 0;
1286
1287 if (cycles == 0)
1288 cycles = get_reg(fast_pool, regs);
1289 c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
1290 j_high = (sizeof(now) > 4) ? now >> 32 : 0;
1291 fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
1292 fast_pool->pool[1] ^= now ^ c_high;
1293 ip = regs ? instruction_pointer(regs) : _RET_IP_;
1294 fast_pool->pool[2] ^= ip;
1295 fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
1296 get_reg(fast_pool, regs);
1297
1298 fast_mix(fast_pool);
1299 add_interrupt_bench(cycles);
1300
1301 if (unlikely(crng_init == 0)) {
1302 if ((fast_pool->count >= 64) &&
1303 crng_fast_load((char *) fast_pool->pool,
1304 sizeof(fast_pool->pool)) > 0) {
1305 fast_pool->count = 0;
1306 fast_pool->last = now;
1307 }
1308 return;
1309 }
1310
1311 if ((fast_pool->count < 64) &&
1312 !time_after(now, fast_pool->last + HZ))
1313 return;
1314
1315 r = &input_pool;
1316 if (!spin_trylock(&r->lock))
1317 return;
1318
1319 fast_pool->last = now;
1320 __mix_pool_bytes(r, &fast_pool->pool, sizeof(fast_pool->pool));
1321
1322 /*
1323 * If we have architectural seed generator, produce a seed and
1324 * add it to the pool. For the sake of paranoia don't let the
1325 * architectural seed generator dominate the input from the
1326 * interrupt noise.
1327 */
1328 if (arch_get_random_seed_long(&seed)) {
1329 __mix_pool_bytes(r, &seed, sizeof(seed));
1330 credit = 1;
1331 }
1332 spin_unlock(&r->lock);
1333
1334 fast_pool->count = 0;
1335
1336 /* award one bit for the contents of the fast pool */
1337 credit_entropy_bits(r, credit + 1);
1338 }
1339 EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1340
1341 #ifdef CONFIG_BLOCK
add_disk_randomness(struct gendisk * disk)1342 void add_disk_randomness(struct gendisk *disk)
1343 {
1344 if (!disk || !disk->random)
1345 return;
1346 /* first major is 1, so we get >= 0x200 here */
1347 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1348 trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
1349 }
1350 EXPORT_SYMBOL_GPL(add_disk_randomness);
1351 #endif
1352
1353 /*********************************************************************
1354 *
1355 * Entropy extraction routines
1356 *
1357 *********************************************************************/
1358
1359 /*
1360 * This function decides how many bytes to actually take from the
1361 * given pool, and also debits the entropy count accordingly.
1362 */
account(struct entropy_store * r,size_t nbytes,int min,int reserved)1363 static size_t account(struct entropy_store *r, size_t nbytes, int min,
1364 int reserved)
1365 {
1366 int entropy_count, orig, have_bytes;
1367 size_t ibytes, nfrac;
1368
1369 BUG_ON(r->entropy_count > r->poolinfo->poolfracbits);
1370
1371 /* Can we pull enough? */
1372 retry:
1373 entropy_count = orig = READ_ONCE(r->entropy_count);
1374 ibytes = nbytes;
1375 /* never pull more than available */
1376 have_bytes = entropy_count >> (ENTROPY_SHIFT + 3);
1377
1378 if ((have_bytes -= reserved) < 0)
1379 have_bytes = 0;
1380 ibytes = min_t(size_t, ibytes, have_bytes);
1381 if (ibytes < min)
1382 ibytes = 0;
1383
1384 if (WARN_ON(entropy_count < 0)) {
1385 pr_warn("negative entropy count: pool %s count %d\n",
1386 r->name, entropy_count);
1387 entropy_count = 0;
1388 }
1389 nfrac = ibytes << (ENTROPY_SHIFT + 3);
1390 if ((size_t) entropy_count > nfrac)
1391 entropy_count -= nfrac;
1392 else
1393 entropy_count = 0;
1394
1395 if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
1396 goto retry;
1397
1398 trace_debit_entropy(r->name, 8 * ibytes);
1399 if (ibytes && ENTROPY_BITS(r) < random_write_wakeup_bits) {
1400 wake_up_interruptible(&random_write_wait);
1401 kill_fasync(&fasync, SIGIO, POLL_OUT);
1402 }
1403
1404 return ibytes;
1405 }
1406
1407 /*
1408 * This function does the actual extraction for extract_entropy and
1409 * extract_entropy_user.
1410 *
1411 * Note: we assume that .poolwords is a multiple of 16 words.
1412 */
extract_buf(struct entropy_store * r,__u8 * out)1413 static void extract_buf(struct entropy_store *r, __u8 *out)
1414 {
1415 int i;
1416 union {
1417 __u32 w[5];
1418 unsigned long l[LONGS(20)];
1419 } hash;
1420 __u32 workspace[SHA1_WORKSPACE_WORDS];
1421 unsigned long flags;
1422
1423 /*
1424 * If we have an architectural hardware random number
1425 * generator, use it for SHA's initial vector
1426 */
1427 sha1_init(hash.w);
1428 for (i = 0; i < LONGS(20); i++) {
1429 unsigned long v;
1430 if (!arch_get_random_long(&v))
1431 break;
1432 hash.l[i] = v;
1433 }
1434
1435 /* Generate a hash across the pool, 16 words (512 bits) at a time */
1436 spin_lock_irqsave(&r->lock, flags);
1437 for (i = 0; i < r->poolinfo->poolwords; i += 16)
1438 sha1_transform(hash.w, (__u8 *)(r->pool + i), workspace);
1439
1440 /*
1441 * We mix the hash back into the pool to prevent backtracking
1442 * attacks (where the attacker knows the state of the pool
1443 * plus the current outputs, and attempts to find previous
1444 * ouputs), unless the hash function can be inverted. By
1445 * mixing at least a SHA1 worth of hash data back, we make
1446 * brute-forcing the feedback as hard as brute-forcing the
1447 * hash.
1448 */
1449 __mix_pool_bytes(r, hash.w, sizeof(hash.w));
1450 spin_unlock_irqrestore(&r->lock, flags);
1451
1452 memzero_explicit(workspace, sizeof(workspace));
1453
1454 /*
1455 * In case the hash function has some recognizable output
1456 * pattern, we fold it in half. Thus, we always feed back
1457 * twice as much data as we output.
1458 */
1459 hash.w[0] ^= hash.w[3];
1460 hash.w[1] ^= hash.w[4];
1461 hash.w[2] ^= rol32(hash.w[2], 16);
1462
1463 memcpy(out, &hash, EXTRACT_SIZE);
1464 memzero_explicit(&hash, sizeof(hash));
1465 }
1466
_extract_entropy(struct entropy_store * r,void * buf,size_t nbytes,int fips)1467 static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
1468 size_t nbytes, int fips)
1469 {
1470 ssize_t ret = 0, i;
1471 __u8 tmp[EXTRACT_SIZE];
1472 unsigned long flags;
1473
1474 while (nbytes) {
1475 extract_buf(r, tmp);
1476
1477 if (fips) {
1478 spin_lock_irqsave(&r->lock, flags);
1479 if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
1480 panic("Hardware RNG duplicated output!\n");
1481 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1482 spin_unlock_irqrestore(&r->lock, flags);
1483 }
1484 i = min_t(int, nbytes, EXTRACT_SIZE);
1485 memcpy(buf, tmp, i);
1486 nbytes -= i;
1487 buf += i;
1488 ret += i;
1489 }
1490
1491 /* Wipe data just returned from memory */
1492 memzero_explicit(tmp, sizeof(tmp));
1493
1494 return ret;
1495 }
1496
1497 /*
1498 * This function extracts randomness from the "entropy pool", and
1499 * returns it in a buffer.
1500 *
1501 * The min parameter specifies the minimum amount we can pull before
1502 * failing to avoid races that defeat catastrophic reseeding while the
1503 * reserved parameter indicates how much entropy we must leave in the
1504 * pool after each pull to avoid starving other readers.
1505 */
extract_entropy(struct entropy_store * r,void * buf,size_t nbytes,int min,int reserved)1506 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1507 size_t nbytes, int min, int reserved)
1508 {
1509 __u8 tmp[EXTRACT_SIZE];
1510 unsigned long flags;
1511
1512 /* if last_data isn't primed, we need EXTRACT_SIZE extra bytes */
1513 if (fips_enabled) {
1514 spin_lock_irqsave(&r->lock, flags);
1515 if (!r->last_data_init) {
1516 r->last_data_init = 1;
1517 spin_unlock_irqrestore(&r->lock, flags);
1518 trace_extract_entropy(r->name, EXTRACT_SIZE,
1519 ENTROPY_BITS(r), _RET_IP_);
1520 extract_buf(r, tmp);
1521 spin_lock_irqsave(&r->lock, flags);
1522 memcpy(r->last_data, tmp, EXTRACT_SIZE);
1523 }
1524 spin_unlock_irqrestore(&r->lock, flags);
1525 }
1526
1527 trace_extract_entropy(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1528 nbytes = account(r, nbytes, min, reserved);
1529
1530 return _extract_entropy(r, buf, nbytes, fips_enabled);
1531 }
1532
1533 #define warn_unseeded_randomness(previous) \
1534 _warn_unseeded_randomness(__func__, (void *) _RET_IP_, (previous))
1535
_warn_unseeded_randomness(const char * func_name,void * caller,void ** previous)1536 static void _warn_unseeded_randomness(const char *func_name, void *caller,
1537 void **previous)
1538 {
1539 #ifdef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1540 const bool print_once = false;
1541 #else
1542 static bool print_once __read_mostly;
1543 #endif
1544
1545 if (print_once ||
1546 crng_ready() ||
1547 (previous && (caller == READ_ONCE(*previous))))
1548 return;
1549 WRITE_ONCE(*previous, caller);
1550 #ifndef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1551 print_once = true;
1552 #endif
1553 if (__ratelimit(&unseeded_warning))
1554 printk_deferred(KERN_NOTICE "random: %s called from %pS "
1555 "with crng_init=%d\n", func_name, caller,
1556 crng_init);
1557 }
1558
1559 /*
1560 * This function is the exported kernel interface. It returns some
1561 * number of good random numbers, suitable for key generation, seeding
1562 * TCP sequence numbers, etc. It does not rely on the hardware random
1563 * number generator. For random bytes direct from the hardware RNG
1564 * (when available), use get_random_bytes_arch(). In order to ensure
1565 * that the randomness provided by this function is okay, the function
1566 * wait_for_random_bytes() should be called and return 0 at least once
1567 * at any point prior.
1568 */
_get_random_bytes(void * buf,int nbytes)1569 static void _get_random_bytes(void *buf, int nbytes)
1570 {
1571 __u8 tmp[CHACHA_BLOCK_SIZE] __aligned(4);
1572
1573 trace_get_random_bytes(nbytes, _RET_IP_);
1574
1575 while (nbytes >= CHACHA_BLOCK_SIZE) {
1576 extract_crng(buf);
1577 buf += CHACHA_BLOCK_SIZE;
1578 nbytes -= CHACHA_BLOCK_SIZE;
1579 }
1580
1581 if (nbytes > 0) {
1582 extract_crng(tmp);
1583 memcpy(buf, tmp, nbytes);
1584 crng_backtrack_protect(tmp, nbytes);
1585 } else
1586 crng_backtrack_protect(tmp, CHACHA_BLOCK_SIZE);
1587 memzero_explicit(tmp, sizeof(tmp));
1588 }
1589
get_random_bytes(void * buf,int nbytes)1590 void get_random_bytes(void *buf, int nbytes)
1591 {
1592 static void *previous;
1593
1594 warn_unseeded_randomness(&previous);
1595 _get_random_bytes(buf, nbytes);
1596 }
1597 EXPORT_SYMBOL(get_random_bytes);
1598
1599
1600 /*
1601 * Each time the timer fires, we expect that we got an unpredictable
1602 * jump in the cycle counter. Even if the timer is running on another
1603 * CPU, the timer activity will be touching the stack of the CPU that is
1604 * generating entropy..
1605 *
1606 * Note that we don't re-arm the timer in the timer itself - we are
1607 * happy to be scheduled away, since that just makes the load more
1608 * complex, but we do not want the timer to keep ticking unless the
1609 * entropy loop is running.
1610 *
1611 * So the re-arming always happens in the entropy loop itself.
1612 */
entropy_timer(struct timer_list * t)1613 static void entropy_timer(struct timer_list *t)
1614 {
1615 credit_entropy_bits(&input_pool, 1);
1616 }
1617
1618 /*
1619 * If we have an actual cycle counter, see if we can
1620 * generate enough entropy with timing noise
1621 */
try_to_generate_entropy(void)1622 static void try_to_generate_entropy(void)
1623 {
1624 struct {
1625 unsigned long now;
1626 struct timer_list timer;
1627 } stack;
1628
1629 stack.now = random_get_entropy();
1630
1631 /* Slow counter - or none. Don't even bother */
1632 if (stack.now == random_get_entropy())
1633 return;
1634
1635 timer_setup_on_stack(&stack.timer, entropy_timer, 0);
1636 while (!crng_ready()) {
1637 if (!timer_pending(&stack.timer))
1638 mod_timer(&stack.timer, jiffies+1);
1639 mix_pool_bytes(&input_pool, &stack.now, sizeof(stack.now));
1640 schedule();
1641 stack.now = random_get_entropy();
1642 }
1643
1644 del_timer_sync(&stack.timer);
1645 destroy_timer_on_stack(&stack.timer);
1646 mix_pool_bytes(&input_pool, &stack.now, sizeof(stack.now));
1647 }
1648
1649 /*
1650 * Wait for the urandom pool to be seeded and thus guaranteed to supply
1651 * cryptographically secure random numbers. This applies to: the /dev/urandom
1652 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
1653 * family of functions. Using any of these functions without first calling
1654 * this function forfeits the guarantee of security.
1655 *
1656 * Returns: 0 if the urandom pool has been seeded.
1657 * -ERESTARTSYS if the function was interrupted by a signal.
1658 */
wait_for_random_bytes(void)1659 int wait_for_random_bytes(void)
1660 {
1661 if (likely(crng_ready()))
1662 return 0;
1663
1664 do {
1665 int ret;
1666 ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ);
1667 if (ret)
1668 return ret > 0 ? 0 : ret;
1669
1670 try_to_generate_entropy();
1671 } while (!crng_ready());
1672
1673 return 0;
1674 }
1675 EXPORT_SYMBOL(wait_for_random_bytes);
1676
1677 /*
1678 * Returns whether or not the urandom pool has been seeded and thus guaranteed
1679 * to supply cryptographically secure random numbers. This applies to: the
1680 * /dev/urandom device, the get_random_bytes function, and the get_random_{u32,
1681 * ,u64,int,long} family of functions.
1682 *
1683 * Returns: true if the urandom pool has been seeded.
1684 * false if the urandom pool has not been seeded.
1685 */
rng_is_initialized(void)1686 bool rng_is_initialized(void)
1687 {
1688 return crng_ready();
1689 }
1690 EXPORT_SYMBOL(rng_is_initialized);
1691
1692 /*
1693 * Add a callback function that will be invoked when the nonblocking
1694 * pool is initialised.
1695 *
1696 * returns: 0 if callback is successfully added
1697 * -EALREADY if pool is already initialised (callback not called)
1698 * -ENOENT if module for callback is not alive
1699 */
add_random_ready_callback(struct random_ready_callback * rdy)1700 int add_random_ready_callback(struct random_ready_callback *rdy)
1701 {
1702 struct module *owner;
1703 unsigned long flags;
1704 int err = -EALREADY;
1705
1706 if (crng_ready())
1707 return err;
1708
1709 owner = rdy->owner;
1710 if (!try_module_get(owner))
1711 return -ENOENT;
1712
1713 spin_lock_irqsave(&random_ready_list_lock, flags);
1714 if (crng_ready())
1715 goto out;
1716
1717 owner = NULL;
1718
1719 list_add(&rdy->list, &random_ready_list);
1720 err = 0;
1721
1722 out:
1723 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1724
1725 module_put(owner);
1726
1727 return err;
1728 }
1729 EXPORT_SYMBOL(add_random_ready_callback);
1730
1731 /*
1732 * Delete a previously registered readiness callback function.
1733 */
del_random_ready_callback(struct random_ready_callback * rdy)1734 void del_random_ready_callback(struct random_ready_callback *rdy)
1735 {
1736 unsigned long flags;
1737 struct module *owner = NULL;
1738
1739 spin_lock_irqsave(&random_ready_list_lock, flags);
1740 if (!list_empty(&rdy->list)) {
1741 list_del_init(&rdy->list);
1742 owner = rdy->owner;
1743 }
1744 spin_unlock_irqrestore(&random_ready_list_lock, flags);
1745
1746 module_put(owner);
1747 }
1748 EXPORT_SYMBOL(del_random_ready_callback);
1749
1750 /*
1751 * This function will use the architecture-specific hardware random
1752 * number generator if it is available. The arch-specific hw RNG will
1753 * almost certainly be faster than what we can do in software, but it
1754 * is impossible to verify that it is implemented securely (as
1755 * opposed, to, say, the AES encryption of a sequence number using a
1756 * key known by the NSA). So it's useful if we need the speed, but
1757 * only if we're willing to trust the hardware manufacturer not to
1758 * have put in a back door.
1759 *
1760 * Return number of bytes filled in.
1761 */
get_random_bytes_arch(void * buf,int nbytes)1762 int __must_check get_random_bytes_arch(void *buf, int nbytes)
1763 {
1764 int left = nbytes;
1765 char *p = buf;
1766
1767 trace_get_random_bytes_arch(left, _RET_IP_);
1768 while (left) {
1769 unsigned long v;
1770 int chunk = min_t(int, left, sizeof(unsigned long));
1771
1772 if (!arch_get_random_long(&v))
1773 break;
1774
1775 memcpy(p, &v, chunk);
1776 p += chunk;
1777 left -= chunk;
1778 }
1779
1780 return nbytes - left;
1781 }
1782 EXPORT_SYMBOL(get_random_bytes_arch);
1783
1784 /*
1785 * init_std_data - initialize pool with system data
1786 *
1787 * @r: pool to initialize
1788 *
1789 * This function clears the pool's entropy count and mixes some system
1790 * data into the pool to prepare it for use. The pool is not cleared
1791 * as that can only decrease the entropy in the pool.
1792 */
init_std_data(struct entropy_store * r)1793 static void __init init_std_data(struct entropy_store *r)
1794 {
1795 int i;
1796 ktime_t now = ktime_get_real();
1797 unsigned long rv;
1798
1799 mix_pool_bytes(r, &now, sizeof(now));
1800 for (i = r->poolinfo->poolbytes; i > 0; i -= sizeof(rv)) {
1801 if (!arch_get_random_seed_long(&rv) &&
1802 !arch_get_random_long(&rv))
1803 rv = random_get_entropy();
1804 mix_pool_bytes(r, &rv, sizeof(rv));
1805 }
1806 mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1807 }
1808
1809 /*
1810 * Note that setup_arch() may call add_device_randomness()
1811 * long before we get here. This allows seeding of the pools
1812 * with some platform dependent data very early in the boot
1813 * process. But it limits our options here. We must use
1814 * statically allocated structures that already have all
1815 * initializations complete at compile time. We should also
1816 * take care not to overwrite the precious per platform data
1817 * we were given.
1818 */
rand_initialize(void)1819 int __init rand_initialize(void)
1820 {
1821 init_std_data(&input_pool);
1822 if (crng_need_final_init)
1823 crng_finalize_init(&primary_crng);
1824 crng_initialize_primary(&primary_crng);
1825 crng_global_init_time = jiffies;
1826 if (ratelimit_disable) {
1827 urandom_warning.interval = 0;
1828 unseeded_warning.interval = 0;
1829 }
1830 return 0;
1831 }
1832
1833 #ifdef CONFIG_BLOCK
rand_initialize_disk(struct gendisk * disk)1834 void rand_initialize_disk(struct gendisk *disk)
1835 {
1836 struct timer_rand_state *state;
1837
1838 /*
1839 * If kzalloc returns null, we just won't use that entropy
1840 * source.
1841 */
1842 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1843 if (state) {
1844 state->last_time = INITIAL_JIFFIES;
1845 disk->random = state;
1846 }
1847 }
1848 #endif
1849
1850 static ssize_t
urandom_read_nowarn(struct file * file,char __user * buf,size_t nbytes,loff_t * ppos)1851 urandom_read_nowarn(struct file *file, char __user *buf, size_t nbytes,
1852 loff_t *ppos)
1853 {
1854 int ret;
1855
1856 nbytes = min_t(size_t, nbytes, INT_MAX >> (ENTROPY_SHIFT + 3));
1857 ret = extract_crng_user(buf, nbytes);
1858 trace_urandom_read(8 * nbytes, 0, ENTROPY_BITS(&input_pool));
1859 return ret;
1860 }
1861
1862 static ssize_t
urandom_read(struct file * file,char __user * buf,size_t nbytes,loff_t * ppos)1863 urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1864 {
1865 unsigned long flags;
1866 static int maxwarn = 10;
1867
1868 if (!crng_ready() && maxwarn > 0) {
1869 maxwarn--;
1870 if (__ratelimit(&urandom_warning))
1871 pr_notice("%s: uninitialized urandom read (%zd bytes read)\n",
1872 current->comm, nbytes);
1873 spin_lock_irqsave(&primary_crng.lock, flags);
1874 crng_init_cnt = 0;
1875 spin_unlock_irqrestore(&primary_crng.lock, flags);
1876 }
1877
1878 return urandom_read_nowarn(file, buf, nbytes, ppos);
1879 }
1880
1881 static ssize_t
random_read(struct file * file,char __user * buf,size_t nbytes,loff_t * ppos)1882 random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1883 {
1884 int ret;
1885
1886 ret = wait_for_random_bytes();
1887 if (ret != 0)
1888 return ret;
1889 return urandom_read_nowarn(file, buf, nbytes, ppos);
1890 }
1891
1892 static __poll_t
random_poll(struct file * file,poll_table * wait)1893 random_poll(struct file *file, poll_table * wait)
1894 {
1895 __poll_t mask;
1896
1897 poll_wait(file, &crng_init_wait, wait);
1898 poll_wait(file, &random_write_wait, wait);
1899 mask = 0;
1900 if (crng_ready())
1901 mask |= EPOLLIN | EPOLLRDNORM;
1902 if (ENTROPY_BITS(&input_pool) < random_write_wakeup_bits)
1903 mask |= EPOLLOUT | EPOLLWRNORM;
1904 return mask;
1905 }
1906
1907 static int
write_pool(struct entropy_store * r,const char __user * buffer,size_t count)1908 write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1909 {
1910 size_t bytes;
1911 __u32 t, buf[16];
1912 const char __user *p = buffer;
1913
1914 while (count > 0) {
1915 int b, i = 0;
1916
1917 bytes = min(count, sizeof(buf));
1918 if (copy_from_user(&buf, p, bytes))
1919 return -EFAULT;
1920
1921 for (b = bytes ; b > 0 ; b -= sizeof(__u32), i++) {
1922 if (!arch_get_random_int(&t))
1923 break;
1924 buf[i] ^= t;
1925 }
1926
1927 count -= bytes;
1928 p += bytes;
1929
1930 mix_pool_bytes(r, buf, bytes);
1931 cond_resched();
1932 }
1933
1934 return 0;
1935 }
1936
random_write(struct file * file,const char __user * buffer,size_t count,loff_t * ppos)1937 static ssize_t random_write(struct file *file, const char __user *buffer,
1938 size_t count, loff_t *ppos)
1939 {
1940 size_t ret;
1941
1942 ret = write_pool(&input_pool, buffer, count);
1943 if (ret)
1944 return ret;
1945
1946 return (ssize_t)count;
1947 }
1948
random_ioctl(struct file * f,unsigned int cmd,unsigned long arg)1949 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1950 {
1951 int size, ent_count;
1952 int __user *p = (int __user *)arg;
1953 int retval;
1954
1955 switch (cmd) {
1956 case RNDGETENTCNT:
1957 /* inherently racy, no point locking */
1958 ent_count = ENTROPY_BITS(&input_pool);
1959 if (put_user(ent_count, p))
1960 return -EFAULT;
1961 return 0;
1962 case RNDADDTOENTCNT:
1963 if (!capable(CAP_SYS_ADMIN))
1964 return -EPERM;
1965 if (get_user(ent_count, p))
1966 return -EFAULT;
1967 return credit_entropy_bits_safe(&input_pool, ent_count);
1968 case RNDADDENTROPY:
1969 if (!capable(CAP_SYS_ADMIN))
1970 return -EPERM;
1971 if (get_user(ent_count, p++))
1972 return -EFAULT;
1973 if (ent_count < 0)
1974 return -EINVAL;
1975 if (get_user(size, p++))
1976 return -EFAULT;
1977 retval = write_pool(&input_pool, (const char __user *)p,
1978 size);
1979 if (retval < 0)
1980 return retval;
1981 return credit_entropy_bits_safe(&input_pool, ent_count);
1982 case RNDZAPENTCNT:
1983 case RNDCLEARPOOL:
1984 /*
1985 * Clear the entropy pool counters. We no longer clear
1986 * the entropy pool, as that's silly.
1987 */
1988 if (!capable(CAP_SYS_ADMIN))
1989 return -EPERM;
1990 input_pool.entropy_count = 0;
1991 return 0;
1992 case RNDRESEEDCRNG:
1993 if (!capable(CAP_SYS_ADMIN))
1994 return -EPERM;
1995 if (crng_init < 2)
1996 return -ENODATA;
1997 crng_reseed(&primary_crng, &input_pool);
1998 WRITE_ONCE(crng_global_init_time, jiffies - 1);
1999 return 0;
2000 default:
2001 return -EINVAL;
2002 }
2003 }
2004
random_fasync(int fd,struct file * filp,int on)2005 static int random_fasync(int fd, struct file *filp, int on)
2006 {
2007 return fasync_helper(fd, filp, on, &fasync);
2008 }
2009
2010 const struct file_operations random_fops = {
2011 .read = random_read,
2012 .write = random_write,
2013 .poll = random_poll,
2014 .unlocked_ioctl = random_ioctl,
2015 .compat_ioctl = compat_ptr_ioctl,
2016 .fasync = random_fasync,
2017 .llseek = noop_llseek,
2018 };
2019
2020 const struct file_operations urandom_fops = {
2021 .read = urandom_read,
2022 .write = random_write,
2023 .unlocked_ioctl = random_ioctl,
2024 .compat_ioctl = compat_ptr_ioctl,
2025 .fasync = random_fasync,
2026 .llseek = noop_llseek,
2027 };
2028
SYSCALL_DEFINE3(getrandom,char __user *,buf,size_t,count,unsigned int,flags)2029 SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count,
2030 unsigned int, flags)
2031 {
2032 int ret;
2033
2034 if (flags & ~(GRND_NONBLOCK|GRND_RANDOM|GRND_INSECURE))
2035 return -EINVAL;
2036
2037 /*
2038 * Requesting insecure and blocking randomness at the same time makes
2039 * no sense.
2040 */
2041 if ((flags & (GRND_INSECURE|GRND_RANDOM)) == (GRND_INSECURE|GRND_RANDOM))
2042 return -EINVAL;
2043
2044 if (count > INT_MAX)
2045 count = INT_MAX;
2046
2047 if (!(flags & GRND_INSECURE) && !crng_ready()) {
2048 if (flags & GRND_NONBLOCK)
2049 return -EAGAIN;
2050 ret = wait_for_random_bytes();
2051 if (unlikely(ret))
2052 return ret;
2053 }
2054 return urandom_read_nowarn(NULL, buf, count, NULL);
2055 }
2056
2057 /********************************************************************
2058 *
2059 * Sysctl interface
2060 *
2061 ********************************************************************/
2062
2063 #ifdef CONFIG_SYSCTL
2064
2065 #include <linux/sysctl.h>
2066
2067 static int min_write_thresh;
2068 static int max_write_thresh = INPUT_POOL_WORDS * 32;
2069 static int random_min_urandom_seed = 60;
2070 static char sysctl_bootid[16];
2071
2072 /*
2073 * This function is used to return both the bootid UUID, and random
2074 * UUID. The difference is in whether table->data is NULL; if it is,
2075 * then a new UUID is generated and returned to the user.
2076 *
2077 * If the user accesses this via the proc interface, the UUID will be
2078 * returned as an ASCII string in the standard UUID format; if via the
2079 * sysctl system call, as 16 bytes of binary data.
2080 */
proc_do_uuid(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)2081 static int proc_do_uuid(struct ctl_table *table, int write,
2082 void *buffer, size_t *lenp, loff_t *ppos)
2083 {
2084 struct ctl_table fake_table;
2085 unsigned char buf[64], tmp_uuid[16], *uuid;
2086
2087 uuid = table->data;
2088 if (!uuid) {
2089 uuid = tmp_uuid;
2090 generate_random_uuid(uuid);
2091 } else {
2092 static DEFINE_SPINLOCK(bootid_spinlock);
2093
2094 spin_lock(&bootid_spinlock);
2095 if (!uuid[8])
2096 generate_random_uuid(uuid);
2097 spin_unlock(&bootid_spinlock);
2098 }
2099
2100 sprintf(buf, "%pU", uuid);
2101
2102 fake_table.data = buf;
2103 fake_table.maxlen = sizeof(buf);
2104
2105 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
2106 }
2107
2108 /*
2109 * Return entropy available scaled to integral bits
2110 */
proc_do_entropy(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)2111 static int proc_do_entropy(struct ctl_table *table, int write,
2112 void *buffer, size_t *lenp, loff_t *ppos)
2113 {
2114 struct ctl_table fake_table;
2115 int entropy_count;
2116
2117 entropy_count = *(int *)table->data >> ENTROPY_SHIFT;
2118
2119 fake_table.data = &entropy_count;
2120 fake_table.maxlen = sizeof(entropy_count);
2121
2122 return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
2123 }
2124
2125 static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
2126 extern struct ctl_table random_table[];
2127 struct ctl_table random_table[] = {
2128 {
2129 .procname = "poolsize",
2130 .data = &sysctl_poolsize,
2131 .maxlen = sizeof(int),
2132 .mode = 0444,
2133 .proc_handler = proc_dointvec,
2134 },
2135 {
2136 .procname = "entropy_avail",
2137 .maxlen = sizeof(int),
2138 .mode = 0444,
2139 .proc_handler = proc_do_entropy,
2140 .data = &input_pool.entropy_count,
2141 },
2142 {
2143 .procname = "write_wakeup_threshold",
2144 .data = &random_write_wakeup_bits,
2145 .maxlen = sizeof(int),
2146 .mode = 0644,
2147 .proc_handler = proc_dointvec_minmax,
2148 .extra1 = &min_write_thresh,
2149 .extra2 = &max_write_thresh,
2150 },
2151 {
2152 .procname = "urandom_min_reseed_secs",
2153 .data = &random_min_urandom_seed,
2154 .maxlen = sizeof(int),
2155 .mode = 0644,
2156 .proc_handler = proc_dointvec,
2157 },
2158 {
2159 .procname = "boot_id",
2160 .data = &sysctl_bootid,
2161 .maxlen = 16,
2162 .mode = 0444,
2163 .proc_handler = proc_do_uuid,
2164 },
2165 {
2166 .procname = "uuid",
2167 .maxlen = 16,
2168 .mode = 0444,
2169 .proc_handler = proc_do_uuid,
2170 },
2171 #ifdef ADD_INTERRUPT_BENCH
2172 {
2173 .procname = "add_interrupt_avg_cycles",
2174 .data = &avg_cycles,
2175 .maxlen = sizeof(avg_cycles),
2176 .mode = 0444,
2177 .proc_handler = proc_doulongvec_minmax,
2178 },
2179 {
2180 .procname = "add_interrupt_avg_deviation",
2181 .data = &avg_deviation,
2182 .maxlen = sizeof(avg_deviation),
2183 .mode = 0444,
2184 .proc_handler = proc_doulongvec_minmax,
2185 },
2186 #endif
2187 { }
2188 };
2189 #endif /* CONFIG_SYSCTL */
2190
2191 struct batched_entropy {
2192 union {
2193 u64 entropy_u64[CHACHA_BLOCK_SIZE / sizeof(u64)];
2194 u32 entropy_u32[CHACHA_BLOCK_SIZE / sizeof(u32)];
2195 };
2196 unsigned int position;
2197 spinlock_t batch_lock;
2198 };
2199
2200 /*
2201 * Get a random word for internal kernel use only. The quality of the random
2202 * number is good as /dev/urandom, but there is no backtrack protection, with
2203 * the goal of being quite fast and not depleting entropy. In order to ensure
2204 * that the randomness provided by this function is okay, the function
2205 * wait_for_random_bytes() should be called and return 0 at least once at any
2206 * point prior.
2207 */
2208 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64) = {
2209 .batch_lock = __SPIN_LOCK_UNLOCKED(batched_entropy_u64.lock),
2210 };
2211
get_random_u64(void)2212 u64 get_random_u64(void)
2213 {
2214 u64 ret;
2215 unsigned long flags;
2216 struct batched_entropy *batch;
2217 static void *previous;
2218
2219 warn_unseeded_randomness(&previous);
2220
2221 batch = raw_cpu_ptr(&batched_entropy_u64);
2222 spin_lock_irqsave(&batch->batch_lock, flags);
2223 if (batch->position % ARRAY_SIZE(batch->entropy_u64) == 0) {
2224 extract_crng((u8 *)batch->entropy_u64);
2225 batch->position = 0;
2226 }
2227 ret = batch->entropy_u64[batch->position++];
2228 spin_unlock_irqrestore(&batch->batch_lock, flags);
2229 return ret;
2230 }
2231 EXPORT_SYMBOL(get_random_u64);
2232
2233 static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32) = {
2234 .batch_lock = __SPIN_LOCK_UNLOCKED(batched_entropy_u32.lock),
2235 };
get_random_u32(void)2236 u32 get_random_u32(void)
2237 {
2238 u32 ret;
2239 unsigned long flags;
2240 struct batched_entropy *batch;
2241 static void *previous;
2242
2243 warn_unseeded_randomness(&previous);
2244
2245 batch = raw_cpu_ptr(&batched_entropy_u32);
2246 spin_lock_irqsave(&batch->batch_lock, flags);
2247 if (batch->position % ARRAY_SIZE(batch->entropy_u32) == 0) {
2248 extract_crng((u8 *)batch->entropy_u32);
2249 batch->position = 0;
2250 }
2251 ret = batch->entropy_u32[batch->position++];
2252 spin_unlock_irqrestore(&batch->batch_lock, flags);
2253 return ret;
2254 }
2255 EXPORT_SYMBOL(get_random_u32);
2256
2257 /* It's important to invalidate all potential batched entropy that might
2258 * be stored before the crng is initialized, which we can do lazily by
2259 * simply resetting the counter to zero so that it's re-extracted on the
2260 * next usage. */
invalidate_batched_entropy(void)2261 static void invalidate_batched_entropy(void)
2262 {
2263 int cpu;
2264 unsigned long flags;
2265
2266 for_each_possible_cpu (cpu) {
2267 struct batched_entropy *batched_entropy;
2268
2269 batched_entropy = per_cpu_ptr(&batched_entropy_u32, cpu);
2270 spin_lock_irqsave(&batched_entropy->batch_lock, flags);
2271 batched_entropy->position = 0;
2272 spin_unlock(&batched_entropy->batch_lock);
2273
2274 batched_entropy = per_cpu_ptr(&batched_entropy_u64, cpu);
2275 spin_lock(&batched_entropy->batch_lock);
2276 batched_entropy->position = 0;
2277 spin_unlock_irqrestore(&batched_entropy->batch_lock, flags);
2278 }
2279 }
2280
2281 /**
2282 * randomize_page - Generate a random, page aligned address
2283 * @start: The smallest acceptable address the caller will take.
2284 * @range: The size of the area, starting at @start, within which the
2285 * random address must fall.
2286 *
2287 * If @start + @range would overflow, @range is capped.
2288 *
2289 * NOTE: Historical use of randomize_range, which this replaces, presumed that
2290 * @start was already page aligned. We now align it regardless.
2291 *
2292 * Return: A page aligned address within [start, start + range). On error,
2293 * @start is returned.
2294 */
2295 unsigned long
randomize_page(unsigned long start,unsigned long range)2296 randomize_page(unsigned long start, unsigned long range)
2297 {
2298 if (!PAGE_ALIGNED(start)) {
2299 range -= PAGE_ALIGN(start) - start;
2300 start = PAGE_ALIGN(start);
2301 }
2302
2303 if (start > ULONG_MAX - range)
2304 range = ULONG_MAX - start;
2305
2306 range >>= PAGE_SHIFT;
2307
2308 if (range == 0)
2309 return start;
2310
2311 return start + (get_random_long() % range << PAGE_SHIFT);
2312 }
2313
2314 /* Interface for in-kernel drivers of true hardware RNGs.
2315 * Those devices may produce endless random bits and will be throttled
2316 * when our pool is full.
2317 */
add_hwgenerator_randomness(const char * buffer,size_t count,size_t entropy)2318 void add_hwgenerator_randomness(const char *buffer, size_t count,
2319 size_t entropy)
2320 {
2321 struct entropy_store *poolp = &input_pool;
2322
2323 if (unlikely(crng_init == 0)) {
2324 size_t ret = crng_fast_load(buffer, count);
2325 count -= ret;
2326 buffer += ret;
2327 if (!count || crng_init == 0)
2328 return;
2329 }
2330
2331 /* Suspend writing if we're above the trickle threshold.
2332 * We'll be woken up again once below random_write_wakeup_thresh,
2333 * or when the calling thread is about to terminate.
2334 */
2335 wait_event_interruptible(random_write_wait,
2336 !system_wq || kthread_should_stop() ||
2337 ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
2338 mix_pool_bytes(poolp, buffer, count);
2339 credit_entropy_bits(poolp, entropy);
2340 }
2341 EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
2342
2343 /* Handle random seed passed by bootloader.
2344 * If the seed is trustworthy, it would be regarded as hardware RNGs. Otherwise
2345 * it would be regarded as device data.
2346 * The decision is controlled by CONFIG_RANDOM_TRUST_BOOTLOADER.
2347 */
add_bootloader_randomness(const void * buf,unsigned int size)2348 void add_bootloader_randomness(const void *buf, unsigned int size)
2349 {
2350 if (IS_ENABLED(CONFIG_RANDOM_TRUST_BOOTLOADER))
2351 add_hwgenerator_randomness(buf, size, size * 8);
2352 else
2353 add_device_randomness(buf, size);
2354 }
2355 EXPORT_SYMBOL_GPL(add_bootloader_randomness);
2356