• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * This is a maximally equidistributed combined Tausworthe generator
4  * based on code from GNU Scientific Library 1.5 (30 Jun 2004)
5  *
6  * lfsr113 version:
7  *
8  * x_n = (s1_n ^ s2_n ^ s3_n ^ s4_n)
9  *
10  * s1_{n+1} = (((s1_n & 4294967294) << 18) ^ (((s1_n <<  6) ^ s1_n) >> 13))
11  * s2_{n+1} = (((s2_n & 4294967288) <<  2) ^ (((s2_n <<  2) ^ s2_n) >> 27))
12  * s3_{n+1} = (((s3_n & 4294967280) <<  7) ^ (((s3_n << 13) ^ s3_n) >> 21))
13  * s4_{n+1} = (((s4_n & 4294967168) << 13) ^ (((s4_n <<  3) ^ s4_n) >> 12))
14  *
15  * The period of this generator is about 2^113 (see erratum paper).
16  *
17  * From: P. L'Ecuyer, "Maximally Equidistributed Combined Tausworthe
18  * Generators", Mathematics of Computation, 65, 213 (1996), 203--213:
19  * http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme.ps
20  * ftp://ftp.iro.umontreal.ca/pub/simulation/lecuyer/papers/tausme.ps
21  *
22  * There is an erratum in the paper "Tables of Maximally Equidistributed
23  * Combined LFSR Generators", Mathematics of Computation, 68, 225 (1999),
24  * 261--269: http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme2.ps
25  *
26  *      ... the k_j most significant bits of z_j must be non-zero,
27  *      for each j. (Note: this restriction also applies to the
28  *      computer code given in [4], but was mistakenly not mentioned
29  *      in that paper.)
30  *
31  * This affects the seeding procedure by imposing the requirement
32  * s1 > 1, s2 > 7, s3 > 15, s4 > 127.
33  */
34 
35 #include <linux/types.h>
36 #include <linux/percpu.h>
37 #include <linux/export.h>
38 #include <linux/jiffies.h>
39 #include <linux/random.h>
40 #include <linux/sched.h>
41 #include <linux/bitops.h>
42 #include <linux/slab.h>
43 #include <linux/notifier.h>
44 #include <asm/unaligned.h>
45 
46 /**
47  *	prandom_u32_state - seeded pseudo-random number generator.
48  *	@state: pointer to state structure holding seeded state.
49  *
50  *	This is used for pseudo-randomness with no outside seeding.
51  *	For more random results, use prandom_u32().
52  */
prandom_u32_state(struct rnd_state * state)53 u32 prandom_u32_state(struct rnd_state *state)
54 {
55 #define TAUSWORTHE(s, a, b, c, d) ((s & c) << d) ^ (((s << a) ^ s) >> b)
56 	state->s1 = TAUSWORTHE(state->s1,  6U, 13U, 4294967294U, 18U);
57 	state->s2 = TAUSWORTHE(state->s2,  2U, 27U, 4294967288U,  2U);
58 	state->s3 = TAUSWORTHE(state->s3, 13U, 21U, 4294967280U,  7U);
59 	state->s4 = TAUSWORTHE(state->s4,  3U, 12U, 4294967168U, 13U);
60 
61 	return (state->s1 ^ state->s2 ^ state->s3 ^ state->s4);
62 }
63 EXPORT_SYMBOL(prandom_u32_state);
64 
65 /**
66  *	prandom_bytes_state - get the requested number of pseudo-random bytes
67  *
68  *	@state: pointer to state structure holding seeded state.
69  *	@buf: where to copy the pseudo-random bytes to
70  *	@bytes: the requested number of bytes
71  *
72  *	This is used for pseudo-randomness with no outside seeding.
73  *	For more random results, use prandom_bytes().
74  */
prandom_bytes_state(struct rnd_state * state,void * buf,size_t bytes)75 void prandom_bytes_state(struct rnd_state *state, void *buf, size_t bytes)
76 {
77 	u8 *ptr = buf;
78 
79 	while (bytes >= sizeof(u32)) {
80 		put_unaligned(prandom_u32_state(state), (u32 *) ptr);
81 		ptr += sizeof(u32);
82 		bytes -= sizeof(u32);
83 	}
84 
85 	if (bytes > 0) {
86 		u32 rem = prandom_u32_state(state);
87 		do {
88 			*ptr++ = (u8) rem;
89 			bytes--;
90 			rem >>= BITS_PER_BYTE;
91 		} while (bytes > 0);
92 	}
93 }
94 EXPORT_SYMBOL(prandom_bytes_state);
95 
prandom_warmup(struct rnd_state * state)96 static void prandom_warmup(struct rnd_state *state)
97 {
98 	/* Calling RNG ten times to satisfy recurrence condition */
99 	prandom_u32_state(state);
100 	prandom_u32_state(state);
101 	prandom_u32_state(state);
102 	prandom_u32_state(state);
103 	prandom_u32_state(state);
104 	prandom_u32_state(state);
105 	prandom_u32_state(state);
106 	prandom_u32_state(state);
107 	prandom_u32_state(state);
108 	prandom_u32_state(state);
109 }
110 
prandom_seed_full_state(struct rnd_state __percpu * pcpu_state)111 void prandom_seed_full_state(struct rnd_state __percpu *pcpu_state)
112 {
113 	int i;
114 
115 	for_each_possible_cpu(i) {
116 		struct rnd_state *state = per_cpu_ptr(pcpu_state, i);
117 		u32 seeds[4];
118 
119 		get_random_bytes(&seeds, sizeof(seeds));
120 		state->s1 = __seed(seeds[0],   2U);
121 		state->s2 = __seed(seeds[1],   8U);
122 		state->s3 = __seed(seeds[2],  16U);
123 		state->s4 = __seed(seeds[3], 128U);
124 
125 		prandom_warmup(state);
126 	}
127 }
128 EXPORT_SYMBOL(prandom_seed_full_state);
129 
130 #ifdef CONFIG_RANDOM32_SELFTEST
131 static struct prandom_test1 {
132 	u32 seed;
133 	u32 result;
134 } test1[] = {
135 	{ 1U, 3484351685U },
136 	{ 2U, 2623130059U },
137 	{ 3U, 3125133893U },
138 	{ 4U,  984847254U },
139 };
140 
141 static struct prandom_test2 {
142 	u32 seed;
143 	u32 iteration;
144 	u32 result;
145 } test2[] = {
146 	/* Test cases against taus113 from GSL library. */
147 	{  931557656U, 959U, 2975593782U },
148 	{ 1339693295U, 876U, 3887776532U },
149 	{ 1545556285U, 961U, 1615538833U },
150 	{  601730776U, 723U, 1776162651U },
151 	{ 1027516047U, 687U,  511983079U },
152 	{  416526298U, 700U,  916156552U },
153 	{ 1395522032U, 652U, 2222063676U },
154 	{  366221443U, 617U, 2992857763U },
155 	{ 1539836965U, 714U, 3783265725U },
156 	{  556206671U, 994U,  799626459U },
157 	{  684907218U, 799U,  367789491U },
158 	{ 2121230701U, 931U, 2115467001U },
159 	{ 1668516451U, 644U, 3620590685U },
160 	{  768046066U, 883U, 2034077390U },
161 	{ 1989159136U, 833U, 1195767305U },
162 	{  536585145U, 996U, 3577259204U },
163 	{ 1008129373U, 642U, 1478080776U },
164 	{ 1740775604U, 939U, 1264980372U },
165 	{ 1967883163U, 508U,   10734624U },
166 	{ 1923019697U, 730U, 3821419629U },
167 	{  442079932U, 560U, 3440032343U },
168 	{ 1961302714U, 845U,  841962572U },
169 	{ 2030205964U, 962U, 1325144227U },
170 	{ 1160407529U, 507U,  240940858U },
171 	{  635482502U, 779U, 4200489746U },
172 	{ 1252788931U, 699U,  867195434U },
173 	{ 1961817131U, 719U,  668237657U },
174 	{ 1071468216U, 983U,  917876630U },
175 	{ 1281848367U, 932U, 1003100039U },
176 	{  582537119U, 780U, 1127273778U },
177 	{ 1973672777U, 853U, 1071368872U },
178 	{ 1896756996U, 762U, 1127851055U },
179 	{  847917054U, 500U, 1717499075U },
180 	{ 1240520510U, 951U, 2849576657U },
181 	{ 1685071682U, 567U, 1961810396U },
182 	{ 1516232129U, 557U,    3173877U },
183 	{ 1208118903U, 612U, 1613145022U },
184 	{ 1817269927U, 693U, 4279122573U },
185 	{ 1510091701U, 717U,  638191229U },
186 	{  365916850U, 807U,  600424314U },
187 	{  399324359U, 702U, 1803598116U },
188 	{ 1318480274U, 779U, 2074237022U },
189 	{  697758115U, 840U, 1483639402U },
190 	{ 1696507773U, 840U,  577415447U },
191 	{ 2081979121U, 981U, 3041486449U },
192 	{  955646687U, 742U, 3846494357U },
193 	{ 1250683506U, 749U,  836419859U },
194 	{  595003102U, 534U,  366794109U },
195 	{   47485338U, 558U, 3521120834U },
196 	{  619433479U, 610U, 3991783875U },
197 	{  704096520U, 518U, 4139493852U },
198 	{ 1712224984U, 606U, 2393312003U },
199 	{ 1318233152U, 922U, 3880361134U },
200 	{  855572992U, 761U, 1472974787U },
201 	{   64721421U, 703U,  683860550U },
202 	{  678931758U, 840U,  380616043U },
203 	{  692711973U, 778U, 1382361947U },
204 	{  677703619U, 530U, 2826914161U },
205 	{   92393223U, 586U, 1522128471U },
206 	{ 1222592920U, 743U, 3466726667U },
207 	{  358288986U, 695U, 1091956998U },
208 	{ 1935056945U, 958U,  514864477U },
209 	{  735675993U, 990U, 1294239989U },
210 	{ 1560089402U, 897U, 2238551287U },
211 	{   70616361U, 829U,   22483098U },
212 	{  368234700U, 731U, 2913875084U },
213 	{   20221190U, 879U, 1564152970U },
214 	{  539444654U, 682U, 1835141259U },
215 	{ 1314987297U, 840U, 1801114136U },
216 	{ 2019295544U, 645U, 3286438930U },
217 	{  469023838U, 716U, 1637918202U },
218 	{ 1843754496U, 653U, 2562092152U },
219 	{  400672036U, 809U, 4264212785U },
220 	{  404722249U, 965U, 2704116999U },
221 	{  600702209U, 758U,  584979986U },
222 	{  519953954U, 667U, 2574436237U },
223 	{ 1658071126U, 694U, 2214569490U },
224 	{  420480037U, 749U, 3430010866U },
225 	{  690103647U, 969U, 3700758083U },
226 	{ 1029424799U, 937U, 3787746841U },
227 	{ 2012608669U, 506U, 3362628973U },
228 	{ 1535432887U, 998U,   42610943U },
229 	{ 1330635533U, 857U, 3040806504U },
230 	{ 1223800550U, 539U, 3954229517U },
231 	{ 1322411537U, 680U, 3223250324U },
232 	{ 1877847898U, 945U, 2915147143U },
233 	{ 1646356099U, 874U,  965988280U },
234 	{  805687536U, 744U, 4032277920U },
235 	{ 1948093210U, 633U, 1346597684U },
236 	{  392609744U, 783U, 1636083295U },
237 	{  690241304U, 770U, 1201031298U },
238 	{ 1360302965U, 696U, 1665394461U },
239 	{ 1220090946U, 780U, 1316922812U },
240 	{  447092251U, 500U, 3438743375U },
241 	{ 1613868791U, 592U,  828546883U },
242 	{  523430951U, 548U, 2552392304U },
243 	{  726692899U, 810U, 1656872867U },
244 	{ 1364340021U, 836U, 3710513486U },
245 	{ 1986257729U, 931U,  935013962U },
246 	{  407983964U, 921U,  728767059U },
247 };
248 
__extract_hwseed(void)249 static u32 __extract_hwseed(void)
250 {
251 	unsigned int val = 0;
252 
253 	(void)(arch_get_random_seed_int(&val) ||
254 	       arch_get_random_int(&val));
255 
256 	return val;
257 }
258 
prandom_seed_early(struct rnd_state * state,u32 seed,bool mix_with_hwseed)259 static void prandom_seed_early(struct rnd_state *state, u32 seed,
260 			       bool mix_with_hwseed)
261 {
262 #define LCG(x)	 ((x) * 69069U)	/* super-duper LCG */
263 #define HWSEED() (mix_with_hwseed ? __extract_hwseed() : 0)
264 	state->s1 = __seed(HWSEED() ^ LCG(seed),        2U);
265 	state->s2 = __seed(HWSEED() ^ LCG(state->s1),   8U);
266 	state->s3 = __seed(HWSEED() ^ LCG(state->s2),  16U);
267 	state->s4 = __seed(HWSEED() ^ LCG(state->s3), 128U);
268 }
269 
prandom_state_selftest(void)270 static int __init prandom_state_selftest(void)
271 {
272 	int i, j, errors = 0, runs = 0;
273 	bool error = false;
274 
275 	for (i = 0; i < ARRAY_SIZE(test1); i++) {
276 		struct rnd_state state;
277 
278 		prandom_seed_early(&state, test1[i].seed, false);
279 		prandom_warmup(&state);
280 
281 		if (test1[i].result != prandom_u32_state(&state))
282 			error = true;
283 	}
284 
285 	if (error)
286 		pr_warn("prandom: seed boundary self test failed\n");
287 	else
288 		pr_info("prandom: seed boundary self test passed\n");
289 
290 	for (i = 0; i < ARRAY_SIZE(test2); i++) {
291 		struct rnd_state state;
292 
293 		prandom_seed_early(&state, test2[i].seed, false);
294 		prandom_warmup(&state);
295 
296 		for (j = 0; j < test2[i].iteration - 1; j++)
297 			prandom_u32_state(&state);
298 
299 		if (test2[i].result != prandom_u32_state(&state))
300 			errors++;
301 
302 		runs++;
303 		cond_resched();
304 	}
305 
306 	if (errors)
307 		pr_warn("prandom: %d/%d self tests failed\n", errors, runs);
308 	else
309 		pr_info("prandom: %d self tests passed\n", runs);
310 	return 0;
311 }
312 core_initcall(prandom_state_selftest);
313 #endif
314 
315 /*
316  * The prandom_u32() implementation is now completely separate from the
317  * prandom_state() functions, which are retained (for now) for compatibility.
318  *
319  * Because of (ab)use in the networking code for choosing random TCP/UDP port
320  * numbers, which open DoS possibilities if guessable, we want something
321  * stronger than a standard PRNG.  But the performance requirements of
322  * the network code do not allow robust crypto for this application.
323  *
324  * So this is a homebrew Junior Spaceman implementation, based on the
325  * lowest-latency trustworthy crypto primitive available, SipHash.
326  * (The authors of SipHash have not been consulted about this abuse of
327  * their work.)
328  *
329  * Standard SipHash-2-4 uses 2n+4 rounds to hash n words of input to
330  * one word of output.  This abbreviated version uses 2 rounds per word
331  * of output.
332  */
333 
334 struct siprand_state {
335 	unsigned long v0;
336 	unsigned long v1;
337 	unsigned long v2;
338 	unsigned long v3;
339 };
340 
341 static DEFINE_PER_CPU(struct siprand_state, net_rand_state) __latent_entropy;
342 
343 /*
344  * This is the core CPRNG function.  As "pseudorandom", this is not used
345  * for truly valuable things, just intended to be a PITA to guess.
346  * For maximum speed, we do just two SipHash rounds per word.  This is
347  * the same rate as 4 rounds per 64 bits that SipHash normally uses,
348  * so hopefully it's reasonably secure.
349  *
350  * There are two changes from the official SipHash finalization:
351  * - We omit some constants XORed with v2 in the SipHash spec as irrelevant;
352  *   they are there only to make the output rounds distinct from the input
353  *   rounds, and this application has no input rounds.
354  * - Rather than returning v0^v1^v2^v3, return v1+v3.
355  *   If you look at the SipHash round, the last operation on v3 is
356  *   "v3 ^= v0", so "v0 ^ v3" just undoes that, a waste of time.
357  *   Likewise "v1 ^= v2".  (The rotate of v2 makes a difference, but
358  *   it still cancels out half of the bits in v2 for no benefit.)
359  *   Second, since the last combining operation was xor, continue the
360  *   pattern of alternating xor/add for a tiny bit of extra non-linearity.
361  */
siprand_u32(struct siprand_state * s)362 static inline u32 siprand_u32(struct siprand_state *s)
363 {
364 	unsigned long v0 = s->v0, v1 = s->v1, v2 = s->v2, v3 = s->v3;
365 
366 	PRND_SIPROUND(v0, v1, v2, v3);
367 	PRND_SIPROUND(v0, v1, v2, v3);
368 	s->v0 = v0;  s->v1 = v1;  s->v2 = v2;  s->v3 = v3;
369 	return v1 + v3;
370 }
371 
372 
373 /**
374  *	prandom_u32 - pseudo random number generator
375  *
376  *	A 32 bit pseudo-random number is generated using a fast
377  *	algorithm suitable for simulation. This algorithm is NOT
378  *	considered safe for cryptographic use.
379  */
prandom_u32(void)380 u32 prandom_u32(void)
381 {
382 	struct siprand_state *state = get_cpu_ptr(&net_rand_state);
383 	u32 res = siprand_u32(state);
384 
385 	put_cpu_ptr(&net_rand_state);
386 	return res;
387 }
388 EXPORT_SYMBOL(prandom_u32);
389 
390 /**
391  *	prandom_bytes - get the requested number of pseudo-random bytes
392  *	@buf: where to copy the pseudo-random bytes to
393  *	@bytes: the requested number of bytes
394  */
prandom_bytes(void * buf,size_t bytes)395 void prandom_bytes(void *buf, size_t bytes)
396 {
397 	struct siprand_state *state = get_cpu_ptr(&net_rand_state);
398 	u8 *ptr = buf;
399 
400 	while (bytes >= sizeof(u32)) {
401 		put_unaligned(siprand_u32(state), (u32 *)ptr);
402 		ptr += sizeof(u32);
403 		bytes -= sizeof(u32);
404 	}
405 
406 	if (bytes > 0) {
407 		u32 rem = siprand_u32(state);
408 
409 		do {
410 			*ptr++ = (u8)rem;
411 			rem >>= BITS_PER_BYTE;
412 		} while (--bytes > 0);
413 	}
414 	put_cpu_ptr(&net_rand_state);
415 }
416 EXPORT_SYMBOL(prandom_bytes);
417 
418 /**
419  *	prandom_seed - add entropy to pseudo random number generator
420  *	@entropy: entropy value
421  *
422  *	Add some additional seed material to the prandom pool.
423  *	The "entropy" is actually our IP address (the only caller is
424  *	the network code), not for unpredictability, but to ensure that
425  *	different machines are initialized differently.
426  */
prandom_seed(u32 entropy)427 void prandom_seed(u32 entropy)
428 {
429 	int i;
430 
431 	add_device_randomness(&entropy, sizeof(entropy));
432 
433 	for_each_possible_cpu(i) {
434 		struct siprand_state *state = per_cpu_ptr(&net_rand_state, i);
435 		unsigned long v0 = state->v0, v1 = state->v1;
436 		unsigned long v2 = state->v2, v3 = state->v3;
437 
438 		do {
439 			v3 ^= entropy;
440 			PRND_SIPROUND(v0, v1, v2, v3);
441 			PRND_SIPROUND(v0, v1, v2, v3);
442 			v0 ^= entropy;
443 		} while (unlikely(!v0 || !v1 || !v2 || !v3));
444 
445 		WRITE_ONCE(state->v0, v0);
446 		WRITE_ONCE(state->v1, v1);
447 		WRITE_ONCE(state->v2, v2);
448 		WRITE_ONCE(state->v3, v3);
449 	}
450 }
451 EXPORT_SYMBOL(prandom_seed);
452 
453 /*
454  *	Generate some initially weak seeding values to allow
455  *	the prandom_u32() engine to be started.
456  */
prandom_init_early(void)457 static int __init prandom_init_early(void)
458 {
459 	int i;
460 	unsigned long v0, v1, v2, v3;
461 
462 	if (!arch_get_random_long(&v0))
463 		v0 = jiffies;
464 	if (!arch_get_random_long(&v1))
465 		v1 = random_get_entropy();
466 	v2 = v0 ^ PRND_K0;
467 	v3 = v1 ^ PRND_K1;
468 
469 	for_each_possible_cpu(i) {
470 		struct siprand_state *state;
471 
472 		v3 ^= i;
473 		PRND_SIPROUND(v0, v1, v2, v3);
474 		PRND_SIPROUND(v0, v1, v2, v3);
475 		v0 ^= i;
476 
477 		state = per_cpu_ptr(&net_rand_state, i);
478 		state->v0 = v0;  state->v1 = v1;
479 		state->v2 = v2;  state->v3 = v3;
480 	}
481 
482 	return 0;
483 }
484 core_initcall(prandom_init_early);
485 
486 
487 /* Stronger reseeding when available, and periodically thereafter. */
488 static void prandom_reseed(struct timer_list *unused);
489 
490 static DEFINE_TIMER(seed_timer, prandom_reseed);
491 
prandom_reseed(struct timer_list * unused)492 static void prandom_reseed(struct timer_list *unused)
493 {
494 	unsigned long expires;
495 	int i;
496 
497 	/*
498 	 * Reinitialize each CPU's PRNG with 128 bits of key.
499 	 * No locking on the CPUs, but then somewhat random results are,
500 	 * well, expected.
501 	 */
502 	for_each_possible_cpu(i) {
503 		struct siprand_state *state;
504 		unsigned long v0 = get_random_long(), v2 = v0 ^ PRND_K0;
505 		unsigned long v1 = get_random_long(), v3 = v1 ^ PRND_K1;
506 #if BITS_PER_LONG == 32
507 		int j;
508 
509 		/*
510 		 * On 32-bit machines, hash in two extra words to
511 		 * approximate 128-bit key length.  Not that the hash
512 		 * has that much security, but this prevents a trivial
513 		 * 64-bit brute force.
514 		 */
515 		for (j = 0; j < 2; j++) {
516 			unsigned long m = get_random_long();
517 
518 			v3 ^= m;
519 			PRND_SIPROUND(v0, v1, v2, v3);
520 			PRND_SIPROUND(v0, v1, v2, v3);
521 			v0 ^= m;
522 		}
523 #endif
524 		/*
525 		 * Probably impossible in practice, but there is a
526 		 * theoretical risk that a race between this reseeding
527 		 * and the target CPU writing its state back could
528 		 * create the all-zero SipHash fixed point.
529 		 *
530 		 * To ensure that never happens, ensure the state
531 		 * we write contains no zero words.
532 		 */
533 		state = per_cpu_ptr(&net_rand_state, i);
534 		WRITE_ONCE(state->v0, v0 ? v0 : -1ul);
535 		WRITE_ONCE(state->v1, v1 ? v1 : -1ul);
536 		WRITE_ONCE(state->v2, v2 ? v2 : -1ul);
537 		WRITE_ONCE(state->v3, v3 ? v3 : -1ul);
538 	}
539 
540 	/* reseed every ~60 seconds, in [40 .. 80) interval with slack */
541 	expires = round_jiffies(jiffies + 40 * HZ + prandom_u32_max(40 * HZ));
542 	mod_timer(&seed_timer, expires);
543 }
544 
545 /*
546  * The random ready callback can be called from almost any interrupt.
547  * To avoid worrying about whether it's safe to delay that interrupt
548  * long enough to seed all CPUs, just schedule an immediate timer event.
549  */
prandom_timer_start(struct notifier_block * nb,unsigned long action,void * data)550 static int prandom_timer_start(struct notifier_block *nb,
551 			       unsigned long action, void *data)
552 {
553 	mod_timer(&seed_timer, jiffies);
554 	return 0;
555 }
556 
557 /*
558  * Start periodic full reseeding as soon as strong
559  * random numbers are available.
560  */
prandom_init_late(void)561 static int __init prandom_init_late(void)
562 {
563 	static struct notifier_block random_ready = {
564 		.notifier_call = prandom_timer_start
565 	};
566 	int ret = register_random_ready_notifier(&random_ready);
567 
568 	if (ret == -EALREADY) {
569 		prandom_timer_start(&random_ready, 0, NULL);
570 		ret = 0;
571 	}
572 	return ret;
573 }
574 late_initcall(prandom_init_late);
575