xref: /illumos-gate/usr/src/uts/common/crypto/api/kcf_random.c (revision 69b1fd3f24d0ee2e682883606201c61f52085805)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright 2012 Nexenta Systems, Inc.  All rights reserved.
24  * Copyright 2017 Joyent, Inc.
25  */
26 
27 /*
28  * This file implements the interfaces that the /dev/random
29  * driver uses for read(2), write(2) and poll(2) on /dev/random or
30  * /dev/urandom. It also implements the kernel API - random_add_entropy(),
31  * random_add_pseudo_entropy(), random_get_pseudo_bytes()
32  * and random_get_bytes().
33  *
34  * We periodically collect random bits from providers which are registered
35  * with the Kernel Cryptographic Framework (kCF) as capable of random
36  * number generation. The random bits are maintained in a cache and
37  * it is used for high quality random numbers (/dev/random) requests.
38  * We pick a provider and call its SPI routine, if the cache does not have
39  * enough bytes to satisfy a request.
40  *
41  * /dev/urandom requests use a software-based generator algorithm that uses the
42  * random bits in the cache as a seed. We create one pseudo-random generator
43  * (for /dev/urandom) per possible CPU on the system, and use it,
44  * kmem-magazine-style, to avoid cache line contention.
45  *
46  * LOCKING HIERARCHY:
47  *	1) rmp->rm_mag.rm_lock protects the per-cpu pseudo-random generators.
48  * 	2) rndpool_lock protects the high-quality randomness pool.
49  *		It may be locked while a rmp->rm_mag.rm_lock is held.
50  *
51  * A history note: The kernel API and the software-based algorithms in this
52  * file used to be part of the /dev/random driver.
53  */
54 
55 #include <sys/types.h>
56 #include <sys/conf.h>
57 #include <sys/sunddi.h>
58 #include <sys/disp.h>
59 #include <sys/modctl.h>
60 #include <sys/ddi.h>
61 #include <sys/crypto/common.h>
62 #include <sys/crypto/api.h>
63 #include <sys/crypto/impl.h>
64 #include <sys/crypto/sched_impl.h>
65 #include <sys/crypto/ioctladmin.h>
66 #include <sys/random.h>
67 #include <sys/sha1.h>
68 #include <sys/time.h>
69 #include <sys/sysmacros.h>
70 #include <sys/cpuvar.h>
71 #include <sys/taskq.h>
72 #include <rng/fips_random.h>
73 
74 #define	RNDPOOLSIZE		1024	/* Pool size in bytes */
75 #define	MINEXTRACTBYTES		20
76 #define	MAXEXTRACTBYTES		1024
77 #define	PRNG_MAXOBLOCKS		1310720	/* Max output block per prng key */
78 #define	TIMEOUT_INTERVAL	5	/* Periodic mixing interval in secs */
79 
80 typedef enum    extract_type {
81 	NONBLOCK_EXTRACT,
82 	BLOCKING_EXTRACT,
83 	ALWAYS_EXTRACT
84 } extract_type_t;
85 
86 /*
87  * Hash-algo generic definitions. For now, they are SHA1's. We use SHA1
88  * routines directly instead of using k-API because we can't return any
89  * error code in /dev/urandom case and we can get an error using k-API
90  * if a mechanism is disabled.
91  */
92 #define	HASHSIZE		20
93 #define	HASH_CTX		SHA1_CTX
94 #define	HashInit(ctx)		SHA1Init((ctx))
95 #define	HashUpdate(ctx, p, s)	SHA1Update((ctx), (p), (s))
96 #define	HashFinal(d, ctx)	SHA1Final((d), (ctx))
97 
98 /* HMAC-SHA1 */
99 #define	HMAC_KEYSIZE			20
100 
101 /*
102  * Cache of random bytes implemented as a circular buffer. findex and rindex
103  * track the front and back of the circular buffer.
104  */
105 uint8_t rndpool[RNDPOOLSIZE];
106 static int findex, rindex;
107 static int rnbyte_cnt;		/* Number of bytes in the cache */
108 
109 static kmutex_t rndpool_lock;	/* protects r/w accesses to the cache, */
110 				/* and the global variables */
111 static kcondvar_t rndpool_read_cv; /* serializes poll/read syscalls */
112 static int num_waiters;		/* #threads waiting to read from /dev/random */
113 
114 static struct pollhead rnd_pollhead;
115 /* LINTED E_STATIC_UNUSED */
116 static timeout_id_t kcf_rndtimeout_id;
117 static crypto_mech_type_t rngmech_type = CRYPTO_MECH_INVALID;
118 rnd_stats_t rnd_stats;
119 static boolean_t rng_prov_found = B_TRUE;
120 static boolean_t rng_ok_to_log = B_TRUE;
121 static boolean_t rngprov_task_idle = B_TRUE;
122 
123 static void rndc_addbytes(uint8_t *, size_t);
124 static void rndc_getbytes(uint8_t *ptr, size_t len);
125 static void rnd_handler(void *);
126 static void rnd_alloc_magazines(void);
127 static void rnd_fips_discard_initial(void);
128 static void rnd_init2(void *);
129 static void rnd_schedule_timeout(void);
130 
131 /*
132  * Called from kcf:_init()
133  */
134 void
135 kcf_rnd_init()
136 {
137 	hrtime_t ts;
138 	time_t now;
139 
140 	mutex_init(&rndpool_lock, NULL, MUTEX_DEFAULT, NULL);
141 	cv_init(&rndpool_read_cv, NULL, CV_DEFAULT, NULL);
142 
143 	/*
144 	 * Add bytes to the cache using
145 	 * . 2 unpredictable times: high resolution time since the boot-time,
146 	 *   and the current time-of-the day.
147 	 * This is used only to make the timeout value in the timer
148 	 * unpredictable.
149 	 */
150 	ts = gethrtime();
151 	rndc_addbytes((uint8_t *)&ts, sizeof (ts));
152 
153 	(void) drv_getparm(TIME, &now);
154 	rndc_addbytes((uint8_t *)&now, sizeof (now));
155 
156 	rnbyte_cnt = 0;
157 	findex = rindex = 0;
158 	num_waiters = 0;
159 
160 	rnd_alloc_magazines();
161 
162 	(void) taskq_dispatch(system_taskq, rnd_init2, NULL, TQ_SLEEP);
163 }
164 
165 /*
166  * This is called via the system taskq, so that we can do further
167  * initializations that have to wait until the kcf module itself is
168  * done loading.  (After kcf:_init returns.)
169  */
170 static void
171 rnd_init2(void *unused)
172 {
173 
174 	_NOTE(ARGUNUSED(unused));
175 
176 	/*
177 	 * This will load a randomness provider; typically "swrand",
178 	 * but could be another provider if so configured.
179 	 */
180 	rngmech_type = crypto_mech2id(SUN_RANDOM);
181 
182 	/* Update rng_prov_found etc. */
183 	(void) kcf_rngprov_check();
184 
185 	/* FIPS 140-2 init. */
186 	rnd_fips_discard_initial();
187 
188 	/* Start rnd_handler calls. */
189 	rnd_schedule_timeout();
190 }
191 
192 /*
193  * Return TRUE if at least one provider exists that can
194  * supply random numbers.
195  */
196 boolean_t
197 kcf_rngprov_check(void)
198 {
199 	int rv;
200 	kcf_provider_desc_t *pd;
201 
202 	if ((pd = kcf_get_mech_provider(rngmech_type, NULL, NULL, &rv,
203 	    NULL, CRYPTO_FG_RANDOM, 0)) != NULL) {
204 		KCF_PROV_REFRELE(pd);
205 		/*
206 		 * We logged a warning once about no provider being available
207 		 * and now a provider became available. So, set the flag so
208 		 * that we can log again if the problem recurs.
209 		 */
210 		rng_ok_to_log = B_TRUE;
211 		rng_prov_found = B_TRUE;
212 		return (B_TRUE);
213 	} else {
214 		rng_prov_found = B_FALSE;
215 		return (B_FALSE);
216 	}
217 }
218 
219 /*
220  * Pick a software-based provider and submit a request to seed
221  * its random number generator.
222  */
223 static void
224 rngprov_seed(uint8_t *buf, int len, uint_t entropy_est, uint32_t flags)
225 {
226 	kcf_provider_desc_t *pd = NULL;
227 
228 	if (kcf_get_sw_prov(rngmech_type, &pd, NULL, B_FALSE) ==
229 	    CRYPTO_SUCCESS) {
230 		(void) KCF_PROV_SEED_RANDOM(pd, pd->pd_sid, buf, len,
231 		    entropy_est, flags, NULL);
232 		KCF_PROV_REFRELE(pd);
233 	}
234 }
235 
236 /*
237  * This routine is called for blocking reads.
238  *
239  * The argument is_taskq_thr indicates whether the caller is
240  * the taskq thread dispatched by the timeout handler routine.
241  * In this case, we cycle through all the providers
242  * submitting a request to each provider to generate random numbers.
243  *
244  * For other cases, we pick a provider and submit a request to generate
245  * random numbers. We retry using another provider if we get an error.
246  *
247  * Returns the number of bytes that are written to 'ptr'. Returns -1
248  * if no provider is found. ptr and need are unchanged.
249  */
250 static int
251 rngprov_getbytes(uint8_t *ptr, size_t need, boolean_t is_taskq_thr)
252 {
253 	int rv;
254 	int prov_cnt = 0;
255 	int total_bytes = 0;
256 	kcf_provider_desc_t *pd;
257 	kcf_req_params_t params;
258 	kcf_prov_tried_t *list = NULL;
259 
260 	while ((pd = kcf_get_mech_provider(rngmech_type, NULL, NULL, &rv,
261 	    list, CRYPTO_FG_RANDOM, 0)) != NULL) {
262 
263 		prov_cnt++;
264 
265 		KCF_WRAP_RANDOM_OPS_PARAMS(&params, KCF_OP_RANDOM_GENERATE,
266 		    pd->pd_sid, ptr, need, 0, 0);
267 		rv = kcf_submit_request(pd, NULL, NULL, &params, B_FALSE);
268 		ASSERT(rv != CRYPTO_QUEUED);
269 
270 		if (rv == CRYPTO_SUCCESS) {
271 			total_bytes += need;
272 			if (is_taskq_thr)
273 				rndc_addbytes(ptr, need);
274 			else {
275 				KCF_PROV_REFRELE(pd);
276 				break;
277 			}
278 		}
279 
280 		if (is_taskq_thr || rv != CRYPTO_SUCCESS) {
281 			/* Add pd to the linked list of providers tried. */
282 			if (kcf_insert_triedlist(&list, pd, KM_SLEEP) == NULL) {
283 				KCF_PROV_REFRELE(pd);
284 				break;
285 			}
286 		}
287 
288 	}
289 
290 	if (list != NULL)
291 		kcf_free_triedlist(list);
292 
293 	if (prov_cnt == 0) { /* no provider could be found. */
294 		rng_prov_found = B_FALSE;
295 		return (-1);
296 	} else {
297 		rng_prov_found = B_TRUE;
298 		/* See comments in kcf_rngprov_check() */
299 		rng_ok_to_log = B_TRUE;
300 	}
301 
302 	return (total_bytes);
303 }
304 
305 static void
306 notify_done(void *arg, int rv)
307 {
308 	uchar_t *rndbuf = arg;
309 
310 	if (rv == CRYPTO_SUCCESS)
311 		rndc_addbytes(rndbuf, MINEXTRACTBYTES);
312 
313 	bzero(rndbuf, MINEXTRACTBYTES);
314 	kmem_free(rndbuf, MINEXTRACTBYTES);
315 }
316 
317 /*
318  * Cycle through all the providers submitting a request to each provider
319  * to generate random numbers. This is called for the modes - NONBLOCK_EXTRACT
320  * and ALWAYS_EXTRACT.
321  *
322  * Returns the number of bytes that are written to 'ptr'. Returns -1
323  * if no provider is found. ptr and len are unchanged.
324  */
325 static int
326 rngprov_getbytes_nblk(uint8_t *ptr, size_t len)
327 {
328 	int rv, total_bytes;
329 	size_t blen;
330 	uchar_t *rndbuf;
331 	kcf_provider_desc_t *pd;
332 	kcf_req_params_t params;
333 	crypto_call_req_t req;
334 	kcf_prov_tried_t *list = NULL;
335 	int prov_cnt = 0;
336 
337 	blen = 0;
338 	total_bytes = 0;
339 	req.cr_flag = CRYPTO_SKIP_REQID;
340 	req.cr_callback_func = notify_done;
341 
342 	while ((pd = kcf_get_mech_provider(rngmech_type, NULL, NULL, &rv,
343 	    list, CRYPTO_FG_RANDOM, 0)) != NULL) {
344 
345 		prov_cnt ++;
346 		switch (pd->pd_prov_type) {
347 		case CRYPTO_HW_PROVIDER:
348 			/*
349 			 * We have to allocate a buffer here as we can not
350 			 * assume that the input buffer will remain valid
351 			 * when the callback comes. We use a fixed size buffer
352 			 * to simplify the book keeping.
353 			 */
354 			rndbuf = kmem_alloc(MINEXTRACTBYTES, KM_NOSLEEP);
355 			if (rndbuf == NULL) {
356 				KCF_PROV_REFRELE(pd);
357 				if (list != NULL)
358 					kcf_free_triedlist(list);
359 				return (total_bytes);
360 			}
361 			req.cr_callback_arg = rndbuf;
362 			KCF_WRAP_RANDOM_OPS_PARAMS(&params,
363 			    KCF_OP_RANDOM_GENERATE,
364 			    pd->pd_sid, rndbuf, MINEXTRACTBYTES, 0, 0);
365 			break;
366 
367 		case CRYPTO_SW_PROVIDER:
368 			/*
369 			 * We do not need to allocate a buffer in the software
370 			 * provider case as there is no callback involved. We
371 			 * avoid any extra data copy by directly passing 'ptr'.
372 			 */
373 			KCF_WRAP_RANDOM_OPS_PARAMS(&params,
374 			    KCF_OP_RANDOM_GENERATE,
375 			    pd->pd_sid, ptr, len, 0, 0);
376 			break;
377 		}
378 
379 		rv = kcf_submit_request(pd, NULL, &req, &params, B_FALSE);
380 		if (rv == CRYPTO_SUCCESS) {
381 			switch (pd->pd_prov_type) {
382 			case CRYPTO_HW_PROVIDER:
383 				/*
384 				 * Since we have the input buffer handy,
385 				 * we directly copy to it rather than
386 				 * adding to the pool.
387 				 */
388 				blen = min(MINEXTRACTBYTES, len);
389 				bcopy(rndbuf, ptr, blen);
390 				if (len < MINEXTRACTBYTES)
391 					rndc_addbytes(rndbuf + len,
392 					    MINEXTRACTBYTES - len);
393 				ptr += blen;
394 				len -= blen;
395 				total_bytes += blen;
396 				break;
397 
398 			case CRYPTO_SW_PROVIDER:
399 				total_bytes += len;
400 				len = 0;
401 				break;
402 			}
403 		}
404 
405 		/*
406 		 * We free the buffer in the callback routine
407 		 * for the CRYPTO_QUEUED case.
408 		 */
409 		if (pd->pd_prov_type == CRYPTO_HW_PROVIDER &&
410 		    rv != CRYPTO_QUEUED) {
411 			bzero(rndbuf, MINEXTRACTBYTES);
412 			kmem_free(rndbuf, MINEXTRACTBYTES);
413 		}
414 
415 		if (len == 0) {
416 			KCF_PROV_REFRELE(pd);
417 			break;
418 		}
419 
420 		if (rv != CRYPTO_SUCCESS) {
421 			/* Add pd to the linked list of providers tried. */
422 			if (kcf_insert_triedlist(&list, pd, KM_NOSLEEP) ==
423 			    NULL) {
424 				KCF_PROV_REFRELE(pd);
425 				break;
426 			}
427 		}
428 	}
429 
430 	if (list != NULL) {
431 		kcf_free_triedlist(list);
432 	}
433 
434 	if (prov_cnt == 0) { /* no provider could be found. */
435 		rng_prov_found = B_FALSE;
436 		return (-1);
437 	} else {
438 		rng_prov_found = B_TRUE;
439 		/* See comments in kcf_rngprov_check() */
440 		rng_ok_to_log = B_TRUE;
441 	}
442 
443 	return (total_bytes);
444 }
445 
446 static void
447 rngprov_task(void *arg)
448 {
449 	int len = (int)(uintptr_t)arg;
450 	uchar_t tbuf[MAXEXTRACTBYTES];
451 
452 	ASSERT(len <= MAXEXTRACTBYTES);
453 	(void) rngprov_getbytes(tbuf, len, B_TRUE);
454 	rngprov_task_idle = B_TRUE;
455 }
456 
457 /*
458  * Returns "len" random or pseudo-random bytes in *ptr.
459  * Will block if not enough random bytes are available and the
460  * call is blocking.
461  *
462  * Called with rndpool_lock held (allowing caller to do optimistic locking;
463  * releases the lock before return).
464  */
465 static int
466 rnd_get_bytes(uint8_t *ptr, size_t len, extract_type_t how)
467 {
468 	size_t	bytes;
469 	int	got;
470 
471 	ASSERT(mutex_owned(&rndpool_lock));
472 	/*
473 	 * Check if the request can be satisfied from the cache
474 	 * of random bytes.
475 	 */
476 	if (len <= rnbyte_cnt) {
477 		rndc_getbytes(ptr, len);
478 		mutex_exit(&rndpool_lock);
479 		return (0);
480 	}
481 	mutex_exit(&rndpool_lock);
482 
483 	switch (how) {
484 	case BLOCKING_EXTRACT:
485 		if ((got = rngprov_getbytes(ptr, len, B_FALSE)) == -1)
486 			break;	/* No provider found */
487 
488 		if (got == len)
489 			return (0);
490 		len -= got;
491 		ptr += got;
492 		break;
493 
494 	case NONBLOCK_EXTRACT:
495 	case ALWAYS_EXTRACT:
496 		if ((got = rngprov_getbytes_nblk(ptr, len)) == -1) {
497 			/* No provider found */
498 			if (how == NONBLOCK_EXTRACT) {
499 				return (EAGAIN);
500 			}
501 		} else {
502 			if (got == len)
503 				return (0);
504 			len -= got;
505 			ptr += got;
506 		}
507 		if (how == NONBLOCK_EXTRACT && (rnbyte_cnt < len))
508 			return (EAGAIN);
509 		break;
510 	}
511 
512 	mutex_enter(&rndpool_lock);
513 	while (len > 0) {
514 		if (how == BLOCKING_EXTRACT) {
515 			/* Check if there is enough */
516 			while (rnbyte_cnt < MINEXTRACTBYTES) {
517 				num_waiters++;
518 				if (cv_wait_sig(&rndpool_read_cv,
519 				    &rndpool_lock) == 0) {
520 					num_waiters--;
521 					mutex_exit(&rndpool_lock);
522 					return (EINTR);
523 				}
524 				num_waiters--;
525 			}
526 		}
527 
528 		/* Figure out how many bytes to extract */
529 		bytes = min(len, rnbyte_cnt);
530 		rndc_getbytes(ptr, bytes);
531 
532 		len -= bytes;
533 		ptr += bytes;
534 
535 		if (len > 0 && how == ALWAYS_EXTRACT) {
536 			/*
537 			 * There are not enough bytes, but we can not block.
538 			 * This only happens in the case of /dev/urandom which
539 			 * runs an additional generation algorithm. So, there
540 			 * is no problem.
541 			 */
542 			while (len > 0) {
543 				*ptr = rndpool[findex];
544 				ptr++; len--;
545 				rindex = findex = (findex + 1) &
546 				    (RNDPOOLSIZE - 1);
547 			}
548 			break;
549 		}
550 	}
551 
552 	mutex_exit(&rndpool_lock);
553 	return (0);
554 }
555 
556 int
557 kcf_rnd_get_bytes(uint8_t *ptr, size_t len, boolean_t noblock)
558 {
559 	extract_type_t how;
560 	int error;
561 
562 	how = noblock ? NONBLOCK_EXTRACT : BLOCKING_EXTRACT;
563 	mutex_enter(&rndpool_lock);
564 	if ((error = rnd_get_bytes(ptr, len, how)) != 0)
565 		return (error);
566 
567 	BUMP_RND_STATS(rs_rndOut, len);
568 	return (0);
569 }
570 
571 /*
572  * Revisit this if the structs grow or we come up with a better way
573  * of cache-line-padding structures.
574  */
575 #define	RND_CPU_CACHE_SIZE	64
576 #define	RND_CPU_PAD_SIZE	RND_CPU_CACHE_SIZE*6
577 #define	RND_CPU_PAD (RND_CPU_PAD_SIZE - \
578 	sizeof (rndmag_t))
579 /*
580  * Per-CPU random state.  Somewhat like like kmem's magazines, this provides
581  * a per-CPU instance of the pseudo-random generator.  We have it much easier
582  * than kmem, as we can afford to "leak" random bits if a CPU is DR'ed out.
583  *
584  * Note that this usage is preemption-safe; a thread
585  * entering a critical section remembers which generator it locked
586  * and unlocks the same one; should it be preempted and wind up running on
587  * a different CPU, there will be a brief period of increased contention
588  * before it exits the critical section but nothing will melt.
589  */
590 typedef struct rndmag_s
591 {
592 	kmutex_t	rm_lock;
593 	uint8_t		*rm_buffer;	/* Start of buffer */
594 	uint8_t		*rm_eptr;	/* End of buffer */
595 	uint8_t		*rm_rptr;	/* Current read pointer */
596 	uint32_t	rm_oblocks;	/* time to rekey? */
597 	uint32_t	rm_ofuzz;	/* Rekey backoff state */
598 	uint32_t	rm_olimit;	/* Hard rekey limit */
599 	rnd_stats_t	rm_stats;	/* Per-CPU Statistics */
600 	uint32_t	rm_key[HASHSIZE/BYTES_IN_WORD];	/* FIPS XKEY */
601 	uint32_t	rm_seed[HASHSIZE/BYTES_IN_WORD]; /* seed for rekey */
602 	uint32_t	rm_previous[HASHSIZE/BYTES_IN_WORD]; /* prev random */
603 } rndmag_t;
604 
605 typedef struct rndmag_pad_s
606 {
607 	rndmag_t	rm_mag;
608 	uint8_t		rm_pad[RND_CPU_PAD];
609 } rndmag_pad_t;
610 
611 /*
612  * Generate random bytes for /dev/urandom by applying the
613  * FIPS 186-2 algorithm with a key created from bytes extracted
614  * from the pool.  A maximum of PRNG_MAXOBLOCKS output blocks
615  * is generated before a new key is obtained.
616  *
617  * Note that callers to this routine are likely to assume it can't fail.
618  *
619  * Called with rmp locked; releases lock.
620  */
621 static int
622 rnd_generate_pseudo_bytes(rndmag_pad_t *rmp, uint8_t *ptr, size_t len)
623 {
624 	size_t bytes = len, size;
625 	int nblock;
626 	uint32_t oblocks;
627 	uint32_t tempout[HASHSIZE/BYTES_IN_WORD];
628 	uint32_t seed[HASHSIZE/BYTES_IN_WORD];
629 	int i;
630 	hrtime_t timestamp;
631 	uint8_t *src, *dst;
632 
633 	ASSERT(mutex_owned(&rmp->rm_mag.rm_lock));
634 
635 	/* Nothing is being asked */
636 	if (len == 0) {
637 		mutex_exit(&rmp->rm_mag.rm_lock);
638 		return (0);
639 	}
640 
641 	nblock = howmany(len, HASHSIZE);
642 
643 	rmp->rm_mag.rm_oblocks += nblock;
644 	oblocks = rmp->rm_mag.rm_oblocks;
645 
646 	do {
647 		if (oblocks >= rmp->rm_mag.rm_olimit) {
648 
649 			/*
650 			 * Contention-avoiding rekey: see if
651 			 * the pool is locked, and if so, wait a bit.
652 			 * Do an 'exponential back-in' to ensure we don't
653 			 * run too long without rekey.
654 			 */
655 			if (rmp->rm_mag.rm_ofuzz) {
656 				/*
657 				 * Decaying exponential back-in for rekey.
658 				 */
659 				if ((rnbyte_cnt < MINEXTRACTBYTES) ||
660 				    (!mutex_tryenter(&rndpool_lock))) {
661 					rmp->rm_mag.rm_olimit +=
662 					    rmp->rm_mag.rm_ofuzz;
663 					rmp->rm_mag.rm_ofuzz >>= 1;
664 					goto punt;
665 				}
666 			} else {
667 				mutex_enter(&rndpool_lock);
668 			}
669 
670 			/* Get a new chunk of entropy */
671 			(void) rnd_get_bytes((uint8_t *)rmp->rm_mag.rm_key,
672 			    HMAC_KEYSIZE, ALWAYS_EXTRACT);
673 
674 			rmp->rm_mag.rm_olimit = PRNG_MAXOBLOCKS/2;
675 			rmp->rm_mag.rm_ofuzz = PRNG_MAXOBLOCKS/4;
676 			oblocks = 0;
677 			rmp->rm_mag.rm_oblocks = nblock;
678 		}
679 punt:
680 		timestamp = gethrtime();
681 
682 		src = (uint8_t *)&timestamp;
683 		dst = (uint8_t *)rmp->rm_mag.rm_seed;
684 
685 		for (i = 0; i < HASHSIZE; i++) {
686 			dst[i] ^= src[i % sizeof (timestamp)];
687 		}
688 
689 		bcopy(rmp->rm_mag.rm_seed, seed, HASHSIZE);
690 
691 		fips_random_inner(rmp->rm_mag.rm_key, tempout,
692 		    seed);
693 
694 		if (bytes >= HASHSIZE) {
695 			size = HASHSIZE;
696 		} else {
697 			size = min(bytes, HASHSIZE);
698 		}
699 
700 		/*
701 		 * FIPS 140-2: Continuous RNG test - each generation
702 		 * of an n-bit block shall be compared with the previously
703 		 * generated block. Test shall fail if any two compared
704 		 * n-bit blocks are equal.
705 		 */
706 		for (i = 0; i < HASHSIZE/BYTES_IN_WORD; i++) {
707 			if (tempout[i] != rmp->rm_mag.rm_previous[i])
708 				break;
709 		}
710 		if (i == HASHSIZE/BYTES_IN_WORD) {
711 			cmn_err(CE_WARN, "kcf_random: The value of 160-bit "
712 			    "block random bytes are same as the previous "
713 			    "one.\n");
714 			/* discard random bytes and return error */
715 			mutex_exit(&rmp->rm_mag.rm_lock);
716 			return (EIO);
717 		}
718 
719 		bcopy(tempout, rmp->rm_mag.rm_previous,
720 		    HASHSIZE);
721 
722 		bcopy(tempout, ptr, size);
723 		ptr += size;
724 		bytes -= size;
725 		oblocks++;
726 		nblock--;
727 	} while (bytes > 0);
728 
729 	/* Zero out sensitive information */
730 	bzero(seed, HASHSIZE);
731 	bzero(tempout, HASHSIZE);
732 	mutex_exit(&rmp->rm_mag.rm_lock);
733 	return (0);
734 }
735 
736 /*
737  * Per-CPU Random magazines.
738  */
739 static rndmag_pad_t *rndmag;
740 static uint8_t	*rndbuf;
741 static size_t 	rndmag_total;
742 /*
743  * common/os/cpu.c says that platform support code can shrinkwrap
744  * max_ncpus.  On the off chance that we get loaded very early, we
745  * read it exactly once, to copy it here.
746  */
747 static uint32_t	random_max_ncpus = 0;
748 
749 /*
750  * Boot-time tunables, for experimentation.
751  */
752 size_t	rndmag_threshold = 2560;
753 size_t	rndbuf_len = 5120;
754 size_t	rndmag_size = 1280;
755 
756 
757 int
758 kcf_rnd_get_pseudo_bytes(uint8_t *ptr, size_t len)
759 {
760 	rndmag_pad_t *rmp;
761 	uint8_t *cptr, *eptr;
762 
763 	/*
764 	 * Anyone who asks for zero bytes of randomness should get slapped.
765 	 */
766 	ASSERT(len > 0);
767 
768 	/*
769 	 * Fast path.
770 	 */
771 	for (;;) {
772 		rmp = &rndmag[CPU->cpu_seqid];
773 		mutex_enter(&rmp->rm_mag.rm_lock);
774 
775 		/*
776 		 * Big requests bypass buffer and tail-call the
777 		 * generate routine directly.
778 		 */
779 		if (len > rndmag_threshold) {
780 			BUMP_CPU_RND_STATS(rmp, rs_urndOut, len);
781 			return (rnd_generate_pseudo_bytes(rmp, ptr, len));
782 		}
783 
784 		cptr = rmp->rm_mag.rm_rptr;
785 		eptr = cptr + len;
786 
787 		if (eptr <= rmp->rm_mag.rm_eptr) {
788 			rmp->rm_mag.rm_rptr = eptr;
789 			bcopy(cptr, ptr, len);
790 			BUMP_CPU_RND_STATS(rmp, rs_urndOut, len);
791 			mutex_exit(&rmp->rm_mag.rm_lock);
792 
793 			return (0);
794 		}
795 		/*
796 		 * End fast path.
797 		 */
798 		rmp->rm_mag.rm_rptr = rmp->rm_mag.rm_buffer;
799 		/*
800 		 * Note:  We assume the generate routine always succeeds
801 		 * in this case (because it does at present..)
802 		 * It also always releases rm_lock.
803 		 */
804 		(void) rnd_generate_pseudo_bytes(rmp, rmp->rm_mag.rm_buffer,
805 		    rndbuf_len);
806 	}
807 }
808 
809 /*
810  * We set up (empty) magazines for all of max_ncpus, possibly wasting a
811  * little memory on big systems that don't have the full set installed.
812  * See above;  "empty" means "rptr equal to eptr"; this will trigger the
813  * refill path in rnd_get_pseudo_bytes above on the first call for each CPU.
814  *
815  * TODO: make rndmag_size tunable at run time!
816  */
817 static void
818 rnd_alloc_magazines()
819 {
820 	rndmag_pad_t *rmp;
821 	int i;
822 
823 	rndbuf_len = roundup(rndbuf_len, HASHSIZE);
824 	if (rndmag_size < rndbuf_len)
825 		rndmag_size = rndbuf_len;
826 	rndmag_size = roundup(rndmag_size, RND_CPU_CACHE_SIZE);
827 
828 	random_max_ncpus = max_ncpus;
829 	rndmag_total = rndmag_size * random_max_ncpus;
830 
831 	rndbuf = kmem_alloc(rndmag_total, KM_SLEEP);
832 	rndmag = kmem_zalloc(sizeof (rndmag_pad_t) * random_max_ncpus,
833 	    KM_SLEEP);
834 
835 	for (i = 0; i < random_max_ncpus; i++) {
836 		uint8_t *buf;
837 
838 		rmp = &rndmag[i];
839 		mutex_init(&rmp->rm_mag.rm_lock, NULL, MUTEX_DRIVER, NULL);
840 
841 		buf = rndbuf + i * rndmag_size;
842 
843 		rmp->rm_mag.rm_buffer = buf;
844 		rmp->rm_mag.rm_eptr = buf + rndbuf_len;
845 		rmp->rm_mag.rm_rptr = buf + rndbuf_len;
846 		rmp->rm_mag.rm_oblocks = 1;
847 	}
848 }
849 
850 /*
851  * FIPS 140-2: the first n-bit (n > 15) block generated
852  * after power-up, initialization, or reset shall not
853  * be used, but shall be saved for comparison.
854  */
855 static void
856 rnd_fips_discard_initial(void)
857 {
858 	uint8_t discard_buf[HASHSIZE];
859 	rndmag_pad_t *rmp;
860 	int i;
861 
862 	for (i = 0; i < random_max_ncpus; i++) {
863 		rmp = &rndmag[i];
864 
865 		/* rnd_get_bytes() will call mutex_exit(&rndpool_lock) */
866 		mutex_enter(&rndpool_lock);
867 		(void) rnd_get_bytes(discard_buf,
868 		    HMAC_KEYSIZE, ALWAYS_EXTRACT);
869 		bcopy(discard_buf, rmp->rm_mag.rm_previous,
870 		    HMAC_KEYSIZE);
871 		/* rnd_get_bytes() will call mutex_exit(&rndpool_lock) */
872 		mutex_enter(&rndpool_lock);
873 		(void) rnd_get_bytes((uint8_t *)rmp->rm_mag.rm_key,
874 		    HMAC_KEYSIZE, ALWAYS_EXTRACT);
875 		/* rnd_get_bytes() will call mutex_exit(&rndpool_lock) */
876 		mutex_enter(&rndpool_lock);
877 		(void) rnd_get_bytes((uint8_t *)rmp->rm_mag.rm_seed,
878 		    HMAC_KEYSIZE, ALWAYS_EXTRACT);
879 	}
880 }
881 
882 static void
883 rnd_schedule_timeout(void)
884 {
885 	clock_t ut;	/* time in microseconds */
886 
887 	/*
888 	 * The new timeout value is taken from the buffer of random bytes.
889 	 * We're merely reading the first 32 bits from the buffer here, not
890 	 * consuming any random bytes.
891 	 * The timeout multiplier value is a random value between 0.5 sec and
892 	 * 1.544480 sec (0.5 sec + 0xFF000 microseconds).
893 	 * The new timeout is TIMEOUT_INTERVAL times that multiplier.
894 	 */
895 	ut = 500000 + (clock_t)((((uint32_t)rndpool[findex]) << 12) & 0xFF000);
896 	kcf_rndtimeout_id = timeout(rnd_handler, NULL,
897 	    TIMEOUT_INTERVAL * drv_usectohz(ut));
898 }
899 
900 /*
901  * Called from the driver for a poll on /dev/random
902  * . POLLOUT always succeeds.
903  * . POLLIN and POLLRDNORM will block until a
904  *   minimum amount of entropy is available.
905  *
906  * &rnd_pollhead is passed in *phpp in order to indicate the calling thread
907  * will block. When enough random bytes are available, later, the timeout
908  * handler routine will issue the pollwakeup() calls.
909  */
910 void
911 kcf_rnd_chpoll(short events, int anyyet, short *reventsp,
912     struct pollhead **phpp)
913 {
914 	*reventsp = events & POLLOUT;
915 
916 	if (events & (POLLIN | POLLRDNORM)) {
917 		/*
918 		 * Sampling of rnbyte_cnt is an atomic
919 		 * operation. Hence we do not need any locking.
920 		 */
921 		if (rnbyte_cnt >= MINEXTRACTBYTES)
922 			*reventsp |= (events & (POLLIN | POLLRDNORM));
923 	}
924 
925 	if ((*reventsp == 0 && !anyyet) || (events & POLLET))
926 		*phpp = &rnd_pollhead;
927 }
928 
929 /*ARGSUSED*/
930 static void
931 rnd_handler(void *arg)
932 {
933 	int len = 0;
934 
935 	if (!rng_prov_found && rng_ok_to_log) {
936 		cmn_err(CE_WARN, "No randomness provider enabled for "
937 		    "/dev/random. Use cryptoadm(1M) to enable a provider.");
938 		rng_ok_to_log = B_FALSE;
939 	}
940 
941 	if (num_waiters > 0)
942 		/*
943 		 * Note: len has no relationship with how many bytes
944 		 * a poll thread needs.
945 		 */
946 		len = MAXEXTRACTBYTES;
947 	else if (rnbyte_cnt < RNDPOOLSIZE)
948 		len = MINEXTRACTBYTES;
949 
950 	/*
951 	 * Only one thread gets to set rngprov_task_idle at a given point
952 	 * of time and the order of the writes is defined. Also, it is OK
953 	 * if we read an older value of it and skip the dispatch once
954 	 * since we will get the correct value during the next time here.
955 	 * So, no locking is needed here.
956 	 */
957 	if (len > 0 && rngprov_task_idle) {
958 		rngprov_task_idle = B_FALSE;
959 
960 		/*
961 		 * It is OK if taskq_dispatch fails here. We will retry
962 		 * the next time around. Meanwhile, a thread doing a
963 		 * read() will go to the provider directly, if the
964 		 * cache becomes empty.
965 		 */
966 		if (taskq_dispatch(system_taskq, rngprov_task,
967 		    (void *)(uintptr_t)len, TQ_NOSLEEP | TQ_NOQUEUE) ==
968 		    TASKQID_INVALID) {
969 			rngprov_task_idle = B_TRUE;
970 		}
971 	}
972 
973 	mutex_enter(&rndpool_lock);
974 	/*
975 	 * Wake up threads waiting in poll() or for enough accumulated
976 	 * random bytes to read from /dev/random. In case a poll() is
977 	 * concurrent with a read(), the polling process may be woken up
978 	 * indicating that enough randomness is now available for reading,
979 	 * and another process *steals* the bits from the pool, causing the
980 	 * subsequent read() from the first process to block. It is acceptable
981 	 * since the blocking will eventually end, after the timeout
982 	 * has expired enough times to honor the read.
983 	 *
984 	 * Note - Since we hold the rndpool_lock across the pollwakeup() call
985 	 * we MUST NOT grab the rndpool_lock in kcf_rndchpoll().
986 	 */
987 	if (rnbyte_cnt >= MINEXTRACTBYTES)
988 		pollwakeup(&rnd_pollhead, POLLIN | POLLRDNORM);
989 
990 	if (num_waiters > 0)
991 		cv_broadcast(&rndpool_read_cv);
992 	mutex_exit(&rndpool_lock);
993 
994 	rnd_schedule_timeout();
995 }
996 
997 static void
998 rndc_addbytes(uint8_t *ptr, size_t len)
999 {
1000 	ASSERT(ptr != NULL && len > 0);
1001 	ASSERT(rnbyte_cnt <= RNDPOOLSIZE);
1002 
1003 	mutex_enter(&rndpool_lock);
1004 	while ((len > 0) && (rnbyte_cnt < RNDPOOLSIZE)) {
1005 		rndpool[rindex] ^= *ptr;
1006 		ptr++; len--;
1007 		rindex = (rindex + 1) & (RNDPOOLSIZE - 1);
1008 		rnbyte_cnt++;
1009 	}
1010 
1011 	/* Handle buffer full case */
1012 	while (len > 0) {
1013 		rndpool[rindex] ^= *ptr;
1014 		ptr++; len--;
1015 		findex = rindex = (rindex + 1) & (RNDPOOLSIZE - 1);
1016 	}
1017 	mutex_exit(&rndpool_lock);
1018 }
1019 
1020 /*
1021  * Caller should check len <= rnbyte_cnt under the
1022  * rndpool_lock before calling.
1023  */
1024 static void
1025 rndc_getbytes(uint8_t *ptr, size_t len)
1026 {
1027 	ASSERT(MUTEX_HELD(&rndpool_lock));
1028 	ASSERT(len <= rnbyte_cnt && rnbyte_cnt <= RNDPOOLSIZE);
1029 
1030 	BUMP_RND_STATS(rs_rndcOut, len);
1031 
1032 	while (len > 0) {
1033 		*ptr = rndpool[findex];
1034 		ptr++; len--;
1035 		findex = (findex + 1) & (RNDPOOLSIZE - 1);
1036 		rnbyte_cnt--;
1037 	}
1038 }
1039 
1040 /* Random number exported entry points */
1041 
1042 /*
1043  * Mix the supplied bytes into the entropy pool of a kCF
1044  * RNG provider.
1045  */
1046 int
1047 random_add_pseudo_entropy(uint8_t *ptr, size_t len, uint_t entropy_est)
1048 {
1049 	if (len < 1)
1050 		return (-1);
1051 
1052 	rngprov_seed(ptr, len, entropy_est, 0);
1053 
1054 	return (0);
1055 }
1056 
1057 /*
1058  * Mix the supplied bytes into the entropy pool of a kCF
1059  * RNG provider. Mix immediately.
1060  */
1061 int
1062 random_add_entropy(uint8_t *ptr, size_t len, uint_t entropy_est)
1063 {
1064 	if (len < 1)
1065 		return (-1);
1066 
1067 	rngprov_seed(ptr, len, entropy_est, CRYPTO_SEED_NOW);
1068 
1069 	return (0);
1070 }
1071 
1072 /*
1073  * Get bytes from the /dev/urandom generator. This function
1074  * always succeeds. Returns 0.
1075  */
1076 int
1077 random_get_pseudo_bytes(uint8_t *ptr, size_t len)
1078 {
1079 	ASSERT(!mutex_owned(&rndpool_lock));
1080 
1081 	if (len < 1)
1082 		return (0);
1083 	return (kcf_rnd_get_pseudo_bytes(ptr, len));
1084 }
1085 
1086 /*
1087  * Get bytes from the /dev/random generator. Returns 0
1088  * on success. Returns EAGAIN if there is insufficient entropy.
1089  */
1090 int
1091 random_get_bytes(uint8_t *ptr, size_t len)
1092 {
1093 	ASSERT(!mutex_owned(&rndpool_lock));
1094 
1095 	if (len < 1)
1096 		return (0);
1097 	return (kcf_rnd_get_bytes(ptr, len, B_TRUE));
1098 }
1099 
1100 int
1101 random_get_blocking_bytes(uint8_t *ptr, size_t len)
1102 {
1103 	ASSERT(!mutex_owned(&rndpool_lock));
1104 
1105 	if (len < 1)
1106 		return (0);
1107 	return (kcf_rnd_get_bytes(ptr, len, B_FALSE));
1108 }
1109