xref: /netbsd-src/lib/libc/gen/arc4random.c (revision 9ddb6ab554e70fb9bbd90c3d96b812bc57755a14)
1 /*	$NetBSD: arc4random.c,v 1.13 2012/03/05 19:40:08 christos Exp $	*/
2 /*	$OpenBSD: arc4random.c,v 1.6 2001/06/05 05:05:38 pvalchev Exp $	*/
3 
4 /*
5  * Arc4 random number generator for OpenBSD.
6  * Copyright 1996 David Mazieres <dm@lcs.mit.edu>.
7  *
8  * Modification and redistribution in source and binary forms is
9  * permitted provided that due credit is given to the author and the
10  * OpenBSD project by leaving this copyright notice intact.
11  */
12 
13 /*
14  * This code is derived from section 17.1 of Applied Cryptography,
15  * second edition, which describes a stream cipher allegedly
16  * compatible with RSA Labs "RC4" cipher (the actual description of
17  * which is a trade secret).  The same algorithm is used as a stream
18  * cipher called "arcfour" in Tatu Ylonen's ssh package.
19  *
20  * Here the stream cipher has been modified always to include the time
21  * when initializing the state.  That makes it impossible to
22  * regenerate the same random sequence twice, so this can't be used
23  * for encryption, but will generate good random numbers.
24  *
25  * RC4 is a registered trademark of RSA Laboratories.
26  */
27 
28 #include <sys/cdefs.h>
29 #if defined(LIBC_SCCS) && !defined(lint)
30 __RCSID("$NetBSD: arc4random.c,v 1.13 2012/03/05 19:40:08 christos Exp $");
31 #endif /* LIBC_SCCS and not lint */
32 
33 #include "namespace.h"
34 #include "reentrant.h"
35 #include <fcntl.h>
36 #include <stdlib.h>
37 #include <unistd.h>
38 #include <sys/types.h>
39 #include <sys/param.h>
40 #include <sys/time.h>
41 #include <sys/sysctl.h>
42 
43 #ifdef __weak_alias
44 __weak_alias(arc4random,_arc4random)
45 #endif
46 
47 #define RSIZE 256
48 struct arc4_stream {
49 	mutex_t mtx;
50 	int initialized;
51 	uint8_t i;
52 	uint8_t j;
53 	uint8_t s[RSIZE];
54 };
55 
56 /* XXX lint explodes with an internal error if only mtx is initialized! */
57 static struct arc4_stream rs = { .i = 0, .mtx = MUTEX_INITIALIZER };
58 
59 static inline void arc4_init(struct arc4_stream *);
60 static inline void arc4_addrandom(struct arc4_stream *, u_char *, int);
61 static void arc4_stir(struct arc4_stream *);
62 static inline uint8_t arc4_getbyte(struct arc4_stream *);
63 static inline uint32_t arc4_getword(struct arc4_stream *);
64 
65 static inline void
66 arc4_init(struct arc4_stream *as)
67 {
68 	for (int n = 0; n < RSIZE; n++)
69 		as->s[n] = n;
70 	as->i = 0;
71 	as->j = 0;
72 
73 	as->initialized = 1;
74 	arc4_stir(as);
75 }
76 
77 static inline void
78 arc4_addrandom(struct arc4_stream *as, u_char *dat, int datlen)
79 {
80 	uint8_t si;
81 
82 	as->i--;
83 	for (int n = 0; n < RSIZE; n++) {
84 		as->i = (as->i + 1);
85 		si = as->s[as->i];
86 		as->j = (as->j + si + dat[n % datlen]);
87 		as->s[as->i] = as->s[as->j];
88 		as->s[as->j] = si;
89 	}
90 	as->j = as->i;
91 }
92 
93 static void
94 arc4_stir(struct arc4_stream *as)
95 {
96 	int rdat[32];
97 	static const int mib[] = { CTL_KERN, KERN_URND };
98 	size_t len;
99 
100 	/*
101 	 * This code once opened and read /dev/urandom on each
102 	 * call.  That causes repeated rekeying of the kernel stream
103 	 * generator, which is very wasteful.  Because of application
104 	 * behavior, caching the fd doesn't really help.  So we just
105 	 * fill up the tank from sysctl, which is a tiny bit slower
106 	 * for us but much friendlier to other entropy consumers.
107 	 */
108 
109 	for (size_t i = 0; i < __arraycount(rdat); i++) {
110 		len = sizeof(rdat[i]);
111 		if (sysctl(mib, 2, &rdat[i], &len, NULL, 0) == -1)
112 			abort();
113 	}
114 
115 	arc4_addrandom(as, (void *) &rdat, (int)sizeof(rdat));
116 
117 	/*
118 	 * Throw away the first N words of output, as suggested in the
119 	 * paper "Weaknesses in the Key Scheduling Algorithm of RC4"
120 	 * by Fluher, Mantin, and Shamir.  (N = 256 in our case.)
121 	 */
122 	for (size_t j = 0; j < RSIZE * 4; j++)
123 		arc4_getbyte(as);
124 }
125 
126 static inline uint8_t
127 arc4_getbyte(struct arc4_stream *as)
128 {
129 	uint8_t si, sj;
130 
131 	as->i = (as->i + 1);
132 	si = as->s[as->i];
133 	as->j = (as->j + si);
134 	sj = as->s[as->j];
135 	as->s[as->i] = sj;
136 	as->s[as->j] = si;
137 	return (as->s[(si + sj) & 0xff]);
138 }
139 
140 static inline uint32_t
141 arc4_getword(struct arc4_stream *as)
142 {
143 	uint32_t val;
144 	val = arc4_getbyte(as) << 24;
145 	val |= arc4_getbyte(as) << 16;
146 	val |= arc4_getbyte(as) << 8;
147 	val |= arc4_getbyte(as);
148 	return val;
149 }
150 
151 static inline void
152 _arc4random_stir_unlocked(void)
153 {
154 	if (__predict_false(!rs.initialized)) {
155 		arc4_init(&rs);				/* stirs */
156 	} else {
157 		arc4_stir(&rs);
158 	}
159 }
160 
161 void
162 arc4random_stir(void)
163 {
164 #ifdef _REENTRANT
165 	if (__isthreaded) {
166 		mutex_lock(&rs.mtx);
167                 _arc4random_stir_unlocked();
168 		mutex_unlock(&rs.mtx);
169 		return;
170         }
171 #endif
172 	_arc4random_stir_unlocked();
173 }
174 
175 static inline void
176 _arc4random_addrandom_unlocked(u_char *dat, int datlen)
177 {
178 	if (__predict_false(rs.initialized)) {
179 		arc4_init(&rs);
180 	}
181 	arc4_addrandom(&rs, dat, datlen);
182 }
183 
184 void
185 arc4random_addrandom(u_char *dat, int datlen)
186 {
187 #ifdef _REENTRANT
188 	if (__isthreaded) {
189 		mutex_lock(&rs.mtx);
190 		_arc4random_addrandom_unlocked(dat, datlen);
191 		mutex_unlock(&rs.mtx);
192 		return;
193 	}
194 #endif
195 	_arc4random_addrandom_unlocked(dat, datlen);
196 }
197 
198 static inline uint32_t
199 _arc4random_unlocked(void)
200 {
201 	if (__predict_false(!rs.initialized)) {
202 		arc4_init(&rs);
203 	}
204 	return arc4_getword(&rs);
205 }
206 
207 uint32_t
208 arc4random(void)
209 {
210 	uint32_t v;
211 #ifdef _REENTRANT
212 	if (__isthreaded) {
213 		mutex_lock(&rs.mtx);
214 		v = _arc4random_unlocked();
215 		mutex_unlock(&rs.mtx);
216 		return v;
217 	}
218 #endif
219 	v = _arc4random_unlocked();
220 	return v;
221 }
222 
223 static void
224 _arc4random_buf_unlocked(void *buf, size_t len)
225 {
226 	uint8_t *bp = buf;
227 	uint8_t *ep = bp + len;
228 
229 	if (__predict_false(!rs.initialized)) {
230 		arc4_init(&rs);
231 	}
232 
233 	bp[0] = arc4_getbyte(&rs) % 3;
234 	while (bp[0]--)
235 		(void)arc4_getbyte(&rs);
236 
237 	while (bp < ep)
238 		*bp++ = arc4_getbyte(&rs);
239 }
240 
241 void
242 arc4random_buf(void *buf, size_t len)
243 {
244 #ifdef _REENTRANT
245 	if (__isthreaded) {
246 		mutex_lock(&rs.mtx);
247 		_arc4random_buf_unlocked(buf, len);
248 		mutex_unlock(&rs.mtx);
249 		return;
250 	} else
251 #endif
252 	_arc4random_buf_unlocked(buf, len);
253 }
254 
255 /*-
256  * Written by Damien Miller.
257  * With simplifications by Jinmei Tatuya.
258  */
259 
260 /*
261  * Calculate a uniformly distributed random number less than
262  * upper_bound avoiding "modulo bias".
263  *
264  * Uniformity is achieved by generating new random numbers
265  * until the one returned is outside the range
266  * [0, 2^32 % upper_bound[. This guarantees the selected
267  * random number will be inside the range
268  * [2^32 % upper_bound, 2^32[ which maps back to
269  * [0, upper_bound[ after reduction modulo upper_bound.
270  */
271 static uint32_t
272 _arc4random_uniform_unlocked(uint32_t upper_bound)
273 {
274 	uint32_t r, min;
275 
276 	if (upper_bound < 2)
277 		return 0;
278 
279 #if defined(ULONG_MAX) && (ULONG_MAX > 0xFFFFFFFFUL)
280 	min = (uint32_t)(0x100000000U % upper_bound);
281 #else
282 	/* calculate (2^32 % upper_bound) avoiding 64-bit math */
283 	if (upper_bound > 0x80000000U)
284 		/* 2^32 - upper_bound (only one "value area") */
285 		min = 1 + ~upper_bound;
286 	else
287 		/* ((2^32 - x) % x) == (2^32 % x) when x <= 2^31 */
288 		min = (0xFFFFFFFFU - upper_bound + 1) % upper_bound;
289 #endif
290 
291 	/*
292 	 * This could theoretically loop forever but each retry has
293 	 * p > 0.5 (worst case, usually far better) of selecting a
294 	 * number inside the range we need, so it should rarely need
295 	 * to re-roll (at all).
296 	 */
297 	if (__predict_false(!rs.initialized)) {
298 		arc4_init(&rs);
299 	}
300 	if (arc4_getbyte(&rs) & 1)
301 		(void)arc4_getbyte(&rs);
302 	do
303 		r = arc4_getword(&rs);
304 	while (r < min);
305 
306 	return r % upper_bound;
307 }
308 
309 uint32_t
310 arc4random_uniform(uint32_t upper_bound)
311 {
312 	uint32_t v;
313 #ifdef _REENTRANT
314 	if (__isthreaded) {
315 		mutex_lock(&rs.mtx);
316 		v = _arc4random_uniform_unlocked(upper_bound);
317 		mutex_unlock(&rs.mtx);
318 		return v;
319 	}
320 #endif
321 	v = _arc4random_uniform_unlocked(upper_bound);
322 	return v;
323 }
324