1 /*
2 * random.c -- A strong random number generator
3 *
4 * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005
5 *
6 * Copyright Theodore Ts'o, 1994, 1995, 1996, 1997, 1998, 1999. All
7 * rights reserved.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, and the entire permission notice in its entirety,
14 * including the disclaimer of warranties.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. The name of the author may not be used to endorse or promote
19 * products derived from this software without specific prior
20 * written permission.
21 *
22 * ALTERNATIVELY, this product may be distributed under the terms of
23 * the GNU General Public License, in which case the provisions of the GPL are
24 * required INSTEAD OF the above restrictions. (This clause is
25 * necessary due to a potential bad interaction between the GPL and
26 * the restrictions contained in a BSD-style copyright.)
27 *
28 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
29 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
30 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ALL OF
31 * WHICH ARE HEREBY DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE
32 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
33 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT
34 * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
35 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
36 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
37 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
38 * USE OF THIS SOFTWARE, EVEN IF NOT ADVISED OF THE POSSIBILITY OF SUCH
39 * DAMAGE.
40 */
41
42 /*
43 * (now, with legal B.S. out of the way.....)
44 *
45 * This routine gathers environmental noise from device drivers, etc.,
46 * and returns good random numbers, suitable for cryptographic use.
47 * Besides the obvious cryptographic uses, these numbers are also good
48 * for seeding TCP sequence numbers, and other places where it is
49 * desirable to have numbers which are not only random, but hard to
50 * predict by an attacker.
51 *
52 * Theory of operation
53 * ===================
54 *
55 * Computers are very predictable devices. Hence it is extremely hard
56 * to produce truly random numbers on a computer --- as opposed to
57 * pseudo-random numbers, which can easily generated by using a
58 * algorithm. Unfortunately, it is very easy for attackers to guess
59 * the sequence of pseudo-random number generators, and for some
60 * applications this is not acceptable. So instead, we must try to
61 * gather "environmental noise" from the computer's environment, which
62 * must be hard for outside attackers to observe, and use that to
63 * generate random numbers. In a Unix environment, this is best done
64 * from inside the kernel.
65 *
66 * Sources of randomness from the environment include inter-keyboard
67 * timings, inter-interrupt timings from some interrupts, and other
68 * events which are both (a) non-deterministic and (b) hard for an
69 * outside observer to measure. Randomness from these sources are
70 * added to an "entropy pool", which is mixed using a CRC-like function.
71 * This is not cryptographically strong, but it is adequate assuming
72 * the randomness is not chosen maliciously, and it is fast enough that
73 * the overhead of doing it on every interrupt is very reasonable.
74 * As random bytes are mixed into the entropy pool, the routines keep
75 * an *estimate* of how many bits of randomness have been stored into
76 * the random number generator's internal state.
77 *
78 * When random bytes are desired, they are obtained by taking the SHA
79 * hash of the contents of the "entropy pool". The SHA hash avoids
80 * exposing the internal state of the entropy pool. It is believed to
81 * be computationally infeasible to derive any useful information
82 * about the input of SHA from its output. Even if it is possible to
83 * analyze SHA in some clever way, as long as the amount of data
84 * returned from the generator is less than the inherent entropy in
85 * the pool, the output data is totally unpredictable. For this
86 * reason, the routine decreases its internal estimate of how many
87 * bits of "true randomness" are contained in the entropy pool as it
88 * outputs random numbers.
89 *
90 * If this estimate goes to zero, the routine can still generate
91 * random numbers; however, an attacker may (at least in theory) be
92 * able to infer the future output of the generator from prior
93 * outputs. This requires successful cryptanalysis of SHA, which is
94 * not believed to be feasible, but there is a remote possibility.
95 * Nonetheless, these numbers should be useful for the vast majority
96 * of purposes.
97 *
98 * Exported interfaces ---- output
99 * ===============================
100 *
101 * There are three exported interfaces; the first is one designed to
102 * be used from within the kernel:
103 *
104 * void get_random_bytes(void *buf, int nbytes);
105 *
106 * This interface will return the requested number of random bytes,
107 * and place it in the requested buffer.
108 *
109 * The two other interfaces are two character devices /dev/random and
110 * /dev/urandom. /dev/random is suitable for use when very high
111 * quality randomness is desired (for example, for key generation or
112 * one-time pads), as it will only return a maximum of the number of
113 * bits of randomness (as estimated by the random number generator)
114 * contained in the entropy pool.
115 *
116 * The /dev/urandom device does not have this limit, and will return
117 * as many bytes as are requested. As more and more random bytes are
118 * requested without giving time for the entropy pool to recharge,
119 * this will result in random numbers that are merely cryptographically
120 * strong. For many applications, however, this is acceptable.
121 *
122 * Exported interfaces ---- input
123 * ==============================
124 *
125 * The current exported interfaces for gathering environmental noise
126 * from the devices are:
127 *
128 * void add_input_randomness(unsigned int type, unsigned int code,
129 * unsigned int value);
130 * void add_interrupt_randomness(int irq);
131 *
132 * add_input_randomness() uses the input layer interrupt timing, as well as
133 * the event type information from the hardware.
134 *
135 * add_interrupt_randomness() uses the inter-interrupt timing as random
136 * inputs to the entropy pool. Note that not all interrupts are good
137 * sources of randomness! For example, the timer interrupts is not a
138 * good choice, because the periodicity of the interrupts is too
139 * regular, and hence predictable to an attacker. Disk interrupts are
140 * a better measure, since the timing of the disk interrupts are more
141 * unpredictable.
142 *
143 * All of these routines try to estimate how many bits of randomness a
144 * particular randomness source. They do this by keeping track of the
145 * first and second order deltas of the event timings.
146 *
147 * Ensuring unpredictability at system startup
148 * ============================================
149 *
150 * When any operating system starts up, it will go through a sequence
151 * of actions that are fairly predictable by an adversary, especially
152 * if the start-up does not involve interaction with a human operator.
153 * This reduces the actual number of bits of unpredictability in the
154 * entropy pool below the value in entropy_count. In order to
155 * counteract this effect, it helps to carry information in the
156 * entropy pool across shut-downs and start-ups. To do this, put the
157 * following lines an appropriate script which is run during the boot
158 * sequence:
159 *
160 * echo "Initializing random number generator..."
161 * random_seed=/var/run/random-seed
162 * # Carry a random seed from start-up to start-up
163 * # Load and then save the whole entropy pool
164 * if [ -f $random_seed ]; then
165 * cat $random_seed >/dev/urandom
166 * else
167 * touch $random_seed
168 * fi
169 * chmod 600 $random_seed
170 * dd if=/dev/urandom of=$random_seed count=1 bs=512
171 *
172 * and the following lines in an appropriate script which is run as
173 * the system is shutdown:
174 *
175 * # Carry a random seed from shut-down to start-up
176 * # Save the whole entropy pool
177 * echo "Saving random seed..."
178 * random_seed=/var/run/random-seed
179 * touch $random_seed
180 * chmod 600 $random_seed
181 * dd if=/dev/urandom of=$random_seed count=1 bs=512
182 *
183 * For example, on most modern systems using the System V init
184 * scripts, such code fragments would be found in
185 * /etc/rc.d/init.d/random. On older Linux systems, the correct script
186 * location might be in /etc/rcb.d/rc.local or /etc/rc.d/rc.0.
187 *
188 * Effectively, these commands cause the contents of the entropy pool
189 * to be saved at shut-down time and reloaded into the entropy pool at
190 * start-up. (The 'dd' in the addition to the bootup script is to
191 * make sure that /etc/random-seed is different for every start-up,
192 * even if the system crashes without executing rc.0.) Even with
193 * complete knowledge of the start-up activities, predicting the state
194 * of the entropy pool requires knowledge of the previous history of
195 * the system.
196 *
197 * Configuring the /dev/random driver under Linux
198 * ==============================================
199 *
200 * The /dev/random driver under Linux uses minor numbers 8 and 9 of
201 * the /dev/mem major number (#1). So if your system does not have
202 * /dev/random and /dev/urandom created already, they can be created
203 * by using the commands:
204 *
205 * mknod /dev/random c 1 8
206 * mknod /dev/urandom c 1 9
207 *
208 * Acknowledgements:
209 * =================
210 *
211 * Ideas for constructing this random number generator were derived
212 * from Pretty Good Privacy's random number generator, and from private
213 * discussions with Phil Karn. Colin Plumb provided a faster random
214 * number generator, which speed up the mixing function of the entropy
215 * pool, taken from PGPfone. Dale Worley has also contributed many
216 * useful ideas and suggestions to improve this driver.
217 *
218 * Any flaws in the design are solely my responsibility, and should
219 * not be attributed to the Phil, Colin, or any of authors of PGP.
220 *
221 * Further background information on this topic may be obtained from
222 * RFC 1750, "Randomness Recommendations for Security", by Donald
223 * Eastlake, Steve Crocker, and Jeff Schiller.
224 */
225
226 #include <linux/utsname.h>
227 #include <linux/module.h>
228 #include <linux/kernel.h>
229 #include <linux/major.h>
230 #include <linux/string.h>
231 #include <linux/fcntl.h>
232 #include <linux/slab.h>
233 #include <linux/random.h>
234 #include <linux/poll.h>
235 #include <linux/init.h>
236 #include <linux/fs.h>
237 #include <linux/genhd.h>
238 #include <linux/interrupt.h>
239 #include <linux/mm.h>
240 #include <linux/spinlock.h>
241 #include <linux/percpu.h>
242 #include <linux/cryptohash.h>
243
244 #ifdef CONFIG_GENERIC_HARDIRQS
245 # include <linux/irq.h>
246 #endif
247
248 #include <asm/processor.h>
249 #include <asm/uaccess.h>
250 #include <asm/irq.h>
251 #include <asm/io.h>
252
253 /*
254 * Configuration information
255 */
256 #define INPUT_POOL_WORDS 128
257 #define OUTPUT_POOL_WORDS 32
258 #define SEC_XFER_SIZE 512
259
260 /*
261 * The minimum number of bits of entropy before we wake up a read on
262 * /dev/random. Should be enough to do a significant reseed.
263 */
264 static int random_read_wakeup_thresh = 64;
265
266 /*
267 * If the entropy count falls under this number of bits, then we
268 * should wake up processes which are selecting or polling on write
269 * access to /dev/random.
270 */
271 static int random_write_wakeup_thresh = 128;
272
273 /*
274 * When the input pool goes over trickle_thresh, start dropping most
275 * samples to avoid wasting CPU time and reduce lock contention.
276 */
277
278 static int trickle_thresh __read_mostly = INPUT_POOL_WORDS * 28;
279
280 static DEFINE_PER_CPU(int, trickle_count);
281
282 /*
283 * A pool of size .poolwords is stirred with a primitive polynomial
284 * of degree .poolwords over GF(2). The taps for various sizes are
285 * defined below. They are chosen to be evenly spaced (minimum RMS
286 * distance from evenly spaced; the numbers in the comments are a
287 * scaled squared error sum) except for the last tap, which is 1 to
288 * get the twisting happening as fast as possible.
289 */
290 static struct poolinfo {
291 int poolwords;
292 int tap1, tap2, tap3, tap4, tap5;
293 } poolinfo_table[] = {
294 /* x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 -- 105 */
295 { 128, 103, 76, 51, 25, 1 },
296 /* x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 -- 15 */
297 { 32, 26, 20, 14, 7, 1 },
298 #if 0
299 /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1 -- 115 */
300 { 2048, 1638, 1231, 819, 411, 1 },
301
302 /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
303 { 1024, 817, 615, 412, 204, 1 },
304
305 /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
306 { 1024, 819, 616, 410, 207, 2 },
307
308 /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
309 { 512, 411, 308, 208, 104, 1 },
310
311 /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
312 { 512, 409, 307, 206, 102, 2 },
313 /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
314 { 512, 409, 309, 205, 103, 2 },
315
316 /* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
317 { 256, 205, 155, 101, 52, 1 },
318
319 /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
320 { 128, 103, 78, 51, 27, 2 },
321
322 /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
323 { 64, 52, 39, 26, 14, 1 },
324 #endif
325 };
326
327 #define POOLBITS poolwords*32
328 #define POOLBYTES poolwords*4
329
330 /*
331 * For the purposes of better mixing, we use the CRC-32 polynomial as
332 * well to make a twisted Generalized Feedback Shift Reigster
333 *
334 * (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR generators. ACM
335 * Transactions on Modeling and Computer Simulation 2(3):179-194.
336 * Also see M. Matsumoto & Y. Kurita, 1994. Twisted GFSR generators
337 * II. ACM Transactions on Mdeling and Computer Simulation 4:254-266)
338 *
339 * Thanks to Colin Plumb for suggesting this.
340 *
341 * We have not analyzed the resultant polynomial to prove it primitive;
342 * in fact it almost certainly isn't. Nonetheless, the irreducible factors
343 * of a random large-degree polynomial over GF(2) are more than large enough
344 * that periodicity is not a concern.
345 *
346 * The input hash is much less sensitive than the output hash. All
347 * that we want of it is that it be a good non-cryptographic hash;
348 * i.e. it not produce collisions when fed "random" data of the sort
349 * we expect to see. As long as the pool state differs for different
350 * inputs, we have preserved the input entropy and done a good job.
351 * The fact that an intelligent attacker can construct inputs that
352 * will produce controlled alterations to the pool's state is not
353 * important because we don't consider such inputs to contribute any
354 * randomness. The only property we need with respect to them is that
355 * the attacker can't increase his/her knowledge of the pool's state.
356 * Since all additions are reversible (knowing the final state and the
357 * input, you can reconstruct the initial state), if an attacker has
358 * any uncertainty about the initial state, he/she can only shuffle
359 * that uncertainty about, but never cause any collisions (which would
360 * decrease the uncertainty).
361 *
362 * The chosen system lets the state of the pool be (essentially) the input
363 * modulo the generator polymnomial. Now, for random primitive polynomials,
364 * this is a universal class of hash functions, meaning that the chance
365 * of a collision is limited by the attacker's knowledge of the generator
366 * polynomail, so if it is chosen at random, an attacker can never force
367 * a collision. Here, we use a fixed polynomial, but we *can* assume that
368 * ###--> it is unknown to the processes generating the input entropy. <-###
369 * Because of this important property, this is a good, collision-resistant
370 * hash; hash collisions will occur no more often than chance.
371 */
372
373 /*
374 * Static global variables
375 */
376 static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
377 static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
378 static struct fasync_struct *fasync;
379
380 #if 0
381 static int debug;
382 module_param(debug, bool, 0644);
383 #define DEBUG_ENT(fmt, arg...) do { \
384 if (debug) \
385 printk(KERN_DEBUG "random %04d %04d %04d: " \
386 fmt,\
387 input_pool.entropy_count,\
388 blocking_pool.entropy_count,\
389 nonblocking_pool.entropy_count,\
390 ## arg); } while (0)
391 #else
392 #define DEBUG_ENT(fmt, arg...) do {} while (0)
393 #endif
394
395 /**********************************************************************
396 *
397 * OS independent entropy store. Here are the functions which handle
398 * storing entropy in an entropy pool.
399 *
400 **********************************************************************/
401
402 struct entropy_store;
403 struct entropy_store {
404 /* read-only data: */
405 struct poolinfo *poolinfo;
406 __u32 *pool;
407 const char *name;
408 int limit;
409 struct entropy_store *pull;
410
411 /* read-write data: */
412 spinlock_t lock;
413 unsigned add_ptr;
414 int entropy_count;
415 int input_rotate;
416 };
417
418 static __u32 input_pool_data[INPUT_POOL_WORDS];
419 static __u32 blocking_pool_data[OUTPUT_POOL_WORDS];
420 static __u32 nonblocking_pool_data[OUTPUT_POOL_WORDS];
421
422 static struct entropy_store input_pool = {
423 .poolinfo = &poolinfo_table[0],
424 .name = "input",
425 .limit = 1,
426 .lock = __SPIN_LOCK_UNLOCKED(&input_pool.lock),
427 .pool = input_pool_data
428 };
429
430 static struct entropy_store blocking_pool = {
431 .poolinfo = &poolinfo_table[1],
432 .name = "blocking",
433 .limit = 1,
434 .pull = &input_pool,
435 .lock = __SPIN_LOCK_UNLOCKED(&blocking_pool.lock),
436 .pool = blocking_pool_data
437 };
438
439 static struct entropy_store nonblocking_pool = {
440 .poolinfo = &poolinfo_table[1],
441 .name = "nonblocking",
442 .pull = &input_pool,
443 .lock = __SPIN_LOCK_UNLOCKED(&nonblocking_pool.lock),
444 .pool = nonblocking_pool_data
445 };
446
447 /*
448 * This function adds bytes into the entropy "pool". It does not
449 * update the entropy estimate. The caller should call
450 * credit_entropy_bits if this is appropriate.
451 *
452 * The pool is stirred with a primitive polynomial of the appropriate
453 * degree, and then twisted. We twist by three bits at a time because
454 * it's cheap to do so and helps slightly in the expected case where
455 * the entropy is concentrated in the low-order bits.
456 */
457 static void mix_pool_bytes_extract(struct entropy_store *r, const void *in,
458 int nbytes, __u8 out[64])
459 {
460 static __u32 const twist_table[8] = {
461 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
462 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
463 unsigned long i, j, tap1, tap2, tap3, tap4, tap5;
464 int input_rotate;
465 int wordmask = r->poolinfo->poolwords - 1;
466 const char *bytes = in;
467 __u32 w;
468 unsigned long flags;
469
470 /* Taps are constant, so we can load them without holding r->lock. */
471 tap1 = r->poolinfo->tap1;
472 tap2 = r->poolinfo->tap2;
473 tap3 = r->poolinfo->tap3;
474 tap4 = r->poolinfo->tap4;
475 tap5 = r->poolinfo->tap5;
476
477 spin_lock_irqsave(&r->lock, flags);
478 input_rotate = r->input_rotate;
479 i = r->add_ptr;
480
481 /* mix one byte at a time to simplify size handling and churn faster */
482 while (nbytes--) {
483 w = rol32(*bytes++, input_rotate & 31);
484 i = (i - 1) & wordmask;
485
486 /* XOR in the various taps */
487 w ^= r->pool[i];
488 w ^= r->pool[(i + tap1) & wordmask];
489 w ^= r->pool[(i + tap2) & wordmask];
490 w ^= r->pool[(i + tap3) & wordmask];
491 w ^= r->pool[(i + tap4) & wordmask];
492 w ^= r->pool[(i + tap5) & wordmask];
493
494 /* Mix the result back in with a twist */
495 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
496
497 /*
498 * Normally, we add 7 bits of rotation to the pool.
499 * At the beginning of the pool, add an extra 7 bits
500 * rotation, so that successive passes spread the
501 * input bits across the pool evenly.
502 */
503 input_rotate += i ? 7 : 14;
504 }
505
506 r->input_rotate = input_rotate;
507 r->add_ptr = i;
508
509 if (out)
510 for (j = 0; j < 16; j++)
511 ((__u32 *)out)[j] = r->pool[(i - j) & wordmask];
512
513 spin_unlock_irqrestore(&r->lock, flags);
514 }
515
516 static void mix_pool_bytes(struct entropy_store *r, const void *in, int bytes)
517 {
518 mix_pool_bytes_extract(r, in, bytes, NULL);
519 }
520
521 /*
522 * Credit (or debit) the entropy store with n bits of entropy
523 */
524 static void credit_entropy_bits(struct entropy_store *r, int nbits)
525 {
526 unsigned long flags;
527 int entropy_count;
528
529 if (!nbits)
530 return;
531
532 spin_lock_irqsave(&r->lock, flags);
533
534 DEBUG_ENT("added %d entropy credits to %s\n", nbits, r->name);
535 entropy_count = r->entropy_count;
536 entropy_count += nbits;
537 if (entropy_count < 0) {
538 DEBUG_ENT("negative entropy/overflow\n");
539 entropy_count = 0;
540 } else if (entropy_count > r->poolinfo->POOLBITS)
541 entropy_count = r->poolinfo->POOLBITS;
542 r->entropy_count = entropy_count;
543
544 /* should we wake readers? */
545 if (r == &input_pool && entropy_count >= random_read_wakeup_thresh) {
546 wake_up_interruptible(&random_read_wait);
547 kill_fasync(&fasync, SIGIO, POLL_IN);
548 }
549 spin_unlock_irqrestore(&r->lock, flags);
550 }
551
552 /*********************************************************************
553 *
554 * Entropy input management
555 *
556 *********************************************************************/
557
558 /* There is one of these per entropy source */
559 struct timer_rand_state {
560 cycles_t last_time;
561 long last_delta, last_delta2;
562 unsigned dont_count_entropy:1;
563 };
564
565 #ifndef CONFIG_GENERIC_HARDIRQS
566
567 static struct timer_rand_state *irq_timer_state[NR_IRQS];
568
569 static struct timer_rand_state *get_timer_rand_state(unsigned int irq)
570 {
571 return irq_timer_state[irq];
572 }
573
574 static void set_timer_rand_state(unsigned int irq,
575 struct timer_rand_state *state)
576 {
577 irq_timer_state[irq] = state;
578 }
579
580 #else
581
582 static struct timer_rand_state *get_timer_rand_state(unsigned int irq)
583 {
584 struct irq_desc *desc;
585
586 desc = irq_to_desc(irq);
587
588 return desc->timer_rand_state;
589 }
590
591 static void set_timer_rand_state(unsigned int irq,
592 struct timer_rand_state *state)
593 {
594 struct irq_desc *desc;
595
596 desc = irq_to_desc(irq);
597
598 desc->timer_rand_state = state;
599 }
600 #endif
601
602 static struct timer_rand_state input_timer_state;
603
604 /*
605 * This function adds entropy to the entropy "pool" by using timing
606 * delays. It uses the timer_rand_state structure to make an estimate
607 * of how many bits of entropy this call has added to the pool.
608 *
609 * The number "num" is also added to the pool - it should somehow describe
610 * the type of event which just happened. This is currently 0-255 for
611 * keyboard scan codes, and 256 upwards for interrupts.
612 *
613 */
614 static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
615 {
616 struct {
617 cycles_t cycles;
618 long jiffies;
619 unsigned num;
620 } sample;
621 long delta, delta2, delta3;
622
623 preempt_disable();
624 /* if over the trickle threshold, use only 1 in 4096 samples */
625 if (input_pool.entropy_count > trickle_thresh &&
626 (__get_cpu_var(trickle_count)++ & 0xfff))
627 goto out;
628
629 sample.jiffies = jiffies;
630 sample.cycles = get_cycles();
631 sample.num = num;
632 mix_pool_bytes(&input_pool, &sample, sizeof(sample));
633
634 /*
635 * Calculate number of bits of randomness we probably added.
636 * We take into account the first, second and third-order deltas
637 * in order to make our estimate.
638 */
639
640 if (!state->dont_count_entropy) {
641 delta = sample.jiffies - state->last_time;
642 state->last_time = sample.jiffies;
643
644 delta2 = delta - state->last_delta;
645 state->last_delta = delta;
646
647 delta3 = delta2 - state->last_delta2;
648 state->last_delta2 = delta2;
649
650 if (delta < 0)
651 delta = -delta;
652 if (delta2 < 0)
653 delta2 = -delta2;
654 if (delta3 < 0)
655 delta3 = -delta3;
656 if (delta > delta2)
657 delta = delta2;
658 if (delta > delta3)
659 delta = delta3;
660
661 /*
662 * delta is now minimum absolute delta.
663 * Round down by 1 bit on general principles,
664 * and limit entropy entimate to 12 bits.
665 */
666 credit_entropy_bits(&input_pool,
667 min_t(int, fls(delta>>1), 11));
668 }
669 out:
670 preempt_enable();
671 }
672
673 void add_input_randomness(unsigned int type, unsigned int code,
674 unsigned int value)
675 {
676 static unsigned char last_value;
677
678 /* ignore autorepeat and the like */
679 if (value == last_value)
680 return;
681
682 DEBUG_ENT("input event\n");
683 last_value = value;
684 add_timer_randomness(&input_timer_state,
685 (type << 4) ^ code ^ (code >> 4) ^ value);
686 }
687 EXPORT_SYMBOL_GPL(add_input_randomness);
688
689 void add_interrupt_randomness(int irq)
690 {
691 struct timer_rand_state *state;
692
693 state = get_timer_rand_state(irq);
694
695 if (state == NULL)
696 return;
697
698 DEBUG_ENT("irq event %d\n", irq);
699 add_timer_randomness(state, 0x100 + irq);
700 }
701
702 #ifdef CONFIG_BLOCK
703 void add_disk_randomness(struct gendisk *disk)
704 {
705 if (!disk || !disk->random)
706 return;
707 /* first major is 1, so we get >= 0x200 here */
708 DEBUG_ENT("disk event %d:%d\n",
709 MAJOR(disk_devt(disk)), MINOR(disk_devt(disk)));
710
711 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
712 }
713 #endif
714
715 #define EXTRACT_SIZE 10
716
717 /*********************************************************************
718 *
719 * Entropy extraction routines
720 *
721 *********************************************************************/
722
723 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
724 size_t nbytes, int min, int rsvd);
725
726 /*
727 * This utility inline function is responsible for transfering entropy
728 * from the primary pool to the secondary extraction pool. We make
729 * sure we pull enough for a 'catastrophic reseed'.
730 */
731 static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
732 {
733 __u32 tmp[OUTPUT_POOL_WORDS];
734
735 if (r->pull && r->entropy_count < nbytes * 8 &&
736 r->entropy_count < r->poolinfo->POOLBITS) {
737 /* If we're limited, always leave two wakeup worth's BITS */
738 int rsvd = r->limit ? 0 : random_read_wakeup_thresh/4;
739 int bytes = nbytes;
740
741 /* pull at least as many as BYTES as wakeup BITS */
742 bytes = max_t(int, bytes, random_read_wakeup_thresh / 8);
743 /* but never more than the buffer size */
744 bytes = min_t(int, bytes, sizeof(tmp));
745
746 DEBUG_ENT("going to reseed %s with %d bits "
747 "(%d of %d requested)\n",
748 r->name, bytes * 8, nbytes * 8, r->entropy_count);
749
750 bytes = extract_entropy(r->pull, tmp, bytes,
751 random_read_wakeup_thresh / 8, rsvd);
752 mix_pool_bytes(r, tmp, bytes);
753 credit_entropy_bits(r, bytes*8);
754 }
755 }
756
757 /*
758 * These functions extracts randomness from the "entropy pool", and
759 * returns it in a buffer.
760 *
761 * The min parameter specifies the minimum amount we can pull before
762 * failing to avoid races that defeat catastrophic reseeding while the
763 * reserved parameter indicates how much entropy we must leave in the
764 * pool after each pull to avoid starving other readers.
765 *
766 * Note: extract_entropy() assumes that .poolwords is a multiple of 16 words.
767 */
768
769 static size_t account(struct entropy_store *r, size_t nbytes, int min,
770 int reserved)
771 {
772 unsigned long flags;
773
774 /* Hold lock while accounting */
775 spin_lock_irqsave(&r->lock, flags);
776
777 BUG_ON(r->entropy_count > r->poolinfo->POOLBITS);
778 DEBUG_ENT("trying to extract %d bits from %s\n",
779 nbytes * 8, r->name);
780
781 /* Can we pull enough? */
782 if (r->entropy_count / 8 < min + reserved) {
783 nbytes = 0;
784 } else {
785 /* If limited, never pull more than available */
786 if (r->limit && nbytes + reserved >= r->entropy_count / 8)
787 nbytes = r->entropy_count/8 - reserved;
788
789 if (r->entropy_count / 8 >= nbytes + reserved)
790 r->entropy_count -= nbytes*8;
791 else
792 r->entropy_count = reserved;
793
794 if (r->entropy_count < random_write_wakeup_thresh) {
795 wake_up_interruptible(&random_write_wait);
796 kill_fasync(&fasync, SIGIO, POLL_OUT);
797 }
798 }
799
800 DEBUG_ENT("debiting %d entropy credits from %s%s\n",
801 nbytes * 8, r->name, r->limit ? "" : " (unlimited)");
802
803 spin_unlock_irqrestore(&r->lock, flags);
804
805 return nbytes;
806 }
807
808 static void extract_buf(struct entropy_store *r, __u8 *out)
809 {
810 int i;
811 __u32 hash[5], workspace[SHA_WORKSPACE_WORDS];
812 __u8 extract[64];
813
814 /* Generate a hash across the pool, 16 words (512 bits) at a time */
815 sha_init(hash);
816 for (i = 0; i < r->poolinfo->poolwords; i += 16)
817 sha_transform(hash, (__u8 *)(r->pool + i), workspace);
818
819 /*
820 * We mix the hash back into the pool to prevent backtracking
821 * attacks (where the attacker knows the state of the pool
822 * plus the current outputs, and attempts to find previous
823 * ouputs), unless the hash function can be inverted. By
824 * mixing at least a SHA1 worth of hash data back, we make
825 * brute-forcing the feedback as hard as brute-forcing the
826 * hash.
827 */
828 mix_pool_bytes_extract(r, hash, sizeof(hash), extract);
829
830 /*
831 * To avoid duplicates, we atomically extract a portion of the
832 * pool while mixing, and hash one final time.
833 */
834 sha_transform(hash, extract, workspace);
835 memset(extract, 0, sizeof(extract));
836 memset(workspace, 0, sizeof(workspace));
837
838 /*
839 * In case the hash function has some recognizable output
840 * pattern, we fold it in half. Thus, we always feed back
841 * twice as much data as we output.
842 */
843 hash[0] ^= hash[3];
844 hash[1] ^= hash[4];
845 hash[2] ^= rol32(hash[2], 16);
846 memcpy(out, hash, EXTRACT_SIZE);
847 memset(hash, 0, sizeof(hash));
848 }
849
850 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
851 size_t nbytes, int min, int reserved)
852 {
853 ssize_t ret = 0, i;
854 __u8 tmp[EXTRACT_SIZE];
855
856 xfer_secondary_pool(r, nbytes);
857 nbytes = account(r, nbytes, min, reserved);
858
859 while (nbytes) {
860 extract_buf(r, tmp);
861 i = min_t(int, nbytes, EXTRACT_SIZE);
862 memcpy(buf, tmp, i);
863 nbytes -= i;
864 buf += i;
865 ret += i;
866 }
867
868 /* Wipe data just returned from memory */
869 memset(tmp, 0, sizeof(tmp));
870
871 return ret;
872 }
873
874 static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
875 size_t nbytes)
876 {
877 ssize_t ret = 0, i;
878 __u8 tmp[EXTRACT_SIZE];
879
880 xfer_secondary_pool(r, nbytes);
881 nbytes = account(r, nbytes, 0, 0);
882
883 while (nbytes) {
884 if (need_resched()) {
885 if (signal_pending(current)) {
886 if (ret == 0)
887 ret = -ERESTARTSYS;
888 break;
889 }
890 schedule();
891 }
892
893 extract_buf(r, tmp);
894 i = min_t(int, nbytes, EXTRACT_SIZE);
895 if (copy_to_user(buf, tmp, i)) {
896 ret = -EFAULT;
897 break;
898 }
899
900 nbytes -= i;
901 buf += i;
902 ret += i;
903 }
904
905 /* Wipe data just returned from memory */
906 memset(tmp, 0, sizeof(tmp));
907
908 return ret;
909 }
910
911 /*
912 * This function is the exported kernel interface. It returns some
913 * number of good random numbers, suitable for seeding TCP sequence
914 * numbers, etc.
915 */
916 void get_random_bytes(void *buf, int nbytes)
917 {
918 extract_entropy(&nonblocking_pool, buf, nbytes, 0, 0);
919 }
920 EXPORT_SYMBOL(get_random_bytes);
921
922 /*
923 * init_std_data - initialize pool with system data
924 *
925 * @r: pool to initialize
926 *
927 * This function clears the pool's entropy count and mixes some system
928 * data into the pool to prepare it for use. The pool is not cleared
929 * as that can only decrease the entropy in the pool.
930 */
931 static void init_std_data(struct entropy_store *r)
932 {
933 ktime_t now;
934 unsigned long flags;
935
936 spin_lock_irqsave(&r->lock, flags);
937 r->entropy_count = 0;
938 spin_unlock_irqrestore(&r->lock, flags);
939
940 now = ktime_get_real();
941 mix_pool_bytes(r, &now, sizeof(now));
942 mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
943 }
944
945 static int rand_initialize(void)
946 {
947 init_std_data(&input_pool);
948 init_std_data(&blocking_pool);
949 init_std_data(&nonblocking_pool);
950 return 0;
951 }
952 module_init(rand_initialize);
953
954 void rand_initialize_irq(int irq)
955 {
956 struct timer_rand_state *state;
957
958 state = get_timer_rand_state(irq);
959
960 if (state)
961 return;
962
963 /*
964 * If kzalloc returns null, we just won't use that entropy
965 * source.
966 */
967 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
968 if (state)
969 set_timer_rand_state(irq, state);
970 }
971
972 #ifdef CONFIG_BLOCK
973 void rand_initialize_disk(struct gendisk *disk)
974 {
975 struct timer_rand_state *state;
976
977 /*
978 * If kzalloc returns null, we just won't use that entropy
979 * source.
980 */
981 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
982 if (state)
983 disk->random = state;
984 }
985 #endif
986
987 static ssize_t
988 random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
989 {
990 ssize_t n, retval = 0, count = 0;
991
992 if (nbytes == 0)
993 return 0;
994
995 while (nbytes > 0) {
996 n = nbytes;
997 if (n > SEC_XFER_SIZE)
998 n = SEC_XFER_SIZE;
999
1000 DEBUG_ENT("reading %d bits\n", n*8);
1001
1002 n = extract_entropy_user(&blocking_pool, buf, n);
1003
1004 DEBUG_ENT("read got %d bits (%d still needed)\n",
1005 n*8, (nbytes-n)*8);
1006
1007 if (n == 0) {
1008 if (file->f_flags & O_NONBLOCK) {
1009 retval = -EAGAIN;
1010 break;
1011 }
1012
1013 DEBUG_ENT("sleeping?\n");
1014
1015 wait_event_interruptible(random_read_wait,
1016 input_pool.entropy_count >=
1017 random_read_wakeup_thresh);
1018
1019 DEBUG_ENT("awake\n");
1020
1021 if (signal_pending(current)) {
1022 retval = -ERESTARTSYS;
1023 break;
1024 }
1025
1026 continue;
1027 }
1028
1029 if (n < 0) {
1030 retval = n;
1031 break;
1032 }
1033 count += n;
1034 buf += n;
1035 nbytes -= n;
1036 break; /* This break makes the device work */
1037 /* like a named pipe */
1038 }
1039
1040 /*
1041 * If we gave the user some bytes, update the access time.
1042 */
1043 if (count)
1044 file_accessed(file);
1045
1046 return (count ? count : retval);
1047 }
1048
1049 static ssize_t
1050 urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1051 {
1052 return extract_entropy_user(&nonblocking_pool, buf, nbytes);
1053 }
1054
1055 static unsigned int
1056 random_poll(struct file *file, poll_table * wait)
1057 {
1058 unsigned int mask;
1059
1060 poll_wait(file, &random_read_wait, wait);
1061 poll_wait(file, &random_write_wait, wait);
1062 mask = 0;
1063 if (input_pool.entropy_count >= random_read_wakeup_thresh)
1064 mask |= POLLIN | POLLRDNORM;
1065 if (input_pool.entropy_count < random_write_wakeup_thresh)
1066 mask |= POLLOUT | POLLWRNORM;
1067 return mask;
1068 }
1069
1070 static int
1071 write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1072 {
1073 size_t bytes;
1074 __u32 buf[16];
1075 const char __user *p = buffer;
1076
1077 while (count > 0) {
1078 bytes = min(count, sizeof(buf));
1079 if (copy_from_user(&buf, p, bytes))
1080 return -EFAULT;
1081
1082 count -= bytes;
1083 p += bytes;
1084
1085 mix_pool_bytes(r, buf, bytes);
1086 cond_resched();
1087 }
1088
1089 return 0;
1090 }
1091
1092 static ssize_t random_write(struct file *file, const char __user *buffer,
1093 size_t count, loff_t *ppos)
1094 {
1095 size_t ret;
1096 struct inode *inode = file->f_path.dentry->d_inode;
1097
1098 ret = write_pool(&blocking_pool, buffer, count);
1099 if (ret)
1100 return ret;
1101 ret = write_pool(&nonblocking_pool, buffer, count);
1102 if (ret)
1103 return ret;
1104
1105 inode->i_mtime = current_fs_time(inode->i_sb);
1106 mark_inode_dirty(inode);
1107 return (ssize_t)count;
1108 }
1109
1110 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1111 {
1112 int size, ent_count;
1113 int __user *p = (int __user *)arg;
1114 int retval;
1115
1116 switch (cmd) {
1117 case RNDGETENTCNT:
1118 /* inherently racy, no point locking */
1119 if (put_user(input_pool.entropy_count, p))
1120 return -EFAULT;
1121 return 0;
1122 case RNDADDTOENTCNT:
1123 if (!capable(CAP_SYS_ADMIN))
1124 return -EPERM;
1125 if (get_user(ent_count, p))
1126 return -EFAULT;
1127 credit_entropy_bits(&input_pool, ent_count);
1128 return 0;
1129 case RNDADDENTROPY:
1130 if (!capable(CAP_SYS_ADMIN))
1131 return -EPERM;
1132 if (get_user(ent_count, p++))
1133 return -EFAULT;
1134 if (ent_count < 0)
1135 return -EINVAL;
1136 if (get_user(size, p++))
1137 return -EFAULT;
1138 retval = write_pool(&input_pool, (const char __user *)p,
1139 size);
1140 if (retval < 0)
1141 return retval;
1142 credit_entropy_bits(&input_pool, ent_count);
1143 return 0;
1144 case RNDZAPENTCNT:
1145 case RNDCLEARPOOL:
1146 /* Clear the entropy pool counters. */
1147 if (!capable(CAP_SYS_ADMIN))
1148 return -EPERM;
1149 rand_initialize();
1150 return 0;
1151 default:
1152 return -EINVAL;
1153 }
1154 }
1155
1156 static int random_fasync(int fd, struct file *filp, int on)
1157 {
1158 return fasync_helper(fd, filp, on, &fasync);
1159 }
1160
1161 const struct file_operations random_fops = {
1162 .read = random_read,
1163 .write = random_write,
1164 .poll = random_poll,
1165 .unlocked_ioctl = random_ioctl,
1166 .fasync = random_fasync,
1167 };
1168
1169 const struct file_operations urandom_fops = {
1170 .read = urandom_read,
1171 .write = random_write,
1172 .unlocked_ioctl = random_ioctl,
1173 .fasync = random_fasync,
1174 };
1175
1176 /***************************************************************
1177 * Random UUID interface
1178 *
1179 * Used here for a Boot ID, but can be useful for other kernel
1180 * drivers.
1181 ***************************************************************/
1182
1183 /*
1184 * Generate random UUID
1185 */
1186 void generate_random_uuid(unsigned char uuid_out[16])
1187 {
1188 get_random_bytes(uuid_out, 16);
1189 /* Set UUID version to 4 --- truely random generation */
1190 uuid_out[6] = (uuid_out[6] & 0x0F) | 0x40;
1191 /* Set the UUID variant to DCE */
1192 uuid_out[8] = (uuid_out[8] & 0x3F) | 0x80;
1193 }
1194 EXPORT_SYMBOL(generate_random_uuid);
1195
1196 /********************************************************************
1197 *
1198 * Sysctl interface
1199 *
1200 ********************************************************************/
1201
1202 #ifdef CONFIG_SYSCTL
1203
1204 #include <linux/sysctl.h>
1205
1206 static int min_read_thresh = 8, min_write_thresh;
1207 static int max_read_thresh = INPUT_POOL_WORDS * 32;
1208 static int max_write_thresh = INPUT_POOL_WORDS * 32;
1209 static char sysctl_bootid[16];
1210
1211 /*
1212 * These functions is used to return both the bootid UUID, and random
1213 * UUID. The difference is in whether table->data is NULL; if it is,
1214 * then a new UUID is generated and returned to the user.
1215 *
1216 * If the user accesses this via the proc interface, it will be returned
1217 * as an ASCII string in the standard UUID format. If accesses via the
1218 * sysctl system call, it is returned as 16 bytes of binary data.
1219 */
1220 static int proc_do_uuid(ctl_table *table, int write, struct file *filp,
1221 void __user *buffer, size_t *lenp, loff_t *ppos)
1222 {
1223 ctl_table fake_table;
1224 unsigned char buf[64], tmp_uuid[16], *uuid;
1225
1226 uuid = table->data;
1227 if (!uuid) {
1228 uuid = tmp_uuid;
1229 uuid[8] = 0;
1230 }
1231 if (uuid[8] == 0)
1232 generate_random_uuid(uuid);
1233
1234 sprintf(buf, "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-"
1235 "%02x%02x%02x%02x%02x%02x",
1236 uuid[0], uuid[1], uuid[2], uuid[3],
1237 uuid[4], uuid[5], uuid[6], uuid[7],
1238 uuid[8], uuid[9], uuid[10], uuid[11],
1239 uuid[12], uuid[13], uuid[14], uuid[15]);
1240 fake_table.data = buf;
1241 fake_table.maxlen = sizeof(buf);
1242
1243 return proc_dostring(&fake_table, write, filp, buffer, lenp, ppos);
1244 }
1245
1246 static int uuid_strategy(ctl_table *table,
1247 void __user *oldval, size_t __user *oldlenp,
1248 void __user *newval, size_t newlen)
1249 {
1250 unsigned char tmp_uuid[16], *uuid;
1251 unsigned int len;
1252
1253 if (!oldval || !oldlenp)
1254 return 1;
1255
1256 uuid = table->data;
1257 if (!uuid) {
1258 uuid = tmp_uuid;
1259 uuid[8] = 0;
1260 }
1261 if (uuid[8] == 0)
1262 generate_random_uuid(uuid);
1263
1264 if (get_user(len, oldlenp))
1265 return -EFAULT;
1266 if (len) {
1267 if (len > 16)
1268 len = 16;
1269 if (copy_to_user(oldval, uuid, len) ||
1270 put_user(len, oldlenp))
1271 return -EFAULT;
1272 }
1273 return 1;
1274 }
1275
1276 static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
1277 ctl_table random_table[] = {
1278 {
1279 .ctl_name = RANDOM_POOLSIZE,
1280 .procname = "poolsize",
1281 .data = &sysctl_poolsize,
1282 .maxlen = sizeof(int),
1283 .mode = 0444,
1284 .proc_handler = &proc_dointvec,
1285 },
1286 {
1287 .ctl_name = RANDOM_ENTROPY_COUNT,
1288 .procname = "entropy_avail",
1289 .maxlen = sizeof(int),
1290 .mode = 0444,
1291 .proc_handler = &proc_dointvec,
1292 .data = &input_pool.entropy_count,
1293 },
1294 {
1295 .ctl_name = RANDOM_READ_THRESH,
1296 .procname = "read_wakeup_threshold",
1297 .data = &random_read_wakeup_thresh,
1298 .maxlen = sizeof(int),
1299 .mode = 0644,
1300 .proc_handler = &proc_dointvec_minmax,
1301 .strategy = &sysctl_intvec,
1302 .extra1 = &min_read_thresh,
1303 .extra2 = &max_read_thresh,
1304 },
1305 {
1306 .ctl_name = RANDOM_WRITE_THRESH,
1307 .procname = "write_wakeup_threshold",
1308 .data = &random_write_wakeup_thresh,
1309 .maxlen = sizeof(int),
1310 .mode = 0644,
1311 .proc_handler = &proc_dointvec_minmax,
1312 .strategy = &sysctl_intvec,
1313 .extra1 = &min_write_thresh,
1314 .extra2 = &max_write_thresh,
1315 },
1316 {
1317 .ctl_name = RANDOM_BOOT_ID,
1318 .procname = "boot_id",
1319 .data = &sysctl_bootid,
1320 .maxlen = 16,
1321 .mode = 0444,
1322 .proc_handler = &proc_do_uuid,
1323 .strategy = &uuid_strategy,
1324 },
1325 {
1326 .ctl_name = RANDOM_UUID,
1327 .procname = "uuid",
1328 .maxlen = 16,
1329 .mode = 0444,
1330 .proc_handler = &proc_do_uuid,
1331 .strategy = &uuid_strategy,
1332 },
1333 { .ctl_name = 0 }
1334 };
1335 #endif /* CONFIG_SYSCTL */
1336
1337 /********************************************************************
1338 *
1339 * Random funtions for networking
1340 *
1341 ********************************************************************/
1342
1343 /*
1344 * TCP initial sequence number picking. This uses the random number
1345 * generator to pick an initial secret value. This value is hashed
1346 * along with the TCP endpoint information to provide a unique
1347 * starting point for each pair of TCP endpoints. This defeats
1348 * attacks which rely on guessing the initial TCP sequence number.
1349 * This algorithm was suggested by Steve Bellovin.
1350 *
1351 * Using a very strong hash was taking an appreciable amount of the total
1352 * TCP connection establishment time, so this is a weaker hash,
1353 * compensated for by changing the secret periodically.
1354 */
1355
1356 /* F, G and H are basic MD4 functions: selection, majority, parity */
1357 #define F(x, y, z) ((z) ^ ((x) & ((y) ^ (z))))
1358 #define G(x, y, z) (((x) & (y)) + (((x) ^ (y)) & (z)))
1359 #define H(x, y, z) ((x) ^ (y) ^ (z))
1360
1361 /*
1362 * The generic round function. The application is so specific that
1363 * we don't bother protecting all the arguments with parens, as is generally
1364 * good macro practice, in favor of extra legibility.
1365 * Rotation is separate from addition to prevent recomputation
1366 */
1367 #define ROUND(f, a, b, c, d, x, s) \
1368 (a += f(b, c, d) + x, a = (a << s) | (a >> (32 - s)))
1369 #define K1 0
1370 #define K2 013240474631UL
1371 #define K3 015666365641UL
1372
1373 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1374
1375 static __u32 twothirdsMD4Transform(__u32 const buf[4], __u32 const in[12])
1376 {
1377 __u32 a = buf[0], b = buf[1], c = buf[2], d = buf[3];
1378
1379 /* Round 1 */
1380 ROUND(F, a, b, c, d, in[ 0] + K1, 3);
1381 ROUND(F, d, a, b, c, in[ 1] + K1, 7);
1382 ROUND(F, c, d, a, b, in[ 2] + K1, 11);
1383 ROUND(F, b, c, d, a, in[ 3] + K1, 19);
1384 ROUND(F, a, b, c, d, in[ 4] + K1, 3);
1385 ROUND(F, d, a, b, c, in[ 5] + K1, 7);
1386 ROUND(F, c, d, a, b, in[ 6] + K1, 11);
1387 ROUND(F, b, c, d, a, in[ 7] + K1, 19);
1388 ROUND(F, a, b, c, d, in[ 8] + K1, 3);
1389 ROUND(F, d, a, b, c, in[ 9] + K1, 7);
1390 ROUND(F, c, d, a, b, in[10] + K1, 11);
1391 ROUND(F, b, c, d, a, in[11] + K1, 19);
1392
1393 /* Round 2 */
1394 ROUND(G, a, b, c, d, in[ 1] + K2, 3);
1395 ROUND(G, d, a, b, c, in[ 3] + K2, 5);
1396 ROUND(G, c, d, a, b, in[ 5] + K2, 9);
1397 ROUND(G, b, c, d, a, in[ 7] + K2, 13);
1398 ROUND(G, a, b, c, d, in[ 9] + K2, 3);
1399 ROUND(G, d, a, b, c, in[11] + K2, 5);
1400 ROUND(G, c, d, a, b, in[ 0] + K2, 9);
1401 ROUND(G, b, c, d, a, in[ 2] + K2, 13);
1402 ROUND(G, a, b, c, d, in[ 4] + K2, 3);
1403 ROUND(G, d, a, b, c, in[ 6] + K2, 5);
1404 ROUND(G, c, d, a, b, in[ 8] + K2, 9);
1405 ROUND(G, b, c, d, a, in[10] + K2, 13);
1406
1407 /* Round 3 */
1408 ROUND(H, a, b, c, d, in[ 3] + K3, 3);
1409 ROUND(H, d, a, b, c, in[ 7] + K3, 9);
1410 ROUND(H, c, d, a, b, in[11] + K3, 11);
1411 ROUND(H, b, c, d, a, in[ 2] + K3, 15);
1412 ROUND(H, a, b, c, d, in[ 6] + K3, 3);
1413 ROUND(H, d, a, b, c, in[10] + K3, 9);
1414 ROUND(H, c, d, a, b, in[ 1] + K3, 11);
1415 ROUND(H, b, c, d, a, in[ 5] + K3, 15);
1416 ROUND(H, a, b, c, d, in[ 9] + K3, 3);
1417 ROUND(H, d, a, b, c, in[ 0] + K3, 9);
1418 ROUND(H, c, d, a, b, in[ 4] + K3, 11);
1419 ROUND(H, b, c, d, a, in[ 8] + K3, 15);
1420
1421 return buf[1] + b; /* "most hashed" word */
1422 /* Alternative: return sum of all words? */
1423 }
1424 #endif
1425
1426 #undef ROUND
1427 #undef F
1428 #undef G
1429 #undef H
1430 #undef K1
1431 #undef K2
1432 #undef K3
1433
1434 /* This should not be decreased so low that ISNs wrap too fast. */
1435 #define REKEY_INTERVAL (300 * HZ)
1436 /*
1437 * Bit layout of the tcp sequence numbers (before adding current time):
1438 * bit 24-31: increased after every key exchange
1439 * bit 0-23: hash(source,dest)
1440 *
1441 * The implementation is similar to the algorithm described
1442 * in the Appendix of RFC 1185, except that
1443 * - it uses a 1 MHz clock instead of a 250 kHz clock
1444 * - it performs a rekey every 5 minutes, which is equivalent
1445 * to a (source,dest) tulple dependent forward jump of the
1446 * clock by 0..2^(HASH_BITS+1)
1447 *
1448 * Thus the average ISN wraparound time is 68 minutes instead of
1449 * 4.55 hours.
1450 *
1451 * SMP cleanup and lock avoidance with poor man's RCU.
1452 * Manfred Spraul <manfred@colorfullife.com>
1453 *
1454 */
1455 #define COUNT_BITS 8
1456 #define COUNT_MASK ((1 << COUNT_BITS) - 1)
1457 #define HASH_BITS 24
1458 #define HASH_MASK ((1 << HASH_BITS) - 1)
1459
1460 static struct keydata {
1461 __u32 count; /* already shifted to the final position */
1462 __u32 secret[12];
1463 } ____cacheline_aligned ip_keydata[2];
1464
1465 static unsigned int ip_cnt;
1466
1467 static void rekey_seq_generator(struct work_struct *work);
1468
1469 static DECLARE_DELAYED_WORK(rekey_work, rekey_seq_generator);
1470
1471 /*
1472 * Lock avoidance:
1473 * The ISN generation runs lockless - it's just a hash over random data.
1474 * State changes happen every 5 minutes when the random key is replaced.
1475 * Synchronization is performed by having two copies of the hash function
1476 * state and rekey_seq_generator always updates the inactive copy.
1477 * The copy is then activated by updating ip_cnt.
1478 * The implementation breaks down if someone blocks the thread
1479 * that processes SYN requests for more than 5 minutes. Should never
1480 * happen, and even if that happens only a not perfectly compliant
1481 * ISN is generated, nothing fatal.
1482 */
1483 static void rekey_seq_generator(struct work_struct *work)
1484 {
1485 struct keydata *keyptr = &ip_keydata[1 ^ (ip_cnt & 1)];
1486
1487 get_random_bytes(keyptr->secret, sizeof(keyptr->secret));
1488 keyptr->count = (ip_cnt & COUNT_MASK) << HASH_BITS;
1489 smp_wmb();
1490 ip_cnt++;
1491 schedule_delayed_work(&rekey_work,
1492 round_jiffies_relative(REKEY_INTERVAL));
1493 }
1494
1495 static inline struct keydata *get_keyptr(void)
1496 {
1497 struct keydata *keyptr = &ip_keydata[ip_cnt & 1];
1498
1499 smp_rmb();
1500
1501 return keyptr;
1502 }
1503
1504 static __init int seqgen_init(void)
1505 {
1506 rekey_seq_generator(NULL);
1507 return 0;
1508 }
1509 late_initcall(seqgen_init);
1510
1511 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1512 __u32 secure_tcpv6_sequence_number(__be32 *saddr, __be32 *daddr,
1513 __be16 sport, __be16 dport)
1514 {
1515 __u32 seq;
1516 __u32 hash[12];
1517 struct keydata *keyptr = get_keyptr();
1518
1519 /* The procedure is the same as for IPv4, but addresses are longer.
1520 * Thus we must use twothirdsMD4Transform.
1521 */
1522
1523 memcpy(hash, saddr, 16);
1524 hash[4] = ((__force u16)sport << 16) + (__force u16)dport;
1525 memcpy(&hash[5], keyptr->secret, sizeof(__u32) * 7);
1526
1527 seq = twothirdsMD4Transform((const __u32 *)daddr, hash) & HASH_MASK;
1528 seq += keyptr->count;
1529
1530 seq += ktime_to_ns(ktime_get_real());
1531
1532 return seq;
1533 }
1534 EXPORT_SYMBOL(secure_tcpv6_sequence_number);
1535 #endif
1536
1537 /* The code below is shamelessly stolen from secure_tcp_sequence_number().
1538 * All blames to Andrey V. Savochkin <saw@msu.ru>.
1539 */
1540 __u32 secure_ip_id(__be32 daddr)
1541 {
1542 struct keydata *keyptr;
1543 __u32 hash[4];
1544
1545 keyptr = get_keyptr();
1546
1547 /*
1548 * Pick a unique starting offset for each IP destination.
1549 * The dest ip address is placed in the starting vector,
1550 * which is then hashed with random data.
1551 */
1552 hash[0] = (__force __u32)daddr;
1553 hash[1] = keyptr->secret[9];
1554 hash[2] = keyptr->secret[10];
1555 hash[3] = keyptr->secret[11];
1556
1557 return half_md4_transform(hash, keyptr->secret);
1558 }
1559
1560 #ifdef CONFIG_INET
1561
1562 __u32 secure_tcp_sequence_number(__be32 saddr, __be32 daddr,
1563 __be16 sport, __be16 dport)
1564 {
1565 __u32 seq;
1566 __u32 hash[4];
1567 struct keydata *keyptr = get_keyptr();
1568
1569 /*
1570 * Pick a unique starting offset for each TCP connection endpoints
1571 * (saddr, daddr, sport, dport).
1572 * Note that the words are placed into the starting vector, which is
1573 * then mixed with a partial MD4 over random data.
1574 */
1575 hash[0] = (__force u32)saddr;
1576 hash[1] = (__force u32)daddr;
1577 hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
1578 hash[3] = keyptr->secret[11];
1579
1580 seq = half_md4_transform(hash, keyptr->secret) & HASH_MASK;
1581 seq += keyptr->count;
1582 /*
1583 * As close as possible to RFC 793, which
1584 * suggests using a 250 kHz clock.
1585 * Further reading shows this assumes 2 Mb/s networks.
1586 * For 10 Mb/s Ethernet, a 1 MHz clock is appropriate.
1587 * For 10 Gb/s Ethernet, a 1 GHz clock should be ok, but
1588 * we also need to limit the resolution so that the u32 seq
1589 * overlaps less than one time per MSL (2 minutes).
1590 * Choosing a clock of 64 ns period is OK. (period of 274 s)
1591 */
1592 seq += ktime_to_ns(ktime_get_real()) >> 6;
1593
1594 return seq;
1595 }
1596
1597 /* Generate secure starting point for ephemeral IPV4 transport port search */
1598 u32 secure_ipv4_port_ephemeral(__be32 saddr, __be32 daddr, __be16 dport)
1599 {
1600 struct keydata *keyptr = get_keyptr();
1601 u32 hash[4];
1602
1603 /*
1604 * Pick a unique starting offset for each ephemeral port search
1605 * (saddr, daddr, dport) and 48bits of random data.
1606 */
1607 hash[0] = (__force u32)saddr;
1608 hash[1] = (__force u32)daddr;
1609 hash[2] = (__force u32)dport ^ keyptr->secret[10];
1610 hash[3] = keyptr->secret[11];
1611
1612 return half_md4_transform(hash, keyptr->secret);
1613 }
1614 EXPORT_SYMBOL_GPL(secure_ipv4_port_ephemeral);
1615
1616 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1617 u32 secure_ipv6_port_ephemeral(const __be32 *saddr, const __be32 *daddr,
1618 __be16 dport)
1619 {
1620 struct keydata *keyptr = get_keyptr();
1621 u32 hash[12];
1622
1623 memcpy(hash, saddr, 16);
1624 hash[4] = (__force u32)dport;
1625 memcpy(&hash[5], keyptr->secret, sizeof(__u32) * 7);
1626
1627 return twothirdsMD4Transform((const __u32 *)daddr, hash);
1628 }
1629 #endif
1630
1631 #if defined(CONFIG_IP_DCCP) || defined(CONFIG_IP_DCCP_MODULE)
1632 /* Similar to secure_tcp_sequence_number but generate a 48 bit value
1633 * bit's 32-47 increase every key exchange
1634 * 0-31 hash(source, dest)
1635 */
1636 u64 secure_dccp_sequence_number(__be32 saddr, __be32 daddr,
1637 __be16 sport, __be16 dport)
1638 {
1639 u64 seq;
1640 __u32 hash[4];
1641 struct keydata *keyptr = get_keyptr();
1642
1643 hash[0] = (__force u32)saddr;
1644 hash[1] = (__force u32)daddr;
1645 hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
1646 hash[3] = keyptr->secret[11];
1647
1648 seq = half_md4_transform(hash, keyptr->secret);
1649 seq |= ((u64)keyptr->count) << (32 - HASH_BITS);
1650
1651 seq += ktime_to_ns(ktime_get_real());
1652 seq &= (1ull << 48) - 1;
1653
1654 return seq;
1655 }
1656 EXPORT_SYMBOL(secure_dccp_sequence_number);
1657 #endif
1658
1659 #endif /* CONFIG_INET */
1660
1661
1662 /*
1663 * Get a random word for internal kernel use only. Similar to urandom but
1664 * with the goal of minimal entropy pool depletion. As a result, the random
1665 * value is not cryptographically secure but for several uses the cost of
1666 * depleting entropy is too high
1667 */
1668 DEFINE_PER_CPU(__u32 [4], get_random_int_hash);
1669 unsigned int get_random_int(void)
1670 {
1671 struct keydata *keyptr;
1672 __u32 *hash = get_cpu_var(get_random_int_hash);
1673 int ret;
1674
1675 keyptr = get_keyptr();
1676 hash[0] += current->pid + jiffies + get_cycles();
1677
1678 ret = half_md4_transform(hash, keyptr->secret);
1679 put_cpu_var(get_random_int_hash);
1680
1681 return ret;
1682 }
1683
1684 /*
1685 * randomize_range() returns a start address such that
1686 *
1687 * [...... <range> .....]
1688 * start end
1689 *
1690 * a <range> with size "len" starting at the return value is inside in the
1691 * area defined by [start, end], but is otherwise randomized.
1692 */
1693 unsigned long
1694 randomize_range(unsigned long start, unsigned long end, unsigned long len)
1695 {
1696 unsigned long range = end - len - start;
1697
1698 if (end <= start + len)
1699 return 0;
1700 return PAGE_ALIGN(get_random_int() % range + start);
1701 }
1702
|
This page was automatically generated by the
LXR engine.
|