blob: b9dc7aa1dfb31032b0d83dc5ad7dcb2d9f4cc7d6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * random.c -- A strong random number generator
3 *
Matt Mackall9e95ce22005-04-16 15:25:56 -07004 * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
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>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227#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/spinlock.h>
240#include <linux/percpu.h>
241#include <linux/cryptohash.h>
242
243#include <asm/processor.h>
244#include <asm/uaccess.h>
245#include <asm/irq.h>
246#include <asm/io.h>
247
248/*
249 * Configuration information
250 */
251#define INPUT_POOL_WORDS 128
252#define OUTPUT_POOL_WORDS 32
253#define SEC_XFER_SIZE 512
254
255/*
256 * The minimum number of bits of entropy before we wake up a read on
257 * /dev/random. Should be enough to do a significant reseed.
258 */
259static int random_read_wakeup_thresh = 64;
260
261/*
262 * If the entropy count falls under this number of bits, then we
263 * should wake up processes which are selecting or polling on write
264 * access to /dev/random.
265 */
266static int random_write_wakeup_thresh = 128;
267
268/*
269 * When the input pool goes over trickle_thresh, start dropping most
270 * samples to avoid wasting CPU time and reduce lock contention.
271 */
272
Christoph Lameter6c036522005-07-07 17:56:59 -0700273static int trickle_thresh __read_mostly = INPUT_POOL_WORDS * 28;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
275static DEFINE_PER_CPU(int, trickle_count) = 0;
276
277/*
278 * A pool of size .poolwords is stirred with a primitive polynomial
279 * of degree .poolwords over GF(2). The taps for various sizes are
280 * defined below. They are chosen to be evenly spaced (minimum RMS
281 * distance from evenly spaced; the numbers in the comments are a
282 * scaled squared error sum) except for the last tap, which is 1 to
283 * get the twisting happening as fast as possible.
284 */
285static struct poolinfo {
286 int poolwords;
287 int tap1, tap2, tap3, tap4, tap5;
288} poolinfo_table[] = {
289 /* x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 -- 105 */
290 { 128, 103, 76, 51, 25, 1 },
291 /* x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 -- 15 */
292 { 32, 26, 20, 14, 7, 1 },
293#if 0
294 /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1 -- 115 */
295 { 2048, 1638, 1231, 819, 411, 1 },
296
297 /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
298 { 1024, 817, 615, 412, 204, 1 },
299
300 /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
301 { 1024, 819, 616, 410, 207, 2 },
302
303 /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
304 { 512, 411, 308, 208, 104, 1 },
305
306 /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
307 { 512, 409, 307, 206, 102, 2 },
308 /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
309 { 512, 409, 309, 205, 103, 2 },
310
311 /* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
312 { 256, 205, 155, 101, 52, 1 },
313
314 /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
315 { 128, 103, 78, 51, 27, 2 },
316
317 /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
318 { 64, 52, 39, 26, 14, 1 },
319#endif
320};
321
322#define POOLBITS poolwords*32
323#define POOLBYTES poolwords*4
324
325/*
326 * For the purposes of better mixing, we use the CRC-32 polynomial as
327 * well to make a twisted Generalized Feedback Shift Reigster
328 *
329 * (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR generators. ACM
330 * Transactions on Modeling and Computer Simulation 2(3):179-194.
331 * Also see M. Matsumoto & Y. Kurita, 1994. Twisted GFSR generators
332 * II. ACM Transactions on Mdeling and Computer Simulation 4:254-266)
333 *
334 * Thanks to Colin Plumb for suggesting this.
335 *
336 * We have not analyzed the resultant polynomial to prove it primitive;
337 * in fact it almost certainly isn't. Nonetheless, the irreducible factors
338 * of a random large-degree polynomial over GF(2) are more than large enough
339 * that periodicity is not a concern.
340 *
341 * The input hash is much less sensitive than the output hash. All
342 * that we want of it is that it be a good non-cryptographic hash;
343 * i.e. it not produce collisions when fed "random" data of the sort
344 * we expect to see. As long as the pool state differs for different
345 * inputs, we have preserved the input entropy and done a good job.
346 * The fact that an intelligent attacker can construct inputs that
347 * will produce controlled alterations to the pool's state is not
348 * important because we don't consider such inputs to contribute any
349 * randomness. The only property we need with respect to them is that
350 * the attacker can't increase his/her knowledge of the pool's state.
351 * Since all additions are reversible (knowing the final state and the
352 * input, you can reconstruct the initial state), if an attacker has
353 * any uncertainty about the initial state, he/she can only shuffle
354 * that uncertainty about, but never cause any collisions (which would
355 * decrease the uncertainty).
356 *
357 * The chosen system lets the state of the pool be (essentially) the input
358 * modulo the generator polymnomial. Now, for random primitive polynomials,
359 * this is a universal class of hash functions, meaning that the chance
360 * of a collision is limited by the attacker's knowledge of the generator
361 * polynomail, so if it is chosen at random, an attacker can never force
362 * a collision. Here, we use a fixed polynomial, but we *can* assume that
363 * ###--> it is unknown to the processes generating the input entropy. <-###
364 * Because of this important property, this is a good, collision-resistant
365 * hash; hash collisions will occur no more often than chance.
366 */
367
368/*
369 * Static global variables
370 */
371static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
372static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
373
374#if 0
375static int debug = 0;
376module_param(debug, bool, 0644);
377#define DEBUG_ENT(fmt, arg...) do { if (debug) \
378 printk(KERN_DEBUG "random %04d %04d %04d: " \
379 fmt,\
380 input_pool.entropy_count,\
381 blocking_pool.entropy_count,\
382 nonblocking_pool.entropy_count,\
383 ## arg); } while (0)
384#else
385#define DEBUG_ENT(fmt, arg...) do {} while (0)
386#endif
387
388/**********************************************************************
389 *
390 * OS independent entropy store. Here are the functions which handle
391 * storing entropy in an entropy pool.
392 *
393 **********************************************************************/
394
395struct entropy_store;
396struct entropy_store {
397 /* mostly-read data: */
398 struct poolinfo *poolinfo;
399 __u32 *pool;
400 const char *name;
401 int limit;
402 struct entropy_store *pull;
403
404 /* read-write data: */
405 spinlock_t lock ____cacheline_aligned_in_smp;
406 unsigned add_ptr;
407 int entropy_count;
408 int input_rotate;
409};
410
411static __u32 input_pool_data[INPUT_POOL_WORDS];
412static __u32 blocking_pool_data[OUTPUT_POOL_WORDS];
413static __u32 nonblocking_pool_data[OUTPUT_POOL_WORDS];
414
415static struct entropy_store input_pool = {
416 .poolinfo = &poolinfo_table[0],
417 .name = "input",
418 .limit = 1,
Ingo Molnare4d91912006-07-03 00:24:34 -0700419 .lock = __SPIN_LOCK_UNLOCKED(&input_pool.lock),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 .pool = input_pool_data
421};
422
423static struct entropy_store blocking_pool = {
424 .poolinfo = &poolinfo_table[1],
425 .name = "blocking",
426 .limit = 1,
427 .pull = &input_pool,
Ingo Molnare4d91912006-07-03 00:24:34 -0700428 .lock = __SPIN_LOCK_UNLOCKED(&blocking_pool.lock),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 .pool = blocking_pool_data
430};
431
432static struct entropy_store nonblocking_pool = {
433 .poolinfo = &poolinfo_table[1],
434 .name = "nonblocking",
435 .pull = &input_pool,
Ingo Molnare4d91912006-07-03 00:24:34 -0700436 .lock = __SPIN_LOCK_UNLOCKED(&nonblocking_pool.lock),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 .pool = nonblocking_pool_data
438};
439
440/*
441 * This function adds a byte into the entropy "pool". It does not
442 * update the entropy estimate. The caller should call
443 * credit_entropy_store if this is appropriate.
444 *
445 * The pool is stirred with a primitive polynomial of the appropriate
446 * degree, and then twisted. We twist by three bits at a time because
447 * it's cheap to do so and helps slightly in the expected case where
448 * the entropy is concentrated in the low-order bits.
449 */
450static void __add_entropy_words(struct entropy_store *r, const __u32 *in,
451 int nwords, __u32 out[16])
452{
453 static __u32 const twist_table[8] = {
454 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
455 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
456 unsigned long i, add_ptr, tap1, tap2, tap3, tap4, tap5;
457 int new_rotate, input_rotate;
458 int wordmask = r->poolinfo->poolwords - 1;
459 __u32 w, next_w;
460 unsigned long flags;
461
462 /* Taps are constant, so we can load them without holding r->lock. */
463 tap1 = r->poolinfo->tap1;
464 tap2 = r->poolinfo->tap2;
465 tap3 = r->poolinfo->tap3;
466 tap4 = r->poolinfo->tap4;
467 tap5 = r->poolinfo->tap5;
468 next_w = *in++;
469
470 spin_lock_irqsave(&r->lock, flags);
471 prefetch_range(r->pool, wordmask);
472 input_rotate = r->input_rotate;
473 add_ptr = r->add_ptr;
474
475 while (nwords--) {
476 w = rol32(next_w, input_rotate);
477 if (nwords > 0)
478 next_w = *in++;
479 i = add_ptr = (add_ptr - 1) & wordmask;
480 /*
481 * Normally, we add 7 bits of rotation to the pool.
482 * At the beginning of the pool, add an extra 7 bits
483 * rotation, so that successive passes spread the
484 * input bits across the pool evenly.
485 */
486 new_rotate = input_rotate + 14;
487 if (i)
488 new_rotate = input_rotate + 7;
489 input_rotate = new_rotate & 31;
490
491 /* XOR in the various taps */
492 w ^= r->pool[(i + tap1) & wordmask];
493 w ^= r->pool[(i + tap2) & wordmask];
494 w ^= r->pool[(i + tap3) & wordmask];
495 w ^= r->pool[(i + tap4) & wordmask];
496 w ^= r->pool[(i + tap5) & wordmask];
497 w ^= r->pool[i];
498 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
499 }
500
501 r->input_rotate = input_rotate;
502 r->add_ptr = add_ptr;
503
504 if (out) {
505 for (i = 0; i < 16; i++) {
506 out[i] = r->pool[add_ptr];
507 add_ptr = (add_ptr - 1) & wordmask;
508 }
509 }
510
511 spin_unlock_irqrestore(&r->lock, flags);
512}
513
514static inline void add_entropy_words(struct entropy_store *r, const __u32 *in,
515 int nwords)
516{
517 __add_entropy_words(r, in, nwords, NULL);
518}
519
520/*
521 * Credit (or debit) the entropy store with n bits of entropy
522 */
523static void credit_entropy_store(struct entropy_store *r, int nbits)
524{
525 unsigned long flags;
526
527 spin_lock_irqsave(&r->lock, flags);
528
529 if (r->entropy_count + nbits < 0) {
530 DEBUG_ENT("negative entropy/overflow (%d+%d)\n",
531 r->entropy_count, nbits);
532 r->entropy_count = 0;
533 } else if (r->entropy_count + nbits > r->poolinfo->POOLBITS) {
534 r->entropy_count = r->poolinfo->POOLBITS;
535 } else {
536 r->entropy_count += nbits;
537 if (nbits)
538 DEBUG_ENT("added %d entropy credits to %s\n",
539 nbits, r->name);
540 }
541
542 spin_unlock_irqrestore(&r->lock, flags);
543}
544
545/*********************************************************************
546 *
547 * Entropy input management
548 *
549 *********************************************************************/
550
551/* There is one of these per entropy source */
552struct timer_rand_state {
553 cycles_t last_time;
554 long last_delta,last_delta2;
555 unsigned dont_count_entropy:1;
556};
557
558static struct timer_rand_state input_timer_state;
559static struct timer_rand_state *irq_timer_state[NR_IRQS];
560
561/*
562 * This function adds entropy to the entropy "pool" by using timing
563 * delays. It uses the timer_rand_state structure to make an estimate
564 * of how many bits of entropy this call has added to the pool.
565 *
566 * The number "num" is also added to the pool - it should somehow describe
567 * the type of event which just happened. This is currently 0-255 for
568 * keyboard scan codes, and 256 upwards for interrupts.
569 *
570 */
571static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
572{
573 struct {
574 cycles_t cycles;
575 long jiffies;
576 unsigned num;
577 } sample;
578 long delta, delta2, delta3;
579
580 preempt_disable();
581 /* if over the trickle threshold, use only 1 in 4096 samples */
582 if (input_pool.entropy_count > trickle_thresh &&
583 (__get_cpu_var(trickle_count)++ & 0xfff))
584 goto out;
585
586 sample.jiffies = jiffies;
587 sample.cycles = get_cycles();
588 sample.num = num;
589 add_entropy_words(&input_pool, (u32 *)&sample, sizeof(sample)/4);
590
591 /*
592 * Calculate number of bits of randomness we probably added.
593 * We take into account the first, second and third-order deltas
594 * in order to make our estimate.
595 */
596
597 if (!state->dont_count_entropy) {
598 delta = sample.jiffies - state->last_time;
599 state->last_time = sample.jiffies;
600
601 delta2 = delta - state->last_delta;
602 state->last_delta = delta;
603
604 delta3 = delta2 - state->last_delta2;
605 state->last_delta2 = delta2;
606
607 if (delta < 0)
608 delta = -delta;
609 if (delta2 < 0)
610 delta2 = -delta2;
611 if (delta3 < 0)
612 delta3 = -delta3;
613 if (delta > delta2)
614 delta = delta2;
615 if (delta > delta3)
616 delta = delta3;
617
618 /*
619 * delta is now minimum absolute delta.
620 * Round down by 1 bit on general principles,
621 * and limit entropy entimate to 12 bits.
622 */
623 credit_entropy_store(&input_pool,
624 min_t(int, fls(delta>>1), 11));
625 }
626
627 if(input_pool.entropy_count >= random_read_wakeup_thresh)
628 wake_up_interruptible(&random_read_wait);
629
630out:
631 preempt_enable();
632}
633
Stephen Hemmingerd2515752006-01-11 12:17:38 -0800634void add_input_randomness(unsigned int type, unsigned int code,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 unsigned int value)
636{
637 static unsigned char last_value;
638
639 /* ignore autorepeat and the like */
640 if (value == last_value)
641 return;
642
643 DEBUG_ENT("input event\n");
644 last_value = value;
645 add_timer_randomness(&input_timer_state,
646 (type << 4) ^ code ^ (code >> 4) ^ value);
647}
Dmitry Torokhov80fc9f52006-10-11 01:43:58 -0400648EXPORT_SYMBOL_GPL(add_input_randomness);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649
650void add_interrupt_randomness(int irq)
651{
652 if (irq >= NR_IRQS || irq_timer_state[irq] == 0)
653 return;
654
655 DEBUG_ENT("irq event %d\n", irq);
656 add_timer_randomness(irq_timer_state[irq], 0x100 + irq);
657}
658
David Howells93614012006-09-30 20:45:40 +0200659#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660void add_disk_randomness(struct gendisk *disk)
661{
662 if (!disk || !disk->random)
663 return;
664 /* first major is 1, so we get >= 0x200 here */
665 DEBUG_ENT("disk event %d:%d\n", disk->major, disk->first_minor);
666
667 add_timer_randomness(disk->random,
668 0x100 + MKDEV(disk->major, disk->first_minor));
669}
670
671EXPORT_SYMBOL(add_disk_randomness);
David Howells93614012006-09-30 20:45:40 +0200672#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
674#define EXTRACT_SIZE 10
675
676/*********************************************************************
677 *
678 * Entropy extraction routines
679 *
680 *********************************************************************/
681
682static ssize_t extract_entropy(struct entropy_store *r, void * buf,
683 size_t nbytes, int min, int rsvd);
684
685/*
686 * This utility inline function is responsible for transfering entropy
687 * from the primary pool to the secondary extraction pool. We make
688 * sure we pull enough for a 'catastrophic reseed'.
689 */
690static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
691{
692 __u32 tmp[OUTPUT_POOL_WORDS];
693
694 if (r->pull && r->entropy_count < nbytes * 8 &&
695 r->entropy_count < r->poolinfo->POOLBITS) {
696 int bytes = max_t(int, random_read_wakeup_thresh / 8,
697 min_t(int, nbytes, sizeof(tmp)));
698 int rsvd = r->limit ? 0 : random_read_wakeup_thresh/4;
699
700 DEBUG_ENT("going to reseed %s with %d bits "
701 "(%d of %d requested)\n",
702 r->name, bytes * 8, nbytes * 8, r->entropy_count);
703
704 bytes=extract_entropy(r->pull, tmp, bytes,
705 random_read_wakeup_thresh / 8, rsvd);
706 add_entropy_words(r, tmp, (bytes + 3) / 4);
707 credit_entropy_store(r, bytes*8);
708 }
709}
710
711/*
712 * These functions extracts randomness from the "entropy pool", and
713 * returns it in a buffer.
714 *
715 * The min parameter specifies the minimum amount we can pull before
716 * failing to avoid races that defeat catastrophic reseeding while the
717 * reserved parameter indicates how much entropy we must leave in the
718 * pool after each pull to avoid starving other readers.
719 *
720 * Note: extract_entropy() assumes that .poolwords is a multiple of 16 words.
721 */
722
723static size_t account(struct entropy_store *r, size_t nbytes, int min,
724 int reserved)
725{
726 unsigned long flags;
727
728 BUG_ON(r->entropy_count > r->poolinfo->POOLBITS);
729
730 /* Hold lock while accounting */
731 spin_lock_irqsave(&r->lock, flags);
732
733 DEBUG_ENT("trying to extract %d bits from %s\n",
734 nbytes * 8, r->name);
735
736 /* Can we pull enough? */
737 if (r->entropy_count / 8 < min + reserved) {
738 nbytes = 0;
739 } else {
740 /* If limited, never pull more than available */
741 if (r->limit && nbytes + reserved >= r->entropy_count / 8)
742 nbytes = r->entropy_count/8 - reserved;
743
744 if(r->entropy_count / 8 >= nbytes + reserved)
745 r->entropy_count -= nbytes*8;
746 else
747 r->entropy_count = reserved;
748
749 if (r->entropy_count < random_write_wakeup_thresh)
750 wake_up_interruptible(&random_write_wait);
751 }
752
753 DEBUG_ENT("debiting %d entropy credits from %s%s\n",
754 nbytes * 8, r->name, r->limit ? "" : " (unlimited)");
755
756 spin_unlock_irqrestore(&r->lock, flags);
757
758 return nbytes;
759}
760
761static void extract_buf(struct entropy_store *r, __u8 *out)
762{
763 int i, x;
764 __u32 data[16], buf[5 + SHA_WORKSPACE_WORDS];
765
766 sha_init(buf);
767 /*
768 * As we hash the pool, we mix intermediate values of
769 * the hash back into the pool. This eliminates
770 * backtracking attacks (where the attacker knows
771 * the state of the pool plus the current outputs, and
772 * attempts to find previous ouputs), unless the hash
773 * function can be inverted.
774 */
775 for (i = 0, x = 0; i < r->poolinfo->poolwords; i += 16, x+=2) {
776 sha_transform(buf, (__u8 *)r->pool+i, buf + 5);
777 add_entropy_words(r, &buf[x % 5], 1);
778 }
779
780 /*
781 * To avoid duplicates, we atomically extract a
782 * portion of the pool while mixing, and hash one
783 * final time.
784 */
785 __add_entropy_words(r, &buf[x % 5], 1, data);
786 sha_transform(buf, (__u8 *)data, buf + 5);
787
788 /*
789 * In case the hash function has some recognizable
790 * output pattern, we fold it in half.
791 */
792
793 buf[0] ^= buf[3];
794 buf[1] ^= buf[4];
795 buf[0] ^= rol32(buf[3], 16);
796 memcpy(out, buf, EXTRACT_SIZE);
797 memset(buf, 0, sizeof(buf));
798}
799
800static ssize_t extract_entropy(struct entropy_store *r, void * buf,
801 size_t nbytes, int min, int reserved)
802{
803 ssize_t ret = 0, i;
804 __u8 tmp[EXTRACT_SIZE];
805
806 xfer_secondary_pool(r, nbytes);
807 nbytes = account(r, nbytes, min, reserved);
808
809 while (nbytes) {
810 extract_buf(r, tmp);
811 i = min_t(int, nbytes, EXTRACT_SIZE);
812 memcpy(buf, tmp, i);
813 nbytes -= i;
814 buf += i;
815 ret += i;
816 }
817
818 /* Wipe data just returned from memory */
819 memset(tmp, 0, sizeof(tmp));
820
821 return ret;
822}
823
824static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
825 size_t nbytes)
826{
827 ssize_t ret = 0, i;
828 __u8 tmp[EXTRACT_SIZE];
829
830 xfer_secondary_pool(r, nbytes);
831 nbytes = account(r, nbytes, 0, 0);
832
833 while (nbytes) {
834 if (need_resched()) {
835 if (signal_pending(current)) {
836 if (ret == 0)
837 ret = -ERESTARTSYS;
838 break;
839 }
840 schedule();
841 }
842
843 extract_buf(r, tmp);
844 i = min_t(int, nbytes, EXTRACT_SIZE);
845 if (copy_to_user(buf, tmp, i)) {
846 ret = -EFAULT;
847 break;
848 }
849
850 nbytes -= i;
851 buf += i;
852 ret += i;
853 }
854
855 /* Wipe data just returned from memory */
856 memset(tmp, 0, sizeof(tmp));
857
858 return ret;
859}
860
861/*
862 * This function is the exported kernel interface. It returns some
863 * number of good random numbers, suitable for seeding TCP sequence
864 * numbers, etc.
865 */
866void get_random_bytes(void *buf, int nbytes)
867{
868 extract_entropy(&nonblocking_pool, buf, nbytes, 0, 0);
869}
870
871EXPORT_SYMBOL(get_random_bytes);
872
873/*
874 * init_std_data - initialize pool with system data
875 *
876 * @r: pool to initialize
877 *
878 * This function clears the pool's entropy count and mixes some system
879 * data into the pool to prepare it for use. The pool is not cleared
880 * as that can only decrease the entropy in the pool.
881 */
882static void init_std_data(struct entropy_store *r)
883{
884 struct timeval tv;
885 unsigned long flags;
886
887 spin_lock_irqsave(&r->lock, flags);
888 r->entropy_count = 0;
889 spin_unlock_irqrestore(&r->lock, flags);
890
891 do_gettimeofday(&tv);
892 add_entropy_words(r, (__u32 *)&tv, sizeof(tv)/4);
Serge E. Hallyne9ff3992006-10-02 02:18:11 -0700893 add_entropy_words(r, (__u32 *)utsname(),
894 sizeof(*(utsname()))/4);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895}
896
897static int __init rand_initialize(void)
898{
899 init_std_data(&input_pool);
900 init_std_data(&blocking_pool);
901 init_std_data(&nonblocking_pool);
902 return 0;
903}
904module_init(rand_initialize);
905
906void rand_initialize_irq(int irq)
907{
908 struct timer_rand_state *state;
909
910 if (irq >= NR_IRQS || irq_timer_state[irq])
911 return;
912
913 /*
914 * If kmalloc returns null, we just won't use that entropy
915 * source.
916 */
917 state = kmalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
918 if (state) {
919 memset(state, 0, sizeof(struct timer_rand_state));
920 irq_timer_state[irq] = state;
921 }
922}
923
David Howells93614012006-09-30 20:45:40 +0200924#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925void rand_initialize_disk(struct gendisk *disk)
926{
927 struct timer_rand_state *state;
928
929 /*
930 * If kmalloc returns null, we just won't use that entropy
931 * source.
932 */
933 state = kmalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
934 if (state) {
935 memset(state, 0, sizeof(struct timer_rand_state));
936 disk->random = state;
937 }
938}
David Howells93614012006-09-30 20:45:40 +0200939#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
941static ssize_t
942random_read(struct file * file, char __user * buf, size_t nbytes, loff_t *ppos)
943{
944 ssize_t n, retval = 0, count = 0;
945
946 if (nbytes == 0)
947 return 0;
948
949 while (nbytes > 0) {
950 n = nbytes;
951 if (n > SEC_XFER_SIZE)
952 n = SEC_XFER_SIZE;
953
954 DEBUG_ENT("reading %d bits\n", n*8);
955
956 n = extract_entropy_user(&blocking_pool, buf, n);
957
958 DEBUG_ENT("read got %d bits (%d still needed)\n",
959 n*8, (nbytes-n)*8);
960
961 if (n == 0) {
962 if (file->f_flags & O_NONBLOCK) {
963 retval = -EAGAIN;
964 break;
965 }
966
967 DEBUG_ENT("sleeping?\n");
968
969 wait_event_interruptible(random_read_wait,
970 input_pool.entropy_count >=
971 random_read_wakeup_thresh);
972
973 DEBUG_ENT("awake\n");
974
975 if (signal_pending(current)) {
976 retval = -ERESTARTSYS;
977 break;
978 }
979
980 continue;
981 }
982
983 if (n < 0) {
984 retval = n;
985 break;
986 }
987 count += n;
988 buf += n;
989 nbytes -= n;
990 break; /* This break makes the device work */
991 /* like a named pipe */
992 }
993
994 /*
995 * If we gave the user some bytes, update the access time.
996 */
997 if (count)
998 file_accessed(file);
999
1000 return (count ? count : retval);
1001}
1002
1003static ssize_t
1004urandom_read(struct file * file, char __user * buf,
1005 size_t nbytes, loff_t *ppos)
1006{
1007 return extract_entropy_user(&nonblocking_pool, buf, nbytes);
1008}
1009
1010static unsigned int
1011random_poll(struct file *file, poll_table * wait)
1012{
1013 unsigned int mask;
1014
1015 poll_wait(file, &random_read_wait, wait);
1016 poll_wait(file, &random_write_wait, wait);
1017 mask = 0;
1018 if (input_pool.entropy_count >= random_read_wakeup_thresh)
1019 mask |= POLLIN | POLLRDNORM;
1020 if (input_pool.entropy_count < random_write_wakeup_thresh)
1021 mask |= POLLOUT | POLLWRNORM;
1022 return mask;
1023}
1024
1025static ssize_t
1026random_write(struct file * file, const char __user * buffer,
1027 size_t count, loff_t *ppos)
1028{
1029 int ret = 0;
1030 size_t bytes;
1031 __u32 buf[16];
1032 const char __user *p = buffer;
1033 size_t c = count;
1034
1035 while (c > 0) {
1036 bytes = min(c, sizeof(buf));
1037
1038 bytes -= copy_from_user(&buf, p, bytes);
1039 if (!bytes) {
1040 ret = -EFAULT;
1041 break;
1042 }
1043 c -= bytes;
1044 p += bytes;
1045
1046 add_entropy_words(&input_pool, buf, (bytes + 3) / 4);
1047 }
1048 if (p == buffer) {
1049 return (ssize_t)ret;
1050 } else {
Josef Sipeka7113a92006-12-08 02:36:55 -08001051 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 inode->i_mtime = current_fs_time(inode->i_sb);
1053 mark_inode_dirty(inode);
1054 return (ssize_t)(p - buffer);
1055 }
1056}
1057
1058static int
1059random_ioctl(struct inode * inode, struct file * file,
1060 unsigned int cmd, unsigned long arg)
1061{
1062 int size, ent_count;
1063 int __user *p = (int __user *)arg;
1064 int retval;
1065
1066 switch (cmd) {
1067 case RNDGETENTCNT:
1068 ent_count = input_pool.entropy_count;
1069 if (put_user(ent_count, p))
1070 return -EFAULT;
1071 return 0;
1072 case RNDADDTOENTCNT:
1073 if (!capable(CAP_SYS_ADMIN))
1074 return -EPERM;
1075 if (get_user(ent_count, p))
1076 return -EFAULT;
1077 credit_entropy_store(&input_pool, ent_count);
1078 /*
1079 * Wake up waiting processes if we have enough
1080 * entropy.
1081 */
1082 if (input_pool.entropy_count >= random_read_wakeup_thresh)
1083 wake_up_interruptible(&random_read_wait);
1084 return 0;
1085 case RNDADDENTROPY:
1086 if (!capable(CAP_SYS_ADMIN))
1087 return -EPERM;
1088 if (get_user(ent_count, p++))
1089 return -EFAULT;
1090 if (ent_count < 0)
1091 return -EINVAL;
1092 if (get_user(size, p++))
1093 return -EFAULT;
1094 retval = random_write(file, (const char __user *) p,
1095 size, &file->f_pos);
1096 if (retval < 0)
1097 return retval;
1098 credit_entropy_store(&input_pool, ent_count);
1099 /*
1100 * Wake up waiting processes if we have enough
1101 * entropy.
1102 */
1103 if (input_pool.entropy_count >= random_read_wakeup_thresh)
1104 wake_up_interruptible(&random_read_wait);
1105 return 0;
1106 case RNDZAPENTCNT:
1107 case RNDCLEARPOOL:
1108 /* Clear the entropy pool counters. */
1109 if (!capable(CAP_SYS_ADMIN))
1110 return -EPERM;
1111 init_std_data(&input_pool);
1112 init_std_data(&blocking_pool);
1113 init_std_data(&nonblocking_pool);
1114 return 0;
1115 default:
1116 return -EINVAL;
1117 }
1118}
1119
Arjan van de Ven2b8693c2007-02-12 00:55:32 -08001120const struct file_operations random_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 .read = random_read,
1122 .write = random_write,
1123 .poll = random_poll,
1124 .ioctl = random_ioctl,
1125};
1126
Arjan van de Ven2b8693c2007-02-12 00:55:32 -08001127const struct file_operations urandom_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 .read = urandom_read,
1129 .write = random_write,
1130 .ioctl = random_ioctl,
1131};
1132
1133/***************************************************************
1134 * Random UUID interface
1135 *
1136 * Used here for a Boot ID, but can be useful for other kernel
1137 * drivers.
1138 ***************************************************************/
1139
1140/*
1141 * Generate random UUID
1142 */
1143void generate_random_uuid(unsigned char uuid_out[16])
1144{
1145 get_random_bytes(uuid_out, 16);
1146 /* Set UUID version to 4 --- truely random generation */
1147 uuid_out[6] = (uuid_out[6] & 0x0F) | 0x40;
1148 /* Set the UUID variant to DCE */
1149 uuid_out[8] = (uuid_out[8] & 0x3F) | 0x80;
1150}
1151
1152EXPORT_SYMBOL(generate_random_uuid);
1153
1154/********************************************************************
1155 *
1156 * Sysctl interface
1157 *
1158 ********************************************************************/
1159
1160#ifdef CONFIG_SYSCTL
1161
1162#include <linux/sysctl.h>
1163
1164static int min_read_thresh = 8, min_write_thresh;
1165static int max_read_thresh = INPUT_POOL_WORDS * 32;
1166static int max_write_thresh = INPUT_POOL_WORDS * 32;
1167static char sysctl_bootid[16];
1168
1169/*
1170 * These functions is used to return both the bootid UUID, and random
1171 * UUID. The difference is in whether table->data is NULL; if it is,
1172 * then a new UUID is generated and returned to the user.
1173 *
1174 * If the user accesses this via the proc interface, it will be returned
1175 * as an ASCII string in the standard UUID format. If accesses via the
1176 * sysctl system call, it is returned as 16 bytes of binary data.
1177 */
1178static int proc_do_uuid(ctl_table *table, int write, struct file *filp,
1179 void __user *buffer, size_t *lenp, loff_t *ppos)
1180{
1181 ctl_table fake_table;
1182 unsigned char buf[64], tmp_uuid[16], *uuid;
1183
1184 uuid = table->data;
1185 if (!uuid) {
1186 uuid = tmp_uuid;
1187 uuid[8] = 0;
1188 }
1189 if (uuid[8] == 0)
1190 generate_random_uuid(uuid);
1191
1192 sprintf(buf, "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-"
1193 "%02x%02x%02x%02x%02x%02x",
1194 uuid[0], uuid[1], uuid[2], uuid[3],
1195 uuid[4], uuid[5], uuid[6], uuid[7],
1196 uuid[8], uuid[9], uuid[10], uuid[11],
1197 uuid[12], uuid[13], uuid[14], uuid[15]);
1198 fake_table.data = buf;
1199 fake_table.maxlen = sizeof(buf);
1200
1201 return proc_dostring(&fake_table, write, filp, buffer, lenp, ppos);
1202}
1203
1204static int uuid_strategy(ctl_table *table, int __user *name, int nlen,
1205 void __user *oldval, size_t __user *oldlenp,
Alexey Dobriyan1f29bcd2006-12-10 02:19:10 -08001206 void __user *newval, size_t newlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207{
1208 unsigned char tmp_uuid[16], *uuid;
1209 unsigned int len;
1210
1211 if (!oldval || !oldlenp)
1212 return 1;
1213
1214 uuid = table->data;
1215 if (!uuid) {
1216 uuid = tmp_uuid;
1217 uuid[8] = 0;
1218 }
1219 if (uuid[8] == 0)
1220 generate_random_uuid(uuid);
1221
1222 if (get_user(len, oldlenp))
1223 return -EFAULT;
1224 if (len) {
1225 if (len > 16)
1226 len = 16;
1227 if (copy_to_user(oldval, uuid, len) ||
1228 put_user(len, oldlenp))
1229 return -EFAULT;
1230 }
1231 return 1;
1232}
1233
1234static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
1235ctl_table random_table[] = {
1236 {
1237 .ctl_name = RANDOM_POOLSIZE,
1238 .procname = "poolsize",
1239 .data = &sysctl_poolsize,
1240 .maxlen = sizeof(int),
1241 .mode = 0444,
1242 .proc_handler = &proc_dointvec,
1243 },
1244 {
1245 .ctl_name = RANDOM_ENTROPY_COUNT,
1246 .procname = "entropy_avail",
1247 .maxlen = sizeof(int),
1248 .mode = 0444,
1249 .proc_handler = &proc_dointvec,
1250 .data = &input_pool.entropy_count,
1251 },
1252 {
1253 .ctl_name = RANDOM_READ_THRESH,
1254 .procname = "read_wakeup_threshold",
1255 .data = &random_read_wakeup_thresh,
1256 .maxlen = sizeof(int),
1257 .mode = 0644,
1258 .proc_handler = &proc_dointvec_minmax,
1259 .strategy = &sysctl_intvec,
1260 .extra1 = &min_read_thresh,
1261 .extra2 = &max_read_thresh,
1262 },
1263 {
1264 .ctl_name = RANDOM_WRITE_THRESH,
1265 .procname = "write_wakeup_threshold",
1266 .data = &random_write_wakeup_thresh,
1267 .maxlen = sizeof(int),
1268 .mode = 0644,
1269 .proc_handler = &proc_dointvec_minmax,
1270 .strategy = &sysctl_intvec,
1271 .extra1 = &min_write_thresh,
1272 .extra2 = &max_write_thresh,
1273 },
1274 {
1275 .ctl_name = RANDOM_BOOT_ID,
1276 .procname = "boot_id",
1277 .data = &sysctl_bootid,
1278 .maxlen = 16,
1279 .mode = 0444,
1280 .proc_handler = &proc_do_uuid,
1281 .strategy = &uuid_strategy,
1282 },
1283 {
1284 .ctl_name = RANDOM_UUID,
1285 .procname = "uuid",
1286 .maxlen = 16,
1287 .mode = 0444,
1288 .proc_handler = &proc_do_uuid,
1289 .strategy = &uuid_strategy,
1290 },
1291 { .ctl_name = 0 }
1292};
1293#endif /* CONFIG_SYSCTL */
1294
1295/********************************************************************
1296 *
1297 * Random funtions for networking
1298 *
1299 ********************************************************************/
1300
1301/*
1302 * TCP initial sequence number picking. This uses the random number
1303 * generator to pick an initial secret value. This value is hashed
1304 * along with the TCP endpoint information to provide a unique
1305 * starting point for each pair of TCP endpoints. This defeats
1306 * attacks which rely on guessing the initial TCP sequence number.
1307 * This algorithm was suggested by Steve Bellovin.
1308 *
1309 * Using a very strong hash was taking an appreciable amount of the total
1310 * TCP connection establishment time, so this is a weaker hash,
1311 * compensated for by changing the secret periodically.
1312 */
1313
1314/* F, G and H are basic MD4 functions: selection, majority, parity */
1315#define F(x, y, z) ((z) ^ ((x) & ((y) ^ (z))))
1316#define G(x, y, z) (((x) & (y)) + (((x) ^ (y)) & (z)))
1317#define H(x, y, z) ((x) ^ (y) ^ (z))
1318
1319/*
1320 * The generic round function. The application is so specific that
1321 * we don't bother protecting all the arguments with parens, as is generally
1322 * good macro practice, in favor of extra legibility.
1323 * Rotation is separate from addition to prevent recomputation
1324 */
1325#define ROUND(f, a, b, c, d, x, s) \
1326 (a += f(b, c, d) + x, a = (a << s) | (a >> (32 - s)))
1327#define K1 0
1328#define K2 013240474631UL
1329#define K3 015666365641UL
1330
1331#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1332
1333static __u32 twothirdsMD4Transform (__u32 const buf[4], __u32 const in[12])
1334{
1335 __u32 a = buf[0], b = buf[1], c = buf[2], d = buf[3];
1336
1337 /* Round 1 */
1338 ROUND(F, a, b, c, d, in[ 0] + K1, 3);
1339 ROUND(F, d, a, b, c, in[ 1] + K1, 7);
1340 ROUND(F, c, d, a, b, in[ 2] + K1, 11);
1341 ROUND(F, b, c, d, a, in[ 3] + K1, 19);
1342 ROUND(F, a, b, c, d, in[ 4] + K1, 3);
1343 ROUND(F, d, a, b, c, in[ 5] + K1, 7);
1344 ROUND(F, c, d, a, b, in[ 6] + K1, 11);
1345 ROUND(F, b, c, d, a, in[ 7] + K1, 19);
1346 ROUND(F, a, b, c, d, in[ 8] + K1, 3);
1347 ROUND(F, d, a, b, c, in[ 9] + K1, 7);
1348 ROUND(F, c, d, a, b, in[10] + K1, 11);
1349 ROUND(F, b, c, d, a, in[11] + K1, 19);
1350
1351 /* Round 2 */
1352 ROUND(G, a, b, c, d, in[ 1] + K2, 3);
1353 ROUND(G, d, a, b, c, in[ 3] + K2, 5);
1354 ROUND(G, c, d, a, b, in[ 5] + K2, 9);
1355 ROUND(G, b, c, d, a, in[ 7] + K2, 13);
1356 ROUND(G, a, b, c, d, in[ 9] + K2, 3);
1357 ROUND(G, d, a, b, c, in[11] + K2, 5);
1358 ROUND(G, c, d, a, b, in[ 0] + K2, 9);
1359 ROUND(G, b, c, d, a, in[ 2] + K2, 13);
1360 ROUND(G, a, b, c, d, in[ 4] + K2, 3);
1361 ROUND(G, d, a, b, c, in[ 6] + K2, 5);
1362 ROUND(G, c, d, a, b, in[ 8] + K2, 9);
1363 ROUND(G, b, c, d, a, in[10] + K2, 13);
1364
1365 /* Round 3 */
1366 ROUND(H, a, b, c, d, in[ 3] + K3, 3);
1367 ROUND(H, d, a, b, c, in[ 7] + K3, 9);
1368 ROUND(H, c, d, a, b, in[11] + K3, 11);
1369 ROUND(H, b, c, d, a, in[ 2] + K3, 15);
1370 ROUND(H, a, b, c, d, in[ 6] + K3, 3);
1371 ROUND(H, d, a, b, c, in[10] + K3, 9);
1372 ROUND(H, c, d, a, b, in[ 1] + K3, 11);
1373 ROUND(H, b, c, d, a, in[ 5] + K3, 15);
1374 ROUND(H, a, b, c, d, in[ 9] + K3, 3);
1375 ROUND(H, d, a, b, c, in[ 0] + K3, 9);
1376 ROUND(H, c, d, a, b, in[ 4] + K3, 11);
1377 ROUND(H, b, c, d, a, in[ 8] + K3, 15);
1378
1379 return buf[1] + b; /* "most hashed" word */
1380 /* Alternative: return sum of all words? */
1381}
1382#endif
1383
1384#undef ROUND
1385#undef F
1386#undef G
1387#undef H
1388#undef K1
1389#undef K2
1390#undef K3
1391
1392/* This should not be decreased so low that ISNs wrap too fast. */
1393#define REKEY_INTERVAL (300 * HZ)
1394/*
1395 * Bit layout of the tcp sequence numbers (before adding current time):
1396 * bit 24-31: increased after every key exchange
1397 * bit 0-23: hash(source,dest)
1398 *
1399 * The implementation is similar to the algorithm described
1400 * in the Appendix of RFC 1185, except that
1401 * - it uses a 1 MHz clock instead of a 250 kHz clock
1402 * - it performs a rekey every 5 minutes, which is equivalent
1403 * to a (source,dest) tulple dependent forward jump of the
1404 * clock by 0..2^(HASH_BITS+1)
1405 *
1406 * Thus the average ISN wraparound time is 68 minutes instead of
1407 * 4.55 hours.
1408 *
1409 * SMP cleanup and lock avoidance with poor man's RCU.
1410 * Manfred Spraul <manfred@colorfullife.com>
1411 *
1412 */
1413#define COUNT_BITS 8
1414#define COUNT_MASK ((1 << COUNT_BITS) - 1)
1415#define HASH_BITS 24
1416#define HASH_MASK ((1 << HASH_BITS) - 1)
1417
1418static struct keydata {
1419 __u32 count; /* already shifted to the final position */
1420 __u32 secret[12];
1421} ____cacheline_aligned ip_keydata[2];
1422
1423static unsigned int ip_cnt;
1424
David Howells65f27f32006-11-22 14:55:48 +00001425static void rekey_seq_generator(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
David Howells65f27f32006-11-22 14:55:48 +00001427static DECLARE_DELAYED_WORK(rekey_work, rekey_seq_generator);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
1429/*
1430 * Lock avoidance:
1431 * The ISN generation runs lockless - it's just a hash over random data.
1432 * State changes happen every 5 minutes when the random key is replaced.
1433 * Synchronization is performed by having two copies of the hash function
1434 * state and rekey_seq_generator always updates the inactive copy.
1435 * The copy is then activated by updating ip_cnt.
1436 * The implementation breaks down if someone blocks the thread
1437 * that processes SYN requests for more than 5 minutes. Should never
1438 * happen, and even if that happens only a not perfectly compliant
1439 * ISN is generated, nothing fatal.
1440 */
David Howells65f27f32006-11-22 14:55:48 +00001441static void rekey_seq_generator(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442{
1443 struct keydata *keyptr = &ip_keydata[1 ^ (ip_cnt & 1)];
1444
1445 get_random_bytes(keyptr->secret, sizeof(keyptr->secret));
1446 keyptr->count = (ip_cnt & COUNT_MASK) << HASH_BITS;
1447 smp_wmb();
1448 ip_cnt++;
1449 schedule_delayed_work(&rekey_work, REKEY_INTERVAL);
1450}
1451
1452static inline struct keydata *get_keyptr(void)
1453{
1454 struct keydata *keyptr = &ip_keydata[ip_cnt & 1];
1455
1456 smp_rmb();
1457
1458 return keyptr;
1459}
1460
1461static __init int seqgen_init(void)
1462{
1463 rekey_seq_generator(NULL);
1464 return 0;
1465}
1466late_initcall(seqgen_init);
1467
1468#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
Al Virob09b845c2006-11-14 20:52:19 -08001469__u32 secure_tcpv6_sequence_number(__be32 *saddr, __be32 *daddr,
1470 __be16 sport, __be16 dport)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471{
1472 struct timeval tv;
1473 __u32 seq;
1474 __u32 hash[12];
1475 struct keydata *keyptr = get_keyptr();
1476
1477 /* The procedure is the same as for IPv4, but addresses are longer.
1478 * Thus we must use twothirdsMD4Transform.
1479 */
1480
1481 memcpy(hash, saddr, 16);
Al Virob09b845c2006-11-14 20:52:19 -08001482 hash[4]=((__force u16)sport << 16) + (__force u16)dport;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 memcpy(&hash[5],keyptr->secret,sizeof(__u32) * 7);
1484
Al Virob09b845c2006-11-14 20:52:19 -08001485 seq = twothirdsMD4Transform((const __u32 *)daddr, hash) & HASH_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 seq += keyptr->count;
1487
1488 do_gettimeofday(&tv);
1489 seq += tv.tv_usec + tv.tv_sec * 1000000;
1490
1491 return seq;
1492}
1493EXPORT_SYMBOL(secure_tcpv6_sequence_number);
1494#endif
1495
1496/* The code below is shamelessly stolen from secure_tcp_sequence_number().
1497 * All blames to Andrey V. Savochkin <saw@msu.ru>.
1498 */
Al Virob09b845c2006-11-14 20:52:19 -08001499__u32 secure_ip_id(__be32 daddr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500{
1501 struct keydata *keyptr;
1502 __u32 hash[4];
1503
1504 keyptr = get_keyptr();
1505
1506 /*
1507 * Pick a unique starting offset for each IP destination.
1508 * The dest ip address is placed in the starting vector,
1509 * which is then hashed with random data.
1510 */
Al Virob09b845c2006-11-14 20:52:19 -08001511 hash[0] = (__force __u32)daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 hash[1] = keyptr->secret[9];
1513 hash[2] = keyptr->secret[10];
1514 hash[3] = keyptr->secret[11];
1515
1516 return half_md4_transform(hash, keyptr->secret);
1517}
1518
1519#ifdef CONFIG_INET
1520
Al Virob09b845c2006-11-14 20:52:19 -08001521__u32 secure_tcp_sequence_number(__be32 saddr, __be32 daddr,
1522 __be16 sport, __be16 dport)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523{
1524 struct timeval tv;
1525 __u32 seq;
1526 __u32 hash[4];
1527 struct keydata *keyptr = get_keyptr();
1528
1529 /*
1530 * Pick a unique starting offset for each TCP connection endpoints
1531 * (saddr, daddr, sport, dport).
1532 * Note that the words are placed into the starting vector, which is
1533 * then mixed with a partial MD4 over random data.
1534 */
Al Virob09b845c2006-11-14 20:52:19 -08001535 hash[0]=(__force u32)saddr;
1536 hash[1]=(__force u32)daddr;
1537 hash[2]=((__force u16)sport << 16) + (__force u16)dport;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 hash[3]=keyptr->secret[11];
1539
1540 seq = half_md4_transform(hash, keyptr->secret) & HASH_MASK;
1541 seq += keyptr->count;
1542 /*
1543 * As close as possible to RFC 793, which
1544 * suggests using a 250 kHz clock.
1545 * Further reading shows this assumes 2 Mb/s networks.
1546 * For 10 Mb/s Ethernet, a 1 MHz clock is appropriate.
1547 * That's funny, Linux has one built in! Use it!
1548 * (Networks are faster now - should this be increased?)
1549 */
1550 do_gettimeofday(&tv);
1551 seq += tv.tv_usec + tv.tv_sec * 1000000;
1552#if 0
1553 printk("init_seq(%lx, %lx, %d, %d) = %d\n",
1554 saddr, daddr, sport, dport, seq);
1555#endif
1556 return seq;
1557}
1558
1559EXPORT_SYMBOL(secure_tcp_sequence_number);
1560
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -08001561/* Generate secure starting point for ephemeral IPV4 transport port search */
Al Virob09b845c2006-11-14 20:52:19 -08001562u32 secure_ipv4_port_ephemeral(__be32 saddr, __be32 daddr, __be16 dport)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563{
1564 struct keydata *keyptr = get_keyptr();
1565 u32 hash[4];
1566
1567 /*
1568 * Pick a unique starting offset for each ephemeral port search
1569 * (saddr, daddr, dport) and 48bits of random data.
1570 */
Al Virob09b845c2006-11-14 20:52:19 -08001571 hash[0] = (__force u32)saddr;
1572 hash[1] = (__force u32)daddr;
1573 hash[2] = (__force u32)dport ^ keyptr->secret[10];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 hash[3] = keyptr->secret[11];
1575
1576 return half_md4_transform(hash, keyptr->secret);
1577}
1578
1579#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
Al Virob09b845c2006-11-14 20:52:19 -08001580u32 secure_ipv6_port_ephemeral(const __be32 *saddr, const __be32 *daddr, __be16 dport)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581{
1582 struct keydata *keyptr = get_keyptr();
1583 u32 hash[12];
1584
1585 memcpy(hash, saddr, 16);
Al Virob09b845c2006-11-14 20:52:19 -08001586 hash[4] = (__force u32)dport;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 memcpy(&hash[5],keyptr->secret,sizeof(__u32) * 7);
1588
Al Virob09b845c2006-11-14 20:52:19 -08001589 return twothirdsMD4Transform((const __u32 *)daddr, hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591#endif
1592
Arnaldo Carvalho de Meloc4365c92005-08-09 20:12:30 -07001593#if defined(CONFIG_IP_DCCP) || defined(CONFIG_IP_DCCP_MODULE)
1594/* Similar to secure_tcp_sequence_number but generate a 48 bit value
1595 * bit's 32-47 increase every key exchange
1596 * 0-31 hash(source, dest)
1597 */
Al Virob09b845c2006-11-14 20:52:19 -08001598u64 secure_dccp_sequence_number(__be32 saddr, __be32 daddr,
1599 __be16 sport, __be16 dport)
Arnaldo Carvalho de Meloc4365c92005-08-09 20:12:30 -07001600{
1601 struct timeval tv;
1602 u64 seq;
1603 __u32 hash[4];
1604 struct keydata *keyptr = get_keyptr();
1605
Al Virob09b845c2006-11-14 20:52:19 -08001606 hash[0] = (__force u32)saddr;
1607 hash[1] = (__force u32)daddr;
1608 hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
Arnaldo Carvalho de Meloc4365c92005-08-09 20:12:30 -07001609 hash[3] = keyptr->secret[11];
1610
1611 seq = half_md4_transform(hash, keyptr->secret);
1612 seq |= ((u64)keyptr->count) << (32 - HASH_BITS);
1613
1614 do_gettimeofday(&tv);
1615 seq += tv.tv_usec + tv.tv_sec * 1000000;
1616 seq &= (1ull << 48) - 1;
1617#if 0
1618 printk("dccp init_seq(%lx, %lx, %d, %d) = %d\n",
1619 saddr, daddr, sport, dport, seq);
1620#endif
1621 return seq;
1622}
1623
1624EXPORT_SYMBOL(secure_dccp_sequence_number);
1625#endif
1626
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627#endif /* CONFIG_INET */
1628
1629
1630/*
1631 * Get a random word for internal kernel use only. Similar to urandom but
1632 * with the goal of minimal entropy pool depletion. As a result, the random
1633 * value is not cryptographically secure but for several uses the cost of
1634 * depleting entropy is too high
1635 */
1636unsigned int get_random_int(void)
1637{
1638 /*
1639 * Use IP's RNG. It suits our purpose perfectly: it re-keys itself
1640 * every second, from the entropy pool (and thus creates a limited
1641 * drain on it), and uses halfMD4Transform within the second. We
1642 * also mix it with jiffies and the PID:
1643 */
Al Virob09b845c2006-11-14 20:52:19 -08001644 return secure_ip_id((__force __be32)(current->pid + jiffies));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645}
1646
1647/*
1648 * randomize_range() returns a start address such that
1649 *
1650 * [...... <range> .....]
1651 * start end
1652 *
1653 * a <range> with size "len" starting at the return value is inside in the
1654 * area defined by [start, end], but is otherwise randomized.
1655 */
1656unsigned long
1657randomize_range(unsigned long start, unsigned long end, unsigned long len)
1658{
1659 unsigned long range = end - len - start;
1660
1661 if (end <= start + len)
1662 return 0;
1663 return PAGE_ALIGN(get_random_int() % range + start);
1664}