]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blame - drivers/char/random.c
pm_qos_requirement might sleep
[mirror_ubuntu-hirsute-kernel.git] / drivers / char / random.c
CommitLineData
1da177e4
LT
1/*
2 * random.c -- A strong random number generator
3 *
9e95ce27 4 * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005
1da177e4
LT
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>
1da177e4
LT
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>
27ac792c 239#include <linux/mm.h>
1da177e4
LT
240#include <linux/spinlock.h>
241#include <linux/percpu.h>
242#include <linux/cryptohash.h>
243
244#include <asm/processor.h>
245#include <asm/uaccess.h>
246#include <asm/irq.h>
247#include <asm/io.h>
248
249/*
250 * Configuration information
251 */
252#define INPUT_POOL_WORDS 128
253#define OUTPUT_POOL_WORDS 32
254#define SEC_XFER_SIZE 512
255
256/*
257 * The minimum number of bits of entropy before we wake up a read on
258 * /dev/random. Should be enough to do a significant reseed.
259 */
260static int random_read_wakeup_thresh = 64;
261
262/*
263 * If the entropy count falls under this number of bits, then we
264 * should wake up processes which are selecting or polling on write
265 * access to /dev/random.
266 */
267static int random_write_wakeup_thresh = 128;
268
269/*
270 * When the input pool goes over trickle_thresh, start dropping most
271 * samples to avoid wasting CPU time and reduce lock contention.
272 */
273
6c036527 274static int trickle_thresh __read_mostly = INPUT_POOL_WORDS * 28;
1da177e4 275
90b75ee5 276static DEFINE_PER_CPU(int, trickle_count);
1da177e4
LT
277
278/*
279 * A pool of size .poolwords is stirred with a primitive polynomial
280 * of degree .poolwords over GF(2). The taps for various sizes are
281 * defined below. They are chosen to be evenly spaced (minimum RMS
282 * distance from evenly spaced; the numbers in the comments are a
283 * scaled squared error sum) except for the last tap, which is 1 to
284 * get the twisting happening as fast as possible.
285 */
286static struct poolinfo {
287 int poolwords;
288 int tap1, tap2, tap3, tap4, tap5;
289} poolinfo_table[] = {
290 /* x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 -- 105 */
291 { 128, 103, 76, 51, 25, 1 },
292 /* x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 -- 15 */
293 { 32, 26, 20, 14, 7, 1 },
294#if 0
295 /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1 -- 115 */
296 { 2048, 1638, 1231, 819, 411, 1 },
297
298 /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
299 { 1024, 817, 615, 412, 204, 1 },
300
301 /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
302 { 1024, 819, 616, 410, 207, 2 },
303
304 /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
305 { 512, 411, 308, 208, 104, 1 },
306
307 /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
308 { 512, 409, 307, 206, 102, 2 },
309 /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
310 { 512, 409, 309, 205, 103, 2 },
311
312 /* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
313 { 256, 205, 155, 101, 52, 1 },
314
315 /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
316 { 128, 103, 78, 51, 27, 2 },
317
318 /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
319 { 64, 52, 39, 26, 14, 1 },
320#endif
321};
322
323#define POOLBITS poolwords*32
324#define POOLBYTES poolwords*4
325
326/*
327 * For the purposes of better mixing, we use the CRC-32 polynomial as
328 * well to make a twisted Generalized Feedback Shift Reigster
329 *
330 * (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR generators. ACM
331 * Transactions on Modeling and Computer Simulation 2(3):179-194.
332 * Also see M. Matsumoto & Y. Kurita, 1994. Twisted GFSR generators
333 * II. ACM Transactions on Mdeling and Computer Simulation 4:254-266)
334 *
335 * Thanks to Colin Plumb for suggesting this.
336 *
337 * We have not analyzed the resultant polynomial to prove it primitive;
338 * in fact it almost certainly isn't. Nonetheless, the irreducible factors
339 * of a random large-degree polynomial over GF(2) are more than large enough
340 * that periodicity is not a concern.
341 *
342 * The input hash is much less sensitive than the output hash. All
343 * that we want of it is that it be a good non-cryptographic hash;
344 * i.e. it not produce collisions when fed "random" data of the sort
345 * we expect to see. As long as the pool state differs for different
346 * inputs, we have preserved the input entropy and done a good job.
347 * The fact that an intelligent attacker can construct inputs that
348 * will produce controlled alterations to the pool's state is not
349 * important because we don't consider such inputs to contribute any
350 * randomness. The only property we need with respect to them is that
351 * the attacker can't increase his/her knowledge of the pool's state.
352 * Since all additions are reversible (knowing the final state and the
353 * input, you can reconstruct the initial state), if an attacker has
354 * any uncertainty about the initial state, he/she can only shuffle
355 * that uncertainty about, but never cause any collisions (which would
356 * decrease the uncertainty).
357 *
358 * The chosen system lets the state of the pool be (essentially) the input
359 * modulo the generator polymnomial. Now, for random primitive polynomials,
360 * this is a universal class of hash functions, meaning that the chance
361 * of a collision is limited by the attacker's knowledge of the generator
362 * polynomail, so if it is chosen at random, an attacker can never force
363 * a collision. Here, we use a fixed polynomial, but we *can* assume that
364 * ###--> it is unknown to the processes generating the input entropy. <-###
365 * Because of this important property, this is a good, collision-resistant
366 * hash; hash collisions will occur no more often than chance.
367 */
368
369/*
370 * Static global variables
371 */
372static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
373static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
9a6f70bb 374static struct fasync_struct *fasync;
1da177e4
LT
375
376#if 0
90b75ee5 377static int debug;
1da177e4 378module_param(debug, bool, 0644);
90b75ee5
MM
379#define DEBUG_ENT(fmt, arg...) do { \
380 if (debug) \
381 printk(KERN_DEBUG "random %04d %04d %04d: " \
382 fmt,\
383 input_pool.entropy_count,\
384 blocking_pool.entropy_count,\
385 nonblocking_pool.entropy_count,\
386 ## arg); } while (0)
1da177e4
LT
387#else
388#define DEBUG_ENT(fmt, arg...) do {} while (0)
389#endif
390
391/**********************************************************************
392 *
393 * OS independent entropy store. Here are the functions which handle
394 * storing entropy in an entropy pool.
395 *
396 **********************************************************************/
397
398struct entropy_store;
399struct entropy_store {
43358209 400 /* read-only data: */
1da177e4
LT
401 struct poolinfo *poolinfo;
402 __u32 *pool;
403 const char *name;
404 int limit;
405 struct entropy_store *pull;
406
407 /* read-write data: */
43358209 408 spinlock_t lock;
1da177e4
LT
409 unsigned add_ptr;
410 int entropy_count;
411 int input_rotate;
412};
413
414static __u32 input_pool_data[INPUT_POOL_WORDS];
415static __u32 blocking_pool_data[OUTPUT_POOL_WORDS];
416static __u32 nonblocking_pool_data[OUTPUT_POOL_WORDS];
417
418static struct entropy_store input_pool = {
419 .poolinfo = &poolinfo_table[0],
420 .name = "input",
421 .limit = 1,
e4d91918 422 .lock = __SPIN_LOCK_UNLOCKED(&input_pool.lock),
1da177e4
LT
423 .pool = input_pool_data
424};
425
426static struct entropy_store blocking_pool = {
427 .poolinfo = &poolinfo_table[1],
428 .name = "blocking",
429 .limit = 1,
430 .pull = &input_pool,
e4d91918 431 .lock = __SPIN_LOCK_UNLOCKED(&blocking_pool.lock),
1da177e4
LT
432 .pool = blocking_pool_data
433};
434
435static struct entropy_store nonblocking_pool = {
436 .poolinfo = &poolinfo_table[1],
437 .name = "nonblocking",
438 .pull = &input_pool,
e4d91918 439 .lock = __SPIN_LOCK_UNLOCKED(&nonblocking_pool.lock),
1da177e4
LT
440 .pool = nonblocking_pool_data
441};
442
443/*
e68e5b66 444 * This function adds bytes into the entropy "pool". It does not
1da177e4 445 * update the entropy estimate. The caller should call
adc782da 446 * credit_entropy_bits if this is appropriate.
1da177e4
LT
447 *
448 * The pool is stirred with a primitive polynomial of the appropriate
449 * degree, and then twisted. We twist by three bits at a time because
450 * it's cheap to do so and helps slightly in the expected case where
451 * the entropy is concentrated in the low-order bits.
452 */
e68e5b66
MM
453static void mix_pool_bytes_extract(struct entropy_store *r, const void *in,
454 int nbytes, __u8 out[64])
1da177e4
LT
455{
456 static __u32 const twist_table[8] = {
457 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
458 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
993ba211 459 unsigned long i, j, tap1, tap2, tap3, tap4, tap5;
feee7697 460 int input_rotate;
1da177e4 461 int wordmask = r->poolinfo->poolwords - 1;
e68e5b66 462 const char *bytes = in;
6d38b827 463 __u32 w;
1da177e4
LT
464 unsigned long flags;
465
466 /* Taps are constant, so we can load them without holding r->lock. */
467 tap1 = r->poolinfo->tap1;
468 tap2 = r->poolinfo->tap2;
469 tap3 = r->poolinfo->tap3;
470 tap4 = r->poolinfo->tap4;
471 tap5 = r->poolinfo->tap5;
1da177e4
LT
472
473 spin_lock_irqsave(&r->lock, flags);
1da177e4 474 input_rotate = r->input_rotate;
993ba211 475 i = r->add_ptr;
1da177e4 476
e68e5b66
MM
477 /* mix one byte at a time to simplify size handling and churn faster */
478 while (nbytes--) {
479 w = rol32(*bytes++, input_rotate & 31);
993ba211 480 i = (i - 1) & wordmask;
1da177e4
LT
481
482 /* XOR in the various taps */
993ba211 483 w ^= r->pool[i];
1da177e4
LT
484 w ^= r->pool[(i + tap1) & wordmask];
485 w ^= r->pool[(i + tap2) & wordmask];
486 w ^= r->pool[(i + tap3) & wordmask];
487 w ^= r->pool[(i + tap4) & wordmask];
488 w ^= r->pool[(i + tap5) & wordmask];
993ba211
MM
489
490 /* Mix the result back in with a twist */
1da177e4 491 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
feee7697
MM
492
493 /*
494 * Normally, we add 7 bits of rotation to the pool.
495 * At the beginning of the pool, add an extra 7 bits
496 * rotation, so that successive passes spread the
497 * input bits across the pool evenly.
498 */
499 input_rotate += i ? 7 : 14;
1da177e4
LT
500 }
501
502 r->input_rotate = input_rotate;
993ba211 503 r->add_ptr = i;
1da177e4 504
993ba211
MM
505 if (out)
506 for (j = 0; j < 16; j++)
e68e5b66 507 ((__u32 *)out)[j] = r->pool[(i - j) & wordmask];
1da177e4
LT
508
509 spin_unlock_irqrestore(&r->lock, flags);
510}
511
e68e5b66 512static void mix_pool_bytes(struct entropy_store *r, const void *in, int bytes)
1da177e4 513{
e68e5b66 514 mix_pool_bytes_extract(r, in, bytes, NULL);
1da177e4
LT
515}
516
517/*
518 * Credit (or debit) the entropy store with n bits of entropy
519 */
adc782da 520static void credit_entropy_bits(struct entropy_store *r, int nbits)
1da177e4
LT
521{
522 unsigned long flags;
523
adc782da
MM
524 if (!nbits)
525 return;
526
1da177e4
LT
527 spin_lock_irqsave(&r->lock, flags);
528
adc782da
MM
529 DEBUG_ENT("added %d entropy credits to %s\n", nbits, r->name);
530 r->entropy_count += nbits;
531 if (r->entropy_count < 0) {
532 DEBUG_ENT("negative entropy/overflow\n");
1da177e4 533 r->entropy_count = 0;
adc782da 534 } else if (r->entropy_count > r->poolinfo->POOLBITS)
1da177e4 535 r->entropy_count = r->poolinfo->POOLBITS;
1da177e4 536
88c730da 537 /* should we wake readers? */
9a6f70bb
JD
538 if (r == &input_pool &&
539 r->entropy_count >= random_read_wakeup_thresh) {
88c730da 540 wake_up_interruptible(&random_read_wait);
9a6f70bb
JD
541 kill_fasync(&fasync, SIGIO, POLL_IN);
542 }
88c730da 543
1da177e4
LT
544 spin_unlock_irqrestore(&r->lock, flags);
545}
546
547/*********************************************************************
548 *
549 * Entropy input management
550 *
551 *********************************************************************/
552
553/* There is one of these per entropy source */
554struct timer_rand_state {
555 cycles_t last_time;
90b75ee5 556 long last_delta, last_delta2;
1da177e4
LT
557 unsigned dont_count_entropy:1;
558};
559
560static struct timer_rand_state input_timer_state;
561static struct timer_rand_state *irq_timer_state[NR_IRQS];
562
563/*
564 * This function adds entropy to the entropy "pool" by using timing
565 * delays. It uses the timer_rand_state structure to make an estimate
566 * of how many bits of entropy this call has added to the pool.
567 *
568 * The number "num" is also added to the pool - it should somehow describe
569 * the type of event which just happened. This is currently 0-255 for
570 * keyboard scan codes, and 256 upwards for interrupts.
571 *
572 */
573static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
574{
575 struct {
576 cycles_t cycles;
577 long jiffies;
578 unsigned num;
579 } sample;
580 long delta, delta2, delta3;
581
582 preempt_disable();
583 /* if over the trickle threshold, use only 1 in 4096 samples */
584 if (input_pool.entropy_count > trickle_thresh &&
585 (__get_cpu_var(trickle_count)++ & 0xfff))
586 goto out;
587
588 sample.jiffies = jiffies;
589 sample.cycles = get_cycles();
590 sample.num = num;
e68e5b66 591 mix_pool_bytes(&input_pool, &sample, sizeof(sample));
1da177e4
LT
592
593 /*
594 * Calculate number of bits of randomness we probably added.
595 * We take into account the first, second and third-order deltas
596 * in order to make our estimate.
597 */
598
599 if (!state->dont_count_entropy) {
600 delta = sample.jiffies - state->last_time;
601 state->last_time = sample.jiffies;
602
603 delta2 = delta - state->last_delta;
604 state->last_delta = delta;
605
606 delta3 = delta2 - state->last_delta2;
607 state->last_delta2 = delta2;
608
609 if (delta < 0)
610 delta = -delta;
611 if (delta2 < 0)
612 delta2 = -delta2;
613 if (delta3 < 0)
614 delta3 = -delta3;
615 if (delta > delta2)
616 delta = delta2;
617 if (delta > delta3)
618 delta = delta3;
619
620 /*
621 * delta is now minimum absolute delta.
622 * Round down by 1 bit on general principles,
623 * and limit entropy entimate to 12 bits.
624 */
adc782da
MM
625 credit_entropy_bits(&input_pool,
626 min_t(int, fls(delta>>1), 11));
1da177e4 627 }
1da177e4
LT
628out:
629 preempt_enable();
630}
631
d251575a 632void add_input_randomness(unsigned int type, unsigned int code,
1da177e4
LT
633 unsigned int value)
634{
635 static unsigned char last_value;
636
637 /* ignore autorepeat and the like */
638 if (value == last_value)
639 return;
640
641 DEBUG_ENT("input event\n");
642 last_value = value;
643 add_timer_randomness(&input_timer_state,
644 (type << 4) ^ code ^ (code >> 4) ^ value);
645}
80fc9f53 646EXPORT_SYMBOL_GPL(add_input_randomness);
1da177e4
LT
647
648void add_interrupt_randomness(int irq)
649{
c80544dc 650 if (irq >= NR_IRQS || irq_timer_state[irq] == NULL)
1da177e4
LT
651 return;
652
653 DEBUG_ENT("irq event %d\n", irq);
654 add_timer_randomness(irq_timer_state[irq], 0x100 + irq);
655}
656
9361401e 657#ifdef CONFIG_BLOCK
1da177e4
LT
658void add_disk_randomness(struct gendisk *disk)
659{
660 if (!disk || !disk->random)
661 return;
662 /* first major is 1, so we get >= 0x200 here */
663 DEBUG_ENT("disk event %d:%d\n", disk->major, disk->first_minor);
664
665 add_timer_randomness(disk->random,
666 0x100 + MKDEV(disk->major, disk->first_minor));
667}
9361401e 668#endif
1da177e4
LT
669
670#define EXTRACT_SIZE 10
671
672/*********************************************************************
673 *
674 * Entropy extraction routines
675 *
676 *********************************************************************/
677
90b75ee5 678static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1da177e4
LT
679 size_t nbytes, int min, int rsvd);
680
681/*
682 * This utility inline function is responsible for transfering entropy
683 * from the primary pool to the secondary extraction pool. We make
684 * sure we pull enough for a 'catastrophic reseed'.
685 */
686static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
687{
688 __u32 tmp[OUTPUT_POOL_WORDS];
689
690 if (r->pull && r->entropy_count < nbytes * 8 &&
691 r->entropy_count < r->poolinfo->POOLBITS) {
5a021e9f 692 /* If we're limited, always leave two wakeup worth's BITS */
1da177e4 693 int rsvd = r->limit ? 0 : random_read_wakeup_thresh/4;
5a021e9f
MM
694 int bytes = nbytes;
695
696 /* pull at least as many as BYTES as wakeup BITS */
697 bytes = max_t(int, bytes, random_read_wakeup_thresh / 8);
698 /* but never more than the buffer size */
699 bytes = min_t(int, bytes, sizeof(tmp));
1da177e4
LT
700
701 DEBUG_ENT("going to reseed %s with %d bits "
702 "(%d of %d requested)\n",
703 r->name, bytes * 8, nbytes * 8, r->entropy_count);
704
90b75ee5
MM
705 bytes = extract_entropy(r->pull, tmp, bytes,
706 random_read_wakeup_thresh / 8, rsvd);
e68e5b66 707 mix_pool_bytes(r, tmp, bytes);
adc782da 708 credit_entropy_bits(r, bytes*8);
1da177e4
LT
709 }
710}
711
712/*
713 * These functions extracts randomness from the "entropy pool", and
714 * returns it in a buffer.
715 *
716 * The min parameter specifies the minimum amount we can pull before
717 * failing to avoid races that defeat catastrophic reseeding while the
718 * reserved parameter indicates how much entropy we must leave in the
719 * pool after each pull to avoid starving other readers.
720 *
721 * Note: extract_entropy() assumes that .poolwords is a multiple of 16 words.
722 */
723
724static size_t account(struct entropy_store *r, size_t nbytes, int min,
725 int reserved)
726{
727 unsigned long flags;
728
729 BUG_ON(r->entropy_count > r->poolinfo->POOLBITS);
730
731 /* Hold lock while accounting */
732 spin_lock_irqsave(&r->lock, flags);
733
734 DEBUG_ENT("trying to extract %d bits from %s\n",
735 nbytes * 8, r->name);
736
737 /* Can we pull enough? */
738 if (r->entropy_count / 8 < min + reserved) {
739 nbytes = 0;
740 } else {
741 /* If limited, never pull more than available */
742 if (r->limit && nbytes + reserved >= r->entropy_count / 8)
743 nbytes = r->entropy_count/8 - reserved;
744
90b75ee5 745 if (r->entropy_count / 8 >= nbytes + reserved)
1da177e4
LT
746 r->entropy_count -= nbytes*8;
747 else
748 r->entropy_count = reserved;
749
9a6f70bb 750 if (r->entropy_count < random_write_wakeup_thresh) {
1da177e4 751 wake_up_interruptible(&random_write_wait);
9a6f70bb
JD
752 kill_fasync(&fasync, SIGIO, POLL_OUT);
753 }
1da177e4
LT
754 }
755
756 DEBUG_ENT("debiting %d entropy credits from %s%s\n",
757 nbytes * 8, r->name, r->limit ? "" : " (unlimited)");
758
759 spin_unlock_irqrestore(&r->lock, flags);
760
761 return nbytes;
762}
763
764static void extract_buf(struct entropy_store *r, __u8 *out)
765{
602b6aee 766 int i;
e68e5b66
MM
767 __u32 hash[5], workspace[SHA_WORKSPACE_WORDS];
768 __u8 extract[64];
1da177e4 769
1c0ad3d4 770 /* Generate a hash across the pool, 16 words (512 bits) at a time */
ffd8d3fa 771 sha_init(hash);
1c0ad3d4
MM
772 for (i = 0; i < r->poolinfo->poolwords; i += 16)
773 sha_transform(hash, (__u8 *)(r->pool + i), workspace);
774
1da177e4 775 /*
1c0ad3d4
MM
776 * We mix the hash back into the pool to prevent backtracking
777 * attacks (where the attacker knows the state of the pool
778 * plus the current outputs, and attempts to find previous
779 * ouputs), unless the hash function can be inverted. By
780 * mixing at least a SHA1 worth of hash data back, we make
781 * brute-forcing the feedback as hard as brute-forcing the
782 * hash.
1da177e4 783 */
e68e5b66 784 mix_pool_bytes_extract(r, hash, sizeof(hash), extract);
1da177e4
LT
785
786 /*
1c0ad3d4
MM
787 * To avoid duplicates, we atomically extract a portion of the
788 * pool while mixing, and hash one final time.
1da177e4 789 */
e68e5b66 790 sha_transform(hash, extract, workspace);
ffd8d3fa
MM
791 memset(extract, 0, sizeof(extract));
792 memset(workspace, 0, sizeof(workspace));
1da177e4
LT
793
794 /*
1c0ad3d4
MM
795 * In case the hash function has some recognizable output
796 * pattern, we fold it in half. Thus, we always feed back
797 * twice as much data as we output.
1da177e4 798 */
ffd8d3fa
MM
799 hash[0] ^= hash[3];
800 hash[1] ^= hash[4];
801 hash[2] ^= rol32(hash[2], 16);
802 memcpy(out, hash, EXTRACT_SIZE);
803 memset(hash, 0, sizeof(hash));
1da177e4
LT
804}
805
90b75ee5 806static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1da177e4
LT
807 size_t nbytes, int min, int reserved)
808{
809 ssize_t ret = 0, i;
810 __u8 tmp[EXTRACT_SIZE];
811
812 xfer_secondary_pool(r, nbytes);
813 nbytes = account(r, nbytes, min, reserved);
814
815 while (nbytes) {
816 extract_buf(r, tmp);
817 i = min_t(int, nbytes, EXTRACT_SIZE);
818 memcpy(buf, tmp, i);
819 nbytes -= i;
820 buf += i;
821 ret += i;
822 }
823
824 /* Wipe data just returned from memory */
825 memset(tmp, 0, sizeof(tmp));
826
827 return ret;
828}
829
830static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
831 size_t nbytes)
832{
833 ssize_t ret = 0, i;
834 __u8 tmp[EXTRACT_SIZE];
835
836 xfer_secondary_pool(r, nbytes);
837 nbytes = account(r, nbytes, 0, 0);
838
839 while (nbytes) {
840 if (need_resched()) {
841 if (signal_pending(current)) {
842 if (ret == 0)
843 ret = -ERESTARTSYS;
844 break;
845 }
846 schedule();
847 }
848
849 extract_buf(r, tmp);
850 i = min_t(int, nbytes, EXTRACT_SIZE);
851 if (copy_to_user(buf, tmp, i)) {
852 ret = -EFAULT;
853 break;
854 }
855
856 nbytes -= i;
857 buf += i;
858 ret += i;
859 }
860
861 /* Wipe data just returned from memory */
862 memset(tmp, 0, sizeof(tmp));
863
864 return ret;
865}
866
867/*
868 * This function is the exported kernel interface. It returns some
869 * number of good random numbers, suitable for seeding TCP sequence
870 * numbers, etc.
871 */
872void get_random_bytes(void *buf, int nbytes)
873{
874 extract_entropy(&nonblocking_pool, buf, nbytes, 0, 0);
875}
1da177e4
LT
876EXPORT_SYMBOL(get_random_bytes);
877
878/*
879 * init_std_data - initialize pool with system data
880 *
881 * @r: pool to initialize
882 *
883 * This function clears the pool's entropy count and mixes some system
884 * data into the pool to prepare it for use. The pool is not cleared
885 * as that can only decrease the entropy in the pool.
886 */
887static void init_std_data(struct entropy_store *r)
888{
f8595815 889 ktime_t now;
1da177e4
LT
890 unsigned long flags;
891
892 spin_lock_irqsave(&r->lock, flags);
893 r->entropy_count = 0;
894 spin_unlock_irqrestore(&r->lock, flags);
895
f8595815 896 now = ktime_get_real();
e68e5b66
MM
897 mix_pool_bytes(r, &now, sizeof(now));
898 mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1da177e4
LT
899}
900
53c3f63e 901static int rand_initialize(void)
1da177e4
LT
902{
903 init_std_data(&input_pool);
904 init_std_data(&blocking_pool);
905 init_std_data(&nonblocking_pool);
906 return 0;
907}
908module_init(rand_initialize);
909
910void rand_initialize_irq(int irq)
911{
912 struct timer_rand_state *state;
913
914 if (irq >= NR_IRQS || irq_timer_state[irq])
915 return;
916
917 /*
f8595815 918 * If kzalloc returns null, we just won't use that entropy
1da177e4
LT
919 * source.
920 */
f8595815
ED
921 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
922 if (state)
1da177e4 923 irq_timer_state[irq] = state;
1da177e4
LT
924}
925
9361401e 926#ifdef CONFIG_BLOCK
1da177e4
LT
927void rand_initialize_disk(struct gendisk *disk)
928{
929 struct timer_rand_state *state;
930
931 /*
f8595815 932 * If kzalloc returns null, we just won't use that entropy
1da177e4
LT
933 * source.
934 */
f8595815
ED
935 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
936 if (state)
1da177e4 937 disk->random = state;
1da177e4 938}
9361401e 939#endif
1da177e4
LT
940
941static ssize_t
90b75ee5 942random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1da177e4
LT
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
90b75ee5 1004urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1da177e4
LT
1005{
1006 return extract_entropy_user(&nonblocking_pool, buf, nbytes);
1007}
1008
1009static unsigned int
1010random_poll(struct file *file, poll_table * wait)
1011{
1012 unsigned int mask;
1013
1014 poll_wait(file, &random_read_wait, wait);
1015 poll_wait(file, &random_write_wait, wait);
1016 mask = 0;
1017 if (input_pool.entropy_count >= random_read_wakeup_thresh)
1018 mask |= POLLIN | POLLRDNORM;
1019 if (input_pool.entropy_count < random_write_wakeup_thresh)
1020 mask |= POLLOUT | POLLWRNORM;
1021 return mask;
1022}
1023
7f397dcd
MM
1024static int
1025write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1da177e4 1026{
1da177e4
LT
1027 size_t bytes;
1028 __u32 buf[16];
1029 const char __user *p = buffer;
1da177e4 1030
7f397dcd
MM
1031 while (count > 0) {
1032 bytes = min(count, sizeof(buf));
1033 if (copy_from_user(&buf, p, bytes))
1034 return -EFAULT;
1da177e4 1035
7f397dcd 1036 count -= bytes;
1da177e4
LT
1037 p += bytes;
1038
e68e5b66 1039 mix_pool_bytes(r, buf, bytes);
91f3f1e3 1040 cond_resched();
1da177e4 1041 }
7f397dcd
MM
1042
1043 return 0;
1044}
1045
90b75ee5
MM
1046static ssize_t random_write(struct file *file, const char __user *buffer,
1047 size_t count, loff_t *ppos)
7f397dcd
MM
1048{
1049 size_t ret;
1050 struct inode *inode = file->f_path.dentry->d_inode;
1051
1052 ret = write_pool(&blocking_pool, buffer, count);
1053 if (ret)
1054 return ret;
1055 ret = write_pool(&nonblocking_pool, buffer, count);
1056 if (ret)
1057 return ret;
1058
1059 inode->i_mtime = current_fs_time(inode->i_sb);
1060 mark_inode_dirty(inode);
1061 return (ssize_t)count;
1da177e4
LT
1062}
1063
43ae4860 1064static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1da177e4
LT
1065{
1066 int size, ent_count;
1067 int __user *p = (int __user *)arg;
1068 int retval;
1069
1070 switch (cmd) {
1071 case RNDGETENTCNT:
43ae4860
MM
1072 /* inherently racy, no point locking */
1073 if (put_user(input_pool.entropy_count, p))
1da177e4
LT
1074 return -EFAULT;
1075 return 0;
1076 case RNDADDTOENTCNT:
1077 if (!capable(CAP_SYS_ADMIN))
1078 return -EPERM;
1079 if (get_user(ent_count, p))
1080 return -EFAULT;
adc782da 1081 credit_entropy_bits(&input_pool, ent_count);
1da177e4
LT
1082 return 0;
1083 case RNDADDENTROPY:
1084 if (!capable(CAP_SYS_ADMIN))
1085 return -EPERM;
1086 if (get_user(ent_count, p++))
1087 return -EFAULT;
1088 if (ent_count < 0)
1089 return -EINVAL;
1090 if (get_user(size, p++))
1091 return -EFAULT;
7f397dcd
MM
1092 retval = write_pool(&input_pool, (const char __user *)p,
1093 size);
1da177e4
LT
1094 if (retval < 0)
1095 return retval;
adc782da 1096 credit_entropy_bits(&input_pool, ent_count);
1da177e4
LT
1097 return 0;
1098 case RNDZAPENTCNT:
1099 case RNDCLEARPOOL:
1100 /* Clear the entropy pool counters. */
1101 if (!capable(CAP_SYS_ADMIN))
1102 return -EPERM;
53c3f63e 1103 rand_initialize();
1da177e4
LT
1104 return 0;
1105 default:
1106 return -EINVAL;
1107 }
1108}
1109
9a6f70bb
JD
1110static int random_fasync(int fd, struct file *filp, int on)
1111{
1112 return fasync_helper(fd, filp, on, &fasync);
1113}
1114
1115static int random_release(struct inode *inode, struct file *filp)
1116{
1117 return fasync_helper(-1, filp, 0, &fasync);
1118}
1119
2b8693c0 1120const struct file_operations random_fops = {
1da177e4
LT
1121 .read = random_read,
1122 .write = random_write,
1123 .poll = random_poll,
43ae4860 1124 .unlocked_ioctl = random_ioctl,
9a6f70bb
JD
1125 .fasync = random_fasync,
1126 .release = random_release,
1da177e4
LT
1127};
1128
2b8693c0 1129const struct file_operations urandom_fops = {
1da177e4
LT
1130 .read = urandom_read,
1131 .write = random_write,
43ae4860 1132 .unlocked_ioctl = random_ioctl,
9a6f70bb
JD
1133 .fasync = random_fasync,
1134 .release = random_release,
1da177e4
LT
1135};
1136
1137/***************************************************************
1138 * Random UUID interface
1139 *
1140 * Used here for a Boot ID, but can be useful for other kernel
1141 * drivers.
1142 ***************************************************************/
1143
1144/*
1145 * Generate random UUID
1146 */
1147void generate_random_uuid(unsigned char uuid_out[16])
1148{
1149 get_random_bytes(uuid_out, 16);
1150 /* Set UUID version to 4 --- truely random generation */
1151 uuid_out[6] = (uuid_out[6] & 0x0F) | 0x40;
1152 /* Set the UUID variant to DCE */
1153 uuid_out[8] = (uuid_out[8] & 0x3F) | 0x80;
1154}
1da177e4
LT
1155EXPORT_SYMBOL(generate_random_uuid);
1156
1157/********************************************************************
1158 *
1159 * Sysctl interface
1160 *
1161 ********************************************************************/
1162
1163#ifdef CONFIG_SYSCTL
1164
1165#include <linux/sysctl.h>
1166
1167static int min_read_thresh = 8, min_write_thresh;
1168static int max_read_thresh = INPUT_POOL_WORDS * 32;
1169static int max_write_thresh = INPUT_POOL_WORDS * 32;
1170static char sysctl_bootid[16];
1171
1172/*
1173 * These functions is used to return both the bootid UUID, and random
1174 * UUID. The difference is in whether table->data is NULL; if it is,
1175 * then a new UUID is generated and returned to the user.
1176 *
1177 * If the user accesses this via the proc interface, it will be returned
1178 * as an ASCII string in the standard UUID format. If accesses via the
1179 * sysctl system call, it is returned as 16 bytes of binary data.
1180 */
1181static int proc_do_uuid(ctl_table *table, int write, struct file *filp,
1182 void __user *buffer, size_t *lenp, loff_t *ppos)
1183{
1184 ctl_table fake_table;
1185 unsigned char buf[64], tmp_uuid[16], *uuid;
1186
1187 uuid = table->data;
1188 if (!uuid) {
1189 uuid = tmp_uuid;
1190 uuid[8] = 0;
1191 }
1192 if (uuid[8] == 0)
1193 generate_random_uuid(uuid);
1194
1195 sprintf(buf, "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-"
1196 "%02x%02x%02x%02x%02x%02x",
1197 uuid[0], uuid[1], uuid[2], uuid[3],
1198 uuid[4], uuid[5], uuid[6], uuid[7],
1199 uuid[8], uuid[9], uuid[10], uuid[11],
1200 uuid[12], uuid[13], uuid[14], uuid[15]);
1201 fake_table.data = buf;
1202 fake_table.maxlen = sizeof(buf);
1203
1204 return proc_dostring(&fake_table, write, filp, buffer, lenp, ppos);
1205}
1206
1207static int uuid_strategy(ctl_table *table, int __user *name, int nlen,
1208 void __user *oldval, size_t __user *oldlenp,
1f29bcd7 1209 void __user *newval, size_t newlen)
1da177e4
LT
1210{
1211 unsigned char tmp_uuid[16], *uuid;
1212 unsigned int len;
1213
1214 if (!oldval || !oldlenp)
1215 return 1;
1216
1217 uuid = table->data;
1218 if (!uuid) {
1219 uuid = tmp_uuid;
1220 uuid[8] = 0;
1221 }
1222 if (uuid[8] == 0)
1223 generate_random_uuid(uuid);
1224
1225 if (get_user(len, oldlenp))
1226 return -EFAULT;
1227 if (len) {
1228 if (len > 16)
1229 len = 16;
1230 if (copy_to_user(oldval, uuid, len) ||
1231 put_user(len, oldlenp))
1232 return -EFAULT;
1233 }
1234 return 1;
1235}
1236
1237static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
1238ctl_table random_table[] = {
1239 {
1240 .ctl_name = RANDOM_POOLSIZE,
1241 .procname = "poolsize",
1242 .data = &sysctl_poolsize,
1243 .maxlen = sizeof(int),
1244 .mode = 0444,
1245 .proc_handler = &proc_dointvec,
1246 },
1247 {
1248 .ctl_name = RANDOM_ENTROPY_COUNT,
1249 .procname = "entropy_avail",
1250 .maxlen = sizeof(int),
1251 .mode = 0444,
1252 .proc_handler = &proc_dointvec,
1253 .data = &input_pool.entropy_count,
1254 },
1255 {
1256 .ctl_name = RANDOM_READ_THRESH,
1257 .procname = "read_wakeup_threshold",
1258 .data = &random_read_wakeup_thresh,
1259 .maxlen = sizeof(int),
1260 .mode = 0644,
1261 .proc_handler = &proc_dointvec_minmax,
1262 .strategy = &sysctl_intvec,
1263 .extra1 = &min_read_thresh,
1264 .extra2 = &max_read_thresh,
1265 },
1266 {
1267 .ctl_name = RANDOM_WRITE_THRESH,
1268 .procname = "write_wakeup_threshold",
1269 .data = &random_write_wakeup_thresh,
1270 .maxlen = sizeof(int),
1271 .mode = 0644,
1272 .proc_handler = &proc_dointvec_minmax,
1273 .strategy = &sysctl_intvec,
1274 .extra1 = &min_write_thresh,
1275 .extra2 = &max_write_thresh,
1276 },
1277 {
1278 .ctl_name = RANDOM_BOOT_ID,
1279 .procname = "boot_id",
1280 .data = &sysctl_bootid,
1281 .maxlen = 16,
1282 .mode = 0444,
1283 .proc_handler = &proc_do_uuid,
1284 .strategy = &uuid_strategy,
1285 },
1286 {
1287 .ctl_name = RANDOM_UUID,
1288 .procname = "uuid",
1289 .maxlen = 16,
1290 .mode = 0444,
1291 .proc_handler = &proc_do_uuid,
1292 .strategy = &uuid_strategy,
1293 },
1294 { .ctl_name = 0 }
1295};
1296#endif /* CONFIG_SYSCTL */
1297
1298/********************************************************************
1299 *
1300 * Random funtions for networking
1301 *
1302 ********************************************************************/
1303
1304/*
1305 * TCP initial sequence number picking. This uses the random number
1306 * generator to pick an initial secret value. This value is hashed
1307 * along with the TCP endpoint information to provide a unique
1308 * starting point for each pair of TCP endpoints. This defeats
1309 * attacks which rely on guessing the initial TCP sequence number.
1310 * This algorithm was suggested by Steve Bellovin.
1311 *
1312 * Using a very strong hash was taking an appreciable amount of the total
1313 * TCP connection establishment time, so this is a weaker hash,
1314 * compensated for by changing the secret periodically.
1315 */
1316
1317/* F, G and H are basic MD4 functions: selection, majority, parity */
1318#define F(x, y, z) ((z) ^ ((x) & ((y) ^ (z))))
1319#define G(x, y, z) (((x) & (y)) + (((x) ^ (y)) & (z)))
1320#define H(x, y, z) ((x) ^ (y) ^ (z))
1321
1322/*
1323 * The generic round function. The application is so specific that
1324 * we don't bother protecting all the arguments with parens, as is generally
1325 * good macro practice, in favor of extra legibility.
1326 * Rotation is separate from addition to prevent recomputation
1327 */
1328#define ROUND(f, a, b, c, d, x, s) \
1329 (a += f(b, c, d) + x, a = (a << s) | (a >> (32 - s)))
1330#define K1 0
1331#define K2 013240474631UL
1332#define K3 015666365641UL
1333
1334#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1335
90b75ee5 1336static __u32 twothirdsMD4Transform(__u32 const buf[4], __u32 const in[12])
1da177e4
LT
1337{
1338 __u32 a = buf[0], b = buf[1], c = buf[2], d = buf[3];
1339
1340 /* Round 1 */
1341 ROUND(F, a, b, c, d, in[ 0] + K1, 3);
1342 ROUND(F, d, a, b, c, in[ 1] + K1, 7);
1343 ROUND(F, c, d, a, b, in[ 2] + K1, 11);
1344 ROUND(F, b, c, d, a, in[ 3] + K1, 19);
1345 ROUND(F, a, b, c, d, in[ 4] + K1, 3);
1346 ROUND(F, d, a, b, c, in[ 5] + K1, 7);
1347 ROUND(F, c, d, a, b, in[ 6] + K1, 11);
1348 ROUND(F, b, c, d, a, in[ 7] + K1, 19);
1349 ROUND(F, a, b, c, d, in[ 8] + K1, 3);
1350 ROUND(F, d, a, b, c, in[ 9] + K1, 7);
1351 ROUND(F, c, d, a, b, in[10] + K1, 11);
1352 ROUND(F, b, c, d, a, in[11] + K1, 19);
1353
1354 /* Round 2 */
1355 ROUND(G, a, b, c, d, in[ 1] + K2, 3);
1356 ROUND(G, d, a, b, c, in[ 3] + K2, 5);
1357 ROUND(G, c, d, a, b, in[ 5] + K2, 9);
1358 ROUND(G, b, c, d, a, in[ 7] + K2, 13);
1359 ROUND(G, a, b, c, d, in[ 9] + K2, 3);
1360 ROUND(G, d, a, b, c, in[11] + K2, 5);
1361 ROUND(G, c, d, a, b, in[ 0] + K2, 9);
1362 ROUND(G, b, c, d, a, in[ 2] + K2, 13);
1363 ROUND(G, a, b, c, d, in[ 4] + K2, 3);
1364 ROUND(G, d, a, b, c, in[ 6] + K2, 5);
1365 ROUND(G, c, d, a, b, in[ 8] + K2, 9);
1366 ROUND(G, b, c, d, a, in[10] + K2, 13);
1367
1368 /* Round 3 */
1369 ROUND(H, a, b, c, d, in[ 3] + K3, 3);
1370 ROUND(H, d, a, b, c, in[ 7] + K3, 9);
1371 ROUND(H, c, d, a, b, in[11] + K3, 11);
1372 ROUND(H, b, c, d, a, in[ 2] + K3, 15);
1373 ROUND(H, a, b, c, d, in[ 6] + K3, 3);
1374 ROUND(H, d, a, b, c, in[10] + K3, 9);
1375 ROUND(H, c, d, a, b, in[ 1] + K3, 11);
1376 ROUND(H, b, c, d, a, in[ 5] + K3, 15);
1377 ROUND(H, a, b, c, d, in[ 9] + K3, 3);
1378 ROUND(H, d, a, b, c, in[ 0] + K3, 9);
1379 ROUND(H, c, d, a, b, in[ 4] + K3, 11);
1380 ROUND(H, b, c, d, a, in[ 8] + K3, 15);
1381
1382 return buf[1] + b; /* "most hashed" word */
1383 /* Alternative: return sum of all words? */
1384}
1385#endif
1386
1387#undef ROUND
1388#undef F
1389#undef G
1390#undef H
1391#undef K1
1392#undef K2
1393#undef K3
1394
1395/* This should not be decreased so low that ISNs wrap too fast. */
1396#define REKEY_INTERVAL (300 * HZ)
1397/*
1398 * Bit layout of the tcp sequence numbers (before adding current time):
1399 * bit 24-31: increased after every key exchange
1400 * bit 0-23: hash(source,dest)
1401 *
1402 * The implementation is similar to the algorithm described
1403 * in the Appendix of RFC 1185, except that
1404 * - it uses a 1 MHz clock instead of a 250 kHz clock
1405 * - it performs a rekey every 5 minutes, which is equivalent
1406 * to a (source,dest) tulple dependent forward jump of the
1407 * clock by 0..2^(HASH_BITS+1)
1408 *
1409 * Thus the average ISN wraparound time is 68 minutes instead of
1410 * 4.55 hours.
1411 *
1412 * SMP cleanup and lock avoidance with poor man's RCU.
1413 * Manfred Spraul <manfred@colorfullife.com>
1414 *
1415 */
1416#define COUNT_BITS 8
1417#define COUNT_MASK ((1 << COUNT_BITS) - 1)
1418#define HASH_BITS 24
1419#define HASH_MASK ((1 << HASH_BITS) - 1)
1420
1421static struct keydata {
1422 __u32 count; /* already shifted to the final position */
1423 __u32 secret[12];
1424} ____cacheline_aligned ip_keydata[2];
1425
1426static unsigned int ip_cnt;
1427
65f27f38 1428static void rekey_seq_generator(struct work_struct *work);
1da177e4 1429
65f27f38 1430static DECLARE_DELAYED_WORK(rekey_work, rekey_seq_generator);
1da177e4
LT
1431
1432/*
1433 * Lock avoidance:
1434 * The ISN generation runs lockless - it's just a hash over random data.
1435 * State changes happen every 5 minutes when the random key is replaced.
1436 * Synchronization is performed by having two copies of the hash function
1437 * state and rekey_seq_generator always updates the inactive copy.
1438 * The copy is then activated by updating ip_cnt.
1439 * The implementation breaks down if someone blocks the thread
1440 * that processes SYN requests for more than 5 minutes. Should never
1441 * happen, and even if that happens only a not perfectly compliant
1442 * ISN is generated, nothing fatal.
1443 */
65f27f38 1444static void rekey_seq_generator(struct work_struct *work)
1da177e4
LT
1445{
1446 struct keydata *keyptr = &ip_keydata[1 ^ (ip_cnt & 1)];
1447
1448 get_random_bytes(keyptr->secret, sizeof(keyptr->secret));
1449 keyptr->count = (ip_cnt & COUNT_MASK) << HASH_BITS;
1450 smp_wmb();
1451 ip_cnt++;
1452 schedule_delayed_work(&rekey_work, REKEY_INTERVAL);
1453}
1454
1455static inline struct keydata *get_keyptr(void)
1456{
1457 struct keydata *keyptr = &ip_keydata[ip_cnt & 1];
1458
1459 smp_rmb();
1460
1461 return keyptr;
1462}
1463
1464static __init int seqgen_init(void)
1465{
1466 rekey_seq_generator(NULL);
1467 return 0;
1468}
1469late_initcall(seqgen_init);
1470
1471#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
b09b845c
AV
1472__u32 secure_tcpv6_sequence_number(__be32 *saddr, __be32 *daddr,
1473 __be16 sport, __be16 dport)
1da177e4 1474{
1da177e4
LT
1475 __u32 seq;
1476 __u32 hash[12];
1477 struct keydata *keyptr = get_keyptr();
1478
1479 /* The procedure is the same as for IPv4, but addresses are longer.
1480 * Thus we must use twothirdsMD4Transform.
1481 */
1482
1483 memcpy(hash, saddr, 16);
90b75ee5
MM
1484 hash[4] = ((__force u16)sport << 16) + (__force u16)dport;
1485 memcpy(&hash[5], keyptr->secret, sizeof(__u32) * 7);
1da177e4 1486
b09b845c 1487 seq = twothirdsMD4Transform((const __u32 *)daddr, hash) & HASH_MASK;
1da177e4
LT
1488 seq += keyptr->count;
1489
6dd10a62 1490 seq += ktime_to_ns(ktime_get_real());
1da177e4
LT
1491
1492 return seq;
1493}
1494EXPORT_SYMBOL(secure_tcpv6_sequence_number);
1495#endif
1496
1497/* The code below is shamelessly stolen from secure_tcp_sequence_number().
1498 * All blames to Andrey V. Savochkin <saw@msu.ru>.
1499 */
b09b845c 1500__u32 secure_ip_id(__be32 daddr)
1da177e4
LT
1501{
1502 struct keydata *keyptr;
1503 __u32 hash[4];
1504
1505 keyptr = get_keyptr();
1506
1507 /*
1508 * Pick a unique starting offset for each IP destination.
1509 * The dest ip address is placed in the starting vector,
1510 * which is then hashed with random data.
1511 */
b09b845c 1512 hash[0] = (__force __u32)daddr;
1da177e4
LT
1513 hash[1] = keyptr->secret[9];
1514 hash[2] = keyptr->secret[10];
1515 hash[3] = keyptr->secret[11];
1516
1517 return half_md4_transform(hash, keyptr->secret);
1518}
1519
1520#ifdef CONFIG_INET
1521
b09b845c
AV
1522__u32 secure_tcp_sequence_number(__be32 saddr, __be32 daddr,
1523 __be16 sport, __be16 dport)
1da177e4 1524{
1da177e4
LT
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 */
90b75ee5
MM
1535 hash[0] = (__force u32)saddr;
1536 hash[1] = (__force u32)daddr;
1537 hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
1538 hash[3] = keyptr->secret[11];
1da177e4
LT
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.
9b42c336
ED
1546 * For 10 Mb/s Ethernet, a 1 MHz clock is appropriate.
1547 * For 10 Gb/s Ethernet, a 1 GHz clock should be ok, but
1548 * we also need to limit the resolution so that the u32 seq
1549 * overlaps less than one time per MSL (2 minutes).
1550 * Choosing a clock of 64 ns period is OK. (period of 274 s)
1da177e4 1551 */
6dd10a62 1552 seq += ktime_to_ns(ktime_get_real()) >> 6;
90b75ee5 1553
1da177e4
LT
1554 return seq;
1555}
1556
a7f5e7f1 1557/* Generate secure starting point for ephemeral IPV4 transport port search */
b09b845c 1558u32 secure_ipv4_port_ephemeral(__be32 saddr, __be32 daddr, __be16 dport)
1da177e4
LT
1559{
1560 struct keydata *keyptr = get_keyptr();
1561 u32 hash[4];
1562
1563 /*
1564 * Pick a unique starting offset for each ephemeral port search
1565 * (saddr, daddr, dport) and 48bits of random data.
1566 */
b09b845c
AV
1567 hash[0] = (__force u32)saddr;
1568 hash[1] = (__force u32)daddr;
1569 hash[2] = (__force u32)dport ^ keyptr->secret[10];
1da177e4
LT
1570 hash[3] = keyptr->secret[11];
1571
1572 return half_md4_transform(hash, keyptr->secret);
1573}
9f593653 1574EXPORT_SYMBOL_GPL(secure_ipv4_port_ephemeral);
1da177e4
LT
1575
1576#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
90b75ee5
MM
1577u32 secure_ipv6_port_ephemeral(const __be32 *saddr, const __be32 *daddr,
1578 __be16 dport)
1da177e4
LT
1579{
1580 struct keydata *keyptr = get_keyptr();
1581 u32 hash[12];
1582
1583 memcpy(hash, saddr, 16);
b09b845c 1584 hash[4] = (__force u32)dport;
90b75ee5 1585 memcpy(&hash[5], keyptr->secret, sizeof(__u32) * 7);
1da177e4 1586
b09b845c 1587 return twothirdsMD4Transform((const __u32 *)daddr, hash);
1da177e4 1588}
1da177e4
LT
1589#endif
1590
c4365c92
ACM
1591#if defined(CONFIG_IP_DCCP) || defined(CONFIG_IP_DCCP_MODULE)
1592/* Similar to secure_tcp_sequence_number but generate a 48 bit value
1593 * bit's 32-47 increase every key exchange
1594 * 0-31 hash(source, dest)
1595 */
b09b845c
AV
1596u64 secure_dccp_sequence_number(__be32 saddr, __be32 daddr,
1597 __be16 sport, __be16 dport)
c4365c92 1598{
c4365c92
ACM
1599 u64 seq;
1600 __u32 hash[4];
1601 struct keydata *keyptr = get_keyptr();
1602
b09b845c
AV
1603 hash[0] = (__force u32)saddr;
1604 hash[1] = (__force u32)daddr;
1605 hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
c4365c92
ACM
1606 hash[3] = keyptr->secret[11];
1607
1608 seq = half_md4_transform(hash, keyptr->secret);
1609 seq |= ((u64)keyptr->count) << (32 - HASH_BITS);
1610
6dd10a62 1611 seq += ktime_to_ns(ktime_get_real());
c4365c92 1612 seq &= (1ull << 48) - 1;
90b75ee5 1613
c4365c92
ACM
1614 return seq;
1615}
c4365c92
ACM
1616EXPORT_SYMBOL(secure_dccp_sequence_number);
1617#endif
1618
1da177e4
LT
1619#endif /* CONFIG_INET */
1620
1621
1622/*
1623 * Get a random word for internal kernel use only. Similar to urandom but
1624 * with the goal of minimal entropy pool depletion. As a result, the random
1625 * value is not cryptographically secure but for several uses the cost of
1626 * depleting entropy is too high
1627 */
1628unsigned int get_random_int(void)
1629{
1630 /*
1631 * Use IP's RNG. It suits our purpose perfectly: it re-keys itself
1632 * every second, from the entropy pool (and thus creates a limited
1633 * drain on it), and uses halfMD4Transform within the second. We
1634 * also mix it with jiffies and the PID:
1635 */
b09b845c 1636 return secure_ip_id((__force __be32)(current->pid + jiffies));
1da177e4
LT
1637}
1638
1639/*
1640 * randomize_range() returns a start address such that
1641 *
1642 * [...... <range> .....]
1643 * start end
1644 *
1645 * a <range> with size "len" starting at the return value is inside in the
1646 * area defined by [start, end], but is otherwise randomized.
1647 */
1648unsigned long
1649randomize_range(unsigned long start, unsigned long end, unsigned long len)
1650{
1651 unsigned long range = end - len - start;
1652
1653 if (end <= start + len)
1654 return 0;
1655 return PAGE_ALIGN(get_random_int() % range + start);
1656}