]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blame - drivers/char/random.c
mm: make register_page_bootmem_info_section() static
[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>
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
6c036527 273static int trickle_thresh __read_mostly = INPUT_POOL_WORDS * 28;
1da177e4 274
90b75ee5 275static DEFINE_PER_CPU(int, trickle_count);
1da177e4
LT
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);
9a6f70bb 373static struct fasync_struct *fasync;
1da177e4
LT
374
375#if 0
90b75ee5 376static int debug;
1da177e4 377module_param(debug, bool, 0644);
90b75ee5
MM
378#define DEBUG_ENT(fmt, arg...) do { \
379 if (debug) \
380 printk(KERN_DEBUG "random %04d %04d %04d: " \
381 fmt,\
382 input_pool.entropy_count,\
383 blocking_pool.entropy_count,\
384 nonblocking_pool.entropy_count,\
385 ## arg); } while (0)
1da177e4
LT
386#else
387#define DEBUG_ENT(fmt, arg...) do {} while (0)
388#endif
389
390/**********************************************************************
391 *
392 * OS independent entropy store. Here are the functions which handle
393 * storing entropy in an entropy pool.
394 *
395 **********************************************************************/
396
397struct entropy_store;
398struct entropy_store {
43358209 399 /* read-only data: */
1da177e4
LT
400 struct poolinfo *poolinfo;
401 __u32 *pool;
402 const char *name;
403 int limit;
404 struct entropy_store *pull;
405
406 /* read-write data: */
43358209 407 spinlock_t lock;
1da177e4
LT
408 unsigned add_ptr;
409 int entropy_count;
410 int input_rotate;
411};
412
413static __u32 input_pool_data[INPUT_POOL_WORDS];
414static __u32 blocking_pool_data[OUTPUT_POOL_WORDS];
415static __u32 nonblocking_pool_data[OUTPUT_POOL_WORDS];
416
417static struct entropy_store input_pool = {
418 .poolinfo = &poolinfo_table[0],
419 .name = "input",
420 .limit = 1,
e4d91918 421 .lock = __SPIN_LOCK_UNLOCKED(&input_pool.lock),
1da177e4
LT
422 .pool = input_pool_data
423};
424
425static struct entropy_store blocking_pool = {
426 .poolinfo = &poolinfo_table[1],
427 .name = "blocking",
428 .limit = 1,
429 .pull = &input_pool,
e4d91918 430 .lock = __SPIN_LOCK_UNLOCKED(&blocking_pool.lock),
1da177e4
LT
431 .pool = blocking_pool_data
432};
433
434static struct entropy_store nonblocking_pool = {
435 .poolinfo = &poolinfo_table[1],
436 .name = "nonblocking",
437 .pull = &input_pool,
e4d91918 438 .lock = __SPIN_LOCK_UNLOCKED(&nonblocking_pool.lock),
1da177e4
LT
439 .pool = nonblocking_pool_data
440};
441
442/*
e68e5b66 443 * This function adds bytes into the entropy "pool". It does not
1da177e4 444 * update the entropy estimate. The caller should call
adc782da 445 * credit_entropy_bits if this is appropriate.
1da177e4
LT
446 *
447 * The pool is stirred with a primitive polynomial of the appropriate
448 * degree, and then twisted. We twist by three bits at a time because
449 * it's cheap to do so and helps slightly in the expected case where
450 * the entropy is concentrated in the low-order bits.
451 */
e68e5b66
MM
452static void mix_pool_bytes_extract(struct entropy_store *r, const void *in,
453 int nbytes, __u8 out[64])
1da177e4
LT
454{
455 static __u32 const twist_table[8] = {
456 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
457 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
993ba211 458 unsigned long i, j, tap1, tap2, tap3, tap4, tap5;
feee7697 459 int input_rotate;
1da177e4 460 int wordmask = r->poolinfo->poolwords - 1;
e68e5b66 461 const char *bytes = in;
6d38b827 462 __u32 w;
1da177e4
LT
463 unsigned long flags;
464
465 /* Taps are constant, so we can load them without holding r->lock. */
466 tap1 = r->poolinfo->tap1;
467 tap2 = r->poolinfo->tap2;
468 tap3 = r->poolinfo->tap3;
469 tap4 = r->poolinfo->tap4;
470 tap5 = r->poolinfo->tap5;
1da177e4
LT
471
472 spin_lock_irqsave(&r->lock, flags);
1da177e4 473 input_rotate = r->input_rotate;
993ba211 474 i = r->add_ptr;
1da177e4 475
e68e5b66
MM
476 /* mix one byte at a time to simplify size handling and churn faster */
477 while (nbytes--) {
478 w = rol32(*bytes++, input_rotate & 31);
993ba211 479 i = (i - 1) & wordmask;
1da177e4
LT
480
481 /* XOR in the various taps */
993ba211 482 w ^= r->pool[i];
1da177e4
LT
483 w ^= r->pool[(i + tap1) & wordmask];
484 w ^= r->pool[(i + tap2) & wordmask];
485 w ^= r->pool[(i + tap3) & wordmask];
486 w ^= r->pool[(i + tap4) & wordmask];
487 w ^= r->pool[(i + tap5) & wordmask];
993ba211
MM
488
489 /* Mix the result back in with a twist */
1da177e4 490 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
feee7697
MM
491
492 /*
493 * Normally, we add 7 bits of rotation to the pool.
494 * At the beginning of the pool, add an extra 7 bits
495 * rotation, so that successive passes spread the
496 * input bits across the pool evenly.
497 */
498 input_rotate += i ? 7 : 14;
1da177e4
LT
499 }
500
501 r->input_rotate = input_rotate;
993ba211 502 r->add_ptr = i;
1da177e4 503
993ba211
MM
504 if (out)
505 for (j = 0; j < 16; j++)
e68e5b66 506 ((__u32 *)out)[j] = r->pool[(i - j) & wordmask];
1da177e4
LT
507
508 spin_unlock_irqrestore(&r->lock, flags);
509}
510
e68e5b66 511static void mix_pool_bytes(struct entropy_store *r, const void *in, int bytes)
1da177e4 512{
e68e5b66 513 mix_pool_bytes_extract(r, in, bytes, NULL);
1da177e4
LT
514}
515
516/*
517 * Credit (or debit) the entropy store with n bits of entropy
518 */
adc782da 519static void credit_entropy_bits(struct entropy_store *r, int nbits)
1da177e4
LT
520{
521 unsigned long flags;
522
adc782da
MM
523 if (!nbits)
524 return;
525
1da177e4
LT
526 spin_lock_irqsave(&r->lock, flags);
527
adc782da
MM
528 DEBUG_ENT("added %d entropy credits to %s\n", nbits, r->name);
529 r->entropy_count += nbits;
530 if (r->entropy_count < 0) {
531 DEBUG_ENT("negative entropy/overflow\n");
1da177e4 532 r->entropy_count = 0;
adc782da 533 } else if (r->entropy_count > r->poolinfo->POOLBITS)
1da177e4 534 r->entropy_count = r->poolinfo->POOLBITS;
1da177e4 535
88c730da 536 /* should we wake readers? */
9a6f70bb
JD
537 if (r == &input_pool &&
538 r->entropy_count >= random_read_wakeup_thresh) {
88c730da 539 wake_up_interruptible(&random_read_wait);
9a6f70bb
JD
540 kill_fasync(&fasync, SIGIO, POLL_IN);
541 }
88c730da 542
1da177e4
LT
543 spin_unlock_irqrestore(&r->lock, flags);
544}
545
546/*********************************************************************
547 *
548 * Entropy input management
549 *
550 *********************************************************************/
551
552/* There is one of these per entropy source */
553struct timer_rand_state {
554 cycles_t last_time;
90b75ee5 555 long last_delta, last_delta2;
1da177e4
LT
556 unsigned dont_count_entropy:1;
557};
558
559static struct timer_rand_state input_timer_state;
560static struct timer_rand_state *irq_timer_state[NR_IRQS];
561
562/*
563 * This function adds entropy to the entropy "pool" by using timing
564 * delays. It uses the timer_rand_state structure to make an estimate
565 * of how many bits of entropy this call has added to the pool.
566 *
567 * The number "num" is also added to the pool - it should somehow describe
568 * the type of event which just happened. This is currently 0-255 for
569 * keyboard scan codes, and 256 upwards for interrupts.
570 *
571 */
572static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
573{
574 struct {
575 cycles_t cycles;
576 long jiffies;
577 unsigned num;
578 } sample;
579 long delta, delta2, delta3;
580
581 preempt_disable();
582 /* if over the trickle threshold, use only 1 in 4096 samples */
583 if (input_pool.entropy_count > trickle_thresh &&
584 (__get_cpu_var(trickle_count)++ & 0xfff))
585 goto out;
586
587 sample.jiffies = jiffies;
588 sample.cycles = get_cycles();
589 sample.num = num;
e68e5b66 590 mix_pool_bytes(&input_pool, &sample, sizeof(sample));
1da177e4
LT
591
592 /*
593 * Calculate number of bits of randomness we probably added.
594 * We take into account the first, second and third-order deltas
595 * in order to make our estimate.
596 */
597
598 if (!state->dont_count_entropy) {
599 delta = sample.jiffies - state->last_time;
600 state->last_time = sample.jiffies;
601
602 delta2 = delta - state->last_delta;
603 state->last_delta = delta;
604
605 delta3 = delta2 - state->last_delta2;
606 state->last_delta2 = delta2;
607
608 if (delta < 0)
609 delta = -delta;
610 if (delta2 < 0)
611 delta2 = -delta2;
612 if (delta3 < 0)
613 delta3 = -delta3;
614 if (delta > delta2)
615 delta = delta2;
616 if (delta > delta3)
617 delta = delta3;
618
619 /*
620 * delta is now minimum absolute delta.
621 * Round down by 1 bit on general principles,
622 * and limit entropy entimate to 12 bits.
623 */
adc782da
MM
624 credit_entropy_bits(&input_pool,
625 min_t(int, fls(delta>>1), 11));
1da177e4 626 }
1da177e4
LT
627out:
628 preempt_enable();
629}
630
d251575a 631void add_input_randomness(unsigned int type, unsigned int code,
1da177e4
LT
632 unsigned int value)
633{
634 static unsigned char last_value;
635
636 /* ignore autorepeat and the like */
637 if (value == last_value)
638 return;
639
640 DEBUG_ENT("input event\n");
641 last_value = value;
642 add_timer_randomness(&input_timer_state,
643 (type << 4) ^ code ^ (code >> 4) ^ value);
644}
80fc9f53 645EXPORT_SYMBOL_GPL(add_input_randomness);
1da177e4
LT
646
647void add_interrupt_randomness(int irq)
648{
c80544dc 649 if (irq >= NR_IRQS || irq_timer_state[irq] == NULL)
1da177e4
LT
650 return;
651
652 DEBUG_ENT("irq event %d\n", irq);
653 add_timer_randomness(irq_timer_state[irq], 0x100 + irq);
654}
655
9361401e 656#ifdef CONFIG_BLOCK
1da177e4
LT
657void add_disk_randomness(struct gendisk *disk)
658{
659 if (!disk || !disk->random)
660 return;
661 /* first major is 1, so we get >= 0x200 here */
662 DEBUG_ENT("disk event %d:%d\n", disk->major, disk->first_minor);
663
664 add_timer_randomness(disk->random,
665 0x100 + MKDEV(disk->major, disk->first_minor));
666}
9361401e 667#endif
1da177e4
LT
668
669#define EXTRACT_SIZE 10
670
671/*********************************************************************
672 *
673 * Entropy extraction routines
674 *
675 *********************************************************************/
676
90b75ee5 677static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1da177e4
LT
678 size_t nbytes, int min, int rsvd);
679
680/*
681 * This utility inline function is responsible for transfering entropy
682 * from the primary pool to the secondary extraction pool. We make
683 * sure we pull enough for a 'catastrophic reseed'.
684 */
685static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
686{
687 __u32 tmp[OUTPUT_POOL_WORDS];
688
689 if (r->pull && r->entropy_count < nbytes * 8 &&
690 r->entropy_count < r->poolinfo->POOLBITS) {
5a021e9f 691 /* If we're limited, always leave two wakeup worth's BITS */
1da177e4 692 int rsvd = r->limit ? 0 : random_read_wakeup_thresh/4;
5a021e9f
MM
693 int bytes = nbytes;
694
695 /* pull at least as many as BYTES as wakeup BITS */
696 bytes = max_t(int, bytes, random_read_wakeup_thresh / 8);
697 /* but never more than the buffer size */
698 bytes = min_t(int, bytes, sizeof(tmp));
1da177e4
LT
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
90b75ee5
MM
704 bytes = extract_entropy(r->pull, tmp, bytes,
705 random_read_wakeup_thresh / 8, rsvd);
e68e5b66 706 mix_pool_bytes(r, tmp, bytes);
adc782da 707 credit_entropy_bits(r, bytes*8);
1da177e4
LT
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
90b75ee5 744 if (r->entropy_count / 8 >= nbytes + reserved)
1da177e4
LT
745 r->entropy_count -= nbytes*8;
746 else
747 r->entropy_count = reserved;
748
9a6f70bb 749 if (r->entropy_count < random_write_wakeup_thresh) {
1da177e4 750 wake_up_interruptible(&random_write_wait);
9a6f70bb
JD
751 kill_fasync(&fasync, SIGIO, POLL_OUT);
752 }
1da177e4
LT
753 }
754
755 DEBUG_ENT("debiting %d entropy credits from %s%s\n",
756 nbytes * 8, r->name, r->limit ? "" : " (unlimited)");
757
758 spin_unlock_irqrestore(&r->lock, flags);
759
760 return nbytes;
761}
762
763static void extract_buf(struct entropy_store *r, __u8 *out)
764{
602b6aee 765 int i;
e68e5b66
MM
766 __u32 hash[5], workspace[SHA_WORKSPACE_WORDS];
767 __u8 extract[64];
1da177e4 768
1c0ad3d4 769 /* Generate a hash across the pool, 16 words (512 bits) at a time */
ffd8d3fa 770 sha_init(hash);
1c0ad3d4
MM
771 for (i = 0; i < r->poolinfo->poolwords; i += 16)
772 sha_transform(hash, (__u8 *)(r->pool + i), workspace);
773
1da177e4 774 /*
1c0ad3d4
MM
775 * We mix the hash back into the pool to prevent backtracking
776 * attacks (where the attacker knows the state of the pool
777 * plus the current outputs, and attempts to find previous
778 * ouputs), unless the hash function can be inverted. By
779 * mixing at least a SHA1 worth of hash data back, we make
780 * brute-forcing the feedback as hard as brute-forcing the
781 * hash.
1da177e4 782 */
e68e5b66 783 mix_pool_bytes_extract(r, hash, sizeof(hash), extract);
1da177e4
LT
784
785 /*
1c0ad3d4
MM
786 * To avoid duplicates, we atomically extract a portion of the
787 * pool while mixing, and hash one final time.
1da177e4 788 */
e68e5b66 789 sha_transform(hash, extract, workspace);
ffd8d3fa
MM
790 memset(extract, 0, sizeof(extract));
791 memset(workspace, 0, sizeof(workspace));
1da177e4
LT
792
793 /*
1c0ad3d4
MM
794 * In case the hash function has some recognizable output
795 * pattern, we fold it in half. Thus, we always feed back
796 * twice as much data as we output.
1da177e4 797 */
ffd8d3fa
MM
798 hash[0] ^= hash[3];
799 hash[1] ^= hash[4];
800 hash[2] ^= rol32(hash[2], 16);
801 memcpy(out, hash, EXTRACT_SIZE);
802 memset(hash, 0, sizeof(hash));
1da177e4
LT
803}
804
90b75ee5 805static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1da177e4
LT
806 size_t nbytes, int min, int reserved)
807{
808 ssize_t ret = 0, i;
809 __u8 tmp[EXTRACT_SIZE];
810
811 xfer_secondary_pool(r, nbytes);
812 nbytes = account(r, nbytes, min, reserved);
813
814 while (nbytes) {
815 extract_buf(r, tmp);
816 i = min_t(int, nbytes, EXTRACT_SIZE);
817 memcpy(buf, tmp, i);
818 nbytes -= i;
819 buf += i;
820 ret += i;
821 }
822
823 /* Wipe data just returned from memory */
824 memset(tmp, 0, sizeof(tmp));
825
826 return ret;
827}
828
829static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
830 size_t nbytes)
831{
832 ssize_t ret = 0, i;
833 __u8 tmp[EXTRACT_SIZE];
834
835 xfer_secondary_pool(r, nbytes);
836 nbytes = account(r, nbytes, 0, 0);
837
838 while (nbytes) {
839 if (need_resched()) {
840 if (signal_pending(current)) {
841 if (ret == 0)
842 ret = -ERESTARTSYS;
843 break;
844 }
845 schedule();
846 }
847
848 extract_buf(r, tmp);
849 i = min_t(int, nbytes, EXTRACT_SIZE);
850 if (copy_to_user(buf, tmp, i)) {
851 ret = -EFAULT;
852 break;
853 }
854
855 nbytes -= i;
856 buf += i;
857 ret += i;
858 }
859
860 /* Wipe data just returned from memory */
861 memset(tmp, 0, sizeof(tmp));
862
863 return ret;
864}
865
866/*
867 * This function is the exported kernel interface. It returns some
868 * number of good random numbers, suitable for seeding TCP sequence
869 * numbers, etc.
870 */
871void get_random_bytes(void *buf, int nbytes)
872{
873 extract_entropy(&nonblocking_pool, buf, nbytes, 0, 0);
874}
1da177e4
LT
875EXPORT_SYMBOL(get_random_bytes);
876
877/*
878 * init_std_data - initialize pool with system data
879 *
880 * @r: pool to initialize
881 *
882 * This function clears the pool's entropy count and mixes some system
883 * data into the pool to prepare it for use. The pool is not cleared
884 * as that can only decrease the entropy in the pool.
885 */
886static void init_std_data(struct entropy_store *r)
887{
f8595815 888 ktime_t now;
1da177e4
LT
889 unsigned long flags;
890
891 spin_lock_irqsave(&r->lock, flags);
892 r->entropy_count = 0;
893 spin_unlock_irqrestore(&r->lock, flags);
894
f8595815 895 now = ktime_get_real();
e68e5b66
MM
896 mix_pool_bytes(r, &now, sizeof(now));
897 mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1da177e4
LT
898}
899
53c3f63e 900static int rand_initialize(void)
1da177e4
LT
901{
902 init_std_data(&input_pool);
903 init_std_data(&blocking_pool);
904 init_std_data(&nonblocking_pool);
905 return 0;
906}
907module_init(rand_initialize);
908
909void rand_initialize_irq(int irq)
910{
911 struct timer_rand_state *state;
912
913 if (irq >= NR_IRQS || irq_timer_state[irq])
914 return;
915
916 /*
f8595815 917 * If kzalloc returns null, we just won't use that entropy
1da177e4
LT
918 * source.
919 */
f8595815
ED
920 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
921 if (state)
1da177e4 922 irq_timer_state[irq] = state;
1da177e4
LT
923}
924
9361401e 925#ifdef CONFIG_BLOCK
1da177e4
LT
926void rand_initialize_disk(struct gendisk *disk)
927{
928 struct timer_rand_state *state;
929
930 /*
f8595815 931 * If kzalloc returns null, we just won't use that entropy
1da177e4
LT
932 * source.
933 */
f8595815
ED
934 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
935 if (state)
1da177e4 936 disk->random = state;
1da177e4 937}
9361401e 938#endif
1da177e4
LT
939
940static ssize_t
90b75ee5 941random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1da177e4
LT
942{
943 ssize_t n, retval = 0, count = 0;
944
945 if (nbytes == 0)
946 return 0;
947
948 while (nbytes > 0) {
949 n = nbytes;
950 if (n > SEC_XFER_SIZE)
951 n = SEC_XFER_SIZE;
952
953 DEBUG_ENT("reading %d bits\n", n*8);
954
955 n = extract_entropy_user(&blocking_pool, buf, n);
956
957 DEBUG_ENT("read got %d bits (%d still needed)\n",
958 n*8, (nbytes-n)*8);
959
960 if (n == 0) {
961 if (file->f_flags & O_NONBLOCK) {
962 retval = -EAGAIN;
963 break;
964 }
965
966 DEBUG_ENT("sleeping?\n");
967
968 wait_event_interruptible(random_read_wait,
969 input_pool.entropy_count >=
970 random_read_wakeup_thresh);
971
972 DEBUG_ENT("awake\n");
973
974 if (signal_pending(current)) {
975 retval = -ERESTARTSYS;
976 break;
977 }
978
979 continue;
980 }
981
982 if (n < 0) {
983 retval = n;
984 break;
985 }
986 count += n;
987 buf += n;
988 nbytes -= n;
989 break; /* This break makes the device work */
990 /* like a named pipe */
991 }
992
993 /*
994 * If we gave the user some bytes, update the access time.
995 */
996 if (count)
997 file_accessed(file);
998
999 return (count ? count : retval);
1000}
1001
1002static ssize_t
90b75ee5 1003urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1da177e4
LT
1004{
1005 return extract_entropy_user(&nonblocking_pool, buf, nbytes);
1006}
1007
1008static unsigned int
1009random_poll(struct file *file, poll_table * wait)
1010{
1011 unsigned int mask;
1012
1013 poll_wait(file, &random_read_wait, wait);
1014 poll_wait(file, &random_write_wait, wait);
1015 mask = 0;
1016 if (input_pool.entropy_count >= random_read_wakeup_thresh)
1017 mask |= POLLIN | POLLRDNORM;
1018 if (input_pool.entropy_count < random_write_wakeup_thresh)
1019 mask |= POLLOUT | POLLWRNORM;
1020 return mask;
1021}
1022
7f397dcd
MM
1023static int
1024write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1da177e4 1025{
1da177e4
LT
1026 size_t bytes;
1027 __u32 buf[16];
1028 const char __user *p = buffer;
1da177e4 1029
7f397dcd
MM
1030 while (count > 0) {
1031 bytes = min(count, sizeof(buf));
1032 if (copy_from_user(&buf, p, bytes))
1033 return -EFAULT;
1da177e4 1034
7f397dcd 1035 count -= bytes;
1da177e4
LT
1036 p += bytes;
1037
e68e5b66 1038 mix_pool_bytes(r, buf, bytes);
91f3f1e3 1039 cond_resched();
1da177e4 1040 }
7f397dcd
MM
1041
1042 return 0;
1043}
1044
90b75ee5
MM
1045static ssize_t random_write(struct file *file, const char __user *buffer,
1046 size_t count, loff_t *ppos)
7f397dcd
MM
1047{
1048 size_t ret;
1049 struct inode *inode = file->f_path.dentry->d_inode;
1050
1051 ret = write_pool(&blocking_pool, buffer, count);
1052 if (ret)
1053 return ret;
1054 ret = write_pool(&nonblocking_pool, buffer, count);
1055 if (ret)
1056 return ret;
1057
1058 inode->i_mtime = current_fs_time(inode->i_sb);
1059 mark_inode_dirty(inode);
1060 return (ssize_t)count;
1da177e4
LT
1061}
1062
43ae4860 1063static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1da177e4
LT
1064{
1065 int size, ent_count;
1066 int __user *p = (int __user *)arg;
1067 int retval;
1068
1069 switch (cmd) {
1070 case RNDGETENTCNT:
43ae4860
MM
1071 /* inherently racy, no point locking */
1072 if (put_user(input_pool.entropy_count, p))
1da177e4
LT
1073 return -EFAULT;
1074 return 0;
1075 case RNDADDTOENTCNT:
1076 if (!capable(CAP_SYS_ADMIN))
1077 return -EPERM;
1078 if (get_user(ent_count, p))
1079 return -EFAULT;
adc782da 1080 credit_entropy_bits(&input_pool, ent_count);
1da177e4
LT
1081 return 0;
1082 case RNDADDENTROPY:
1083 if (!capable(CAP_SYS_ADMIN))
1084 return -EPERM;
1085 if (get_user(ent_count, p++))
1086 return -EFAULT;
1087 if (ent_count < 0)
1088 return -EINVAL;
1089 if (get_user(size, p++))
1090 return -EFAULT;
7f397dcd
MM
1091 retval = write_pool(&input_pool, (const char __user *)p,
1092 size);
1da177e4
LT
1093 if (retval < 0)
1094 return retval;
adc782da 1095 credit_entropy_bits(&input_pool, ent_count);
1da177e4
LT
1096 return 0;
1097 case RNDZAPENTCNT:
1098 case RNDCLEARPOOL:
1099 /* Clear the entropy pool counters. */
1100 if (!capable(CAP_SYS_ADMIN))
1101 return -EPERM;
53c3f63e 1102 rand_initialize();
1da177e4
LT
1103 return 0;
1104 default:
1105 return -EINVAL;
1106 }
1107}
1108
9a6f70bb
JD
1109static int random_fasync(int fd, struct file *filp, int on)
1110{
1111 return fasync_helper(fd, filp, on, &fasync);
1112}
1113
1114static int random_release(struct inode *inode, struct file *filp)
1115{
1116 return fasync_helper(-1, filp, 0, &fasync);
1117}
1118
2b8693c0 1119const struct file_operations random_fops = {
1da177e4
LT
1120 .read = random_read,
1121 .write = random_write,
1122 .poll = random_poll,
43ae4860 1123 .unlocked_ioctl = random_ioctl,
9a6f70bb
JD
1124 .fasync = random_fasync,
1125 .release = random_release,
1da177e4
LT
1126};
1127
2b8693c0 1128const struct file_operations urandom_fops = {
1da177e4
LT
1129 .read = urandom_read,
1130 .write = random_write,
43ae4860 1131 .unlocked_ioctl = random_ioctl,
9a6f70bb
JD
1132 .fasync = random_fasync,
1133 .release = random_release,
1da177e4
LT
1134};
1135
1136/***************************************************************
1137 * Random UUID interface
1138 *
1139 * Used here for a Boot ID, but can be useful for other kernel
1140 * drivers.
1141 ***************************************************************/
1142
1143/*
1144 * Generate random UUID
1145 */
1146void generate_random_uuid(unsigned char uuid_out[16])
1147{
1148 get_random_bytes(uuid_out, 16);
1149 /* Set UUID version to 4 --- truely random generation */
1150 uuid_out[6] = (uuid_out[6] & 0x0F) | 0x40;
1151 /* Set the UUID variant to DCE */
1152 uuid_out[8] = (uuid_out[8] & 0x3F) | 0x80;
1153}
1da177e4
LT
1154EXPORT_SYMBOL(generate_random_uuid);
1155
1156/********************************************************************
1157 *
1158 * Sysctl interface
1159 *
1160 ********************************************************************/
1161
1162#ifdef CONFIG_SYSCTL
1163
1164#include <linux/sysctl.h>
1165
1166static int min_read_thresh = 8, min_write_thresh;
1167static int max_read_thresh = INPUT_POOL_WORDS * 32;
1168static int max_write_thresh = INPUT_POOL_WORDS * 32;
1169static char sysctl_bootid[16];
1170
1171/*
1172 * These functions is used to return both the bootid UUID, and random
1173 * UUID. The difference is in whether table->data is NULL; if it is,
1174 * then a new UUID is generated and returned to the user.
1175 *
1176 * If the user accesses this via the proc interface, it will be returned
1177 * as an ASCII string in the standard UUID format. If accesses via the
1178 * sysctl system call, it is returned as 16 bytes of binary data.
1179 */
1180static int proc_do_uuid(ctl_table *table, int write, struct file *filp,
1181 void __user *buffer, size_t *lenp, loff_t *ppos)
1182{
1183 ctl_table fake_table;
1184 unsigned char buf[64], tmp_uuid[16], *uuid;
1185
1186 uuid = table->data;
1187 if (!uuid) {
1188 uuid = tmp_uuid;
1189 uuid[8] = 0;
1190 }
1191 if (uuid[8] == 0)
1192 generate_random_uuid(uuid);
1193
1194 sprintf(buf, "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-"
1195 "%02x%02x%02x%02x%02x%02x",
1196 uuid[0], uuid[1], uuid[2], uuid[3],
1197 uuid[4], uuid[5], uuid[6], uuid[7],
1198 uuid[8], uuid[9], uuid[10], uuid[11],
1199 uuid[12], uuid[13], uuid[14], uuid[15]);
1200 fake_table.data = buf;
1201 fake_table.maxlen = sizeof(buf);
1202
1203 return proc_dostring(&fake_table, write, filp, buffer, lenp, ppos);
1204}
1205
1206static int uuid_strategy(ctl_table *table, int __user *name, int nlen,
1207 void __user *oldval, size_t __user *oldlenp,
1f29bcd7 1208 void __user *newval, size_t newlen)
1da177e4
LT
1209{
1210 unsigned char tmp_uuid[16], *uuid;
1211 unsigned int len;
1212
1213 if (!oldval || !oldlenp)
1214 return 1;
1215
1216 uuid = table->data;
1217 if (!uuid) {
1218 uuid = tmp_uuid;
1219 uuid[8] = 0;
1220 }
1221 if (uuid[8] == 0)
1222 generate_random_uuid(uuid);
1223
1224 if (get_user(len, oldlenp))
1225 return -EFAULT;
1226 if (len) {
1227 if (len > 16)
1228 len = 16;
1229 if (copy_to_user(oldval, uuid, len) ||
1230 put_user(len, oldlenp))
1231 return -EFAULT;
1232 }
1233 return 1;
1234}
1235
1236static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
1237ctl_table random_table[] = {
1238 {
1239 .ctl_name = RANDOM_POOLSIZE,
1240 .procname = "poolsize",
1241 .data = &sysctl_poolsize,
1242 .maxlen = sizeof(int),
1243 .mode = 0444,
1244 .proc_handler = &proc_dointvec,
1245 },
1246 {
1247 .ctl_name = RANDOM_ENTROPY_COUNT,
1248 .procname = "entropy_avail",
1249 .maxlen = sizeof(int),
1250 .mode = 0444,
1251 .proc_handler = &proc_dointvec,
1252 .data = &input_pool.entropy_count,
1253 },
1254 {
1255 .ctl_name = RANDOM_READ_THRESH,
1256 .procname = "read_wakeup_threshold",
1257 .data = &random_read_wakeup_thresh,
1258 .maxlen = sizeof(int),
1259 .mode = 0644,
1260 .proc_handler = &proc_dointvec_minmax,
1261 .strategy = &sysctl_intvec,
1262 .extra1 = &min_read_thresh,
1263 .extra2 = &max_read_thresh,
1264 },
1265 {
1266 .ctl_name = RANDOM_WRITE_THRESH,
1267 .procname = "write_wakeup_threshold",
1268 .data = &random_write_wakeup_thresh,
1269 .maxlen = sizeof(int),
1270 .mode = 0644,
1271 .proc_handler = &proc_dointvec_minmax,
1272 .strategy = &sysctl_intvec,
1273 .extra1 = &min_write_thresh,
1274 .extra2 = &max_write_thresh,
1275 },
1276 {
1277 .ctl_name = RANDOM_BOOT_ID,
1278 .procname = "boot_id",
1279 .data = &sysctl_bootid,
1280 .maxlen = 16,
1281 .mode = 0444,
1282 .proc_handler = &proc_do_uuid,
1283 .strategy = &uuid_strategy,
1284 },
1285 {
1286 .ctl_name = RANDOM_UUID,
1287 .procname = "uuid",
1288 .maxlen = 16,
1289 .mode = 0444,
1290 .proc_handler = &proc_do_uuid,
1291 .strategy = &uuid_strategy,
1292 },
1293 { .ctl_name = 0 }
1294};
1295#endif /* CONFIG_SYSCTL */
1296
1297/********************************************************************
1298 *
1299 * Random funtions for networking
1300 *
1301 ********************************************************************/
1302
1303/*
1304 * TCP initial sequence number picking. This uses the random number
1305 * generator to pick an initial secret value. This value is hashed
1306 * along with the TCP endpoint information to provide a unique
1307 * starting point for each pair of TCP endpoints. This defeats
1308 * attacks which rely on guessing the initial TCP sequence number.
1309 * This algorithm was suggested by Steve Bellovin.
1310 *
1311 * Using a very strong hash was taking an appreciable amount of the total
1312 * TCP connection establishment time, so this is a weaker hash,
1313 * compensated for by changing the secret periodically.
1314 */
1315
1316/* F, G and H are basic MD4 functions: selection, majority, parity */
1317#define F(x, y, z) ((z) ^ ((x) & ((y) ^ (z))))
1318#define G(x, y, z) (((x) & (y)) + (((x) ^ (y)) & (z)))
1319#define H(x, y, z) ((x) ^ (y) ^ (z))
1320
1321/*
1322 * The generic round function. The application is so specific that
1323 * we don't bother protecting all the arguments with parens, as is generally
1324 * good macro practice, in favor of extra legibility.
1325 * Rotation is separate from addition to prevent recomputation
1326 */
1327#define ROUND(f, a, b, c, d, x, s) \
1328 (a += f(b, c, d) + x, a = (a << s) | (a >> (32 - s)))
1329#define K1 0
1330#define K2 013240474631UL
1331#define K3 015666365641UL
1332
1333#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1334
90b75ee5 1335static __u32 twothirdsMD4Transform(__u32 const buf[4], __u32 const in[12])
1da177e4
LT
1336{
1337 __u32 a = buf[0], b = buf[1], c = buf[2], d = buf[3];
1338
1339 /* Round 1 */
1340 ROUND(F, a, b, c, d, in[ 0] + K1, 3);
1341 ROUND(F, d, a, b, c, in[ 1] + K1, 7);
1342 ROUND(F, c, d, a, b, in[ 2] + K1, 11);
1343 ROUND(F, b, c, d, a, in[ 3] + K1, 19);
1344 ROUND(F, a, b, c, d, in[ 4] + K1, 3);
1345 ROUND(F, d, a, b, c, in[ 5] + K1, 7);
1346 ROUND(F, c, d, a, b, in[ 6] + K1, 11);
1347 ROUND(F, b, c, d, a, in[ 7] + K1, 19);
1348 ROUND(F, a, b, c, d, in[ 8] + K1, 3);
1349 ROUND(F, d, a, b, c, in[ 9] + K1, 7);
1350 ROUND(F, c, d, a, b, in[10] + K1, 11);
1351 ROUND(F, b, c, d, a, in[11] + K1, 19);
1352
1353 /* Round 2 */
1354 ROUND(G, a, b, c, d, in[ 1] + K2, 3);
1355 ROUND(G, d, a, b, c, in[ 3] + K2, 5);
1356 ROUND(G, c, d, a, b, in[ 5] + K2, 9);
1357 ROUND(G, b, c, d, a, in[ 7] + K2, 13);
1358 ROUND(G, a, b, c, d, in[ 9] + K2, 3);
1359 ROUND(G, d, a, b, c, in[11] + K2, 5);
1360 ROUND(G, c, d, a, b, in[ 0] + K2, 9);
1361 ROUND(G, b, c, d, a, in[ 2] + K2, 13);
1362 ROUND(G, a, b, c, d, in[ 4] + K2, 3);
1363 ROUND(G, d, a, b, c, in[ 6] + K2, 5);
1364 ROUND(G, c, d, a, b, in[ 8] + K2, 9);
1365 ROUND(G, b, c, d, a, in[10] + K2, 13);
1366
1367 /* Round 3 */
1368 ROUND(H, a, b, c, d, in[ 3] + K3, 3);
1369 ROUND(H, d, a, b, c, in[ 7] + K3, 9);
1370 ROUND(H, c, d, a, b, in[11] + K3, 11);
1371 ROUND(H, b, c, d, a, in[ 2] + K3, 15);
1372 ROUND(H, a, b, c, d, in[ 6] + K3, 3);
1373 ROUND(H, d, a, b, c, in[10] + K3, 9);
1374 ROUND(H, c, d, a, b, in[ 1] + K3, 11);
1375 ROUND(H, b, c, d, a, in[ 5] + K3, 15);
1376 ROUND(H, a, b, c, d, in[ 9] + K3, 3);
1377 ROUND(H, d, a, b, c, in[ 0] + K3, 9);
1378 ROUND(H, c, d, a, b, in[ 4] + K3, 11);
1379 ROUND(H, b, c, d, a, in[ 8] + K3, 15);
1380
1381 return buf[1] + b; /* "most hashed" word */
1382 /* Alternative: return sum of all words? */
1383}
1384#endif
1385
1386#undef ROUND
1387#undef F
1388#undef G
1389#undef H
1390#undef K1
1391#undef K2
1392#undef K3
1393
1394/* This should not be decreased so low that ISNs wrap too fast. */
1395#define REKEY_INTERVAL (300 * HZ)
1396/*
1397 * Bit layout of the tcp sequence numbers (before adding current time):
1398 * bit 24-31: increased after every key exchange
1399 * bit 0-23: hash(source,dest)
1400 *
1401 * The implementation is similar to the algorithm described
1402 * in the Appendix of RFC 1185, except that
1403 * - it uses a 1 MHz clock instead of a 250 kHz clock
1404 * - it performs a rekey every 5 minutes, which is equivalent
1405 * to a (source,dest) tulple dependent forward jump of the
1406 * clock by 0..2^(HASH_BITS+1)
1407 *
1408 * Thus the average ISN wraparound time is 68 minutes instead of
1409 * 4.55 hours.
1410 *
1411 * SMP cleanup and lock avoidance with poor man's RCU.
1412 * Manfred Spraul <manfred@colorfullife.com>
1413 *
1414 */
1415#define COUNT_BITS 8
1416#define COUNT_MASK ((1 << COUNT_BITS) - 1)
1417#define HASH_BITS 24
1418#define HASH_MASK ((1 << HASH_BITS) - 1)
1419
1420static struct keydata {
1421 __u32 count; /* already shifted to the final position */
1422 __u32 secret[12];
1423} ____cacheline_aligned ip_keydata[2];
1424
1425static unsigned int ip_cnt;
1426
65f27f38 1427static void rekey_seq_generator(struct work_struct *work);
1da177e4 1428
65f27f38 1429static DECLARE_DELAYED_WORK(rekey_work, rekey_seq_generator);
1da177e4
LT
1430
1431/*
1432 * Lock avoidance:
1433 * The ISN generation runs lockless - it's just a hash over random data.
1434 * State changes happen every 5 minutes when the random key is replaced.
1435 * Synchronization is performed by having two copies of the hash function
1436 * state and rekey_seq_generator always updates the inactive copy.
1437 * The copy is then activated by updating ip_cnt.
1438 * The implementation breaks down if someone blocks the thread
1439 * that processes SYN requests for more than 5 minutes. Should never
1440 * happen, and even if that happens only a not perfectly compliant
1441 * ISN is generated, nothing fatal.
1442 */
65f27f38 1443static void rekey_seq_generator(struct work_struct *work)
1da177e4
LT
1444{
1445 struct keydata *keyptr = &ip_keydata[1 ^ (ip_cnt & 1)];
1446
1447 get_random_bytes(keyptr->secret, sizeof(keyptr->secret));
1448 keyptr->count = (ip_cnt & COUNT_MASK) << HASH_BITS;
1449 smp_wmb();
1450 ip_cnt++;
1451 schedule_delayed_work(&rekey_work, REKEY_INTERVAL);
1452}
1453
1454static inline struct keydata *get_keyptr(void)
1455{
1456 struct keydata *keyptr = &ip_keydata[ip_cnt & 1];
1457
1458 smp_rmb();
1459
1460 return keyptr;
1461}
1462
1463static __init int seqgen_init(void)
1464{
1465 rekey_seq_generator(NULL);
1466 return 0;
1467}
1468late_initcall(seqgen_init);
1469
1470#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
b09b845c
AV
1471__u32 secure_tcpv6_sequence_number(__be32 *saddr, __be32 *daddr,
1472 __be16 sport, __be16 dport)
1da177e4 1473{
1da177e4
LT
1474 __u32 seq;
1475 __u32 hash[12];
1476 struct keydata *keyptr = get_keyptr();
1477
1478 /* The procedure is the same as for IPv4, but addresses are longer.
1479 * Thus we must use twothirdsMD4Transform.
1480 */
1481
1482 memcpy(hash, saddr, 16);
90b75ee5
MM
1483 hash[4] = ((__force u16)sport << 16) + (__force u16)dport;
1484 memcpy(&hash[5], keyptr->secret, sizeof(__u32) * 7);
1da177e4 1485
b09b845c 1486 seq = twothirdsMD4Transform((const __u32 *)daddr, hash) & HASH_MASK;
1da177e4
LT
1487 seq += keyptr->count;
1488
6dd10a62 1489 seq += ktime_to_ns(ktime_get_real());
1da177e4
LT
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 */
b09b845c 1499__u32 secure_ip_id(__be32 daddr)
1da177e4
LT
1500{
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 */
b09b845c 1511 hash[0] = (__force __u32)daddr;
1da177e4
LT
1512 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
b09b845c
AV
1521__u32 secure_tcp_sequence_number(__be32 saddr, __be32 daddr,
1522 __be16 sport, __be16 dport)
1da177e4 1523{
1da177e4
LT
1524 __u32 seq;
1525 __u32 hash[4];
1526 struct keydata *keyptr = get_keyptr();
1527
1528 /*
1529 * Pick a unique starting offset for each TCP connection endpoints
1530 * (saddr, daddr, sport, dport).
1531 * Note that the words are placed into the starting vector, which is
1532 * then mixed with a partial MD4 over random data.
1533 */
90b75ee5
MM
1534 hash[0] = (__force u32)saddr;
1535 hash[1] = (__force u32)daddr;
1536 hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
1537 hash[3] = keyptr->secret[11];
1da177e4
LT
1538
1539 seq = half_md4_transform(hash, keyptr->secret) & HASH_MASK;
1540 seq += keyptr->count;
1541 /*
1542 * As close as possible to RFC 793, which
1543 * suggests using a 250 kHz clock.
1544 * Further reading shows this assumes 2 Mb/s networks.
9b42c336
ED
1545 * For 10 Mb/s Ethernet, a 1 MHz clock is appropriate.
1546 * For 10 Gb/s Ethernet, a 1 GHz clock should be ok, but
1547 * we also need to limit the resolution so that the u32 seq
1548 * overlaps less than one time per MSL (2 minutes).
1549 * Choosing a clock of 64 ns period is OK. (period of 274 s)
1da177e4 1550 */
6dd10a62 1551 seq += ktime_to_ns(ktime_get_real()) >> 6;
90b75ee5 1552
1da177e4
LT
1553 return seq;
1554}
1555
a7f5e7f1 1556/* Generate secure starting point for ephemeral IPV4 transport port search */
b09b845c 1557u32 secure_ipv4_port_ephemeral(__be32 saddr, __be32 daddr, __be16 dport)
1da177e4
LT
1558{
1559 struct keydata *keyptr = get_keyptr();
1560 u32 hash[4];
1561
1562 /*
1563 * Pick a unique starting offset for each ephemeral port search
1564 * (saddr, daddr, dport) and 48bits of random data.
1565 */
b09b845c
AV
1566 hash[0] = (__force u32)saddr;
1567 hash[1] = (__force u32)daddr;
1568 hash[2] = (__force u32)dport ^ keyptr->secret[10];
1da177e4
LT
1569 hash[3] = keyptr->secret[11];
1570
1571 return half_md4_transform(hash, keyptr->secret);
1572}
1573
1574#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
90b75ee5
MM
1575u32 secure_ipv6_port_ephemeral(const __be32 *saddr, const __be32 *daddr,
1576 __be16 dport)
1da177e4
LT
1577{
1578 struct keydata *keyptr = get_keyptr();
1579 u32 hash[12];
1580
1581 memcpy(hash, saddr, 16);
b09b845c 1582 hash[4] = (__force u32)dport;
90b75ee5 1583 memcpy(&hash[5], keyptr->secret, sizeof(__u32) * 7);
1da177e4 1584
b09b845c 1585 return twothirdsMD4Transform((const __u32 *)daddr, hash);
1da177e4 1586}
1da177e4
LT
1587#endif
1588
c4365c92
ACM
1589#if defined(CONFIG_IP_DCCP) || defined(CONFIG_IP_DCCP_MODULE)
1590/* Similar to secure_tcp_sequence_number but generate a 48 bit value
1591 * bit's 32-47 increase every key exchange
1592 * 0-31 hash(source, dest)
1593 */
b09b845c
AV
1594u64 secure_dccp_sequence_number(__be32 saddr, __be32 daddr,
1595 __be16 sport, __be16 dport)
c4365c92 1596{
c4365c92
ACM
1597 u64 seq;
1598 __u32 hash[4];
1599 struct keydata *keyptr = get_keyptr();
1600
b09b845c
AV
1601 hash[0] = (__force u32)saddr;
1602 hash[1] = (__force u32)daddr;
1603 hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
c4365c92
ACM
1604 hash[3] = keyptr->secret[11];
1605
1606 seq = half_md4_transform(hash, keyptr->secret);
1607 seq |= ((u64)keyptr->count) << (32 - HASH_BITS);
1608
6dd10a62 1609 seq += ktime_to_ns(ktime_get_real());
c4365c92 1610 seq &= (1ull << 48) - 1;
90b75ee5 1611
c4365c92
ACM
1612 return seq;
1613}
c4365c92
ACM
1614EXPORT_SYMBOL(secure_dccp_sequence_number);
1615#endif
1616
1da177e4
LT
1617#endif /* CONFIG_INET */
1618
1619
1620/*
1621 * Get a random word for internal kernel use only. Similar to urandom but
1622 * with the goal of minimal entropy pool depletion. As a result, the random
1623 * value is not cryptographically secure but for several uses the cost of
1624 * depleting entropy is too high
1625 */
1626unsigned int get_random_int(void)
1627{
1628 /*
1629 * Use IP's RNG. It suits our purpose perfectly: it re-keys itself
1630 * every second, from the entropy pool (and thus creates a limited
1631 * drain on it), and uses halfMD4Transform within the second. We
1632 * also mix it with jiffies and the PID:
1633 */
b09b845c 1634 return secure_ip_id((__force __be32)(current->pid + jiffies));
1da177e4
LT
1635}
1636
1637/*
1638 * randomize_range() returns a start address such that
1639 *
1640 * [...... <range> .....]
1641 * start end
1642 *
1643 * a <range> with size "len" starting at the return value is inside in the
1644 * area defined by [start, end], but is otherwise randomized.
1645 */
1646unsigned long
1647randomize_range(unsigned long start, unsigned long end, unsigned long len)
1648{
1649 unsigned long range = end - len - start;
1650
1651 if (end <= start + len)
1652 return 0;
1653 return PAGE_ALIGN(get_random_int() % range + start);
1654}