blob: 7c13581ca9cd6ac1ea4a2d54e80397831cebd75f [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * random.c -- A strong random number generator
3 *
Matt Mackall9e95ce22005-04-16 15:25:56 -07004 * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
6 * Copyright Theodore Ts'o, 1994, 1995, 1996, 1997, 1998, 1999. All
7 * rights reserved.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, and the entire permission notice in its entirety,
14 * including the disclaimer of warranties.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. The name of the author may not be used to endorse or promote
19 * products derived from this software without specific prior
20 * written permission.
21 *
22 * ALTERNATIVELY, this product may be distributed under the terms of
23 * the GNU General Public License, in which case the provisions of the GPL are
24 * required INSTEAD OF the above restrictions. (This clause is
25 * necessary due to a potential bad interaction between the GPL and
26 * the restrictions contained in a BSD-style copyright.)
27 *
28 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
29 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
30 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ALL OF
31 * WHICH ARE HEREBY DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE
32 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
33 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT
34 * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
35 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
36 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
37 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
38 * USE OF THIS SOFTWARE, EVEN IF NOT ADVISED OF THE POSSIBILITY OF SUCH
39 * DAMAGE.
40 */
41
42/*
43 * (now, with legal B.S. out of the way.....)
44 *
45 * This routine gathers environmental noise from device drivers, etc.,
46 * and returns good random numbers, suitable for cryptographic use.
47 * Besides the obvious cryptographic uses, these numbers are also good
48 * for seeding TCP sequence numbers, and other places where it is
49 * desirable to have numbers which are not only random, but hard to
50 * predict by an attacker.
51 *
52 * Theory of operation
53 * ===================
54 *
55 * Computers are very predictable devices. Hence it is extremely hard
56 * to produce truly random numbers on a computer --- as opposed to
57 * pseudo-random numbers, which can easily generated by using a
58 * algorithm. Unfortunately, it is very easy for attackers to guess
59 * the sequence of pseudo-random number generators, and for some
60 * applications this is not acceptable. So instead, we must try to
61 * gather "environmental noise" from the computer's environment, which
62 * must be hard for outside attackers to observe, and use that to
63 * generate random numbers. In a Unix environment, this is best done
64 * from inside the kernel.
65 *
66 * Sources of randomness from the environment include inter-keyboard
67 * timings, inter-interrupt timings from some interrupts, and other
68 * events which are both (a) non-deterministic and (b) hard for an
69 * outside observer to measure. Randomness from these sources are
70 * added to an "entropy pool", which is mixed using a CRC-like function.
71 * This is not cryptographically strong, but it is adequate assuming
72 * the randomness is not chosen maliciously, and it is fast enough that
73 * the overhead of doing it on every interrupt is very reasonable.
74 * As random bytes are mixed into the entropy pool, the routines keep
75 * an *estimate* of how many bits of randomness have been stored into
76 * the random number generator's internal state.
77 *
78 * When random bytes are desired, they are obtained by taking the SHA
79 * hash of the contents of the "entropy pool". The SHA hash avoids
80 * exposing the internal state of the entropy pool. It is believed to
81 * be computationally infeasible to derive any useful information
82 * about the input of SHA from its output. Even if it is possible to
83 * analyze SHA in some clever way, as long as the amount of data
84 * returned from the generator is less than the inherent entropy in
85 * the pool, the output data is totally unpredictable. For this
86 * reason, the routine decreases its internal estimate of how many
87 * bits of "true randomness" are contained in the entropy pool as it
88 * outputs random numbers.
89 *
90 * If this estimate goes to zero, the routine can still generate
91 * random numbers; however, an attacker may (at least in theory) be
92 * able to infer the future output of the generator from prior
93 * outputs. This requires successful cryptanalysis of SHA, which is
94 * not believed to be feasible, but there is a remote possibility.
95 * Nonetheless, these numbers should be useful for the vast majority
96 * of purposes.
97 *
98 * Exported interfaces ---- output
99 * ===============================
100 *
101 * There are three exported interfaces; the first is one designed to
102 * be used from within the kernel:
103 *
104 * void get_random_bytes(void *buf, int nbytes);
105 *
106 * This interface will return the requested number of random bytes,
107 * and place it in the requested buffer.
108 *
109 * The two other interfaces are two character devices /dev/random and
110 * /dev/urandom. /dev/random is suitable for use when very high
111 * quality randomness is desired (for example, for key generation or
112 * one-time pads), as it will only return a maximum of the number of
113 * bits of randomness (as estimated by the random number generator)
114 * contained in the entropy pool.
115 *
116 * The /dev/urandom device does not have this limit, and will return
117 * as many bytes as are requested. As more and more random bytes are
118 * requested without giving time for the entropy pool to recharge,
119 * this will result in random numbers that are merely cryptographically
120 * strong. For many applications, however, this is acceptable.
121 *
122 * Exported interfaces ---- input
123 * ==============================
124 *
125 * The current exported interfaces for gathering environmental noise
126 * from the devices are:
127 *
128 * void add_input_randomness(unsigned int type, unsigned int code,
129 * unsigned int value);
130 * void add_interrupt_randomness(int irq);
131 *
132 * add_input_randomness() uses the input layer interrupt timing, as well as
133 * the event type information from the hardware.
134 *
135 * add_interrupt_randomness() uses the inter-interrupt timing as random
136 * inputs to the entropy pool. Note that not all interrupts are good
137 * sources of randomness! For example, the timer interrupts is not a
138 * good choice, because the periodicity of the interrupts is too
139 * regular, and hence predictable to an attacker. Disk interrupts are
140 * a better measure, since the timing of the disk interrupts are more
141 * unpredictable.
142 *
143 * All of these routines try to estimate how many bits of randomness a
144 * particular randomness source. They do this by keeping track of the
145 * first and second order deltas of the event timings.
146 *
147 * Ensuring unpredictability at system startup
148 * ============================================
149 *
150 * When any operating system starts up, it will go through a sequence
151 * of actions that are fairly predictable by an adversary, especially
152 * if the start-up does not involve interaction with a human operator.
153 * This reduces the actual number of bits of unpredictability in the
154 * entropy pool below the value in entropy_count. In order to
155 * counteract this effect, it helps to carry information in the
156 * entropy pool across shut-downs and start-ups. To do this, put the
157 * following lines an appropriate script which is run during the boot
158 * sequence:
159 *
160 * echo "Initializing random number generator..."
161 * random_seed=/var/run/random-seed
162 * # Carry a random seed from start-up to start-up
163 * # Load and then save the whole entropy pool
164 * if [ -f $random_seed ]; then
165 * cat $random_seed >/dev/urandom
166 * else
167 * touch $random_seed
168 * fi
169 * chmod 600 $random_seed
170 * dd if=/dev/urandom of=$random_seed count=1 bs=512
171 *
172 * and the following lines in an appropriate script which is run as
173 * the system is shutdown:
174 *
175 * # Carry a random seed from shut-down to start-up
176 * # Save the whole entropy pool
177 * echo "Saving random seed..."
178 * random_seed=/var/run/random-seed
179 * touch $random_seed
180 * chmod 600 $random_seed
181 * dd if=/dev/urandom of=$random_seed count=1 bs=512
182 *
183 * For example, on most modern systems using the System V init
184 * scripts, such code fragments would be found in
185 * /etc/rc.d/init.d/random. On older Linux systems, the correct script
186 * location might be in /etc/rcb.d/rc.local or /etc/rc.d/rc.0.
187 *
188 * Effectively, these commands cause the contents of the entropy pool
189 * to be saved at shut-down time and reloaded into the entropy pool at
190 * start-up. (The 'dd' in the addition to the bootup script is to
191 * make sure that /etc/random-seed is different for every start-up,
192 * even if the system crashes without executing rc.0.) Even with
193 * complete knowledge of the start-up activities, predicting the state
194 * of the entropy pool requires knowledge of the previous history of
195 * the system.
196 *
197 * Configuring the /dev/random driver under Linux
198 * ==============================================
199 *
200 * The /dev/random driver under Linux uses minor numbers 8 and 9 of
201 * the /dev/mem major number (#1). So if your system does not have
202 * /dev/random and /dev/urandom created already, they can be created
203 * by using the commands:
204 *
205 * mknod /dev/random c 1 8
206 * mknod /dev/urandom c 1 9
207 *
208 * Acknowledgements:
209 * =================
210 *
211 * Ideas for constructing this random number generator were derived
212 * from Pretty Good Privacy's random number generator, and from private
213 * discussions with Phil Karn. Colin Plumb provided a faster random
214 * number generator, which speed up the mixing function of the entropy
215 * pool, taken from PGPfone. Dale Worley has also contributed many
216 * useful ideas and suggestions to improve this driver.
217 *
218 * Any flaws in the design are solely my responsibility, and should
219 * not be attributed to the Phil, Colin, or any of authors of PGP.
220 *
221 * Further background information on this topic may be obtained from
222 * RFC 1750, "Randomness Recommendations for Security", by Donald
223 * Eastlake, Steve Crocker, and Jeff Schiller.
224 */
225
226#include <linux/utsname.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227#include <linux/module.h>
228#include <linux/kernel.h>
229#include <linux/major.h>
230#include <linux/string.h>
231#include <linux/fcntl.h>
232#include <linux/slab.h>
233#include <linux/random.h>
234#include <linux/poll.h>
235#include <linux/init.h>
236#include <linux/fs.h>
237#include <linux/genhd.h>
238#include <linux/interrupt.h>
Andrea Righi27ac7922008-07-23 21:28:13 -0700239#include <linux/mm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240#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
Christoph Lameter6c036522005-07-07 17:56:59 -0700274static int trickle_thresh __read_mostly = INPUT_POOL_WORDS * 28;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275
Matt Mackall90b75ee2008-04-29 01:02:55 -0700276static DEFINE_PER_CPU(int, trickle_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277
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);
Jeff Dike9a6f70b2008-04-29 01:03:08 -0700374static struct fasync_struct *fasync;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
376#if 0
Matt Mackall90b75ee2008-04-29 01:02:55 -0700377static int debug;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378module_param(debug, bool, 0644);
Matt Mackall90b75ee2008-04-29 01:02:55 -0700379#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)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387#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 {
Matt Mackall43358202008-04-29 01:03:01 -0700400 /* read-only data: */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 struct poolinfo *poolinfo;
402 __u32 *pool;
403 const char *name;
404 int limit;
405 struct entropy_store *pull;
406
407 /* read-write data: */
Matt Mackall43358202008-04-29 01:03:01 -0700408 spinlock_t lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 unsigned add_ptr;
Matt Mackallcda796a2009-01-06 14:42:55 -0800410 int entropy_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 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,
Ingo Molnare4d91912006-07-03 00:24:34 -0700422 .lock = __SPIN_LOCK_UNLOCKED(&input_pool.lock),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 .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,
Ingo Molnare4d91912006-07-03 00:24:34 -0700431 .lock = __SPIN_LOCK_UNLOCKED(&blocking_pool.lock),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 .pool = blocking_pool_data
433};
434
435static struct entropy_store nonblocking_pool = {
436 .poolinfo = &poolinfo_table[1],
437 .name = "nonblocking",
438 .pull = &input_pool,
Ingo Molnare4d91912006-07-03 00:24:34 -0700439 .lock = __SPIN_LOCK_UNLOCKED(&nonblocking_pool.lock),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 .pool = nonblocking_pool_data
441};
442
443/*
Matt Mackalle68e5b62008-04-29 01:03:05 -0700444 * This function adds bytes into the entropy "pool". It does not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 * update the entropy estimate. The caller should call
Matt Mackalladc782d2008-04-29 01:03:07 -0700446 * credit_entropy_bits if this is appropriate.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 *
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 */
Matt Mackalle68e5b62008-04-29 01:03:05 -0700453static void mix_pool_bytes_extract(struct entropy_store *r, const void *in,
454 int nbytes, __u8 out[64])
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455{
456 static __u32 const twist_table[8] = {
457 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
458 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
Matt Mackall993ba212008-04-29 01:03:04 -0700459 unsigned long i, j, tap1, tap2, tap3, tap4, tap5;
Matt Mackallfeee7692008-04-29 01:03:02 -0700460 int input_rotate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 int wordmask = r->poolinfo->poolwords - 1;
Matt Mackalle68e5b62008-04-29 01:03:05 -0700462 const char *bytes = in;
Matt Mackall6d38b822008-04-29 01:03:03 -0700463 __u32 w;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
473 spin_lock_irqsave(&r->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 input_rotate = r->input_rotate;
Matt Mackall993ba212008-04-29 01:03:04 -0700475 i = r->add_ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476
Matt Mackalle68e5b62008-04-29 01:03:05 -0700477 /* mix one byte at a time to simplify size handling and churn faster */
478 while (nbytes--) {
479 w = rol32(*bytes++, input_rotate & 31);
Matt Mackall993ba212008-04-29 01:03:04 -0700480 i = (i - 1) & wordmask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
482 /* XOR in the various taps */
Matt Mackall993ba212008-04-29 01:03:04 -0700483 w ^= r->pool[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 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];
Matt Mackall993ba212008-04-29 01:03:04 -0700489
490 /* Mix the result back in with a twist */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
Matt Mackallfeee7692008-04-29 01:03:02 -0700492
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 }
501
502 r->input_rotate = input_rotate;
Matt Mackall993ba212008-04-29 01:03:04 -0700503 r->add_ptr = i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
Matt Mackall993ba212008-04-29 01:03:04 -0700505 if (out)
506 for (j = 0; j < 16; j++)
Matt Mackalle68e5b62008-04-29 01:03:05 -0700507 ((__u32 *)out)[j] = r->pool[(i - j) & wordmask];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
509 spin_unlock_irqrestore(&r->lock, flags);
510}
511
Matt Mackalle68e5b62008-04-29 01:03:05 -0700512static void mix_pool_bytes(struct entropy_store *r, const void *in, int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513{
Matt Mackalle68e5b62008-04-29 01:03:05 -0700514 mix_pool_bytes_extract(r, in, bytes, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515}
516
517/*
518 * Credit (or debit) the entropy store with n bits of entropy
519 */
Matt Mackalladc782d2008-04-29 01:03:07 -0700520static void credit_entropy_bits(struct entropy_store *r, int nbits)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521{
522 unsigned long flags;
Andrew Morton8b76f462008-09-02 14:36:14 -0700523 int entropy_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
Matt Mackalladc782d2008-04-29 01:03:07 -0700525 if (!nbits)
526 return;
527
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 spin_lock_irqsave(&r->lock, flags);
529
Matt Mackalladc782d2008-04-29 01:03:07 -0700530 DEBUG_ENT("added %d entropy credits to %s\n", nbits, r->name);
Andrew Morton8b76f462008-09-02 14:36:14 -0700531 entropy_count = r->entropy_count;
532 entropy_count += nbits;
533 if (entropy_count < 0) {
Matt Mackalladc782d2008-04-29 01:03:07 -0700534 DEBUG_ENT("negative entropy/overflow\n");
Andrew Morton8b76f462008-09-02 14:36:14 -0700535 entropy_count = 0;
536 } else if (entropy_count > r->poolinfo->POOLBITS)
537 entropy_count = r->poolinfo->POOLBITS;
538 r->entropy_count = entropy_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539
Matt Mackall88c730d2008-04-29 01:02:56 -0700540 /* should we wake readers? */
Andrew Morton8b76f462008-09-02 14:36:14 -0700541 if (r == &input_pool && entropy_count >= random_read_wakeup_thresh) {
Matt Mackall88c730d2008-04-29 01:02:56 -0700542 wake_up_interruptible(&random_read_wait);
Jeff Dike9a6f70b2008-04-29 01:03:08 -0700543 kill_fasync(&fasync, SIGIO, POLL_IN);
544 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 spin_unlock_irqrestore(&r->lock, flags);
546}
547
548/*********************************************************************
549 *
550 * Entropy input management
551 *
552 *********************************************************************/
553
554/* There is one of these per entropy source */
555struct timer_rand_state {
556 cycles_t last_time;
Matt Mackall90b75ee2008-04-29 01:02:55 -0700557 long last_delta, last_delta2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 unsigned dont_count_entropy:1;
559};
560
Yinghai Lu0b8f1ef2008-12-05 18:58:31 -0800561#ifndef CONFIG_SPARSE_IRQ
Yinghai Lu2f983572009-01-03 00:06:34 -0800562
563static struct timer_rand_state *irq_timer_state[NR_IRQS];
564
565static struct timer_rand_state *get_timer_rand_state(unsigned int irq)
566{
567 return irq_timer_state[irq];
568}
569
570static void set_timer_rand_state(unsigned int irq,
571 struct timer_rand_state *state)
572{
573 irq_timer_state[irq] = state;
574}
575
576#else
577
578static struct timer_rand_state *get_timer_rand_state(unsigned int irq)
579{
580 struct irq_desc *desc;
581
582 desc = irq_to_desc(irq);
583
584 return desc->timer_rand_state;
585}
586
587static void set_timer_rand_state(unsigned int irq,
588 struct timer_rand_state *state)
589{
590 struct irq_desc *desc;
591
592 desc = irq_to_desc(irq);
593
594 desc->timer_rand_state = state;
595}
Yinghai Lu0b8f1ef2008-12-05 18:58:31 -0800596#endif
Yinghai Lu3060d6f2008-08-19 20:50:08 -0700597
Yinghai Lu3060d6f2008-08-19 20:50:08 -0700598static struct timer_rand_state input_timer_state;
599
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600/*
601 * This function adds entropy to the entropy "pool" by using timing
602 * delays. It uses the timer_rand_state structure to make an estimate
603 * of how many bits of entropy this call has added to the pool.
604 *
605 * The number "num" is also added to the pool - it should somehow describe
606 * the type of event which just happened. This is currently 0-255 for
607 * keyboard scan codes, and 256 upwards for interrupts.
608 *
609 */
610static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
611{
612 struct {
613 cycles_t cycles;
614 long jiffies;
615 unsigned num;
616 } sample;
617 long delta, delta2, delta3;
618
619 preempt_disable();
620 /* if over the trickle threshold, use only 1 in 4096 samples */
621 if (input_pool.entropy_count > trickle_thresh &&
622 (__get_cpu_var(trickle_count)++ & 0xfff))
623 goto out;
624
625 sample.jiffies = jiffies;
626 sample.cycles = get_cycles();
627 sample.num = num;
Matt Mackalle68e5b62008-04-29 01:03:05 -0700628 mix_pool_bytes(&input_pool, &sample, sizeof(sample));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
630 /*
631 * Calculate number of bits of randomness we probably added.
632 * We take into account the first, second and third-order deltas
633 * in order to make our estimate.
634 */
635
636 if (!state->dont_count_entropy) {
637 delta = sample.jiffies - state->last_time;
638 state->last_time = sample.jiffies;
639
640 delta2 = delta - state->last_delta;
641 state->last_delta = delta;
642
643 delta3 = delta2 - state->last_delta2;
644 state->last_delta2 = delta2;
645
646 if (delta < 0)
647 delta = -delta;
648 if (delta2 < 0)
649 delta2 = -delta2;
650 if (delta3 < 0)
651 delta3 = -delta3;
652 if (delta > delta2)
653 delta = delta2;
654 if (delta > delta3)
655 delta = delta3;
656
657 /*
658 * delta is now minimum absolute delta.
659 * Round down by 1 bit on general principles,
660 * and limit entropy entimate to 12 bits.
661 */
Matt Mackalladc782d2008-04-29 01:03:07 -0700662 credit_entropy_bits(&input_pool,
663 min_t(int, fls(delta>>1), 11));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665out:
666 preempt_enable();
667}
668
Stephen Hemmingerd2515752006-01-11 12:17:38 -0800669void add_input_randomness(unsigned int type, unsigned int code,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 unsigned int value)
671{
672 static unsigned char last_value;
673
674 /* ignore autorepeat and the like */
675 if (value == last_value)
676 return;
677
678 DEBUG_ENT("input event\n");
679 last_value = value;
680 add_timer_randomness(&input_timer_state,
681 (type << 4) ^ code ^ (code >> 4) ^ value);
682}
Dmitry Torokhov80fc9f52006-10-11 01:43:58 -0400683EXPORT_SYMBOL_GPL(add_input_randomness);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
685void add_interrupt_randomness(int irq)
686{
Yinghai Lu3060d6f2008-08-19 20:50:08 -0700687 struct timer_rand_state *state;
688
689 state = get_timer_rand_state(irq);
690
691 if (state == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 return;
693
694 DEBUG_ENT("irq event %d\n", irq);
Yinghai Lu3060d6f2008-08-19 20:50:08 -0700695 add_timer_randomness(state, 0x100 + irq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696}
697
David Howells93614012006-09-30 20:45:40 +0200698#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699void add_disk_randomness(struct gendisk *disk)
700{
701 if (!disk || !disk->random)
702 return;
703 /* first major is 1, so we get >= 0x200 here */
Tejun Heof331c022008-09-03 09:01:48 +0200704 DEBUG_ENT("disk event %d:%d\n",
705 MAJOR(disk_devt(disk)), MINOR(disk_devt(disk)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
Tejun Heof331c022008-09-03 09:01:48 +0200707 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708}
David Howells93614012006-09-30 20:45:40 +0200709#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710
711#define EXTRACT_SIZE 10
712
713/*********************************************************************
714 *
715 * Entropy extraction routines
716 *
717 *********************************************************************/
718
Matt Mackall90b75ee2008-04-29 01:02:55 -0700719static ssize_t extract_entropy(struct entropy_store *r, void *buf,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 size_t nbytes, int min, int rsvd);
721
722/*
723 * This utility inline function is responsible for transfering entropy
724 * from the primary pool to the secondary extraction pool. We make
725 * sure we pull enough for a 'catastrophic reseed'.
726 */
727static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
728{
729 __u32 tmp[OUTPUT_POOL_WORDS];
730
731 if (r->pull && r->entropy_count < nbytes * 8 &&
732 r->entropy_count < r->poolinfo->POOLBITS) {
Matt Mackall5a021e92007-07-19 11:30:14 -0700733 /* If we're limited, always leave two wakeup worth's BITS */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 int rsvd = r->limit ? 0 : random_read_wakeup_thresh/4;
Matt Mackall5a021e92007-07-19 11:30:14 -0700735 int bytes = nbytes;
736
737 /* pull at least as many as BYTES as wakeup BITS */
738 bytes = max_t(int, bytes, random_read_wakeup_thresh / 8);
739 /* but never more than the buffer size */
740 bytes = min_t(int, bytes, sizeof(tmp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
742 DEBUG_ENT("going to reseed %s with %d bits "
743 "(%d of %d requested)\n",
744 r->name, bytes * 8, nbytes * 8, r->entropy_count);
745
Matt Mackall90b75ee2008-04-29 01:02:55 -0700746 bytes = extract_entropy(r->pull, tmp, bytes,
747 random_read_wakeup_thresh / 8, rsvd);
Matt Mackalle68e5b62008-04-29 01:03:05 -0700748 mix_pool_bytes(r, tmp, bytes);
Matt Mackalladc782d2008-04-29 01:03:07 -0700749 credit_entropy_bits(r, bytes*8);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 }
751}
752
753/*
754 * These functions extracts randomness from the "entropy pool", and
755 * returns it in a buffer.
756 *
757 * The min parameter specifies the minimum amount we can pull before
758 * failing to avoid races that defeat catastrophic reseeding while the
759 * reserved parameter indicates how much entropy we must leave in the
760 * pool after each pull to avoid starving other readers.
761 *
762 * Note: extract_entropy() assumes that .poolwords is a multiple of 16 words.
763 */
764
765static size_t account(struct entropy_store *r, size_t nbytes, int min,
766 int reserved)
767{
768 unsigned long flags;
769
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 /* Hold lock while accounting */
771 spin_lock_irqsave(&r->lock, flags);
772
Matt Mackallcda796a2009-01-06 14:42:55 -0800773 BUG_ON(r->entropy_count > r->poolinfo->POOLBITS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 DEBUG_ENT("trying to extract %d bits from %s\n",
775 nbytes * 8, r->name);
776
777 /* Can we pull enough? */
778 if (r->entropy_count / 8 < min + reserved) {
779 nbytes = 0;
780 } else {
781 /* If limited, never pull more than available */
782 if (r->limit && nbytes + reserved >= r->entropy_count / 8)
783 nbytes = r->entropy_count/8 - reserved;
784
Matt Mackall90b75ee2008-04-29 01:02:55 -0700785 if (r->entropy_count / 8 >= nbytes + reserved)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 r->entropy_count -= nbytes*8;
787 else
788 r->entropy_count = reserved;
789
Jeff Dike9a6f70b2008-04-29 01:03:08 -0700790 if (r->entropy_count < random_write_wakeup_thresh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 wake_up_interruptible(&random_write_wait);
Jeff Dike9a6f70b2008-04-29 01:03:08 -0700792 kill_fasync(&fasync, SIGIO, POLL_OUT);
793 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 }
795
796 DEBUG_ENT("debiting %d entropy credits from %s%s\n",
797 nbytes * 8, r->name, r->limit ? "" : " (unlimited)");
798
799 spin_unlock_irqrestore(&r->lock, flags);
800
801 return nbytes;
802}
803
804static void extract_buf(struct entropy_store *r, __u8 *out)
805{
Matt Mackall602b6ae2007-05-29 21:54:27 -0500806 int i;
Matt Mackalle68e5b62008-04-29 01:03:05 -0700807 __u32 hash[5], workspace[SHA_WORKSPACE_WORDS];
808 __u8 extract[64];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809
Matt Mackall1c0ad3d2008-04-29 01:03:00 -0700810 /* Generate a hash across the pool, 16 words (512 bits) at a time */
Matt Mackallffd8d3f2008-04-29 01:02:59 -0700811 sha_init(hash);
Matt Mackall1c0ad3d2008-04-29 01:03:00 -0700812 for (i = 0; i < r->poolinfo->poolwords; i += 16)
Matt Mackallffd8d3f2008-04-29 01:02:59 -0700813 sha_transform(hash, (__u8 *)(r->pool + i), workspace);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
815 /*
Matt Mackall1c0ad3d2008-04-29 01:03:00 -0700816 * We mix the hash back into the pool to prevent backtracking
817 * attacks (where the attacker knows the state of the pool
818 * plus the current outputs, and attempts to find previous
819 * ouputs), unless the hash function can be inverted. By
820 * mixing at least a SHA1 worth of hash data back, we make
821 * brute-forcing the feedback as hard as brute-forcing the
822 * hash.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 */
Matt Mackalle68e5b62008-04-29 01:03:05 -0700824 mix_pool_bytes_extract(r, hash, sizeof(hash), extract);
Matt Mackall1c0ad3d2008-04-29 01:03:00 -0700825
826 /*
827 * To avoid duplicates, we atomically extract a portion of the
828 * pool while mixing, and hash one final time.
829 */
Matt Mackalle68e5b62008-04-29 01:03:05 -0700830 sha_transform(hash, extract, workspace);
Matt Mackallffd8d3f2008-04-29 01:02:59 -0700831 memset(extract, 0, sizeof(extract));
832 memset(workspace, 0, sizeof(workspace));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833
834 /*
Matt Mackall1c0ad3d2008-04-29 01:03:00 -0700835 * In case the hash function has some recognizable output
836 * pattern, we fold it in half. Thus, we always feed back
837 * twice as much data as we output.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 */
Matt Mackallffd8d3f2008-04-29 01:02:59 -0700839 hash[0] ^= hash[3];
840 hash[1] ^= hash[4];
841 hash[2] ^= rol32(hash[2], 16);
842 memcpy(out, hash, EXTRACT_SIZE);
843 memset(hash, 0, sizeof(hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844}
845
Matt Mackall90b75ee2008-04-29 01:02:55 -0700846static ssize_t extract_entropy(struct entropy_store *r, void *buf,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 size_t nbytes, int min, int reserved)
848{
849 ssize_t ret = 0, i;
850 __u8 tmp[EXTRACT_SIZE];
851
852 xfer_secondary_pool(r, nbytes);
853 nbytes = account(r, nbytes, min, reserved);
854
855 while (nbytes) {
856 extract_buf(r, tmp);
857 i = min_t(int, nbytes, EXTRACT_SIZE);
858 memcpy(buf, tmp, i);
859 nbytes -= i;
860 buf += i;
861 ret += i;
862 }
863
864 /* Wipe data just returned from memory */
865 memset(tmp, 0, sizeof(tmp));
866
867 return ret;
868}
869
870static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
871 size_t nbytes)
872{
873 ssize_t ret = 0, i;
874 __u8 tmp[EXTRACT_SIZE];
875
876 xfer_secondary_pool(r, nbytes);
877 nbytes = account(r, nbytes, 0, 0);
878
879 while (nbytes) {
880 if (need_resched()) {
881 if (signal_pending(current)) {
882 if (ret == 0)
883 ret = -ERESTARTSYS;
884 break;
885 }
886 schedule();
887 }
888
889 extract_buf(r, tmp);
890 i = min_t(int, nbytes, EXTRACT_SIZE);
891 if (copy_to_user(buf, tmp, i)) {
892 ret = -EFAULT;
893 break;
894 }
895
896 nbytes -= i;
897 buf += i;
898 ret += i;
899 }
900
901 /* Wipe data just returned from memory */
902 memset(tmp, 0, sizeof(tmp));
903
904 return ret;
905}
906
907/*
908 * This function is the exported kernel interface. It returns some
909 * number of good random numbers, suitable for seeding TCP sequence
910 * numbers, etc.
911 */
912void get_random_bytes(void *buf, int nbytes)
913{
914 extract_entropy(&nonblocking_pool, buf, nbytes, 0, 0);
915}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916EXPORT_SYMBOL(get_random_bytes);
917
918/*
919 * init_std_data - initialize pool with system data
920 *
921 * @r: pool to initialize
922 *
923 * This function clears the pool's entropy count and mixes some system
924 * data into the pool to prepare it for use. The pool is not cleared
925 * as that can only decrease the entropy in the pool.
926 */
927static void init_std_data(struct entropy_store *r)
928{
Eric Dumazetf8595812007-03-28 14:22:33 -0700929 ktime_t now;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 unsigned long flags;
931
932 spin_lock_irqsave(&r->lock, flags);
933 r->entropy_count = 0;
934 spin_unlock_irqrestore(&r->lock, flags);
935
Eric Dumazetf8595812007-03-28 14:22:33 -0700936 now = ktime_get_real();
Matt Mackalle68e5b62008-04-29 01:03:05 -0700937 mix_pool_bytes(r, &now, sizeof(now));
938 mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939}
940
Matt Mackall53c3f632008-04-29 01:02:58 -0700941static int rand_initialize(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942{
943 init_std_data(&input_pool);
944 init_std_data(&blocking_pool);
945 init_std_data(&nonblocking_pool);
946 return 0;
947}
948module_init(rand_initialize);
949
950void rand_initialize_irq(int irq)
951{
952 struct timer_rand_state *state;
953
Yinghai Lu3060d6f2008-08-19 20:50:08 -0700954 state = get_timer_rand_state(irq);
955
956 if (state)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 return;
958
959 /*
Eric Dumazetf8595812007-03-28 14:22:33 -0700960 * If kzalloc returns null, we just won't use that entropy
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 * source.
962 */
Eric Dumazetf8595812007-03-28 14:22:33 -0700963 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
964 if (state)
Yinghai Lu3060d6f2008-08-19 20:50:08 -0700965 set_timer_rand_state(irq, state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966}
967
David Howells93614012006-09-30 20:45:40 +0200968#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969void rand_initialize_disk(struct gendisk *disk)
970{
971 struct timer_rand_state *state;
972
973 /*
Eric Dumazetf8595812007-03-28 14:22:33 -0700974 * If kzalloc returns null, we just won't use that entropy
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 * source.
976 */
Eric Dumazetf8595812007-03-28 14:22:33 -0700977 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
978 if (state)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 disk->random = state;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980}
David Howells93614012006-09-30 20:45:40 +0200981#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
983static ssize_t
Matt Mackall90b75ee2008-04-29 01:02:55 -0700984random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985{
986 ssize_t n, retval = 0, count = 0;
987
988 if (nbytes == 0)
989 return 0;
990
991 while (nbytes > 0) {
992 n = nbytes;
993 if (n > SEC_XFER_SIZE)
994 n = SEC_XFER_SIZE;
995
996 DEBUG_ENT("reading %d bits\n", n*8);
997
998 n = extract_entropy_user(&blocking_pool, buf, n);
999
1000 DEBUG_ENT("read got %d bits (%d still needed)\n",
1001 n*8, (nbytes-n)*8);
1002
1003 if (n == 0) {
1004 if (file->f_flags & O_NONBLOCK) {
1005 retval = -EAGAIN;
1006 break;
1007 }
1008
1009 DEBUG_ENT("sleeping?\n");
1010
1011 wait_event_interruptible(random_read_wait,
1012 input_pool.entropy_count >=
1013 random_read_wakeup_thresh);
1014
1015 DEBUG_ENT("awake\n");
1016
1017 if (signal_pending(current)) {
1018 retval = -ERESTARTSYS;
1019 break;
1020 }
1021
1022 continue;
1023 }
1024
1025 if (n < 0) {
1026 retval = n;
1027 break;
1028 }
1029 count += n;
1030 buf += n;
1031 nbytes -= n;
1032 break; /* This break makes the device work */
1033 /* like a named pipe */
1034 }
1035
1036 /*
1037 * If we gave the user some bytes, update the access time.
1038 */
1039 if (count)
1040 file_accessed(file);
1041
1042 return (count ? count : retval);
1043}
1044
1045static ssize_t
Matt Mackall90b75ee2008-04-29 01:02:55 -07001046urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047{
1048 return extract_entropy_user(&nonblocking_pool, buf, nbytes);
1049}
1050
1051static unsigned int
1052random_poll(struct file *file, poll_table * wait)
1053{
1054 unsigned int mask;
1055
1056 poll_wait(file, &random_read_wait, wait);
1057 poll_wait(file, &random_write_wait, wait);
1058 mask = 0;
1059 if (input_pool.entropy_count >= random_read_wakeup_thresh)
1060 mask |= POLLIN | POLLRDNORM;
1061 if (input_pool.entropy_count < random_write_wakeup_thresh)
1062 mask |= POLLOUT | POLLWRNORM;
1063 return mask;
1064}
1065
Matt Mackall7f397dc2007-05-29 21:58:10 -05001066static int
1067write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1068{
1069 size_t bytes;
1070 __u32 buf[16];
1071 const char __user *p = buffer;
1072
1073 while (count > 0) {
1074 bytes = min(count, sizeof(buf));
1075 if (copy_from_user(&buf, p, bytes))
1076 return -EFAULT;
1077
1078 count -= bytes;
1079 p += bytes;
1080
Matt Mackalle68e5b62008-04-29 01:03:05 -07001081 mix_pool_bytes(r, buf, bytes);
Matt Mackall91f3f1e2008-02-06 01:37:20 -08001082 cond_resched();
Matt Mackall7f397dc2007-05-29 21:58:10 -05001083 }
1084
1085 return 0;
1086}
1087
Matt Mackall90b75ee2008-04-29 01:02:55 -07001088static ssize_t random_write(struct file *file, const char __user *buffer,
1089 size_t count, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090{
Matt Mackall7f397dc2007-05-29 21:58:10 -05001091 size_t ret;
1092 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
Matt Mackall7f397dc2007-05-29 21:58:10 -05001094 ret = write_pool(&blocking_pool, buffer, count);
1095 if (ret)
1096 return ret;
1097 ret = write_pool(&nonblocking_pool, buffer, count);
1098 if (ret)
1099 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100
Matt Mackall7f397dc2007-05-29 21:58:10 -05001101 inode->i_mtime = current_fs_time(inode->i_sb);
1102 mark_inode_dirty(inode);
1103 return (ssize_t)count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104}
1105
Matt Mackall43ae4862008-04-29 01:02:58 -07001106static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107{
1108 int size, ent_count;
1109 int __user *p = (int __user *)arg;
1110 int retval;
1111
1112 switch (cmd) {
1113 case RNDGETENTCNT:
Matt Mackall43ae4862008-04-29 01:02:58 -07001114 /* inherently racy, no point locking */
1115 if (put_user(input_pool.entropy_count, p))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 return -EFAULT;
1117 return 0;
1118 case RNDADDTOENTCNT:
1119 if (!capable(CAP_SYS_ADMIN))
1120 return -EPERM;
1121 if (get_user(ent_count, p))
1122 return -EFAULT;
Matt Mackalladc782d2008-04-29 01:03:07 -07001123 credit_entropy_bits(&input_pool, ent_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 return 0;
1125 case RNDADDENTROPY:
1126 if (!capable(CAP_SYS_ADMIN))
1127 return -EPERM;
1128 if (get_user(ent_count, p++))
1129 return -EFAULT;
1130 if (ent_count < 0)
1131 return -EINVAL;
1132 if (get_user(size, p++))
1133 return -EFAULT;
Matt Mackall7f397dc2007-05-29 21:58:10 -05001134 retval = write_pool(&input_pool, (const char __user *)p,
1135 size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 if (retval < 0)
1137 return retval;
Matt Mackalladc782d2008-04-29 01:03:07 -07001138 credit_entropy_bits(&input_pool, ent_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 return 0;
1140 case RNDZAPENTCNT:
1141 case RNDCLEARPOOL:
1142 /* Clear the entropy pool counters. */
1143 if (!capable(CAP_SYS_ADMIN))
1144 return -EPERM;
Matt Mackall53c3f632008-04-29 01:02:58 -07001145 rand_initialize();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 return 0;
1147 default:
1148 return -EINVAL;
1149 }
1150}
1151
Jeff Dike9a6f70b2008-04-29 01:03:08 -07001152static int random_fasync(int fd, struct file *filp, int on)
1153{
1154 return fasync_helper(fd, filp, on, &fasync);
1155}
1156
Arjan van de Ven2b8693c2007-02-12 00:55:32 -08001157const struct file_operations random_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 .read = random_read,
1159 .write = random_write,
1160 .poll = random_poll,
Matt Mackall43ae4862008-04-29 01:02:58 -07001161 .unlocked_ioctl = random_ioctl,
Jeff Dike9a6f70b2008-04-29 01:03:08 -07001162 .fasync = random_fasync,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163};
1164
Arjan van de Ven2b8693c2007-02-12 00:55:32 -08001165const struct file_operations urandom_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 .read = urandom_read,
1167 .write = random_write,
Matt Mackall43ae4862008-04-29 01:02:58 -07001168 .unlocked_ioctl = random_ioctl,
Jeff Dike9a6f70b2008-04-29 01:03:08 -07001169 .fasync = random_fasync,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170};
1171
1172/***************************************************************
1173 * Random UUID interface
1174 *
1175 * Used here for a Boot ID, but can be useful for other kernel
1176 * drivers.
1177 ***************************************************************/
1178
1179/*
1180 * Generate random UUID
1181 */
1182void generate_random_uuid(unsigned char uuid_out[16])
1183{
1184 get_random_bytes(uuid_out, 16);
1185 /* Set UUID version to 4 --- truely random generation */
1186 uuid_out[6] = (uuid_out[6] & 0x0F) | 0x40;
1187 /* Set the UUID variant to DCE */
1188 uuid_out[8] = (uuid_out[8] & 0x3F) | 0x80;
1189}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190EXPORT_SYMBOL(generate_random_uuid);
1191
1192/********************************************************************
1193 *
1194 * Sysctl interface
1195 *
1196 ********************************************************************/
1197
1198#ifdef CONFIG_SYSCTL
1199
1200#include <linux/sysctl.h>
1201
1202static int min_read_thresh = 8, min_write_thresh;
1203static int max_read_thresh = INPUT_POOL_WORDS * 32;
1204static int max_write_thresh = INPUT_POOL_WORDS * 32;
1205static char sysctl_bootid[16];
1206
1207/*
1208 * These functions is used to return both the bootid UUID, and random
1209 * UUID. The difference is in whether table->data is NULL; if it is,
1210 * then a new UUID is generated and returned to the user.
1211 *
1212 * If the user accesses this via the proc interface, it will be returned
1213 * as an ASCII string in the standard UUID format. If accesses via the
1214 * sysctl system call, it is returned as 16 bytes of binary data.
1215 */
1216static int proc_do_uuid(ctl_table *table, int write, struct file *filp,
1217 void __user *buffer, size_t *lenp, loff_t *ppos)
1218{
1219 ctl_table fake_table;
1220 unsigned char buf[64], tmp_uuid[16], *uuid;
1221
1222 uuid = table->data;
1223 if (!uuid) {
1224 uuid = tmp_uuid;
1225 uuid[8] = 0;
1226 }
1227 if (uuid[8] == 0)
1228 generate_random_uuid(uuid);
1229
1230 sprintf(buf, "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-"
1231 "%02x%02x%02x%02x%02x%02x",
1232 uuid[0], uuid[1], uuid[2], uuid[3],
1233 uuid[4], uuid[5], uuid[6], uuid[7],
1234 uuid[8], uuid[9], uuid[10], uuid[11],
1235 uuid[12], uuid[13], uuid[14], uuid[15]);
1236 fake_table.data = buf;
1237 fake_table.maxlen = sizeof(buf);
1238
1239 return proc_dostring(&fake_table, write, filp, buffer, lenp, ppos);
1240}
1241
Alexey Dobriyanf221e722008-10-15 22:04:23 -07001242static int uuid_strategy(ctl_table *table,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 void __user *oldval, size_t __user *oldlenp,
Alexey Dobriyan1f29bcd2006-12-10 02:19:10 -08001244 void __user *newval, size_t newlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245{
1246 unsigned char tmp_uuid[16], *uuid;
1247 unsigned int len;
1248
1249 if (!oldval || !oldlenp)
1250 return 1;
1251
1252 uuid = table->data;
1253 if (!uuid) {
1254 uuid = tmp_uuid;
1255 uuid[8] = 0;
1256 }
1257 if (uuid[8] == 0)
1258 generate_random_uuid(uuid);
1259
1260 if (get_user(len, oldlenp))
1261 return -EFAULT;
1262 if (len) {
1263 if (len > 16)
1264 len = 16;
1265 if (copy_to_user(oldval, uuid, len) ||
1266 put_user(len, oldlenp))
1267 return -EFAULT;
1268 }
1269 return 1;
1270}
1271
1272static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
1273ctl_table random_table[] = {
1274 {
1275 .ctl_name = RANDOM_POOLSIZE,
1276 .procname = "poolsize",
1277 .data = &sysctl_poolsize,
1278 .maxlen = sizeof(int),
1279 .mode = 0444,
1280 .proc_handler = &proc_dointvec,
1281 },
1282 {
1283 .ctl_name = RANDOM_ENTROPY_COUNT,
1284 .procname = "entropy_avail",
1285 .maxlen = sizeof(int),
1286 .mode = 0444,
1287 .proc_handler = &proc_dointvec,
1288 .data = &input_pool.entropy_count,
1289 },
1290 {
1291 .ctl_name = RANDOM_READ_THRESH,
1292 .procname = "read_wakeup_threshold",
1293 .data = &random_read_wakeup_thresh,
1294 .maxlen = sizeof(int),
1295 .mode = 0644,
1296 .proc_handler = &proc_dointvec_minmax,
1297 .strategy = &sysctl_intvec,
1298 .extra1 = &min_read_thresh,
1299 .extra2 = &max_read_thresh,
1300 },
1301 {
1302 .ctl_name = RANDOM_WRITE_THRESH,
1303 .procname = "write_wakeup_threshold",
1304 .data = &random_write_wakeup_thresh,
1305 .maxlen = sizeof(int),
1306 .mode = 0644,
1307 .proc_handler = &proc_dointvec_minmax,
1308 .strategy = &sysctl_intvec,
1309 .extra1 = &min_write_thresh,
1310 .extra2 = &max_write_thresh,
1311 },
1312 {
1313 .ctl_name = RANDOM_BOOT_ID,
1314 .procname = "boot_id",
1315 .data = &sysctl_bootid,
1316 .maxlen = 16,
1317 .mode = 0444,
1318 .proc_handler = &proc_do_uuid,
1319 .strategy = &uuid_strategy,
1320 },
1321 {
1322 .ctl_name = RANDOM_UUID,
1323 .procname = "uuid",
1324 .maxlen = 16,
1325 .mode = 0444,
1326 .proc_handler = &proc_do_uuid,
1327 .strategy = &uuid_strategy,
1328 },
1329 { .ctl_name = 0 }
1330};
1331#endif /* CONFIG_SYSCTL */
1332
1333/********************************************************************
1334 *
1335 * Random funtions for networking
1336 *
1337 ********************************************************************/
1338
1339/*
1340 * TCP initial sequence number picking. This uses the random number
1341 * generator to pick an initial secret value. This value is hashed
1342 * along with the TCP endpoint information to provide a unique
1343 * starting point for each pair of TCP endpoints. This defeats
1344 * attacks which rely on guessing the initial TCP sequence number.
1345 * This algorithm was suggested by Steve Bellovin.
1346 *
1347 * Using a very strong hash was taking an appreciable amount of the total
1348 * TCP connection establishment time, so this is a weaker hash,
1349 * compensated for by changing the secret periodically.
1350 */
1351
1352/* F, G and H are basic MD4 functions: selection, majority, parity */
1353#define F(x, y, z) ((z) ^ ((x) & ((y) ^ (z))))
1354#define G(x, y, z) (((x) & (y)) + (((x) ^ (y)) & (z)))
1355#define H(x, y, z) ((x) ^ (y) ^ (z))
1356
1357/*
1358 * The generic round function. The application is so specific that
1359 * we don't bother protecting all the arguments with parens, as is generally
1360 * good macro practice, in favor of extra legibility.
1361 * Rotation is separate from addition to prevent recomputation
1362 */
1363#define ROUND(f, a, b, c, d, x, s) \
1364 (a += f(b, c, d) + x, a = (a << s) | (a >> (32 - s)))
1365#define K1 0
1366#define K2 013240474631UL
1367#define K3 015666365641UL
1368
1369#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1370
Matt Mackall90b75ee2008-04-29 01:02:55 -07001371static __u32 twothirdsMD4Transform(__u32 const buf[4], __u32 const in[12])
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372{
1373 __u32 a = buf[0], b = buf[1], c = buf[2], d = buf[3];
1374
1375 /* Round 1 */
1376 ROUND(F, a, b, c, d, in[ 0] + K1, 3);
1377 ROUND(F, d, a, b, c, in[ 1] + K1, 7);
1378 ROUND(F, c, d, a, b, in[ 2] + K1, 11);
1379 ROUND(F, b, c, d, a, in[ 3] + K1, 19);
1380 ROUND(F, a, b, c, d, in[ 4] + K1, 3);
1381 ROUND(F, d, a, b, c, in[ 5] + K1, 7);
1382 ROUND(F, c, d, a, b, in[ 6] + K1, 11);
1383 ROUND(F, b, c, d, a, in[ 7] + K1, 19);
1384 ROUND(F, a, b, c, d, in[ 8] + K1, 3);
1385 ROUND(F, d, a, b, c, in[ 9] + K1, 7);
1386 ROUND(F, c, d, a, b, in[10] + K1, 11);
1387 ROUND(F, b, c, d, a, in[11] + K1, 19);
1388
1389 /* Round 2 */
1390 ROUND(G, a, b, c, d, in[ 1] + K2, 3);
1391 ROUND(G, d, a, b, c, in[ 3] + K2, 5);
1392 ROUND(G, c, d, a, b, in[ 5] + K2, 9);
1393 ROUND(G, b, c, d, a, in[ 7] + K2, 13);
1394 ROUND(G, a, b, c, d, in[ 9] + K2, 3);
1395 ROUND(G, d, a, b, c, in[11] + K2, 5);
1396 ROUND(G, c, d, a, b, in[ 0] + K2, 9);
1397 ROUND(G, b, c, d, a, in[ 2] + K2, 13);
1398 ROUND(G, a, b, c, d, in[ 4] + K2, 3);
1399 ROUND(G, d, a, b, c, in[ 6] + K2, 5);
1400 ROUND(G, c, d, a, b, in[ 8] + K2, 9);
1401 ROUND(G, b, c, d, a, in[10] + K2, 13);
1402
1403 /* Round 3 */
1404 ROUND(H, a, b, c, d, in[ 3] + K3, 3);
1405 ROUND(H, d, a, b, c, in[ 7] + K3, 9);
1406 ROUND(H, c, d, a, b, in[11] + K3, 11);
1407 ROUND(H, b, c, d, a, in[ 2] + K3, 15);
1408 ROUND(H, a, b, c, d, in[ 6] + K3, 3);
1409 ROUND(H, d, a, b, c, in[10] + K3, 9);
1410 ROUND(H, c, d, a, b, in[ 1] + K3, 11);
1411 ROUND(H, b, c, d, a, in[ 5] + K3, 15);
1412 ROUND(H, a, b, c, d, in[ 9] + K3, 3);
1413 ROUND(H, d, a, b, c, in[ 0] + K3, 9);
1414 ROUND(H, c, d, a, b, in[ 4] + K3, 11);
1415 ROUND(H, b, c, d, a, in[ 8] + K3, 15);
1416
1417 return buf[1] + b; /* "most hashed" word */
1418 /* Alternative: return sum of all words? */
1419}
1420#endif
1421
1422#undef ROUND
1423#undef F
1424#undef G
1425#undef H
1426#undef K1
1427#undef K2
1428#undef K3
1429
1430/* This should not be decreased so low that ISNs wrap too fast. */
1431#define REKEY_INTERVAL (300 * HZ)
1432/*
1433 * Bit layout of the tcp sequence numbers (before adding current time):
1434 * bit 24-31: increased after every key exchange
1435 * bit 0-23: hash(source,dest)
1436 *
1437 * The implementation is similar to the algorithm described
1438 * in the Appendix of RFC 1185, except that
1439 * - it uses a 1 MHz clock instead of a 250 kHz clock
1440 * - it performs a rekey every 5 minutes, which is equivalent
1441 * to a (source,dest) tulple dependent forward jump of the
1442 * clock by 0..2^(HASH_BITS+1)
1443 *
1444 * Thus the average ISN wraparound time is 68 minutes instead of
1445 * 4.55 hours.
1446 *
1447 * SMP cleanup and lock avoidance with poor man's RCU.
1448 * Manfred Spraul <manfred@colorfullife.com>
1449 *
1450 */
1451#define COUNT_BITS 8
1452#define COUNT_MASK ((1 << COUNT_BITS) - 1)
1453#define HASH_BITS 24
1454#define HASH_MASK ((1 << HASH_BITS) - 1)
1455
1456static struct keydata {
1457 __u32 count; /* already shifted to the final position */
1458 __u32 secret[12];
1459} ____cacheline_aligned ip_keydata[2];
1460
1461static unsigned int ip_cnt;
1462
David Howells65f27f32006-11-22 14:55:48 +00001463static void rekey_seq_generator(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
David Howells65f27f32006-11-22 14:55:48 +00001465static DECLARE_DELAYED_WORK(rekey_work, rekey_seq_generator);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466
1467/*
1468 * Lock avoidance:
1469 * The ISN generation runs lockless - it's just a hash over random data.
1470 * State changes happen every 5 minutes when the random key is replaced.
1471 * Synchronization is performed by having two copies of the hash function
1472 * state and rekey_seq_generator always updates the inactive copy.
1473 * The copy is then activated by updating ip_cnt.
1474 * The implementation breaks down if someone blocks the thread
1475 * that processes SYN requests for more than 5 minutes. Should never
1476 * happen, and even if that happens only a not perfectly compliant
1477 * ISN is generated, nothing fatal.
1478 */
David Howells65f27f32006-11-22 14:55:48 +00001479static void rekey_seq_generator(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480{
1481 struct keydata *keyptr = &ip_keydata[1 ^ (ip_cnt & 1)];
1482
1483 get_random_bytes(keyptr->secret, sizeof(keyptr->secret));
1484 keyptr->count = (ip_cnt & COUNT_MASK) << HASH_BITS;
1485 smp_wmb();
1486 ip_cnt++;
1487 schedule_delayed_work(&rekey_work, REKEY_INTERVAL);
1488}
1489
1490static inline struct keydata *get_keyptr(void)
1491{
1492 struct keydata *keyptr = &ip_keydata[ip_cnt & 1];
1493
1494 smp_rmb();
1495
1496 return keyptr;
1497}
1498
1499static __init int seqgen_init(void)
1500{
1501 rekey_seq_generator(NULL);
1502 return 0;
1503}
1504late_initcall(seqgen_init);
1505
1506#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
Al Virob09b845c2006-11-14 20:52:19 -08001507__u32 secure_tcpv6_sequence_number(__be32 *saddr, __be32 *daddr,
1508 __be16 sport, __be16 dport)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 __u32 seq;
1511 __u32 hash[12];
1512 struct keydata *keyptr = get_keyptr();
1513
1514 /* The procedure is the same as for IPv4, but addresses are longer.
1515 * Thus we must use twothirdsMD4Transform.
1516 */
1517
1518 memcpy(hash, saddr, 16);
Matt Mackall90b75ee2008-04-29 01:02:55 -07001519 hash[4] = ((__force u16)sport << 16) + (__force u16)dport;
1520 memcpy(&hash[5], keyptr->secret, sizeof(__u32) * 7);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521
Al Virob09b845c2006-11-14 20:52:19 -08001522 seq = twothirdsMD4Transform((const __u32 *)daddr, hash) & HASH_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 seq += keyptr->count;
1524
Eric Dumazet6dd10a62007-11-13 21:12:14 -08001525 seq += ktime_to_ns(ktime_get_real());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526
1527 return seq;
1528}
1529EXPORT_SYMBOL(secure_tcpv6_sequence_number);
1530#endif
1531
1532/* The code below is shamelessly stolen from secure_tcp_sequence_number().
1533 * All blames to Andrey V. Savochkin <saw@msu.ru>.
1534 */
Al Virob09b845c2006-11-14 20:52:19 -08001535__u32 secure_ip_id(__be32 daddr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536{
1537 struct keydata *keyptr;
1538 __u32 hash[4];
1539
1540 keyptr = get_keyptr();
1541
1542 /*
1543 * Pick a unique starting offset for each IP destination.
1544 * The dest ip address is placed in the starting vector,
1545 * which is then hashed with random data.
1546 */
Al Virob09b845c2006-11-14 20:52:19 -08001547 hash[0] = (__force __u32)daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 hash[1] = keyptr->secret[9];
1549 hash[2] = keyptr->secret[10];
1550 hash[3] = keyptr->secret[11];
1551
1552 return half_md4_transform(hash, keyptr->secret);
1553}
1554
1555#ifdef CONFIG_INET
1556
Al Virob09b845c2006-11-14 20:52:19 -08001557__u32 secure_tcp_sequence_number(__be32 saddr, __be32 daddr,
1558 __be16 sport, __be16 dport)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 __u32 seq;
1561 __u32 hash[4];
1562 struct keydata *keyptr = get_keyptr();
1563
1564 /*
1565 * Pick a unique starting offset for each TCP connection endpoints
1566 * (saddr, daddr, sport, dport).
1567 * Note that the words are placed into the starting vector, which is
1568 * then mixed with a partial MD4 over random data.
1569 */
Matt Mackall90b75ee2008-04-29 01:02:55 -07001570 hash[0] = (__force u32)saddr;
1571 hash[1] = (__force u32)daddr;
1572 hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
1573 hash[3] = keyptr->secret[11];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574
1575 seq = half_md4_transform(hash, keyptr->secret) & HASH_MASK;
1576 seq += keyptr->count;
1577 /*
1578 * As close as possible to RFC 793, which
1579 * suggests using a 250 kHz clock.
1580 * Further reading shows this assumes 2 Mb/s networks.
Eric Dumazet9b42c332007-10-01 13:58:36 -07001581 * For 10 Mb/s Ethernet, a 1 MHz clock is appropriate.
1582 * For 10 Gb/s Ethernet, a 1 GHz clock should be ok, but
1583 * we also need to limit the resolution so that the u32 seq
1584 * overlaps less than one time per MSL (2 minutes).
1585 * Choosing a clock of 64 ns period is OK. (period of 274 s)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 */
Eric Dumazet6dd10a62007-11-13 21:12:14 -08001587 seq += ktime_to_ns(ktime_get_real()) >> 6;
Matt Mackall90b75ee2008-04-29 01:02:55 -07001588
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 return seq;
1590}
1591
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -08001592/* Generate secure starting point for ephemeral IPV4 transport port search */
Al Virob09b845c2006-11-14 20:52:19 -08001593u32 secure_ipv4_port_ephemeral(__be32 saddr, __be32 daddr, __be16 dport)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594{
1595 struct keydata *keyptr = get_keyptr();
1596 u32 hash[4];
1597
1598 /*
1599 * Pick a unique starting offset for each ephemeral port search
1600 * (saddr, daddr, dport) and 48bits of random data.
1601 */
Al Virob09b845c2006-11-14 20:52:19 -08001602 hash[0] = (__force u32)saddr;
1603 hash[1] = (__force u32)daddr;
1604 hash[2] = (__force u32)dport ^ keyptr->secret[10];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 hash[3] = keyptr->secret[11];
1606
1607 return half_md4_transform(hash, keyptr->secret);
1608}
Stephen Hemminger9f593652008-08-18 21:32:32 -07001609EXPORT_SYMBOL_GPL(secure_ipv4_port_ephemeral);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610
1611#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
Matt Mackall90b75ee2008-04-29 01:02:55 -07001612u32 secure_ipv6_port_ephemeral(const __be32 *saddr, const __be32 *daddr,
1613 __be16 dport)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614{
1615 struct keydata *keyptr = get_keyptr();
1616 u32 hash[12];
1617
1618 memcpy(hash, saddr, 16);
Al Virob09b845c2006-11-14 20:52:19 -08001619 hash[4] = (__force u32)dport;
Matt Mackall90b75ee2008-04-29 01:02:55 -07001620 memcpy(&hash[5], keyptr->secret, sizeof(__u32) * 7);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621
Al Virob09b845c2006-11-14 20:52:19 -08001622 return twothirdsMD4Transform((const __u32 *)daddr, hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624#endif
1625
Arnaldo Carvalho de Meloc4365c92005-08-09 20:12:30 -07001626#if defined(CONFIG_IP_DCCP) || defined(CONFIG_IP_DCCP_MODULE)
1627/* Similar to secure_tcp_sequence_number but generate a 48 bit value
1628 * bit's 32-47 increase every key exchange
1629 * 0-31 hash(source, dest)
1630 */
Al Virob09b845c2006-11-14 20:52:19 -08001631u64 secure_dccp_sequence_number(__be32 saddr, __be32 daddr,
1632 __be16 sport, __be16 dport)
Arnaldo Carvalho de Meloc4365c92005-08-09 20:12:30 -07001633{
Arnaldo Carvalho de Meloc4365c92005-08-09 20:12:30 -07001634 u64 seq;
1635 __u32 hash[4];
1636 struct keydata *keyptr = get_keyptr();
1637
Al Virob09b845c2006-11-14 20:52:19 -08001638 hash[0] = (__force u32)saddr;
1639 hash[1] = (__force u32)daddr;
1640 hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
Arnaldo Carvalho de Meloc4365c92005-08-09 20:12:30 -07001641 hash[3] = keyptr->secret[11];
1642
1643 seq = half_md4_transform(hash, keyptr->secret);
1644 seq |= ((u64)keyptr->count) << (32 - HASH_BITS);
1645
Eric Dumazet6dd10a62007-11-13 21:12:14 -08001646 seq += ktime_to_ns(ktime_get_real());
Arnaldo Carvalho de Meloc4365c92005-08-09 20:12:30 -07001647 seq &= (1ull << 48) - 1;
Matt Mackall90b75ee2008-04-29 01:02:55 -07001648
Arnaldo Carvalho de Meloc4365c92005-08-09 20:12:30 -07001649 return seq;
1650}
Arnaldo Carvalho de Meloc4365c92005-08-09 20:12:30 -07001651EXPORT_SYMBOL(secure_dccp_sequence_number);
1652#endif
1653
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654#endif /* CONFIG_INET */
1655
1656
1657/*
1658 * Get a random word for internal kernel use only. Similar to urandom but
1659 * with the goal of minimal entropy pool depletion. As a result, the random
1660 * value is not cryptographically secure but for several uses the cost of
1661 * depleting entropy is too high
1662 */
1663unsigned int get_random_int(void)
1664{
1665 /*
1666 * Use IP's RNG. It suits our purpose perfectly: it re-keys itself
1667 * every second, from the entropy pool (and thus creates a limited
1668 * drain on it), and uses halfMD4Transform within the second. We
1669 * also mix it with jiffies and the PID:
1670 */
Al Virob09b845c2006-11-14 20:52:19 -08001671 return secure_ip_id((__force __be32)(current->pid + jiffies));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672}
1673
1674/*
1675 * randomize_range() returns a start address such that
1676 *
1677 * [...... <range> .....]
1678 * start end
1679 *
1680 * a <range> with size "len" starting at the return value is inside in the
1681 * area defined by [start, end], but is otherwise randomized.
1682 */
1683unsigned long
1684randomize_range(unsigned long start, unsigned long end, unsigned long len)
1685{
1686 unsigned long range = end - len - start;
1687
1688 if (end <= start + len)
1689 return 0;
1690 return PAGE_ALIGN(get_random_int() % range + start);
1691}