blob: 60c9c7ee6b2cea1ff8d6854d36fdf8c7d0660276 [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;
Andrew Morton8b76f462008-09-02 14:36:14 -0700410 int entropy_count; /* Must at no time exceed ->POOLBITS! */
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 Lu3060d6f2008-08-19 20:50:08 -0700561#ifndef CONFIG_HAVE_SPARSE_IRQ
Yinghai Lueef1de72008-08-19 20:49:58 -0700562
563#ifdef CONFIG_HAVE_DYN_ARRAY
564static struct timer_rand_state **irq_timer_state;
565DEFINE_DYN_ARRAY(irq_timer_state, sizeof(struct timer_rand_state *), nr_irqs, PAGE_SIZE, NULL);
566#else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567static struct timer_rand_state *irq_timer_state[NR_IRQS];
Yinghai Lueef1de72008-08-19 20:49:58 -0700568#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
Yinghai Lu3060d6f2008-08-19 20:50:08 -0700570static struct timer_rand_state *get_timer_rand_state(unsigned int irq)
571{
572 if (irq >= nr_irqs)
573 return NULL;
574
575 return irq_timer_state[irq];
576}
577
578static void set_timer_rand_state(unsigned int irq, struct timer_rand_state *state)
579{
580 if (irq >= nr_irqs)
581 return;
582
583 irq_timer_state[irq] = state;
584}
585
586#else
587
588static struct timer_rand_state *get_timer_rand_state(unsigned int irq)
589{
590 struct irq_desc *desc;
591
592 desc = irq_to_desc(irq);
593
594 if (!desc)
595 return NULL;
596
597 return desc->timer_rand_state;
598}
599
600static void set_timer_rand_state(unsigned int irq, struct timer_rand_state *state)
601{
602 struct irq_desc *desc;
603
604 desc = irq_to_desc(irq);
605
606 if (!desc)
607 return;
608
609 desc->timer_rand_state = state;
610}
611#endif
612
613static struct timer_rand_state input_timer_state;
614
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615/*
616 * This function adds entropy to the entropy "pool" by using timing
617 * delays. It uses the timer_rand_state structure to make an estimate
618 * of how many bits of entropy this call has added to the pool.
619 *
620 * The number "num" is also added to the pool - it should somehow describe
621 * the type of event which just happened. This is currently 0-255 for
622 * keyboard scan codes, and 256 upwards for interrupts.
623 *
624 */
625static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
626{
627 struct {
628 cycles_t cycles;
629 long jiffies;
630 unsigned num;
631 } sample;
632 long delta, delta2, delta3;
633
634 preempt_disable();
635 /* if over the trickle threshold, use only 1 in 4096 samples */
636 if (input_pool.entropy_count > trickle_thresh &&
637 (__get_cpu_var(trickle_count)++ & 0xfff))
638 goto out;
639
640 sample.jiffies = jiffies;
641 sample.cycles = get_cycles();
642 sample.num = num;
Matt Mackalle68e5b62008-04-29 01:03:05 -0700643 mix_pool_bytes(&input_pool, &sample, sizeof(sample));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
645 /*
646 * Calculate number of bits of randomness we probably added.
647 * We take into account the first, second and third-order deltas
648 * in order to make our estimate.
649 */
650
651 if (!state->dont_count_entropy) {
652 delta = sample.jiffies - state->last_time;
653 state->last_time = sample.jiffies;
654
655 delta2 = delta - state->last_delta;
656 state->last_delta = delta;
657
658 delta3 = delta2 - state->last_delta2;
659 state->last_delta2 = delta2;
660
661 if (delta < 0)
662 delta = -delta;
663 if (delta2 < 0)
664 delta2 = -delta2;
665 if (delta3 < 0)
666 delta3 = -delta3;
667 if (delta > delta2)
668 delta = delta2;
669 if (delta > delta3)
670 delta = delta3;
671
672 /*
673 * delta is now minimum absolute delta.
674 * Round down by 1 bit on general principles,
675 * and limit entropy entimate to 12 bits.
676 */
Matt Mackalladc782d2008-04-29 01:03:07 -0700677 credit_entropy_bits(&input_pool,
678 min_t(int, fls(delta>>1), 11));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680out:
681 preempt_enable();
682}
683
Stephen Hemmingerd2515752006-01-11 12:17:38 -0800684void add_input_randomness(unsigned int type, unsigned int code,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 unsigned int value)
686{
687 static unsigned char last_value;
688
689 /* ignore autorepeat and the like */
690 if (value == last_value)
691 return;
692
693 DEBUG_ENT("input event\n");
694 last_value = value;
695 add_timer_randomness(&input_timer_state,
696 (type << 4) ^ code ^ (code >> 4) ^ value);
697}
Dmitry Torokhov80fc9f52006-10-11 01:43:58 -0400698EXPORT_SYMBOL_GPL(add_input_randomness);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
700void add_interrupt_randomness(int irq)
701{
Yinghai Lu3060d6f2008-08-19 20:50:08 -0700702 struct timer_rand_state *state;
703
704 state = get_timer_rand_state(irq);
705
706 if (state == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 return;
708
709 DEBUG_ENT("irq event %d\n", irq);
Yinghai Lu3060d6f2008-08-19 20:50:08 -0700710 add_timer_randomness(state, 0x100 + irq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711}
712
David Howells93614012006-09-30 20:45:40 +0200713#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714void add_disk_randomness(struct gendisk *disk)
715{
716 if (!disk || !disk->random)
717 return;
718 /* first major is 1, so we get >= 0x200 here */
Tejun Heof331c022008-09-03 09:01:48 +0200719 DEBUG_ENT("disk event %d:%d\n",
720 MAJOR(disk_devt(disk)), MINOR(disk_devt(disk)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
Tejun Heof331c022008-09-03 09:01:48 +0200722 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723}
David Howells93614012006-09-30 20:45:40 +0200724#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
726#define EXTRACT_SIZE 10
727
728/*********************************************************************
729 *
730 * Entropy extraction routines
731 *
732 *********************************************************************/
733
Matt Mackall90b75ee2008-04-29 01:02:55 -0700734static ssize_t extract_entropy(struct entropy_store *r, void *buf,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 size_t nbytes, int min, int rsvd);
736
737/*
738 * This utility inline function is responsible for transfering entropy
739 * from the primary pool to the secondary extraction pool. We make
740 * sure we pull enough for a 'catastrophic reseed'.
741 */
742static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
743{
744 __u32 tmp[OUTPUT_POOL_WORDS];
745
746 if (r->pull && r->entropy_count < nbytes * 8 &&
747 r->entropy_count < r->poolinfo->POOLBITS) {
Matt Mackall5a021e92007-07-19 11:30:14 -0700748 /* If we're limited, always leave two wakeup worth's BITS */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 int rsvd = r->limit ? 0 : random_read_wakeup_thresh/4;
Matt Mackall5a021e92007-07-19 11:30:14 -0700750 int bytes = nbytes;
751
752 /* pull at least as many as BYTES as wakeup BITS */
753 bytes = max_t(int, bytes, random_read_wakeup_thresh / 8);
754 /* but never more than the buffer size */
755 bytes = min_t(int, bytes, sizeof(tmp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
757 DEBUG_ENT("going to reseed %s with %d bits "
758 "(%d of %d requested)\n",
759 r->name, bytes * 8, nbytes * 8, r->entropy_count);
760
Matt Mackall90b75ee2008-04-29 01:02:55 -0700761 bytes = extract_entropy(r->pull, tmp, bytes,
762 random_read_wakeup_thresh / 8, rsvd);
Matt Mackalle68e5b62008-04-29 01:03:05 -0700763 mix_pool_bytes(r, tmp, bytes);
Matt Mackalladc782d2008-04-29 01:03:07 -0700764 credit_entropy_bits(r, bytes*8);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 }
766}
767
768/*
769 * These functions extracts randomness from the "entropy pool", and
770 * returns it in a buffer.
771 *
772 * The min parameter specifies the minimum amount we can pull before
773 * failing to avoid races that defeat catastrophic reseeding while the
774 * reserved parameter indicates how much entropy we must leave in the
775 * pool after each pull to avoid starving other readers.
776 *
777 * Note: extract_entropy() assumes that .poolwords is a multiple of 16 words.
778 */
779
780static size_t account(struct entropy_store *r, size_t nbytes, int min,
781 int reserved)
782{
783 unsigned long flags;
784
785 BUG_ON(r->entropy_count > r->poolinfo->POOLBITS);
786
787 /* Hold lock while accounting */
788 spin_lock_irqsave(&r->lock, flags);
789
790 DEBUG_ENT("trying to extract %d bits from %s\n",
791 nbytes * 8, r->name);
792
793 /* Can we pull enough? */
794 if (r->entropy_count / 8 < min + reserved) {
795 nbytes = 0;
796 } else {
797 /* If limited, never pull more than available */
798 if (r->limit && nbytes + reserved >= r->entropy_count / 8)
799 nbytes = r->entropy_count/8 - reserved;
800
Matt Mackall90b75ee2008-04-29 01:02:55 -0700801 if (r->entropy_count / 8 >= nbytes + reserved)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 r->entropy_count -= nbytes*8;
803 else
804 r->entropy_count = reserved;
805
Jeff Dike9a6f70b2008-04-29 01:03:08 -0700806 if (r->entropy_count < random_write_wakeup_thresh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 wake_up_interruptible(&random_write_wait);
Jeff Dike9a6f70b2008-04-29 01:03:08 -0700808 kill_fasync(&fasync, SIGIO, POLL_OUT);
809 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 }
811
812 DEBUG_ENT("debiting %d entropy credits from %s%s\n",
813 nbytes * 8, r->name, r->limit ? "" : " (unlimited)");
814
815 spin_unlock_irqrestore(&r->lock, flags);
816
817 return nbytes;
818}
819
820static void extract_buf(struct entropy_store *r, __u8 *out)
821{
Matt Mackall602b6ae2007-05-29 21:54:27 -0500822 int i;
Matt Mackalle68e5b62008-04-29 01:03:05 -0700823 __u32 hash[5], workspace[SHA_WORKSPACE_WORDS];
824 __u8 extract[64];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
Matt Mackall1c0ad3d2008-04-29 01:03:00 -0700826 /* Generate a hash across the pool, 16 words (512 bits) at a time */
Matt Mackallffd8d3f2008-04-29 01:02:59 -0700827 sha_init(hash);
Matt Mackall1c0ad3d2008-04-29 01:03:00 -0700828 for (i = 0; i < r->poolinfo->poolwords; i += 16)
Matt Mackallffd8d3f2008-04-29 01:02:59 -0700829 sha_transform(hash, (__u8 *)(r->pool + i), workspace);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
831 /*
Matt Mackall1c0ad3d2008-04-29 01:03:00 -0700832 * We mix the hash back into the pool to prevent backtracking
833 * attacks (where the attacker knows the state of the pool
834 * plus the current outputs, and attempts to find previous
835 * ouputs), unless the hash function can be inverted. By
836 * mixing at least a SHA1 worth of hash data back, we make
837 * brute-forcing the feedback as hard as brute-forcing the
838 * hash.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 */
Matt Mackalle68e5b62008-04-29 01:03:05 -0700840 mix_pool_bytes_extract(r, hash, sizeof(hash), extract);
Matt Mackall1c0ad3d2008-04-29 01:03:00 -0700841
842 /*
843 * To avoid duplicates, we atomically extract a portion of the
844 * pool while mixing, and hash one final time.
845 */
Matt Mackalle68e5b62008-04-29 01:03:05 -0700846 sha_transform(hash, extract, workspace);
Matt Mackallffd8d3f2008-04-29 01:02:59 -0700847 memset(extract, 0, sizeof(extract));
848 memset(workspace, 0, sizeof(workspace));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
850 /*
Matt Mackall1c0ad3d2008-04-29 01:03:00 -0700851 * In case the hash function has some recognizable output
852 * pattern, we fold it in half. Thus, we always feed back
853 * twice as much data as we output.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 */
Matt Mackallffd8d3f2008-04-29 01:02:59 -0700855 hash[0] ^= hash[3];
856 hash[1] ^= hash[4];
857 hash[2] ^= rol32(hash[2], 16);
858 memcpy(out, hash, EXTRACT_SIZE);
859 memset(hash, 0, sizeof(hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860}
861
Matt Mackall90b75ee2008-04-29 01:02:55 -0700862static ssize_t extract_entropy(struct entropy_store *r, void *buf,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 size_t nbytes, int min, int reserved)
864{
865 ssize_t ret = 0, i;
866 __u8 tmp[EXTRACT_SIZE];
867
868 xfer_secondary_pool(r, nbytes);
869 nbytes = account(r, nbytes, min, reserved);
870
871 while (nbytes) {
872 extract_buf(r, tmp);
873 i = min_t(int, nbytes, EXTRACT_SIZE);
874 memcpy(buf, tmp, i);
875 nbytes -= i;
876 buf += i;
877 ret += i;
878 }
879
880 /* Wipe data just returned from memory */
881 memset(tmp, 0, sizeof(tmp));
882
883 return ret;
884}
885
886static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
887 size_t nbytes)
888{
889 ssize_t ret = 0, i;
890 __u8 tmp[EXTRACT_SIZE];
891
892 xfer_secondary_pool(r, nbytes);
893 nbytes = account(r, nbytes, 0, 0);
894
895 while (nbytes) {
896 if (need_resched()) {
897 if (signal_pending(current)) {
898 if (ret == 0)
899 ret = -ERESTARTSYS;
900 break;
901 }
902 schedule();
903 }
904
905 extract_buf(r, tmp);
906 i = min_t(int, nbytes, EXTRACT_SIZE);
907 if (copy_to_user(buf, tmp, i)) {
908 ret = -EFAULT;
909 break;
910 }
911
912 nbytes -= i;
913 buf += i;
914 ret += i;
915 }
916
917 /* Wipe data just returned from memory */
918 memset(tmp, 0, sizeof(tmp));
919
920 return ret;
921}
922
923/*
924 * This function is the exported kernel interface. It returns some
925 * number of good random numbers, suitable for seeding TCP sequence
926 * numbers, etc.
927 */
928void get_random_bytes(void *buf, int nbytes)
929{
930 extract_entropy(&nonblocking_pool, buf, nbytes, 0, 0);
931}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932EXPORT_SYMBOL(get_random_bytes);
933
934/*
935 * init_std_data - initialize pool with system data
936 *
937 * @r: pool to initialize
938 *
939 * This function clears the pool's entropy count and mixes some system
940 * data into the pool to prepare it for use. The pool is not cleared
941 * as that can only decrease the entropy in the pool.
942 */
943static void init_std_data(struct entropy_store *r)
944{
Eric Dumazetf8595812007-03-28 14:22:33 -0700945 ktime_t now;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 unsigned long flags;
947
948 spin_lock_irqsave(&r->lock, flags);
949 r->entropy_count = 0;
950 spin_unlock_irqrestore(&r->lock, flags);
951
Eric Dumazetf8595812007-03-28 14:22:33 -0700952 now = ktime_get_real();
Matt Mackalle68e5b62008-04-29 01:03:05 -0700953 mix_pool_bytes(r, &now, sizeof(now));
954 mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955}
956
Matt Mackall53c3f632008-04-29 01:02:58 -0700957static int rand_initialize(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958{
959 init_std_data(&input_pool);
960 init_std_data(&blocking_pool);
961 init_std_data(&nonblocking_pool);
962 return 0;
963}
964module_init(rand_initialize);
965
966void rand_initialize_irq(int irq)
967{
968 struct timer_rand_state *state;
969
Yinghai Lu3060d6f2008-08-19 20:50:08 -0700970#ifndef CONFIG_HAVE_SPARSE_IRQ
971 if (irq >= nr_irqs)
972 return;
973#endif
974
975 state = get_timer_rand_state(irq);
976
977 if (state)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 return;
979
980 /*
Eric Dumazetf8595812007-03-28 14:22:33 -0700981 * If kzalloc returns null, we just won't use that entropy
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 * source.
983 */
Eric Dumazetf8595812007-03-28 14:22:33 -0700984 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
985 if (state)
Yinghai Lu3060d6f2008-08-19 20:50:08 -0700986 set_timer_rand_state(irq, state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987}
988
David Howells93614012006-09-30 20:45:40 +0200989#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990void rand_initialize_disk(struct gendisk *disk)
991{
992 struct timer_rand_state *state;
993
994 /*
Eric Dumazetf8595812007-03-28 14:22:33 -0700995 * If kzalloc returns null, we just won't use that entropy
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 * source.
997 */
Eric Dumazetf8595812007-03-28 14:22:33 -0700998 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
999 if (state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 disk->random = state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001}
David Howells93614012006-09-30 20:45:40 +02001002#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003
1004static ssize_t
Matt Mackall90b75ee2008-04-29 01:02:55 -07001005random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006{
1007 ssize_t n, retval = 0, count = 0;
1008
1009 if (nbytes == 0)
1010 return 0;
1011
1012 while (nbytes > 0) {
1013 n = nbytes;
1014 if (n > SEC_XFER_SIZE)
1015 n = SEC_XFER_SIZE;
1016
1017 DEBUG_ENT("reading %d bits\n", n*8);
1018
1019 n = extract_entropy_user(&blocking_pool, buf, n);
1020
1021 DEBUG_ENT("read got %d bits (%d still needed)\n",
1022 n*8, (nbytes-n)*8);
1023
1024 if (n == 0) {
1025 if (file->f_flags & O_NONBLOCK) {
1026 retval = -EAGAIN;
1027 break;
1028 }
1029
1030 DEBUG_ENT("sleeping?\n");
1031
1032 wait_event_interruptible(random_read_wait,
1033 input_pool.entropy_count >=
1034 random_read_wakeup_thresh);
1035
1036 DEBUG_ENT("awake\n");
1037
1038 if (signal_pending(current)) {
1039 retval = -ERESTARTSYS;
1040 break;
1041 }
1042
1043 continue;
1044 }
1045
1046 if (n < 0) {
1047 retval = n;
1048 break;
1049 }
1050 count += n;
1051 buf += n;
1052 nbytes -= n;
1053 break; /* This break makes the device work */
1054 /* like a named pipe */
1055 }
1056
1057 /*
1058 * If we gave the user some bytes, update the access time.
1059 */
1060 if (count)
1061 file_accessed(file);
1062
1063 return (count ? count : retval);
1064}
1065
1066static ssize_t
Matt Mackall90b75ee2008-04-29 01:02:55 -07001067urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068{
1069 return extract_entropy_user(&nonblocking_pool, buf, nbytes);
1070}
1071
1072static unsigned int
1073random_poll(struct file *file, poll_table * wait)
1074{
1075 unsigned int mask;
1076
1077 poll_wait(file, &random_read_wait, wait);
1078 poll_wait(file, &random_write_wait, wait);
1079 mask = 0;
1080 if (input_pool.entropy_count >= random_read_wakeup_thresh)
1081 mask |= POLLIN | POLLRDNORM;
1082 if (input_pool.entropy_count < random_write_wakeup_thresh)
1083 mask |= POLLOUT | POLLWRNORM;
1084 return mask;
1085}
1086
Matt Mackall7f397dc2007-05-29 21:58:10 -05001087static int
1088write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1089{
1090 size_t bytes;
1091 __u32 buf[16];
1092 const char __user *p = buffer;
1093
1094 while (count > 0) {
1095 bytes = min(count, sizeof(buf));
1096 if (copy_from_user(&buf, p, bytes))
1097 return -EFAULT;
1098
1099 count -= bytes;
1100 p += bytes;
1101
Matt Mackalle68e5b62008-04-29 01:03:05 -07001102 mix_pool_bytes(r, buf, bytes);
Matt Mackall91f3f1e2008-02-06 01:37:20 -08001103 cond_resched();
Matt Mackall7f397dc2007-05-29 21:58:10 -05001104 }
1105
1106 return 0;
1107}
1108
Matt Mackall90b75ee2008-04-29 01:02:55 -07001109static ssize_t random_write(struct file *file, const char __user *buffer,
1110 size_t count, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111{
Matt Mackall7f397dc2007-05-29 21:58:10 -05001112 size_t ret;
1113 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
Matt Mackall7f397dc2007-05-29 21:58:10 -05001115 ret = write_pool(&blocking_pool, buffer, count);
1116 if (ret)
1117 return ret;
1118 ret = write_pool(&nonblocking_pool, buffer, count);
1119 if (ret)
1120 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121
Matt Mackall7f397dc2007-05-29 21:58:10 -05001122 inode->i_mtime = current_fs_time(inode->i_sb);
1123 mark_inode_dirty(inode);
1124 return (ssize_t)count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125}
1126
Matt Mackall43ae4862008-04-29 01:02:58 -07001127static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128{
1129 int size, ent_count;
1130 int __user *p = (int __user *)arg;
1131 int retval;
1132
1133 switch (cmd) {
1134 case RNDGETENTCNT:
Matt Mackall43ae4862008-04-29 01:02:58 -07001135 /* inherently racy, no point locking */
1136 if (put_user(input_pool.entropy_count, p))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 return -EFAULT;
1138 return 0;
1139 case RNDADDTOENTCNT:
1140 if (!capable(CAP_SYS_ADMIN))
1141 return -EPERM;
1142 if (get_user(ent_count, p))
1143 return -EFAULT;
Matt Mackalladc782d2008-04-29 01:03:07 -07001144 credit_entropy_bits(&input_pool, ent_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 return 0;
1146 case RNDADDENTROPY:
1147 if (!capable(CAP_SYS_ADMIN))
1148 return -EPERM;
1149 if (get_user(ent_count, p++))
1150 return -EFAULT;
1151 if (ent_count < 0)
1152 return -EINVAL;
1153 if (get_user(size, p++))
1154 return -EFAULT;
Matt Mackall7f397dc2007-05-29 21:58:10 -05001155 retval = write_pool(&input_pool, (const char __user *)p,
1156 size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 if (retval < 0)
1158 return retval;
Matt Mackalladc782d2008-04-29 01:03:07 -07001159 credit_entropy_bits(&input_pool, ent_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 return 0;
1161 case RNDZAPENTCNT:
1162 case RNDCLEARPOOL:
1163 /* Clear the entropy pool counters. */
1164 if (!capable(CAP_SYS_ADMIN))
1165 return -EPERM;
Matt Mackall53c3f632008-04-29 01:02:58 -07001166 rand_initialize();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 return 0;
1168 default:
1169 return -EINVAL;
1170 }
1171}
1172
Jeff Dike9a6f70b2008-04-29 01:03:08 -07001173static int random_fasync(int fd, struct file *filp, int on)
1174{
1175 return fasync_helper(fd, filp, on, &fasync);
1176}
1177
1178static int random_release(struct inode *inode, struct file *filp)
1179{
1180 return fasync_helper(-1, filp, 0, &fasync);
1181}
1182
Arjan van de Ven2b8693c2007-02-12 00:55:32 -08001183const struct file_operations random_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 .read = random_read,
1185 .write = random_write,
1186 .poll = random_poll,
Matt Mackall43ae4862008-04-29 01:02:58 -07001187 .unlocked_ioctl = random_ioctl,
Jeff Dike9a6f70b2008-04-29 01:03:08 -07001188 .fasync = random_fasync,
1189 .release = random_release,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190};
1191
Arjan van de Ven2b8693c2007-02-12 00:55:32 -08001192const struct file_operations urandom_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 .read = urandom_read,
1194 .write = random_write,
Matt Mackall43ae4862008-04-29 01:02:58 -07001195 .unlocked_ioctl = random_ioctl,
Jeff Dike9a6f70b2008-04-29 01:03:08 -07001196 .fasync = random_fasync,
1197 .release = random_release,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198};
1199
1200/***************************************************************
1201 * Random UUID interface
1202 *
1203 * Used here for a Boot ID, but can be useful for other kernel
1204 * drivers.
1205 ***************************************************************/
1206
1207/*
1208 * Generate random UUID
1209 */
1210void generate_random_uuid(unsigned char uuid_out[16])
1211{
1212 get_random_bytes(uuid_out, 16);
1213 /* Set UUID version to 4 --- truely random generation */
1214 uuid_out[6] = (uuid_out[6] & 0x0F) | 0x40;
1215 /* Set the UUID variant to DCE */
1216 uuid_out[8] = (uuid_out[8] & 0x3F) | 0x80;
1217}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218EXPORT_SYMBOL(generate_random_uuid);
1219
1220/********************************************************************
1221 *
1222 * Sysctl interface
1223 *
1224 ********************************************************************/
1225
1226#ifdef CONFIG_SYSCTL
1227
1228#include <linux/sysctl.h>
1229
1230static int min_read_thresh = 8, min_write_thresh;
1231static int max_read_thresh = INPUT_POOL_WORDS * 32;
1232static int max_write_thresh = INPUT_POOL_WORDS * 32;
1233static char sysctl_bootid[16];
1234
1235/*
1236 * These functions is used to return both the bootid UUID, and random
1237 * UUID. The difference is in whether table->data is NULL; if it is,
1238 * then a new UUID is generated and returned to the user.
1239 *
1240 * If the user accesses this via the proc interface, it will be returned
1241 * as an ASCII string in the standard UUID format. If accesses via the
1242 * sysctl system call, it is returned as 16 bytes of binary data.
1243 */
1244static int proc_do_uuid(ctl_table *table, int write, struct file *filp,
1245 void __user *buffer, size_t *lenp, loff_t *ppos)
1246{
1247 ctl_table fake_table;
1248 unsigned char buf[64], tmp_uuid[16], *uuid;
1249
1250 uuid = table->data;
1251 if (!uuid) {
1252 uuid = tmp_uuid;
1253 uuid[8] = 0;
1254 }
1255 if (uuid[8] == 0)
1256 generate_random_uuid(uuid);
1257
1258 sprintf(buf, "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-"
1259 "%02x%02x%02x%02x%02x%02x",
1260 uuid[0], uuid[1], uuid[2], uuid[3],
1261 uuid[4], uuid[5], uuid[6], uuid[7],
1262 uuid[8], uuid[9], uuid[10], uuid[11],
1263 uuid[12], uuid[13], uuid[14], uuid[15]);
1264 fake_table.data = buf;
1265 fake_table.maxlen = sizeof(buf);
1266
1267 return proc_dostring(&fake_table, write, filp, buffer, lenp, ppos);
1268}
1269
1270static int uuid_strategy(ctl_table *table, int __user *name, int nlen,
1271 void __user *oldval, size_t __user *oldlenp,
Alexey Dobriyan1f29bcd2006-12-10 02:19:10 -08001272 void __user *newval, size_t newlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273{
1274 unsigned char tmp_uuid[16], *uuid;
1275 unsigned int len;
1276
1277 if (!oldval || !oldlenp)
1278 return 1;
1279
1280 uuid = table->data;
1281 if (!uuid) {
1282 uuid = tmp_uuid;
1283 uuid[8] = 0;
1284 }
1285 if (uuid[8] == 0)
1286 generate_random_uuid(uuid);
1287
1288 if (get_user(len, oldlenp))
1289 return -EFAULT;
1290 if (len) {
1291 if (len > 16)
1292 len = 16;
1293 if (copy_to_user(oldval, uuid, len) ||
1294 put_user(len, oldlenp))
1295 return -EFAULT;
1296 }
1297 return 1;
1298}
1299
1300static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
1301ctl_table random_table[] = {
1302 {
1303 .ctl_name = RANDOM_POOLSIZE,
1304 .procname = "poolsize",
1305 .data = &sysctl_poolsize,
1306 .maxlen = sizeof(int),
1307 .mode = 0444,
1308 .proc_handler = &proc_dointvec,
1309 },
1310 {
1311 .ctl_name = RANDOM_ENTROPY_COUNT,
1312 .procname = "entropy_avail",
1313 .maxlen = sizeof(int),
1314 .mode = 0444,
1315 .proc_handler = &proc_dointvec,
1316 .data = &input_pool.entropy_count,
1317 },
1318 {
1319 .ctl_name = RANDOM_READ_THRESH,
1320 .procname = "read_wakeup_threshold",
1321 .data = &random_read_wakeup_thresh,
1322 .maxlen = sizeof(int),
1323 .mode = 0644,
1324 .proc_handler = &proc_dointvec_minmax,
1325 .strategy = &sysctl_intvec,
1326 .extra1 = &min_read_thresh,
1327 .extra2 = &max_read_thresh,
1328 },
1329 {
1330 .ctl_name = RANDOM_WRITE_THRESH,
1331 .procname = "write_wakeup_threshold",
1332 .data = &random_write_wakeup_thresh,
1333 .maxlen = sizeof(int),
1334 .mode = 0644,
1335 .proc_handler = &proc_dointvec_minmax,
1336 .strategy = &sysctl_intvec,
1337 .extra1 = &min_write_thresh,
1338 .extra2 = &max_write_thresh,
1339 },
1340 {
1341 .ctl_name = RANDOM_BOOT_ID,
1342 .procname = "boot_id",
1343 .data = &sysctl_bootid,
1344 .maxlen = 16,
1345 .mode = 0444,
1346 .proc_handler = &proc_do_uuid,
1347 .strategy = &uuid_strategy,
1348 },
1349 {
1350 .ctl_name = RANDOM_UUID,
1351 .procname = "uuid",
1352 .maxlen = 16,
1353 .mode = 0444,
1354 .proc_handler = &proc_do_uuid,
1355 .strategy = &uuid_strategy,
1356 },
1357 { .ctl_name = 0 }
1358};
1359#endif /* CONFIG_SYSCTL */
1360
1361/********************************************************************
1362 *
1363 * Random funtions for networking
1364 *
1365 ********************************************************************/
1366
1367/*
1368 * TCP initial sequence number picking. This uses the random number
1369 * generator to pick an initial secret value. This value is hashed
1370 * along with the TCP endpoint information to provide a unique
1371 * starting point for each pair of TCP endpoints. This defeats
1372 * attacks which rely on guessing the initial TCP sequence number.
1373 * This algorithm was suggested by Steve Bellovin.
1374 *
1375 * Using a very strong hash was taking an appreciable amount of the total
1376 * TCP connection establishment time, so this is a weaker hash,
1377 * compensated for by changing the secret periodically.
1378 */
1379
1380/* F, G and H are basic MD4 functions: selection, majority, parity */
1381#define F(x, y, z) ((z) ^ ((x) & ((y) ^ (z))))
1382#define G(x, y, z) (((x) & (y)) + (((x) ^ (y)) & (z)))
1383#define H(x, y, z) ((x) ^ (y) ^ (z))
1384
1385/*
1386 * The generic round function. The application is so specific that
1387 * we don't bother protecting all the arguments with parens, as is generally
1388 * good macro practice, in favor of extra legibility.
1389 * Rotation is separate from addition to prevent recomputation
1390 */
1391#define ROUND(f, a, b, c, d, x, s) \
1392 (a += f(b, c, d) + x, a = (a << s) | (a >> (32 - s)))
1393#define K1 0
1394#define K2 013240474631UL
1395#define K3 015666365641UL
1396
1397#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1398
Matt Mackall90b75ee2008-04-29 01:02:55 -07001399static __u32 twothirdsMD4Transform(__u32 const buf[4], __u32 const in[12])
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400{
1401 __u32 a = buf[0], b = buf[1], c = buf[2], d = buf[3];
1402
1403 /* Round 1 */
1404 ROUND(F, a, b, c, d, in[ 0] + K1, 3);
1405 ROUND(F, d, a, b, c, in[ 1] + K1, 7);
1406 ROUND(F, c, d, a, b, in[ 2] + K1, 11);
1407 ROUND(F, b, c, d, a, in[ 3] + K1, 19);
1408 ROUND(F, a, b, c, d, in[ 4] + K1, 3);
1409 ROUND(F, d, a, b, c, in[ 5] + K1, 7);
1410 ROUND(F, c, d, a, b, in[ 6] + K1, 11);
1411 ROUND(F, b, c, d, a, in[ 7] + K1, 19);
1412 ROUND(F, a, b, c, d, in[ 8] + K1, 3);
1413 ROUND(F, d, a, b, c, in[ 9] + K1, 7);
1414 ROUND(F, c, d, a, b, in[10] + K1, 11);
1415 ROUND(F, b, c, d, a, in[11] + K1, 19);
1416
1417 /* Round 2 */
1418 ROUND(G, a, b, c, d, in[ 1] + K2, 3);
1419 ROUND(G, d, a, b, c, in[ 3] + K2, 5);
1420 ROUND(G, c, d, a, b, in[ 5] + K2, 9);
1421 ROUND(G, b, c, d, a, in[ 7] + K2, 13);
1422 ROUND(G, a, b, c, d, in[ 9] + K2, 3);
1423 ROUND(G, d, a, b, c, in[11] + K2, 5);
1424 ROUND(G, c, d, a, b, in[ 0] + K2, 9);
1425 ROUND(G, b, c, d, a, in[ 2] + K2, 13);
1426 ROUND(G, a, b, c, d, in[ 4] + K2, 3);
1427 ROUND(G, d, a, b, c, in[ 6] + K2, 5);
1428 ROUND(G, c, d, a, b, in[ 8] + K2, 9);
1429 ROUND(G, b, c, d, a, in[10] + K2, 13);
1430
1431 /* Round 3 */
1432 ROUND(H, a, b, c, d, in[ 3] + K3, 3);
1433 ROUND(H, d, a, b, c, in[ 7] + K3, 9);
1434 ROUND(H, c, d, a, b, in[11] + K3, 11);
1435 ROUND(H, b, c, d, a, in[ 2] + K3, 15);
1436 ROUND(H, a, b, c, d, in[ 6] + K3, 3);
1437 ROUND(H, d, a, b, c, in[10] + K3, 9);
1438 ROUND(H, c, d, a, b, in[ 1] + K3, 11);
1439 ROUND(H, b, c, d, a, in[ 5] + K3, 15);
1440 ROUND(H, a, b, c, d, in[ 9] + K3, 3);
1441 ROUND(H, d, a, b, c, in[ 0] + K3, 9);
1442 ROUND(H, c, d, a, b, in[ 4] + K3, 11);
1443 ROUND(H, b, c, d, a, in[ 8] + K3, 15);
1444
1445 return buf[1] + b; /* "most hashed" word */
1446 /* Alternative: return sum of all words? */
1447}
1448#endif
1449
1450#undef ROUND
1451#undef F
1452#undef G
1453#undef H
1454#undef K1
1455#undef K2
1456#undef K3
1457
1458/* This should not be decreased so low that ISNs wrap too fast. */
1459#define REKEY_INTERVAL (300 * HZ)
1460/*
1461 * Bit layout of the tcp sequence numbers (before adding current time):
1462 * bit 24-31: increased after every key exchange
1463 * bit 0-23: hash(source,dest)
1464 *
1465 * The implementation is similar to the algorithm described
1466 * in the Appendix of RFC 1185, except that
1467 * - it uses a 1 MHz clock instead of a 250 kHz clock
1468 * - it performs a rekey every 5 minutes, which is equivalent
1469 * to a (source,dest) tulple dependent forward jump of the
1470 * clock by 0..2^(HASH_BITS+1)
1471 *
1472 * Thus the average ISN wraparound time is 68 minutes instead of
1473 * 4.55 hours.
1474 *
1475 * SMP cleanup and lock avoidance with poor man's RCU.
1476 * Manfred Spraul <manfred@colorfullife.com>
1477 *
1478 */
1479#define COUNT_BITS 8
1480#define COUNT_MASK ((1 << COUNT_BITS) - 1)
1481#define HASH_BITS 24
1482#define HASH_MASK ((1 << HASH_BITS) - 1)
1483
1484static struct keydata {
1485 __u32 count; /* already shifted to the final position */
1486 __u32 secret[12];
1487} ____cacheline_aligned ip_keydata[2];
1488
1489static unsigned int ip_cnt;
1490
David Howells65f27f32006-11-22 14:55:48 +00001491static void rekey_seq_generator(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
David Howells65f27f32006-11-22 14:55:48 +00001493static DECLARE_DELAYED_WORK(rekey_work, rekey_seq_generator);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
1495/*
1496 * Lock avoidance:
1497 * The ISN generation runs lockless - it's just a hash over random data.
1498 * State changes happen every 5 minutes when the random key is replaced.
1499 * Synchronization is performed by having two copies of the hash function
1500 * state and rekey_seq_generator always updates the inactive copy.
1501 * The copy is then activated by updating ip_cnt.
1502 * The implementation breaks down if someone blocks the thread
1503 * that processes SYN requests for more than 5 minutes. Should never
1504 * happen, and even if that happens only a not perfectly compliant
1505 * ISN is generated, nothing fatal.
1506 */
David Howells65f27f32006-11-22 14:55:48 +00001507static void rekey_seq_generator(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508{
1509 struct keydata *keyptr = &ip_keydata[1 ^ (ip_cnt & 1)];
1510
1511 get_random_bytes(keyptr->secret, sizeof(keyptr->secret));
1512 keyptr->count = (ip_cnt & COUNT_MASK) << HASH_BITS;
1513 smp_wmb();
1514 ip_cnt++;
1515 schedule_delayed_work(&rekey_work, REKEY_INTERVAL);
1516}
1517
1518static inline struct keydata *get_keyptr(void)
1519{
1520 struct keydata *keyptr = &ip_keydata[ip_cnt & 1];
1521
1522 smp_rmb();
1523
1524 return keyptr;
1525}
1526
1527static __init int seqgen_init(void)
1528{
1529 rekey_seq_generator(NULL);
1530 return 0;
1531}
1532late_initcall(seqgen_init);
1533
1534#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
Al Virob09b845c2006-11-14 20:52:19 -08001535__u32 secure_tcpv6_sequence_number(__be32 *saddr, __be32 *daddr,
1536 __be16 sport, __be16 dport)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 __u32 seq;
1539 __u32 hash[12];
1540 struct keydata *keyptr = get_keyptr();
1541
1542 /* The procedure is the same as for IPv4, but addresses are longer.
1543 * Thus we must use twothirdsMD4Transform.
1544 */
1545
1546 memcpy(hash, saddr, 16);
Matt Mackall90b75ee2008-04-29 01:02:55 -07001547 hash[4] = ((__force u16)sport << 16) + (__force u16)dport;
1548 memcpy(&hash[5], keyptr->secret, sizeof(__u32) * 7);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549
Al Virob09b845c2006-11-14 20:52:19 -08001550 seq = twothirdsMD4Transform((const __u32 *)daddr, hash) & HASH_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 seq += keyptr->count;
1552
Eric Dumazet6dd10a62007-11-13 21:12:14 -08001553 seq += ktime_to_ns(ktime_get_real());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
1555 return seq;
1556}
1557EXPORT_SYMBOL(secure_tcpv6_sequence_number);
1558#endif
1559
1560/* The code below is shamelessly stolen from secure_tcp_sequence_number().
1561 * All blames to Andrey V. Savochkin <saw@msu.ru>.
1562 */
Al Virob09b845c2006-11-14 20:52:19 -08001563__u32 secure_ip_id(__be32 daddr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564{
1565 struct keydata *keyptr;
1566 __u32 hash[4];
1567
1568 keyptr = get_keyptr();
1569
1570 /*
1571 * Pick a unique starting offset for each IP destination.
1572 * The dest ip address is placed in the starting vector,
1573 * which is then hashed with random data.
1574 */
Al Virob09b845c2006-11-14 20:52:19 -08001575 hash[0] = (__force __u32)daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 hash[1] = keyptr->secret[9];
1577 hash[2] = keyptr->secret[10];
1578 hash[3] = keyptr->secret[11];
1579
1580 return half_md4_transform(hash, keyptr->secret);
1581}
1582
1583#ifdef CONFIG_INET
1584
Al Virob09b845c2006-11-14 20:52:19 -08001585__u32 secure_tcp_sequence_number(__be32 saddr, __be32 daddr,
1586 __be16 sport, __be16 dport)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 __u32 seq;
1589 __u32 hash[4];
1590 struct keydata *keyptr = get_keyptr();
1591
1592 /*
1593 * Pick a unique starting offset for each TCP connection endpoints
1594 * (saddr, daddr, sport, dport).
1595 * Note that the words are placed into the starting vector, which is
1596 * then mixed with a partial MD4 over random data.
1597 */
Matt Mackall90b75ee2008-04-29 01:02:55 -07001598 hash[0] = (__force u32)saddr;
1599 hash[1] = (__force u32)daddr;
1600 hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
1601 hash[3] = keyptr->secret[11];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602
1603 seq = half_md4_transform(hash, keyptr->secret) & HASH_MASK;
1604 seq += keyptr->count;
1605 /*
1606 * As close as possible to RFC 793, which
1607 * suggests using a 250 kHz clock.
1608 * Further reading shows this assumes 2 Mb/s networks.
Eric Dumazet9b42c332007-10-01 13:58:36 -07001609 * For 10 Mb/s Ethernet, a 1 MHz clock is appropriate.
1610 * For 10 Gb/s Ethernet, a 1 GHz clock should be ok, but
1611 * we also need to limit the resolution so that the u32 seq
1612 * overlaps less than one time per MSL (2 minutes).
1613 * Choosing a clock of 64 ns period is OK. (period of 274 s)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 */
Eric Dumazet6dd10a62007-11-13 21:12:14 -08001615 seq += ktime_to_ns(ktime_get_real()) >> 6;
Matt Mackall90b75ee2008-04-29 01:02:55 -07001616
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 return seq;
1618}
1619
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -08001620/* Generate secure starting point for ephemeral IPV4 transport port search */
Al Virob09b845c2006-11-14 20:52:19 -08001621u32 secure_ipv4_port_ephemeral(__be32 saddr, __be32 daddr, __be16 dport)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622{
1623 struct keydata *keyptr = get_keyptr();
1624 u32 hash[4];
1625
1626 /*
1627 * Pick a unique starting offset for each ephemeral port search
1628 * (saddr, daddr, dport) and 48bits of random data.
1629 */
Al Virob09b845c2006-11-14 20:52:19 -08001630 hash[0] = (__force u32)saddr;
1631 hash[1] = (__force u32)daddr;
1632 hash[2] = (__force u32)dport ^ keyptr->secret[10];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 hash[3] = keyptr->secret[11];
1634
1635 return half_md4_transform(hash, keyptr->secret);
1636}
Stephen Hemminger9f593652008-08-18 21:32:32 -07001637EXPORT_SYMBOL_GPL(secure_ipv4_port_ephemeral);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638
1639#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
Matt Mackall90b75ee2008-04-29 01:02:55 -07001640u32 secure_ipv6_port_ephemeral(const __be32 *saddr, const __be32 *daddr,
1641 __be16 dport)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642{
1643 struct keydata *keyptr = get_keyptr();
1644 u32 hash[12];
1645
1646 memcpy(hash, saddr, 16);
Al Virob09b845c2006-11-14 20:52:19 -08001647 hash[4] = (__force u32)dport;
Matt Mackall90b75ee2008-04-29 01:02:55 -07001648 memcpy(&hash[5], keyptr->secret, sizeof(__u32) * 7);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649
Al Virob09b845c2006-11-14 20:52:19 -08001650 return twothirdsMD4Transform((const __u32 *)daddr, hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652#endif
1653
Arnaldo Carvalho de Meloc4365c92005-08-09 20:12:30 -07001654#if defined(CONFIG_IP_DCCP) || defined(CONFIG_IP_DCCP_MODULE)
1655/* Similar to secure_tcp_sequence_number but generate a 48 bit value
1656 * bit's 32-47 increase every key exchange
1657 * 0-31 hash(source, dest)
1658 */
Al Virob09b845c2006-11-14 20:52:19 -08001659u64 secure_dccp_sequence_number(__be32 saddr, __be32 daddr,
1660 __be16 sport, __be16 dport)
Arnaldo Carvalho de Meloc4365c92005-08-09 20:12:30 -07001661{
Arnaldo Carvalho de Meloc4365c92005-08-09 20:12:30 -07001662 u64 seq;
1663 __u32 hash[4];
1664 struct keydata *keyptr = get_keyptr();
1665
Al Virob09b845c2006-11-14 20:52:19 -08001666 hash[0] = (__force u32)saddr;
1667 hash[1] = (__force u32)daddr;
1668 hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
Arnaldo Carvalho de Meloc4365c92005-08-09 20:12:30 -07001669 hash[3] = keyptr->secret[11];
1670
1671 seq = half_md4_transform(hash, keyptr->secret);
1672 seq |= ((u64)keyptr->count) << (32 - HASH_BITS);
1673
Eric Dumazet6dd10a62007-11-13 21:12:14 -08001674 seq += ktime_to_ns(ktime_get_real());
Arnaldo Carvalho de Meloc4365c92005-08-09 20:12:30 -07001675 seq &= (1ull << 48) - 1;
Matt Mackall90b75ee2008-04-29 01:02:55 -07001676
Arnaldo Carvalho de Meloc4365c92005-08-09 20:12:30 -07001677 return seq;
1678}
Arnaldo Carvalho de Meloc4365c92005-08-09 20:12:30 -07001679EXPORT_SYMBOL(secure_dccp_sequence_number);
1680#endif
1681
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682#endif /* CONFIG_INET */
1683
1684
1685/*
1686 * Get a random word for internal kernel use only. Similar to urandom but
1687 * with the goal of minimal entropy pool depletion. As a result, the random
1688 * value is not cryptographically secure but for several uses the cost of
1689 * depleting entropy is too high
1690 */
1691unsigned int get_random_int(void)
1692{
1693 /*
1694 * Use IP's RNG. It suits our purpose perfectly: it re-keys itself
1695 * every second, from the entropy pool (and thus creates a limited
1696 * drain on it), and uses halfMD4Transform within the second. We
1697 * also mix it with jiffies and the PID:
1698 */
Al Virob09b845c2006-11-14 20:52:19 -08001699 return secure_ip_id((__force __be32)(current->pid + jiffies));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700}
1701
1702/*
1703 * randomize_range() returns a start address such that
1704 *
1705 * [...... <range> .....]
1706 * start end
1707 *
1708 * a <range> with size "len" starting at the return value is inside in the
1709 * area defined by [start, end], but is otherwise randomized.
1710 */
1711unsigned long
1712randomize_range(unsigned long start, unsigned long end, unsigned long len)
1713{
1714 unsigned long range = end - len - start;
1715
1716 if (end <= start + len)
1717 return 0;
1718 return PAGE_ALIGN(get_random_int() % range + start);
1719}