ALSA: virtuoso: fix silent analog output on Xonar Essence ST Deluxe
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / char / random.c
blobd4ddeba56682f129bbcfab8145c253286f1fc6a2
1 /*
2 * random.c -- A strong random number generator
4 * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005
6 * Copyright Theodore Ts'o, 1994, 1995, 1996, 1997, 1998, 1999. All
7 * rights reserved.
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.
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.)
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.
43 * (now, with legal B.S. out of the way.....)
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.
52 * Theory of operation
53 * ===================
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.
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.
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.
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.
98 * Exported interfaces ---- output
99 * ===============================
101 * There are three exported interfaces; the first is one designed to
102 * be used from within the kernel:
104 * void get_random_bytes(void *buf, int nbytes);
106 * This interface will return the requested number of random bytes,
107 * and place it in the requested buffer.
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.
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.
122 * Exported interfaces ---- input
123 * ==============================
125 * The current exported interfaces for gathering environmental noise
126 * from the devices are:
128 * void add_input_randomness(unsigned int type, unsigned int code,
129 * unsigned int value);
130 * void add_interrupt_randomness(int irq);
131 * void add_disk_randomness(struct gendisk *disk);
133 * add_input_randomness() uses the input layer interrupt timing, as well as
134 * the event type information from the hardware.
136 * add_interrupt_randomness() uses the inter-interrupt timing as random
137 * inputs to the entropy pool. Note that not all interrupts are good
138 * sources of randomness! For example, the timer interrupts is not a
139 * good choice, because the periodicity of the interrupts is too
140 * regular, and hence predictable to an attacker. Network Interface
141 * Controller interrupts are a better measure, since the timing of the
142 * NIC interrupts are more unpredictable.
144 * add_disk_randomness() uses what amounts to the seek time of block
145 * layer request events, on a per-disk_devt basis, as input to the
146 * entropy pool. Note that high-speed solid state drives with very low
147 * seek times do not make for good sources of entropy, as their seek
148 * times are usually fairly consistent.
150 * All of these routines try to estimate how many bits of randomness a
151 * particular randomness source. They do this by keeping track of the
152 * first and second order deltas of the event timings.
154 * Ensuring unpredictability at system startup
155 * ============================================
157 * When any operating system starts up, it will go through a sequence
158 * of actions that are fairly predictable by an adversary, especially
159 * if the start-up does not involve interaction with a human operator.
160 * This reduces the actual number of bits of unpredictability in the
161 * entropy pool below the value in entropy_count. In order to
162 * counteract this effect, it helps to carry information in the
163 * entropy pool across shut-downs and start-ups. To do this, put the
164 * following lines an appropriate script which is run during the boot
165 * sequence:
167 * echo "Initializing random number generator..."
168 * random_seed=/var/run/random-seed
169 * # Carry a random seed from start-up to start-up
170 * # Load and then save the whole entropy pool
171 * if [ -f $random_seed ]; then
172 * cat $random_seed >/dev/urandom
173 * else
174 * touch $random_seed
175 * fi
176 * chmod 600 $random_seed
177 * dd if=/dev/urandom of=$random_seed count=1 bs=512
179 * and the following lines in an appropriate script which is run as
180 * the system is shutdown:
182 * # Carry a random seed from shut-down to start-up
183 * # Save the whole entropy pool
184 * echo "Saving random seed..."
185 * random_seed=/var/run/random-seed
186 * touch $random_seed
187 * chmod 600 $random_seed
188 * dd if=/dev/urandom of=$random_seed count=1 bs=512
190 * For example, on most modern systems using the System V init
191 * scripts, such code fragments would be found in
192 * /etc/rc.d/init.d/random. On older Linux systems, the correct script
193 * location might be in /etc/rcb.d/rc.local or /etc/rc.d/rc.0.
195 * Effectively, these commands cause the contents of the entropy pool
196 * to be saved at shut-down time and reloaded into the entropy pool at
197 * start-up. (The 'dd' in the addition to the bootup script is to
198 * make sure that /etc/random-seed is different for every start-up,
199 * even if the system crashes without executing rc.0.) Even with
200 * complete knowledge of the start-up activities, predicting the state
201 * of the entropy pool requires knowledge of the previous history of
202 * the system.
204 * Configuring the /dev/random driver under Linux
205 * ==============================================
207 * The /dev/random driver under Linux uses minor numbers 8 and 9 of
208 * the /dev/mem major number (#1). So if your system does not have
209 * /dev/random and /dev/urandom created already, they can be created
210 * by using the commands:
212 * mknod /dev/random c 1 8
213 * mknod /dev/urandom c 1 9
215 * Acknowledgements:
216 * =================
218 * Ideas for constructing this random number generator were derived
219 * from Pretty Good Privacy's random number generator, and from private
220 * discussions with Phil Karn. Colin Plumb provided a faster random
221 * number generator, which speed up the mixing function of the entropy
222 * pool, taken from PGPfone. Dale Worley has also contributed many
223 * useful ideas and suggestions to improve this driver.
225 * Any flaws in the design are solely my responsibility, and should
226 * not be attributed to the Phil, Colin, or any of authors of PGP.
228 * Further background information on this topic may be obtained from
229 * RFC 1750, "Randomness Recommendations for Security", by Donald
230 * Eastlake, Steve Crocker, and Jeff Schiller.
233 #include <linux/utsname.h>
234 #include <linux/module.h>
235 #include <linux/kernel.h>
236 #include <linux/major.h>
237 #include <linux/string.h>
238 #include <linux/fcntl.h>
239 #include <linux/slab.h>
240 #include <linux/random.h>
241 #include <linux/poll.h>
242 #include <linux/init.h>
243 #include <linux/fs.h>
244 #include <linux/genhd.h>
245 #include <linux/interrupt.h>
246 #include <linux/mm.h>
247 #include <linux/spinlock.h>
248 #include <linux/percpu.h>
249 #include <linux/cryptohash.h>
250 #include <linux/fips.h>
252 #ifdef CONFIG_GENERIC_HARDIRQS
253 # include <linux/irq.h>
254 #endif
256 #include <asm/processor.h>
257 #include <asm/uaccess.h>
258 #include <asm/irq.h>
259 #include <asm/io.h>
262 * Configuration information
264 #define INPUT_POOL_WORDS 128
265 #define OUTPUT_POOL_WORDS 32
266 #define SEC_XFER_SIZE 512
267 #define EXTRACT_SIZE 10
270 * The minimum number of bits of entropy before we wake up a read on
271 * /dev/random. Should be enough to do a significant reseed.
273 static int random_read_wakeup_thresh = 64;
276 * If the entropy count falls under this number of bits, then we
277 * should wake up processes which are selecting or polling on write
278 * access to /dev/random.
280 static int random_write_wakeup_thresh = 128;
283 * When the input pool goes over trickle_thresh, start dropping most
284 * samples to avoid wasting CPU time and reduce lock contention.
287 static int trickle_thresh __read_mostly = INPUT_POOL_WORDS * 28;
289 static DEFINE_PER_CPU(int, trickle_count);
292 * A pool of size .poolwords is stirred with a primitive polynomial
293 * of degree .poolwords over GF(2). The taps for various sizes are
294 * defined below. They are chosen to be evenly spaced (minimum RMS
295 * distance from evenly spaced; the numbers in the comments are a
296 * scaled squared error sum) except for the last tap, which is 1 to
297 * get the twisting happening as fast as possible.
299 static struct poolinfo {
300 int poolwords;
301 int tap1, tap2, tap3, tap4, tap5;
302 } poolinfo_table[] = {
303 /* x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 -- 105 */
304 { 128, 103, 76, 51, 25, 1 },
305 /* x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 -- 15 */
306 { 32, 26, 20, 14, 7, 1 },
307 #if 0
308 /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1 -- 115 */
309 { 2048, 1638, 1231, 819, 411, 1 },
311 /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
312 { 1024, 817, 615, 412, 204, 1 },
314 /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
315 { 1024, 819, 616, 410, 207, 2 },
317 /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
318 { 512, 411, 308, 208, 104, 1 },
320 /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
321 { 512, 409, 307, 206, 102, 2 },
322 /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
323 { 512, 409, 309, 205, 103, 2 },
325 /* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
326 { 256, 205, 155, 101, 52, 1 },
328 /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
329 { 128, 103, 78, 51, 27, 2 },
331 /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
332 { 64, 52, 39, 26, 14, 1 },
333 #endif
336 #define POOLBITS poolwords*32
337 #define POOLBYTES poolwords*4
340 * For the purposes of better mixing, we use the CRC-32 polynomial as
341 * well to make a twisted Generalized Feedback Shift Reigster
343 * (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR generators. ACM
344 * Transactions on Modeling and Computer Simulation 2(3):179-194.
345 * Also see M. Matsumoto & Y. Kurita, 1994. Twisted GFSR generators
346 * II. ACM Transactions on Mdeling and Computer Simulation 4:254-266)
348 * Thanks to Colin Plumb for suggesting this.
350 * We have not analyzed the resultant polynomial to prove it primitive;
351 * in fact it almost certainly isn't. Nonetheless, the irreducible factors
352 * of a random large-degree polynomial over GF(2) are more than large enough
353 * that periodicity is not a concern.
355 * The input hash is much less sensitive than the output hash. All
356 * that we want of it is that it be a good non-cryptographic hash;
357 * i.e. it not produce collisions when fed "random" data of the sort
358 * we expect to see. As long as the pool state differs for different
359 * inputs, we have preserved the input entropy and done a good job.
360 * The fact that an intelligent attacker can construct inputs that
361 * will produce controlled alterations to the pool's state is not
362 * important because we don't consider such inputs to contribute any
363 * randomness. The only property we need with respect to them is that
364 * the attacker can't increase his/her knowledge of the pool's state.
365 * Since all additions are reversible (knowing the final state and the
366 * input, you can reconstruct the initial state), if an attacker has
367 * any uncertainty about the initial state, he/she can only shuffle
368 * that uncertainty about, but never cause any collisions (which would
369 * decrease the uncertainty).
371 * The chosen system lets the state of the pool be (essentially) the input
372 * modulo the generator polymnomial. Now, for random primitive polynomials,
373 * this is a universal class of hash functions, meaning that the chance
374 * of a collision is limited by the attacker's knowledge of the generator
375 * polynomail, so if it is chosen at random, an attacker can never force
376 * a collision. Here, we use a fixed polynomial, but we *can* assume that
377 * ###--> it is unknown to the processes generating the input entropy. <-###
378 * Because of this important property, this is a good, collision-resistant
379 * hash; hash collisions will occur no more often than chance.
383 * Static global variables
385 static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
386 static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
387 static struct fasync_struct *fasync;
389 #if 0
390 static int debug;
391 module_param(debug, bool, 0644);
392 #define DEBUG_ENT(fmt, arg...) do { \
393 if (debug) \
394 printk(KERN_DEBUG "random %04d %04d %04d: " \
395 fmt,\
396 input_pool.entropy_count,\
397 blocking_pool.entropy_count,\
398 nonblocking_pool.entropy_count,\
399 ## arg); } while (0)
400 #else
401 #define DEBUG_ENT(fmt, arg...) do {} while (0)
402 #endif
404 /**********************************************************************
406 * OS independent entropy store. Here are the functions which handle
407 * storing entropy in an entropy pool.
409 **********************************************************************/
411 struct entropy_store;
412 struct entropy_store {
413 /* read-only data: */
414 struct poolinfo *poolinfo;
415 __u32 *pool;
416 const char *name;
417 struct entropy_store *pull;
418 int limit;
420 /* read-write data: */
421 spinlock_t lock;
422 unsigned add_ptr;
423 int entropy_count;
424 int input_rotate;
425 __u8 last_data[EXTRACT_SIZE];
428 static __u32 input_pool_data[INPUT_POOL_WORDS];
429 static __u32 blocking_pool_data[OUTPUT_POOL_WORDS];
430 static __u32 nonblocking_pool_data[OUTPUT_POOL_WORDS];
432 static struct entropy_store input_pool = {
433 .poolinfo = &poolinfo_table[0],
434 .name = "input",
435 .limit = 1,
436 .lock = __SPIN_LOCK_UNLOCKED(&input_pool.lock),
437 .pool = input_pool_data
440 static struct entropy_store blocking_pool = {
441 .poolinfo = &poolinfo_table[1],
442 .name = "blocking",
443 .limit = 1,
444 .pull = &input_pool,
445 .lock = __SPIN_LOCK_UNLOCKED(&blocking_pool.lock),
446 .pool = blocking_pool_data
449 static struct entropy_store nonblocking_pool = {
450 .poolinfo = &poolinfo_table[1],
451 .name = "nonblocking",
452 .pull = &input_pool,
453 .lock = __SPIN_LOCK_UNLOCKED(&nonblocking_pool.lock),
454 .pool = nonblocking_pool_data
458 * This function adds bytes into the entropy "pool". It does not
459 * update the entropy estimate. The caller should call
460 * credit_entropy_bits if this is appropriate.
462 * The pool is stirred with a primitive polynomial of the appropriate
463 * degree, and then twisted. We twist by three bits at a time because
464 * it's cheap to do so and helps slightly in the expected case where
465 * the entropy is concentrated in the low-order bits.
467 static void mix_pool_bytes_extract(struct entropy_store *r, const void *in,
468 int nbytes, __u8 out[64])
470 static __u32 const twist_table[8] = {
471 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
472 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
473 unsigned long i, j, tap1, tap2, tap3, tap4, tap5;
474 int input_rotate;
475 int wordmask = r->poolinfo->poolwords - 1;
476 const char *bytes = in;
477 __u32 w;
478 unsigned long flags;
480 /* Taps are constant, so we can load them without holding r->lock. */
481 tap1 = r->poolinfo->tap1;
482 tap2 = r->poolinfo->tap2;
483 tap3 = r->poolinfo->tap3;
484 tap4 = r->poolinfo->tap4;
485 tap5 = r->poolinfo->tap5;
487 spin_lock_irqsave(&r->lock, flags);
488 input_rotate = r->input_rotate;
489 i = r->add_ptr;
491 /* mix one byte at a time to simplify size handling and churn faster */
492 while (nbytes--) {
493 w = rol32(*bytes++, input_rotate & 31);
494 i = (i - 1) & wordmask;
496 /* XOR in the various taps */
497 w ^= r->pool[i];
498 w ^= r->pool[(i + tap1) & wordmask];
499 w ^= r->pool[(i + tap2) & wordmask];
500 w ^= r->pool[(i + tap3) & wordmask];
501 w ^= r->pool[(i + tap4) & wordmask];
502 w ^= r->pool[(i + tap5) & wordmask];
504 /* Mix the result back in with a twist */
505 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
508 * Normally, we add 7 bits of rotation to the pool.
509 * At the beginning of the pool, add an extra 7 bits
510 * rotation, so that successive passes spread the
511 * input bits across the pool evenly.
513 input_rotate += i ? 7 : 14;
516 r->input_rotate = input_rotate;
517 r->add_ptr = i;
519 if (out)
520 for (j = 0; j < 16; j++)
521 ((__u32 *)out)[j] = r->pool[(i - j) & wordmask];
523 spin_unlock_irqrestore(&r->lock, flags);
526 static void mix_pool_bytes(struct entropy_store *r, const void *in, int bytes)
528 mix_pool_bytes_extract(r, in, bytes, NULL);
532 * Credit (or debit) the entropy store with n bits of entropy
534 static void credit_entropy_bits(struct entropy_store *r, int nbits)
536 unsigned long flags;
537 int entropy_count;
539 if (!nbits)
540 return;
542 spin_lock_irqsave(&r->lock, flags);
544 DEBUG_ENT("added %d entropy credits to %s\n", nbits, r->name);
545 entropy_count = r->entropy_count;
546 entropy_count += nbits;
547 if (entropy_count < 0) {
548 DEBUG_ENT("negative entropy/overflow\n");
549 entropy_count = 0;
550 } else if (entropy_count > r->poolinfo->POOLBITS)
551 entropy_count = r->poolinfo->POOLBITS;
552 r->entropy_count = entropy_count;
554 /* should we wake readers? */
555 if (r == &input_pool && entropy_count >= random_read_wakeup_thresh) {
556 wake_up_interruptible(&random_read_wait);
557 kill_fasync(&fasync, SIGIO, POLL_IN);
559 spin_unlock_irqrestore(&r->lock, flags);
562 /*********************************************************************
564 * Entropy input management
566 *********************************************************************/
568 /* There is one of these per entropy source */
569 struct timer_rand_state {
570 cycles_t last_time;
571 long last_delta, last_delta2;
572 unsigned dont_count_entropy:1;
575 #ifndef CONFIG_GENERIC_HARDIRQS
577 static struct timer_rand_state *irq_timer_state[NR_IRQS];
579 static struct timer_rand_state *get_timer_rand_state(unsigned int irq)
581 return irq_timer_state[irq];
584 static void set_timer_rand_state(unsigned int irq,
585 struct timer_rand_state *state)
587 irq_timer_state[irq] = state;
590 #else
592 static struct timer_rand_state *get_timer_rand_state(unsigned int irq)
594 struct irq_desc *desc;
596 desc = irq_to_desc(irq);
598 return desc->timer_rand_state;
601 static void set_timer_rand_state(unsigned int irq,
602 struct timer_rand_state *state)
604 struct irq_desc *desc;
606 desc = irq_to_desc(irq);
608 desc->timer_rand_state = state;
610 #endif
612 static struct timer_rand_state input_timer_state;
615 * This function adds entropy to the entropy "pool" by using timing
616 * delays. It uses the timer_rand_state structure to make an estimate
617 * of how many bits of entropy this call has added to the pool.
619 * The number "num" is also added to the pool - it should somehow describe
620 * the type of event which just happened. This is currently 0-255 for
621 * keyboard scan codes, and 256 upwards for interrupts.
624 static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
626 struct {
627 cycles_t cycles;
628 long jiffies;
629 unsigned num;
630 } sample;
631 long delta, delta2, delta3;
633 preempt_disable();
634 /* if over the trickle threshold, use only 1 in 4096 samples */
635 if (input_pool.entropy_count > trickle_thresh &&
636 ((__this_cpu_inc_return(trickle_count) - 1) & 0xfff))
637 goto out;
639 sample.jiffies = jiffies;
640 sample.cycles = get_cycles();
641 sample.num = num;
642 mix_pool_bytes(&input_pool, &sample, sizeof(sample));
645 * Calculate number of bits of randomness we probably added.
646 * We take into account the first, second and third-order deltas
647 * in order to make our estimate.
650 if (!state->dont_count_entropy) {
651 delta = sample.jiffies - state->last_time;
652 state->last_time = sample.jiffies;
654 delta2 = delta - state->last_delta;
655 state->last_delta = delta;
657 delta3 = delta2 - state->last_delta2;
658 state->last_delta2 = delta2;
660 if (delta < 0)
661 delta = -delta;
662 if (delta2 < 0)
663 delta2 = -delta2;
664 if (delta3 < 0)
665 delta3 = -delta3;
666 if (delta > delta2)
667 delta = delta2;
668 if (delta > delta3)
669 delta = delta3;
672 * delta is now minimum absolute delta.
673 * Round down by 1 bit on general principles,
674 * and limit entropy entimate to 12 bits.
676 credit_entropy_bits(&input_pool,
677 min_t(int, fls(delta>>1), 11));
679 out:
680 preempt_enable();
683 void add_input_randomness(unsigned int type, unsigned int code,
684 unsigned int value)
686 static unsigned char last_value;
688 /* ignore autorepeat and the like */
689 if (value == last_value)
690 return;
692 DEBUG_ENT("input event\n");
693 last_value = value;
694 add_timer_randomness(&input_timer_state,
695 (type << 4) ^ code ^ (code >> 4) ^ value);
697 EXPORT_SYMBOL_GPL(add_input_randomness);
699 void add_interrupt_randomness(int irq)
701 struct timer_rand_state *state;
703 state = get_timer_rand_state(irq);
705 if (state == NULL)
706 return;
708 DEBUG_ENT("irq event %d\n", irq);
709 add_timer_randomness(state, 0x100 + irq);
712 #ifdef CONFIG_BLOCK
713 void add_disk_randomness(struct gendisk *disk)
715 if (!disk || !disk->random)
716 return;
717 /* first major is 1, so we get >= 0x200 here */
718 DEBUG_ENT("disk event %d:%d\n",
719 MAJOR(disk_devt(disk)), MINOR(disk_devt(disk)));
721 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
723 #endif
725 /*********************************************************************
727 * Entropy extraction routines
729 *********************************************************************/
731 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
732 size_t nbytes, int min, int rsvd);
735 * This utility inline function is responsible for transferring entropy
736 * from the primary pool to the secondary extraction pool. We make
737 * sure we pull enough for a 'catastrophic reseed'.
739 static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
741 __u32 tmp[OUTPUT_POOL_WORDS];
743 if (r->pull && r->entropy_count < nbytes * 8 &&
744 r->entropy_count < r->poolinfo->POOLBITS) {
745 /* If we're limited, always leave two wakeup worth's BITS */
746 int rsvd = r->limit ? 0 : random_read_wakeup_thresh/4;
747 int bytes = nbytes;
749 /* pull at least as many as BYTES as wakeup BITS */
750 bytes = max_t(int, bytes, random_read_wakeup_thresh / 8);
751 /* but never more than the buffer size */
752 bytes = min_t(int, bytes, sizeof(tmp));
754 DEBUG_ENT("going to reseed %s with %d bits "
755 "(%d of %d requested)\n",
756 r->name, bytes * 8, nbytes * 8, r->entropy_count);
758 bytes = extract_entropy(r->pull, tmp, bytes,
759 random_read_wakeup_thresh / 8, rsvd);
760 mix_pool_bytes(r, tmp, bytes);
761 credit_entropy_bits(r, bytes*8);
766 * These functions extracts randomness from the "entropy pool", and
767 * returns it in a buffer.
769 * The min parameter specifies the minimum amount we can pull before
770 * failing to avoid races that defeat catastrophic reseeding while the
771 * reserved parameter indicates how much entropy we must leave in the
772 * pool after each pull to avoid starving other readers.
774 * Note: extract_entropy() assumes that .poolwords is a multiple of 16 words.
777 static size_t account(struct entropy_store *r, size_t nbytes, int min,
778 int reserved)
780 unsigned long flags;
782 /* Hold lock while accounting */
783 spin_lock_irqsave(&r->lock, flags);
785 BUG_ON(r->entropy_count > r->poolinfo->POOLBITS);
786 DEBUG_ENT("trying to extract %d bits from %s\n",
787 nbytes * 8, r->name);
789 /* Can we pull enough? */
790 if (r->entropy_count / 8 < min + reserved) {
791 nbytes = 0;
792 } else {
793 /* If limited, never pull more than available */
794 if (r->limit && nbytes + reserved >= r->entropy_count / 8)
795 nbytes = r->entropy_count/8 - reserved;
797 if (r->entropy_count / 8 >= nbytes + reserved)
798 r->entropy_count -= nbytes*8;
799 else
800 r->entropy_count = reserved;
802 if (r->entropy_count < random_write_wakeup_thresh) {
803 wake_up_interruptible(&random_write_wait);
804 kill_fasync(&fasync, SIGIO, POLL_OUT);
808 DEBUG_ENT("debiting %d entropy credits from %s%s\n",
809 nbytes * 8, r->name, r->limit ? "" : " (unlimited)");
811 spin_unlock_irqrestore(&r->lock, flags);
813 return nbytes;
816 static void extract_buf(struct entropy_store *r, __u8 *out)
818 int i;
819 __u32 hash[5], workspace[SHA_WORKSPACE_WORDS];
820 __u8 extract[64];
822 /* Generate a hash across the pool, 16 words (512 bits) at a time */
823 sha_init(hash);
824 for (i = 0; i < r->poolinfo->poolwords; i += 16)
825 sha_transform(hash, (__u8 *)(r->pool + i), workspace);
828 * We mix the hash back into the pool to prevent backtracking
829 * attacks (where the attacker knows the state of the pool
830 * plus the current outputs, and attempts to find previous
831 * ouputs), unless the hash function can be inverted. By
832 * mixing at least a SHA1 worth of hash data back, we make
833 * brute-forcing the feedback as hard as brute-forcing the
834 * hash.
836 mix_pool_bytes_extract(r, hash, sizeof(hash), extract);
839 * To avoid duplicates, we atomically extract a portion of the
840 * pool while mixing, and hash one final time.
842 sha_transform(hash, extract, workspace);
843 memset(extract, 0, sizeof(extract));
844 memset(workspace, 0, sizeof(workspace));
847 * In case the hash function has some recognizable output
848 * pattern, we fold it in half. Thus, we always feed back
849 * twice as much data as we output.
851 hash[0] ^= hash[3];
852 hash[1] ^= hash[4];
853 hash[2] ^= rol32(hash[2], 16);
854 memcpy(out, hash, EXTRACT_SIZE);
855 memset(hash, 0, sizeof(hash));
858 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
859 size_t nbytes, int min, int reserved)
861 ssize_t ret = 0, i;
862 __u8 tmp[EXTRACT_SIZE];
863 unsigned long flags;
865 xfer_secondary_pool(r, nbytes);
866 nbytes = account(r, nbytes, min, reserved);
868 while (nbytes) {
869 extract_buf(r, tmp);
871 if (fips_enabled) {
872 spin_lock_irqsave(&r->lock, flags);
873 if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
874 panic("Hardware RNG duplicated output!\n");
875 memcpy(r->last_data, tmp, EXTRACT_SIZE);
876 spin_unlock_irqrestore(&r->lock, flags);
878 i = min_t(int, nbytes, EXTRACT_SIZE);
879 memcpy(buf, tmp, i);
880 nbytes -= i;
881 buf += i;
882 ret += i;
885 /* Wipe data just returned from memory */
886 memset(tmp, 0, sizeof(tmp));
888 return ret;
891 static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
892 size_t nbytes)
894 ssize_t ret = 0, i;
895 __u8 tmp[EXTRACT_SIZE];
897 xfer_secondary_pool(r, nbytes);
898 nbytes = account(r, nbytes, 0, 0);
900 while (nbytes) {
901 if (need_resched()) {
902 if (signal_pending(current)) {
903 if (ret == 0)
904 ret = -ERESTARTSYS;
905 break;
907 schedule();
910 extract_buf(r, tmp);
911 i = min_t(int, nbytes, EXTRACT_SIZE);
912 if (copy_to_user(buf, tmp, i)) {
913 ret = -EFAULT;
914 break;
917 nbytes -= i;
918 buf += i;
919 ret += i;
922 /* Wipe data just returned from memory */
923 memset(tmp, 0, sizeof(tmp));
925 return ret;
929 * This function is the exported kernel interface. It returns some
930 * number of good random numbers, suitable for seeding TCP sequence
931 * numbers, etc.
933 void get_random_bytes(void *buf, int nbytes)
935 extract_entropy(&nonblocking_pool, buf, nbytes, 0, 0);
937 EXPORT_SYMBOL(get_random_bytes);
940 * init_std_data - initialize pool with system data
942 * @r: pool to initialize
944 * This function clears the pool's entropy count and mixes some system
945 * data into the pool to prepare it for use. The pool is not cleared
946 * as that can only decrease the entropy in the pool.
948 static void init_std_data(struct entropy_store *r)
950 ktime_t now;
951 unsigned long flags;
953 spin_lock_irqsave(&r->lock, flags);
954 r->entropy_count = 0;
955 spin_unlock_irqrestore(&r->lock, flags);
957 now = ktime_get_real();
958 mix_pool_bytes(r, &now, sizeof(now));
959 mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
962 static int rand_initialize(void)
964 init_std_data(&input_pool);
965 init_std_data(&blocking_pool);
966 init_std_data(&nonblocking_pool);
967 return 0;
969 module_init(rand_initialize);
971 void rand_initialize_irq(int irq)
973 struct timer_rand_state *state;
975 state = get_timer_rand_state(irq);
977 if (state)
978 return;
981 * If kzalloc returns null, we just won't use that entropy
982 * source.
984 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
985 if (state)
986 set_timer_rand_state(irq, state);
989 #ifdef CONFIG_BLOCK
990 void rand_initialize_disk(struct gendisk *disk)
992 struct timer_rand_state *state;
995 * If kzalloc returns null, we just won't use that entropy
996 * source.
998 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
999 if (state)
1000 disk->random = state;
1002 #endif
1004 static ssize_t
1005 random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1007 ssize_t n, retval = 0, count = 0;
1009 if (nbytes == 0)
1010 return 0;
1012 while (nbytes > 0) {
1013 n = nbytes;
1014 if (n > SEC_XFER_SIZE)
1015 n = SEC_XFER_SIZE;
1017 DEBUG_ENT("reading %d bits\n", n*8);
1019 n = extract_entropy_user(&blocking_pool, buf, n);
1021 DEBUG_ENT("read got %d bits (%d still needed)\n",
1022 n*8, (nbytes-n)*8);
1024 if (n == 0) {
1025 if (file->f_flags & O_NONBLOCK) {
1026 retval = -EAGAIN;
1027 break;
1030 DEBUG_ENT("sleeping?\n");
1032 wait_event_interruptible(random_read_wait,
1033 input_pool.entropy_count >=
1034 random_read_wakeup_thresh);
1036 DEBUG_ENT("awake\n");
1038 if (signal_pending(current)) {
1039 retval = -ERESTARTSYS;
1040 break;
1043 continue;
1046 if (n < 0) {
1047 retval = n;
1048 break;
1050 count += n;
1051 buf += n;
1052 nbytes -= n;
1053 break; /* This break makes the device work */
1054 /* like a named pipe */
1057 return (count ? count : retval);
1060 static ssize_t
1061 urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1063 return extract_entropy_user(&nonblocking_pool, buf, nbytes);
1066 static unsigned int
1067 random_poll(struct file *file, poll_table * wait)
1069 unsigned int mask;
1071 poll_wait(file, &random_read_wait, wait);
1072 poll_wait(file, &random_write_wait, wait);
1073 mask = 0;
1074 if (input_pool.entropy_count >= random_read_wakeup_thresh)
1075 mask |= POLLIN | POLLRDNORM;
1076 if (input_pool.entropy_count < random_write_wakeup_thresh)
1077 mask |= POLLOUT | POLLWRNORM;
1078 return mask;
1081 static int
1082 write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1084 size_t bytes;
1085 __u32 buf[16];
1086 const char __user *p = buffer;
1088 while (count > 0) {
1089 bytes = min(count, sizeof(buf));
1090 if (copy_from_user(&buf, p, bytes))
1091 return -EFAULT;
1093 count -= bytes;
1094 p += bytes;
1096 mix_pool_bytes(r, buf, bytes);
1097 cond_resched();
1100 return 0;
1103 static ssize_t random_write(struct file *file, const char __user *buffer,
1104 size_t count, loff_t *ppos)
1106 size_t ret;
1108 ret = write_pool(&blocking_pool, buffer, count);
1109 if (ret)
1110 return ret;
1111 ret = write_pool(&nonblocking_pool, buffer, count);
1112 if (ret)
1113 return ret;
1115 return (ssize_t)count;
1118 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1120 int size, ent_count;
1121 int __user *p = (int __user *)arg;
1122 int retval;
1124 switch (cmd) {
1125 case RNDGETENTCNT:
1126 /* inherently racy, no point locking */
1127 if (put_user(input_pool.entropy_count, p))
1128 return -EFAULT;
1129 return 0;
1130 case RNDADDTOENTCNT:
1131 if (!capable(CAP_SYS_ADMIN))
1132 return -EPERM;
1133 if (get_user(ent_count, p))
1134 return -EFAULT;
1135 credit_entropy_bits(&input_pool, ent_count);
1136 return 0;
1137 case RNDADDENTROPY:
1138 if (!capable(CAP_SYS_ADMIN))
1139 return -EPERM;
1140 if (get_user(ent_count, p++))
1141 return -EFAULT;
1142 if (ent_count < 0)
1143 return -EINVAL;
1144 if (get_user(size, p++))
1145 return -EFAULT;
1146 retval = write_pool(&input_pool, (const char __user *)p,
1147 size);
1148 if (retval < 0)
1149 return retval;
1150 credit_entropy_bits(&input_pool, ent_count);
1151 return 0;
1152 case RNDZAPENTCNT:
1153 case RNDCLEARPOOL:
1154 /* Clear the entropy pool counters. */
1155 if (!capable(CAP_SYS_ADMIN))
1156 return -EPERM;
1157 rand_initialize();
1158 return 0;
1159 default:
1160 return -EINVAL;
1164 static int random_fasync(int fd, struct file *filp, int on)
1166 return fasync_helper(fd, filp, on, &fasync);
1169 const struct file_operations random_fops = {
1170 .read = random_read,
1171 .write = random_write,
1172 .poll = random_poll,
1173 .unlocked_ioctl = random_ioctl,
1174 .fasync = random_fasync,
1175 .llseek = noop_llseek,
1178 const struct file_operations urandom_fops = {
1179 .read = urandom_read,
1180 .write = random_write,
1181 .unlocked_ioctl = random_ioctl,
1182 .fasync = random_fasync,
1183 .llseek = noop_llseek,
1186 /***************************************************************
1187 * Random UUID interface
1189 * Used here for a Boot ID, but can be useful for other kernel
1190 * drivers.
1191 ***************************************************************/
1194 * Generate random UUID
1196 void generate_random_uuid(unsigned char uuid_out[16])
1198 get_random_bytes(uuid_out, 16);
1199 /* Set UUID version to 4 --- truly random generation */
1200 uuid_out[6] = (uuid_out[6] & 0x0F) | 0x40;
1201 /* Set the UUID variant to DCE */
1202 uuid_out[8] = (uuid_out[8] & 0x3F) | 0x80;
1204 EXPORT_SYMBOL(generate_random_uuid);
1206 /********************************************************************
1208 * Sysctl interface
1210 ********************************************************************/
1212 #ifdef CONFIG_SYSCTL
1214 #include <linux/sysctl.h>
1216 static int min_read_thresh = 8, min_write_thresh;
1217 static int max_read_thresh = INPUT_POOL_WORDS * 32;
1218 static int max_write_thresh = INPUT_POOL_WORDS * 32;
1219 static char sysctl_bootid[16];
1222 * These functions is used to return both the bootid UUID, and random
1223 * UUID. The difference is in whether table->data is NULL; if it is,
1224 * then a new UUID is generated and returned to the user.
1226 * If the user accesses this via the proc interface, it will be returned
1227 * as an ASCII string in the standard UUID format. If accesses via the
1228 * sysctl system call, it is returned as 16 bytes of binary data.
1230 static int proc_do_uuid(ctl_table *table, int write,
1231 void __user *buffer, size_t *lenp, loff_t *ppos)
1233 ctl_table fake_table;
1234 unsigned char buf[64], tmp_uuid[16], *uuid;
1236 uuid = table->data;
1237 if (!uuid) {
1238 uuid = tmp_uuid;
1239 uuid[8] = 0;
1241 if (uuid[8] == 0)
1242 generate_random_uuid(uuid);
1244 sprintf(buf, "%pU", uuid);
1246 fake_table.data = buf;
1247 fake_table.maxlen = sizeof(buf);
1249 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
1252 static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
1253 ctl_table random_table[] = {
1255 .procname = "poolsize",
1256 .data = &sysctl_poolsize,
1257 .maxlen = sizeof(int),
1258 .mode = 0444,
1259 .proc_handler = proc_dointvec,
1262 .procname = "entropy_avail",
1263 .maxlen = sizeof(int),
1264 .mode = 0444,
1265 .proc_handler = proc_dointvec,
1266 .data = &input_pool.entropy_count,
1269 .procname = "read_wakeup_threshold",
1270 .data = &random_read_wakeup_thresh,
1271 .maxlen = sizeof(int),
1272 .mode = 0644,
1273 .proc_handler = proc_dointvec_minmax,
1274 .extra1 = &min_read_thresh,
1275 .extra2 = &max_read_thresh,
1278 .procname = "write_wakeup_threshold",
1279 .data = &random_write_wakeup_thresh,
1280 .maxlen = sizeof(int),
1281 .mode = 0644,
1282 .proc_handler = proc_dointvec_minmax,
1283 .extra1 = &min_write_thresh,
1284 .extra2 = &max_write_thresh,
1287 .procname = "boot_id",
1288 .data = &sysctl_bootid,
1289 .maxlen = 16,
1290 .mode = 0444,
1291 .proc_handler = proc_do_uuid,
1294 .procname = "uuid",
1295 .maxlen = 16,
1296 .mode = 0444,
1297 .proc_handler = proc_do_uuid,
1301 #endif /* CONFIG_SYSCTL */
1303 /********************************************************************
1305 * Random functions for networking
1307 ********************************************************************/
1310 * TCP initial sequence number picking. This uses the random number
1311 * generator to pick an initial secret value. This value is hashed
1312 * along with the TCP endpoint information to provide a unique
1313 * starting point for each pair of TCP endpoints. This defeats
1314 * attacks which rely on guessing the initial TCP sequence number.
1315 * This algorithm was suggested by Steve Bellovin.
1317 * Using a very strong hash was taking an appreciable amount of the total
1318 * TCP connection establishment time, so this is a weaker hash,
1319 * compensated for by changing the secret periodically.
1322 /* F, G and H are basic MD4 functions: selection, majority, parity */
1323 #define F(x, y, z) ((z) ^ ((x) & ((y) ^ (z))))
1324 #define G(x, y, z) (((x) & (y)) + (((x) ^ (y)) & (z)))
1325 #define H(x, y, z) ((x) ^ (y) ^ (z))
1328 * The generic round function. The application is so specific that
1329 * we don't bother protecting all the arguments with parens, as is generally
1330 * good macro practice, in favor of extra legibility.
1331 * Rotation is separate from addition to prevent recomputation
1333 #define ROUND(f, a, b, c, d, x, s) \
1334 (a += f(b, c, d) + x, a = (a << s) | (a >> (32 - s)))
1335 #define K1 0
1336 #define K2 013240474631UL
1337 #define K3 015666365641UL
1339 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1341 static __u32 twothirdsMD4Transform(__u32 const buf[4], __u32 const in[12])
1343 __u32 a = buf[0], b = buf[1], c = buf[2], d = buf[3];
1345 /* Round 1 */
1346 ROUND(F, a, b, c, d, in[ 0] + K1, 3);
1347 ROUND(F, d, a, b, c, in[ 1] + K1, 7);
1348 ROUND(F, c, d, a, b, in[ 2] + K1, 11);
1349 ROUND(F, b, c, d, a, in[ 3] + K1, 19);
1350 ROUND(F, a, b, c, d, in[ 4] + K1, 3);
1351 ROUND(F, d, a, b, c, in[ 5] + K1, 7);
1352 ROUND(F, c, d, a, b, in[ 6] + K1, 11);
1353 ROUND(F, b, c, d, a, in[ 7] + K1, 19);
1354 ROUND(F, a, b, c, d, in[ 8] + K1, 3);
1355 ROUND(F, d, a, b, c, in[ 9] + K1, 7);
1356 ROUND(F, c, d, a, b, in[10] + K1, 11);
1357 ROUND(F, b, c, d, a, in[11] + K1, 19);
1359 /* Round 2 */
1360 ROUND(G, a, b, c, d, in[ 1] + K2, 3);
1361 ROUND(G, d, a, b, c, in[ 3] + K2, 5);
1362 ROUND(G, c, d, a, b, in[ 5] + K2, 9);
1363 ROUND(G, b, c, d, a, in[ 7] + K2, 13);
1364 ROUND(G, a, b, c, d, in[ 9] + K2, 3);
1365 ROUND(G, d, a, b, c, in[11] + K2, 5);
1366 ROUND(G, c, d, a, b, in[ 0] + K2, 9);
1367 ROUND(G, b, c, d, a, in[ 2] + K2, 13);
1368 ROUND(G, a, b, c, d, in[ 4] + K2, 3);
1369 ROUND(G, d, a, b, c, in[ 6] + K2, 5);
1370 ROUND(G, c, d, a, b, in[ 8] + K2, 9);
1371 ROUND(G, b, c, d, a, in[10] + K2, 13);
1373 /* Round 3 */
1374 ROUND(H, a, b, c, d, in[ 3] + K3, 3);
1375 ROUND(H, d, a, b, c, in[ 7] + K3, 9);
1376 ROUND(H, c, d, a, b, in[11] + K3, 11);
1377 ROUND(H, b, c, d, a, in[ 2] + K3, 15);
1378 ROUND(H, a, b, c, d, in[ 6] + K3, 3);
1379 ROUND(H, d, a, b, c, in[10] + K3, 9);
1380 ROUND(H, c, d, a, b, in[ 1] + K3, 11);
1381 ROUND(H, b, c, d, a, in[ 5] + K3, 15);
1382 ROUND(H, a, b, c, d, in[ 9] + K3, 3);
1383 ROUND(H, d, a, b, c, in[ 0] + K3, 9);
1384 ROUND(H, c, d, a, b, in[ 4] + K3, 11);
1385 ROUND(H, b, c, d, a, in[ 8] + K3, 15);
1387 return buf[1] + b; /* "most hashed" word */
1388 /* Alternative: return sum of all words? */
1390 #endif
1392 #undef ROUND
1393 #undef F
1394 #undef G
1395 #undef H
1396 #undef K1
1397 #undef K2
1398 #undef K3
1400 /* This should not be decreased so low that ISNs wrap too fast. */
1401 #define REKEY_INTERVAL (300 * HZ)
1403 * Bit layout of the tcp sequence numbers (before adding current time):
1404 * bit 24-31: increased after every key exchange
1405 * bit 0-23: hash(source,dest)
1407 * The implementation is similar to the algorithm described
1408 * in the Appendix of RFC 1185, except that
1409 * - it uses a 1 MHz clock instead of a 250 kHz clock
1410 * - it performs a rekey every 5 minutes, which is equivalent
1411 * to a (source,dest) tulple dependent forward jump of the
1412 * clock by 0..2^(HASH_BITS+1)
1414 * Thus the average ISN wraparound time is 68 minutes instead of
1415 * 4.55 hours.
1417 * SMP cleanup and lock avoidance with poor man's RCU.
1418 * Manfred Spraul <manfred@colorfullife.com>
1421 #define COUNT_BITS 8
1422 #define COUNT_MASK ((1 << COUNT_BITS) - 1)
1423 #define HASH_BITS 24
1424 #define HASH_MASK ((1 << HASH_BITS) - 1)
1426 static struct keydata {
1427 __u32 count; /* already shifted to the final position */
1428 __u32 secret[12];
1429 } ____cacheline_aligned ip_keydata[2];
1431 static unsigned int ip_cnt;
1433 static void rekey_seq_generator(struct work_struct *work);
1435 static DECLARE_DELAYED_WORK(rekey_work, rekey_seq_generator);
1438 * Lock avoidance:
1439 * The ISN generation runs lockless - it's just a hash over random data.
1440 * State changes happen every 5 minutes when the random key is replaced.
1441 * Synchronization is performed by having two copies of the hash function
1442 * state and rekey_seq_generator always updates the inactive copy.
1443 * The copy is then activated by updating ip_cnt.
1444 * The implementation breaks down if someone blocks the thread
1445 * that processes SYN requests for more than 5 minutes. Should never
1446 * happen, and even if that happens only a not perfectly compliant
1447 * ISN is generated, nothing fatal.
1449 static void rekey_seq_generator(struct work_struct *work)
1451 struct keydata *keyptr = &ip_keydata[1 ^ (ip_cnt & 1)];
1453 get_random_bytes(keyptr->secret, sizeof(keyptr->secret));
1454 keyptr->count = (ip_cnt & COUNT_MASK) << HASH_BITS;
1455 smp_wmb();
1456 ip_cnt++;
1457 schedule_delayed_work(&rekey_work,
1458 round_jiffies_relative(REKEY_INTERVAL));
1461 static inline struct keydata *get_keyptr(void)
1463 struct keydata *keyptr = &ip_keydata[ip_cnt & 1];
1465 smp_rmb();
1467 return keyptr;
1470 static __init int seqgen_init(void)
1472 rekey_seq_generator(NULL);
1473 return 0;
1475 late_initcall(seqgen_init);
1477 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1478 __u32 secure_tcpv6_sequence_number(__be32 *saddr, __be32 *daddr,
1479 __be16 sport, __be16 dport)
1481 __u32 seq;
1482 __u32 hash[12];
1483 struct keydata *keyptr = get_keyptr();
1485 /* The procedure is the same as for IPv4, but addresses are longer.
1486 * Thus we must use twothirdsMD4Transform.
1489 memcpy(hash, saddr, 16);
1490 hash[4] = ((__force u16)sport << 16) + (__force u16)dport;
1491 memcpy(&hash[5], keyptr->secret, sizeof(__u32) * 7);
1493 seq = twothirdsMD4Transform((const __u32 *)daddr, hash) & HASH_MASK;
1494 seq += keyptr->count;
1496 seq += ktime_to_ns(ktime_get_real());
1498 return seq;
1500 EXPORT_SYMBOL(secure_tcpv6_sequence_number);
1501 #endif
1503 /* The code below is shamelessly stolen from secure_tcp_sequence_number().
1504 * All blames to Andrey V. Savochkin <saw@msu.ru>.
1506 __u32 secure_ip_id(__be32 daddr)
1508 struct keydata *keyptr;
1509 __u32 hash[4];
1511 keyptr = get_keyptr();
1514 * Pick a unique starting offset for each IP destination.
1515 * The dest ip address is placed in the starting vector,
1516 * which is then hashed with random data.
1518 hash[0] = (__force __u32)daddr;
1519 hash[1] = keyptr->secret[9];
1520 hash[2] = keyptr->secret[10];
1521 hash[3] = keyptr->secret[11];
1523 return half_md4_transform(hash, keyptr->secret);
1526 #ifdef CONFIG_INET
1528 __u32 secure_tcp_sequence_number(__be32 saddr, __be32 daddr,
1529 __be16 sport, __be16 dport)
1531 __u32 seq;
1532 __u32 hash[4];
1533 struct keydata *keyptr = get_keyptr();
1536 * Pick a unique starting offset for each TCP connection endpoints
1537 * (saddr, daddr, sport, dport).
1538 * Note that the words are placed into the starting vector, which is
1539 * then mixed with a partial MD4 over random data.
1541 hash[0] = (__force u32)saddr;
1542 hash[1] = (__force u32)daddr;
1543 hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
1544 hash[3] = keyptr->secret[11];
1546 seq = half_md4_transform(hash, keyptr->secret) & HASH_MASK;
1547 seq += keyptr->count;
1549 * As close as possible to RFC 793, which
1550 * suggests using a 250 kHz clock.
1551 * Further reading shows this assumes 2 Mb/s networks.
1552 * For 10 Mb/s Ethernet, a 1 MHz clock is appropriate.
1553 * For 10 Gb/s Ethernet, a 1 GHz clock should be ok, but
1554 * we also need to limit the resolution so that the u32 seq
1555 * overlaps less than one time per MSL (2 minutes).
1556 * Choosing a clock of 64 ns period is OK. (period of 274 s)
1558 seq += ktime_to_ns(ktime_get_real()) >> 6;
1560 return seq;
1563 /* Generate secure starting point for ephemeral IPV4 transport port search */
1564 u32 secure_ipv4_port_ephemeral(__be32 saddr, __be32 daddr, __be16 dport)
1566 struct keydata *keyptr = get_keyptr();
1567 u32 hash[4];
1570 * Pick a unique starting offset for each ephemeral port search
1571 * (saddr, daddr, dport) and 48bits of random data.
1573 hash[0] = (__force u32)saddr;
1574 hash[1] = (__force u32)daddr;
1575 hash[2] = (__force u32)dport ^ keyptr->secret[10];
1576 hash[3] = keyptr->secret[11];
1578 return half_md4_transform(hash, keyptr->secret);
1580 EXPORT_SYMBOL_GPL(secure_ipv4_port_ephemeral);
1582 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1583 u32 secure_ipv6_port_ephemeral(const __be32 *saddr, const __be32 *daddr,
1584 __be16 dport)
1586 struct keydata *keyptr = get_keyptr();
1587 u32 hash[12];
1589 memcpy(hash, saddr, 16);
1590 hash[4] = (__force u32)dport;
1591 memcpy(&hash[5], keyptr->secret, sizeof(__u32) * 7);
1593 return twothirdsMD4Transform((const __u32 *)daddr, hash);
1595 #endif
1597 #if defined(CONFIG_IP_DCCP) || defined(CONFIG_IP_DCCP_MODULE)
1598 /* Similar to secure_tcp_sequence_number but generate a 48 bit value
1599 * bit's 32-47 increase every key exchange
1600 * 0-31 hash(source, dest)
1602 u64 secure_dccp_sequence_number(__be32 saddr, __be32 daddr,
1603 __be16 sport, __be16 dport)
1605 u64 seq;
1606 __u32 hash[4];
1607 struct keydata *keyptr = get_keyptr();
1609 hash[0] = (__force u32)saddr;
1610 hash[1] = (__force u32)daddr;
1611 hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
1612 hash[3] = keyptr->secret[11];
1614 seq = half_md4_transform(hash, keyptr->secret);
1615 seq |= ((u64)keyptr->count) << (32 - HASH_BITS);
1617 seq += ktime_to_ns(ktime_get_real());
1618 seq &= (1ull << 48) - 1;
1620 return seq;
1622 EXPORT_SYMBOL(secure_dccp_sequence_number);
1623 #endif
1625 #endif /* CONFIG_INET */
1629 * Get a random word for internal kernel use only. Similar to urandom but
1630 * with the goal of minimal entropy pool depletion. As a result, the random
1631 * value is not cryptographically secure but for several uses the cost of
1632 * depleting entropy is too high
1634 DEFINE_PER_CPU(__u32 [4], get_random_int_hash);
1635 unsigned int get_random_int(void)
1637 struct keydata *keyptr;
1638 __u32 *hash = get_cpu_var(get_random_int_hash);
1639 int ret;
1641 keyptr = get_keyptr();
1642 hash[0] += current->pid + jiffies + get_cycles();
1644 ret = half_md4_transform(hash, keyptr->secret);
1645 put_cpu_var(get_random_int_hash);
1647 return ret;
1651 * randomize_range() returns a start address such that
1653 * [...... <range> .....]
1654 * start end
1656 * a <range> with size "len" starting at the return value is inside in the
1657 * area defined by [start, end], but is otherwise randomized.
1659 unsigned long
1660 randomize_range(unsigned long start, unsigned long end, unsigned long len)
1662 unsigned long range = end - len - start;
1664 if (end <= start + len)
1665 return 0;
1666 return PAGE_ALIGN(get_random_int() % range + start);