Add -Werror-implicit-function-declaration
[linux-2.6/verdex.git] / drivers / char / random.c
blob7f5271272f91400b17d1ac928047d5d363c35be4
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);
132 * add_input_randomness() uses the input layer interrupt timing, as well as
133 * the event type information from the hardware.
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.
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.
147 * Ensuring unpredictability at system startup
148 * ============================================
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:
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
172 * and the following lines in an appropriate script which is run as
173 * the system is shutdown:
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
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.
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.
197 * Configuring the /dev/random driver under Linux
198 * ==============================================
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:
205 * mknod /dev/random c 1 8
206 * mknod /dev/urandom c 1 9
208 * Acknowledgements:
209 * =================
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.
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.
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.
226 #include <linux/utsname.h>
227 #include <linux/module.h>
228 #include <linux/kernel.h>
229 #include <linux/major.h>
230 #include <linux/string.h>
231 #include <linux/fcntl.h>
232 #include <linux/slab.h>
233 #include <linux/random.h>
234 #include <linux/poll.h>
235 #include <linux/init.h>
236 #include <linux/fs.h>
237 #include <linux/genhd.h>
238 #include <linux/interrupt.h>
239 #include <linux/spinlock.h>
240 #include <linux/percpu.h>
241 #include <linux/cryptohash.h>
243 #include <asm/processor.h>
244 #include <asm/uaccess.h>
245 #include <asm/irq.h>
246 #include <asm/io.h>
249 * Configuration information
251 #define INPUT_POOL_WORDS 128
252 #define OUTPUT_POOL_WORDS 32
253 #define SEC_XFER_SIZE 512
256 * The minimum number of bits of entropy before we wake up a read on
257 * /dev/random. Should be enough to do a significant reseed.
259 static int random_read_wakeup_thresh = 64;
262 * If the entropy count falls under this number of bits, then we
263 * should wake up processes which are selecting or polling on write
264 * access to /dev/random.
266 static int random_write_wakeup_thresh = 128;
269 * When the input pool goes over trickle_thresh, start dropping most
270 * samples to avoid wasting CPU time and reduce lock contention.
273 static int trickle_thresh __read_mostly = INPUT_POOL_WORDS * 28;
275 static DEFINE_PER_CPU(int, trickle_count) = 0;
278 * A pool of size .poolwords is stirred with a primitive polynomial
279 * of degree .poolwords over GF(2). The taps for various sizes are
280 * defined below. They are chosen to be evenly spaced (minimum RMS
281 * distance from evenly spaced; the numbers in the comments are a
282 * scaled squared error sum) except for the last tap, which is 1 to
283 * get the twisting happening as fast as possible.
285 static struct poolinfo {
286 int poolwords;
287 int tap1, tap2, tap3, tap4, tap5;
288 } poolinfo_table[] = {
289 /* x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 -- 105 */
290 { 128, 103, 76, 51, 25, 1 },
291 /* x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 -- 15 */
292 { 32, 26, 20, 14, 7, 1 },
293 #if 0
294 /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1 -- 115 */
295 { 2048, 1638, 1231, 819, 411, 1 },
297 /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
298 { 1024, 817, 615, 412, 204, 1 },
300 /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
301 { 1024, 819, 616, 410, 207, 2 },
303 /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
304 { 512, 411, 308, 208, 104, 1 },
306 /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
307 { 512, 409, 307, 206, 102, 2 },
308 /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
309 { 512, 409, 309, 205, 103, 2 },
311 /* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
312 { 256, 205, 155, 101, 52, 1 },
314 /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
315 { 128, 103, 78, 51, 27, 2 },
317 /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
318 { 64, 52, 39, 26, 14, 1 },
319 #endif
322 #define POOLBITS poolwords*32
323 #define POOLBYTES poolwords*4
326 * For the purposes of better mixing, we use the CRC-32 polynomial as
327 * well to make a twisted Generalized Feedback Shift Reigster
329 * (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR generators. ACM
330 * Transactions on Modeling and Computer Simulation 2(3):179-194.
331 * Also see M. Matsumoto & Y. Kurita, 1994. Twisted GFSR generators
332 * II. ACM Transactions on Mdeling and Computer Simulation 4:254-266)
334 * Thanks to Colin Plumb for suggesting this.
336 * We have not analyzed the resultant polynomial to prove it primitive;
337 * in fact it almost certainly isn't. Nonetheless, the irreducible factors
338 * of a random large-degree polynomial over GF(2) are more than large enough
339 * that periodicity is not a concern.
341 * The input hash is much less sensitive than the output hash. All
342 * that we want of it is that it be a good non-cryptographic hash;
343 * i.e. it not produce collisions when fed "random" data of the sort
344 * we expect to see. As long as the pool state differs for different
345 * inputs, we have preserved the input entropy and done a good job.
346 * The fact that an intelligent attacker can construct inputs that
347 * will produce controlled alterations to the pool's state is not
348 * important because we don't consider such inputs to contribute any
349 * randomness. The only property we need with respect to them is that
350 * the attacker can't increase his/her knowledge of the pool's state.
351 * Since all additions are reversible (knowing the final state and the
352 * input, you can reconstruct the initial state), if an attacker has
353 * any uncertainty about the initial state, he/she can only shuffle
354 * that uncertainty about, but never cause any collisions (which would
355 * decrease the uncertainty).
357 * The chosen system lets the state of the pool be (essentially) the input
358 * modulo the generator polymnomial. Now, for random primitive polynomials,
359 * this is a universal class of hash functions, meaning that the chance
360 * of a collision is limited by the attacker's knowledge of the generator
361 * polynomail, so if it is chosen at random, an attacker can never force
362 * a collision. Here, we use a fixed polynomial, but we *can* assume that
363 * ###--> it is unknown to the processes generating the input entropy. <-###
364 * Because of this important property, this is a good, collision-resistant
365 * hash; hash collisions will occur no more often than chance.
369 * Static global variables
371 static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
372 static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
374 #if 0
375 static int debug = 0;
376 module_param(debug, bool, 0644);
377 #define DEBUG_ENT(fmt, arg...) do { if (debug) \
378 printk(KERN_DEBUG "random %04d %04d %04d: " \
379 fmt,\
380 input_pool.entropy_count,\
381 blocking_pool.entropy_count,\
382 nonblocking_pool.entropy_count,\
383 ## arg); } while (0)
384 #else
385 #define DEBUG_ENT(fmt, arg...) do {} while (0)
386 #endif
388 /**********************************************************************
390 * OS independent entropy store. Here are the functions which handle
391 * storing entropy in an entropy pool.
393 **********************************************************************/
395 struct entropy_store;
396 struct entropy_store {
397 /* mostly-read data: */
398 struct poolinfo *poolinfo;
399 __u32 *pool;
400 const char *name;
401 int limit;
402 struct entropy_store *pull;
404 /* read-write data: */
405 spinlock_t lock ____cacheline_aligned_in_smp;
406 unsigned add_ptr;
407 int entropy_count;
408 int input_rotate;
411 static __u32 input_pool_data[INPUT_POOL_WORDS];
412 static __u32 blocking_pool_data[OUTPUT_POOL_WORDS];
413 static __u32 nonblocking_pool_data[OUTPUT_POOL_WORDS];
415 static struct entropy_store input_pool = {
416 .poolinfo = &poolinfo_table[0],
417 .name = "input",
418 .limit = 1,
419 .lock = __SPIN_LOCK_UNLOCKED(&input_pool.lock),
420 .pool = input_pool_data
423 static struct entropy_store blocking_pool = {
424 .poolinfo = &poolinfo_table[1],
425 .name = "blocking",
426 .limit = 1,
427 .pull = &input_pool,
428 .lock = __SPIN_LOCK_UNLOCKED(&blocking_pool.lock),
429 .pool = blocking_pool_data
432 static struct entropy_store nonblocking_pool = {
433 .poolinfo = &poolinfo_table[1],
434 .name = "nonblocking",
435 .pull = &input_pool,
436 .lock = __SPIN_LOCK_UNLOCKED(&nonblocking_pool.lock),
437 .pool = nonblocking_pool_data
441 * This function adds a byte into the entropy "pool". It does not
442 * update the entropy estimate. The caller should call
443 * credit_entropy_store if this is appropriate.
445 * The pool is stirred with a primitive polynomial of the appropriate
446 * degree, and then twisted. We twist by three bits at a time because
447 * it's cheap to do so and helps slightly in the expected case where
448 * the entropy is concentrated in the low-order bits.
450 static void __add_entropy_words(struct entropy_store *r, const __u32 *in,
451 int nwords, __u32 out[16])
453 static __u32 const twist_table[8] = {
454 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
455 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
456 unsigned long i, add_ptr, tap1, tap2, tap3, tap4, tap5;
457 int new_rotate, input_rotate;
458 int wordmask = r->poolinfo->poolwords - 1;
459 __u32 w, next_w;
460 unsigned long flags;
462 /* Taps are constant, so we can load them without holding r->lock. */
463 tap1 = r->poolinfo->tap1;
464 tap2 = r->poolinfo->tap2;
465 tap3 = r->poolinfo->tap3;
466 tap4 = r->poolinfo->tap4;
467 tap5 = r->poolinfo->tap5;
468 next_w = *in++;
470 spin_lock_irqsave(&r->lock, flags);
471 prefetch_range(r->pool, wordmask);
472 input_rotate = r->input_rotate;
473 add_ptr = r->add_ptr;
475 while (nwords--) {
476 w = rol32(next_w, input_rotate);
477 if (nwords > 0)
478 next_w = *in++;
479 i = add_ptr = (add_ptr - 1) & wordmask;
481 * Normally, we add 7 bits of rotation to the pool.
482 * At the beginning of the pool, add an extra 7 bits
483 * rotation, so that successive passes spread the
484 * input bits across the pool evenly.
486 new_rotate = input_rotate + 14;
487 if (i)
488 new_rotate = input_rotate + 7;
489 input_rotate = new_rotate & 31;
491 /* XOR in the various taps */
492 w ^= r->pool[(i + tap1) & wordmask];
493 w ^= r->pool[(i + tap2) & wordmask];
494 w ^= r->pool[(i + tap3) & wordmask];
495 w ^= r->pool[(i + tap4) & wordmask];
496 w ^= r->pool[(i + tap5) & wordmask];
497 w ^= r->pool[i];
498 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
501 r->input_rotate = input_rotate;
502 r->add_ptr = add_ptr;
504 if (out) {
505 for (i = 0; i < 16; i++) {
506 out[i] = r->pool[add_ptr];
507 add_ptr = (add_ptr - 1) & wordmask;
511 spin_unlock_irqrestore(&r->lock, flags);
514 static inline void add_entropy_words(struct entropy_store *r, const __u32 *in,
515 int nwords)
517 __add_entropy_words(r, in, nwords, NULL);
521 * Credit (or debit) the entropy store with n bits of entropy
523 static void credit_entropy_store(struct entropy_store *r, int nbits)
525 unsigned long flags;
527 spin_lock_irqsave(&r->lock, flags);
529 if (r->entropy_count + nbits < 0) {
530 DEBUG_ENT("negative entropy/overflow (%d+%d)\n",
531 r->entropy_count, nbits);
532 r->entropy_count = 0;
533 } else if (r->entropy_count + nbits > r->poolinfo->POOLBITS) {
534 r->entropy_count = r->poolinfo->POOLBITS;
535 } else {
536 r->entropy_count += nbits;
537 if (nbits)
538 DEBUG_ENT("added %d entropy credits to %s\n",
539 nbits, r->name);
542 spin_unlock_irqrestore(&r->lock, flags);
545 /*********************************************************************
547 * Entropy input management
549 *********************************************************************/
551 /* There is one of these per entropy source */
552 struct timer_rand_state {
553 cycles_t last_time;
554 long last_delta,last_delta2;
555 unsigned dont_count_entropy:1;
558 static struct timer_rand_state input_timer_state;
559 static struct timer_rand_state *irq_timer_state[NR_IRQS];
562 * This function adds entropy to the entropy "pool" by using timing
563 * delays. It uses the timer_rand_state structure to make an estimate
564 * of how many bits of entropy this call has added to the pool.
566 * The number "num" is also added to the pool - it should somehow describe
567 * the type of event which just happened. This is currently 0-255 for
568 * keyboard scan codes, and 256 upwards for interrupts.
571 static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
573 struct {
574 cycles_t cycles;
575 long jiffies;
576 unsigned num;
577 } sample;
578 long delta, delta2, delta3;
580 preempt_disable();
581 /* if over the trickle threshold, use only 1 in 4096 samples */
582 if (input_pool.entropy_count > trickle_thresh &&
583 (__get_cpu_var(trickle_count)++ & 0xfff))
584 goto out;
586 sample.jiffies = jiffies;
587 sample.cycles = get_cycles();
588 sample.num = num;
589 add_entropy_words(&input_pool, (u32 *)&sample, sizeof(sample)/4);
592 * Calculate number of bits of randomness we probably added.
593 * We take into account the first, second and third-order deltas
594 * in order to make our estimate.
597 if (!state->dont_count_entropy) {
598 delta = sample.jiffies - state->last_time;
599 state->last_time = sample.jiffies;
601 delta2 = delta - state->last_delta;
602 state->last_delta = delta;
604 delta3 = delta2 - state->last_delta2;
605 state->last_delta2 = delta2;
607 if (delta < 0)
608 delta = -delta;
609 if (delta2 < 0)
610 delta2 = -delta2;
611 if (delta3 < 0)
612 delta3 = -delta3;
613 if (delta > delta2)
614 delta = delta2;
615 if (delta > delta3)
616 delta = delta3;
619 * delta is now minimum absolute delta.
620 * Round down by 1 bit on general principles,
621 * and limit entropy entimate to 12 bits.
623 credit_entropy_store(&input_pool,
624 min_t(int, fls(delta>>1), 11));
627 if(input_pool.entropy_count >= random_read_wakeup_thresh)
628 wake_up_interruptible(&random_read_wait);
630 out:
631 preempt_enable();
634 void add_input_randomness(unsigned int type, unsigned int code,
635 unsigned int value)
637 static unsigned char last_value;
639 /* ignore autorepeat and the like */
640 if (value == last_value)
641 return;
643 DEBUG_ENT("input event\n");
644 last_value = value;
645 add_timer_randomness(&input_timer_state,
646 (type << 4) ^ code ^ (code >> 4) ^ value);
648 EXPORT_SYMBOL_GPL(add_input_randomness);
650 void add_interrupt_randomness(int irq)
652 if (irq >= NR_IRQS || irq_timer_state[irq] == 0)
653 return;
655 DEBUG_ENT("irq event %d\n", irq);
656 add_timer_randomness(irq_timer_state[irq], 0x100 + irq);
659 #ifdef CONFIG_BLOCK
660 void add_disk_randomness(struct gendisk *disk)
662 if (!disk || !disk->random)
663 return;
664 /* first major is 1, so we get >= 0x200 here */
665 DEBUG_ENT("disk event %d:%d\n", disk->major, disk->first_minor);
667 add_timer_randomness(disk->random,
668 0x100 + MKDEV(disk->major, disk->first_minor));
671 EXPORT_SYMBOL(add_disk_randomness);
672 #endif
674 #define EXTRACT_SIZE 10
676 /*********************************************************************
678 * Entropy extraction routines
680 *********************************************************************/
682 static ssize_t extract_entropy(struct entropy_store *r, void * buf,
683 size_t nbytes, int min, int rsvd);
686 * This utility inline function is responsible for transfering entropy
687 * from the primary pool to the secondary extraction pool. We make
688 * sure we pull enough for a 'catastrophic reseed'.
690 static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
692 __u32 tmp[OUTPUT_POOL_WORDS];
694 if (r->pull && r->entropy_count < nbytes * 8 &&
695 r->entropy_count < r->poolinfo->POOLBITS) {
696 int bytes = max_t(int, random_read_wakeup_thresh / 8,
697 min_t(int, nbytes, sizeof(tmp)));
698 int rsvd = r->limit ? 0 : random_read_wakeup_thresh/4;
700 DEBUG_ENT("going to reseed %s with %d bits "
701 "(%d of %d requested)\n",
702 r->name, bytes * 8, nbytes * 8, r->entropy_count);
704 bytes=extract_entropy(r->pull, tmp, bytes,
705 random_read_wakeup_thresh / 8, rsvd);
706 add_entropy_words(r, tmp, (bytes + 3) / 4);
707 credit_entropy_store(r, bytes*8);
712 * These functions extracts randomness from the "entropy pool", and
713 * returns it in a buffer.
715 * The min parameter specifies the minimum amount we can pull before
716 * failing to avoid races that defeat catastrophic reseeding while the
717 * reserved parameter indicates how much entropy we must leave in the
718 * pool after each pull to avoid starving other readers.
720 * Note: extract_entropy() assumes that .poolwords is a multiple of 16 words.
723 static size_t account(struct entropy_store *r, size_t nbytes, int min,
724 int reserved)
726 unsigned long flags;
728 BUG_ON(r->entropy_count > r->poolinfo->POOLBITS);
730 /* Hold lock while accounting */
731 spin_lock_irqsave(&r->lock, flags);
733 DEBUG_ENT("trying to extract %d bits from %s\n",
734 nbytes * 8, r->name);
736 /* Can we pull enough? */
737 if (r->entropy_count / 8 < min + reserved) {
738 nbytes = 0;
739 } else {
740 /* If limited, never pull more than available */
741 if (r->limit && nbytes + reserved >= r->entropy_count / 8)
742 nbytes = r->entropy_count/8 - reserved;
744 if(r->entropy_count / 8 >= nbytes + reserved)
745 r->entropy_count -= nbytes*8;
746 else
747 r->entropy_count = reserved;
749 if (r->entropy_count < random_write_wakeup_thresh)
750 wake_up_interruptible(&random_write_wait);
753 DEBUG_ENT("debiting %d entropy credits from %s%s\n",
754 nbytes * 8, r->name, r->limit ? "" : " (unlimited)");
756 spin_unlock_irqrestore(&r->lock, flags);
758 return nbytes;
761 static void extract_buf(struct entropy_store *r, __u8 *out)
763 int i;
764 __u32 data[16], buf[5 + SHA_WORKSPACE_WORDS];
766 sha_init(buf);
768 * As we hash the pool, we mix intermediate values of
769 * the hash back into the pool. This eliminates
770 * backtracking attacks (where the attacker knows
771 * the state of the pool plus the current outputs, and
772 * attempts to find previous ouputs), unless the hash
773 * function can be inverted.
775 for (i = 0; i < r->poolinfo->poolwords; i += 16) {
776 /* hash blocks of 16 words = 512 bits */
777 sha_transform(buf, (__u8 *)(r->pool + i), buf + 5);
778 /* feed back portion of the resulting hash */
779 add_entropy_words(r, &buf[i % 5], 1);
783 * To avoid duplicates, we atomically extract a
784 * portion of the pool while mixing, and hash one
785 * final time.
787 __add_entropy_words(r, &buf[i % 5], 1, data);
788 sha_transform(buf, (__u8 *)data, buf + 5);
791 * In case the hash function has some recognizable
792 * output pattern, we fold it in half.
795 buf[0] ^= buf[3];
796 buf[1] ^= buf[4];
797 buf[2] ^= rol32(buf[2], 16);
798 memcpy(out, buf, EXTRACT_SIZE);
799 memset(buf, 0, sizeof(buf));
802 static ssize_t extract_entropy(struct entropy_store *r, void * buf,
803 size_t nbytes, int min, int reserved)
805 ssize_t ret = 0, i;
806 __u8 tmp[EXTRACT_SIZE];
808 xfer_secondary_pool(r, nbytes);
809 nbytes = account(r, nbytes, min, reserved);
811 while (nbytes) {
812 extract_buf(r, tmp);
813 i = min_t(int, nbytes, EXTRACT_SIZE);
814 memcpy(buf, tmp, i);
815 nbytes -= i;
816 buf += i;
817 ret += i;
820 /* Wipe data just returned from memory */
821 memset(tmp, 0, sizeof(tmp));
823 return ret;
826 static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
827 size_t nbytes)
829 ssize_t ret = 0, i;
830 __u8 tmp[EXTRACT_SIZE];
832 xfer_secondary_pool(r, nbytes);
833 nbytes = account(r, nbytes, 0, 0);
835 while (nbytes) {
836 if (need_resched()) {
837 if (signal_pending(current)) {
838 if (ret == 0)
839 ret = -ERESTARTSYS;
840 break;
842 schedule();
845 extract_buf(r, tmp);
846 i = min_t(int, nbytes, EXTRACT_SIZE);
847 if (copy_to_user(buf, tmp, i)) {
848 ret = -EFAULT;
849 break;
852 nbytes -= i;
853 buf += i;
854 ret += i;
857 /* Wipe data just returned from memory */
858 memset(tmp, 0, sizeof(tmp));
860 return ret;
864 * This function is the exported kernel interface. It returns some
865 * number of good random numbers, suitable for seeding TCP sequence
866 * numbers, etc.
868 void get_random_bytes(void *buf, int nbytes)
870 extract_entropy(&nonblocking_pool, buf, nbytes, 0, 0);
873 EXPORT_SYMBOL(get_random_bytes);
876 * init_std_data - initialize pool with system data
878 * @r: pool to initialize
880 * This function clears the pool's entropy count and mixes some system
881 * data into the pool to prepare it for use. The pool is not cleared
882 * as that can only decrease the entropy in the pool.
884 static void init_std_data(struct entropy_store *r)
886 ktime_t now;
887 unsigned long flags;
889 spin_lock_irqsave(&r->lock, flags);
890 r->entropy_count = 0;
891 spin_unlock_irqrestore(&r->lock, flags);
893 now = ktime_get_real();
894 add_entropy_words(r, (__u32 *)&now, sizeof(now)/4);
895 add_entropy_words(r, (__u32 *)utsname(),
896 sizeof(*(utsname()))/4);
899 static int __init rand_initialize(void)
901 init_std_data(&input_pool);
902 init_std_data(&blocking_pool);
903 init_std_data(&nonblocking_pool);
904 return 0;
906 module_init(rand_initialize);
908 void rand_initialize_irq(int irq)
910 struct timer_rand_state *state;
912 if (irq >= NR_IRQS || irq_timer_state[irq])
913 return;
916 * If kzalloc returns null, we just won't use that entropy
917 * source.
919 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
920 if (state)
921 irq_timer_state[irq] = state;
924 #ifdef CONFIG_BLOCK
925 void rand_initialize_disk(struct gendisk *disk)
927 struct timer_rand_state *state;
930 * If kzalloc returns null, we just won't use that entropy
931 * source.
933 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
934 if (state)
935 disk->random = state;
937 #endif
939 static ssize_t
940 random_read(struct file * file, char __user * buf, size_t nbytes, loff_t *ppos)
942 ssize_t n, retval = 0, count = 0;
944 if (nbytes == 0)
945 return 0;
947 while (nbytes > 0) {
948 n = nbytes;
949 if (n > SEC_XFER_SIZE)
950 n = SEC_XFER_SIZE;
952 DEBUG_ENT("reading %d bits\n", n*8);
954 n = extract_entropy_user(&blocking_pool, buf, n);
956 DEBUG_ENT("read got %d bits (%d still needed)\n",
957 n*8, (nbytes-n)*8);
959 if (n == 0) {
960 if (file->f_flags & O_NONBLOCK) {
961 retval = -EAGAIN;
962 break;
965 DEBUG_ENT("sleeping?\n");
967 wait_event_interruptible(random_read_wait,
968 input_pool.entropy_count >=
969 random_read_wakeup_thresh);
971 DEBUG_ENT("awake\n");
973 if (signal_pending(current)) {
974 retval = -ERESTARTSYS;
975 break;
978 continue;
981 if (n < 0) {
982 retval = n;
983 break;
985 count += n;
986 buf += n;
987 nbytes -= n;
988 break; /* This break makes the device work */
989 /* like a named pipe */
993 * If we gave the user some bytes, update the access time.
995 if (count)
996 file_accessed(file);
998 return (count ? count : retval);
1001 static ssize_t
1002 urandom_read(struct file * file, char __user * buf,
1003 size_t nbytes, loff_t *ppos)
1005 return extract_entropy_user(&nonblocking_pool, buf, nbytes);
1008 static unsigned int
1009 random_poll(struct file *file, poll_table * wait)
1011 unsigned int mask;
1013 poll_wait(file, &random_read_wait, wait);
1014 poll_wait(file, &random_write_wait, wait);
1015 mask = 0;
1016 if (input_pool.entropy_count >= random_read_wakeup_thresh)
1017 mask |= POLLIN | POLLRDNORM;
1018 if (input_pool.entropy_count < random_write_wakeup_thresh)
1019 mask |= POLLOUT | POLLWRNORM;
1020 return mask;
1023 static int
1024 write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1026 size_t bytes;
1027 __u32 buf[16];
1028 const char __user *p = buffer;
1030 while (count > 0) {
1031 bytes = min(count, sizeof(buf));
1032 if (copy_from_user(&buf, p, bytes))
1033 return -EFAULT;
1035 count -= bytes;
1036 p += bytes;
1038 add_entropy_words(r, buf, (bytes + 3) / 4);
1041 return 0;
1044 static ssize_t
1045 random_write(struct file * file, const char __user * buffer,
1046 size_t count, loff_t *ppos)
1048 size_t ret;
1049 struct inode *inode = file->f_path.dentry->d_inode;
1051 ret = write_pool(&blocking_pool, buffer, count);
1052 if (ret)
1053 return ret;
1054 ret = write_pool(&nonblocking_pool, buffer, count);
1055 if (ret)
1056 return ret;
1058 inode->i_mtime = current_fs_time(inode->i_sb);
1059 mark_inode_dirty(inode);
1060 return (ssize_t)count;
1063 static int
1064 random_ioctl(struct inode * inode, struct file * file,
1065 unsigned int cmd, unsigned long arg)
1067 int size, ent_count;
1068 int __user *p = (int __user *)arg;
1069 int retval;
1071 switch (cmd) {
1072 case RNDGETENTCNT:
1073 ent_count = input_pool.entropy_count;
1074 if (put_user(ent_count, p))
1075 return -EFAULT;
1076 return 0;
1077 case RNDADDTOENTCNT:
1078 if (!capable(CAP_SYS_ADMIN))
1079 return -EPERM;
1080 if (get_user(ent_count, p))
1081 return -EFAULT;
1082 credit_entropy_store(&input_pool, ent_count);
1084 * Wake up waiting processes if we have enough
1085 * entropy.
1087 if (input_pool.entropy_count >= random_read_wakeup_thresh)
1088 wake_up_interruptible(&random_read_wait);
1089 return 0;
1090 case RNDADDENTROPY:
1091 if (!capable(CAP_SYS_ADMIN))
1092 return -EPERM;
1093 if (get_user(ent_count, p++))
1094 return -EFAULT;
1095 if (ent_count < 0)
1096 return -EINVAL;
1097 if (get_user(size, p++))
1098 return -EFAULT;
1099 retval = write_pool(&input_pool, (const char __user *)p,
1100 size);
1101 if (retval < 0)
1102 return retval;
1103 credit_entropy_store(&input_pool, ent_count);
1105 * Wake up waiting processes if we have enough
1106 * entropy.
1108 if (input_pool.entropy_count >= random_read_wakeup_thresh)
1109 wake_up_interruptible(&random_read_wait);
1110 return 0;
1111 case RNDZAPENTCNT:
1112 case RNDCLEARPOOL:
1113 /* Clear the entropy pool counters. */
1114 if (!capable(CAP_SYS_ADMIN))
1115 return -EPERM;
1116 init_std_data(&input_pool);
1117 init_std_data(&blocking_pool);
1118 init_std_data(&nonblocking_pool);
1119 return 0;
1120 default:
1121 return -EINVAL;
1125 const struct file_operations random_fops = {
1126 .read = random_read,
1127 .write = random_write,
1128 .poll = random_poll,
1129 .ioctl = random_ioctl,
1132 const struct file_operations urandom_fops = {
1133 .read = urandom_read,
1134 .write = random_write,
1135 .ioctl = random_ioctl,
1138 /***************************************************************
1139 * Random UUID interface
1141 * Used here for a Boot ID, but can be useful for other kernel
1142 * drivers.
1143 ***************************************************************/
1146 * Generate random UUID
1148 void generate_random_uuid(unsigned char uuid_out[16])
1150 get_random_bytes(uuid_out, 16);
1151 /* Set UUID version to 4 --- truely random generation */
1152 uuid_out[6] = (uuid_out[6] & 0x0F) | 0x40;
1153 /* Set the UUID variant to DCE */
1154 uuid_out[8] = (uuid_out[8] & 0x3F) | 0x80;
1157 EXPORT_SYMBOL(generate_random_uuid);
1159 /********************************************************************
1161 * Sysctl interface
1163 ********************************************************************/
1165 #ifdef CONFIG_SYSCTL
1167 #include <linux/sysctl.h>
1169 static int min_read_thresh = 8, min_write_thresh;
1170 static int max_read_thresh = INPUT_POOL_WORDS * 32;
1171 static int max_write_thresh = INPUT_POOL_WORDS * 32;
1172 static char sysctl_bootid[16];
1175 * These functions is used to return both the bootid UUID, and random
1176 * UUID. The difference is in whether table->data is NULL; if it is,
1177 * then a new UUID is generated and returned to the user.
1179 * If the user accesses this via the proc interface, it will be returned
1180 * as an ASCII string in the standard UUID format. If accesses via the
1181 * sysctl system call, it is returned as 16 bytes of binary data.
1183 static int proc_do_uuid(ctl_table *table, int write, struct file *filp,
1184 void __user *buffer, size_t *lenp, loff_t *ppos)
1186 ctl_table fake_table;
1187 unsigned char buf[64], tmp_uuid[16], *uuid;
1189 uuid = table->data;
1190 if (!uuid) {
1191 uuid = tmp_uuid;
1192 uuid[8] = 0;
1194 if (uuid[8] == 0)
1195 generate_random_uuid(uuid);
1197 sprintf(buf, "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-"
1198 "%02x%02x%02x%02x%02x%02x",
1199 uuid[0], uuid[1], uuid[2], uuid[3],
1200 uuid[4], uuid[5], uuid[6], uuid[7],
1201 uuid[8], uuid[9], uuid[10], uuid[11],
1202 uuid[12], uuid[13], uuid[14], uuid[15]);
1203 fake_table.data = buf;
1204 fake_table.maxlen = sizeof(buf);
1206 return proc_dostring(&fake_table, write, filp, buffer, lenp, ppos);
1209 static int uuid_strategy(ctl_table *table, int __user *name, int nlen,
1210 void __user *oldval, size_t __user *oldlenp,
1211 void __user *newval, size_t newlen)
1213 unsigned char tmp_uuid[16], *uuid;
1214 unsigned int len;
1216 if (!oldval || !oldlenp)
1217 return 1;
1219 uuid = table->data;
1220 if (!uuid) {
1221 uuid = tmp_uuid;
1222 uuid[8] = 0;
1224 if (uuid[8] == 0)
1225 generate_random_uuid(uuid);
1227 if (get_user(len, oldlenp))
1228 return -EFAULT;
1229 if (len) {
1230 if (len > 16)
1231 len = 16;
1232 if (copy_to_user(oldval, uuid, len) ||
1233 put_user(len, oldlenp))
1234 return -EFAULT;
1236 return 1;
1239 static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
1240 ctl_table random_table[] = {
1242 .ctl_name = RANDOM_POOLSIZE,
1243 .procname = "poolsize",
1244 .data = &sysctl_poolsize,
1245 .maxlen = sizeof(int),
1246 .mode = 0444,
1247 .proc_handler = &proc_dointvec,
1250 .ctl_name = RANDOM_ENTROPY_COUNT,
1251 .procname = "entropy_avail",
1252 .maxlen = sizeof(int),
1253 .mode = 0444,
1254 .proc_handler = &proc_dointvec,
1255 .data = &input_pool.entropy_count,
1258 .ctl_name = RANDOM_READ_THRESH,
1259 .procname = "read_wakeup_threshold",
1260 .data = &random_read_wakeup_thresh,
1261 .maxlen = sizeof(int),
1262 .mode = 0644,
1263 .proc_handler = &proc_dointvec_minmax,
1264 .strategy = &sysctl_intvec,
1265 .extra1 = &min_read_thresh,
1266 .extra2 = &max_read_thresh,
1269 .ctl_name = RANDOM_WRITE_THRESH,
1270 .procname = "write_wakeup_threshold",
1271 .data = &random_write_wakeup_thresh,
1272 .maxlen = sizeof(int),
1273 .mode = 0644,
1274 .proc_handler = &proc_dointvec_minmax,
1275 .strategy = &sysctl_intvec,
1276 .extra1 = &min_write_thresh,
1277 .extra2 = &max_write_thresh,
1280 .ctl_name = RANDOM_BOOT_ID,
1281 .procname = "boot_id",
1282 .data = &sysctl_bootid,
1283 .maxlen = 16,
1284 .mode = 0444,
1285 .proc_handler = &proc_do_uuid,
1286 .strategy = &uuid_strategy,
1289 .ctl_name = RANDOM_UUID,
1290 .procname = "uuid",
1291 .maxlen = 16,
1292 .mode = 0444,
1293 .proc_handler = &proc_do_uuid,
1294 .strategy = &uuid_strategy,
1296 { .ctl_name = 0 }
1298 #endif /* CONFIG_SYSCTL */
1300 /********************************************************************
1302 * Random funtions for networking
1304 ********************************************************************/
1307 * TCP initial sequence number picking. This uses the random number
1308 * generator to pick an initial secret value. This value is hashed
1309 * along with the TCP endpoint information to provide a unique
1310 * starting point for each pair of TCP endpoints. This defeats
1311 * attacks which rely on guessing the initial TCP sequence number.
1312 * This algorithm was suggested by Steve Bellovin.
1314 * Using a very strong hash was taking an appreciable amount of the total
1315 * TCP connection establishment time, so this is a weaker hash,
1316 * compensated for by changing the secret periodically.
1319 /* F, G and H are basic MD4 functions: selection, majority, parity */
1320 #define F(x, y, z) ((z) ^ ((x) & ((y) ^ (z))))
1321 #define G(x, y, z) (((x) & (y)) + (((x) ^ (y)) & (z)))
1322 #define H(x, y, z) ((x) ^ (y) ^ (z))
1325 * The generic round function. The application is so specific that
1326 * we don't bother protecting all the arguments with parens, as is generally
1327 * good macro practice, in favor of extra legibility.
1328 * Rotation is separate from addition to prevent recomputation
1330 #define ROUND(f, a, b, c, d, x, s) \
1331 (a += f(b, c, d) + x, a = (a << s) | (a >> (32 - s)))
1332 #define K1 0
1333 #define K2 013240474631UL
1334 #define K3 015666365641UL
1336 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1338 static __u32 twothirdsMD4Transform (__u32 const buf[4], __u32 const in[12])
1340 __u32 a = buf[0], b = buf[1], c = buf[2], d = buf[3];
1342 /* Round 1 */
1343 ROUND(F, a, b, c, d, in[ 0] + K1, 3);
1344 ROUND(F, d, a, b, c, in[ 1] + K1, 7);
1345 ROUND(F, c, d, a, b, in[ 2] + K1, 11);
1346 ROUND(F, b, c, d, a, in[ 3] + K1, 19);
1347 ROUND(F, a, b, c, d, in[ 4] + K1, 3);
1348 ROUND(F, d, a, b, c, in[ 5] + K1, 7);
1349 ROUND(F, c, d, a, b, in[ 6] + K1, 11);
1350 ROUND(F, b, c, d, a, in[ 7] + K1, 19);
1351 ROUND(F, a, b, c, d, in[ 8] + K1, 3);
1352 ROUND(F, d, a, b, c, in[ 9] + K1, 7);
1353 ROUND(F, c, d, a, b, in[10] + K1, 11);
1354 ROUND(F, b, c, d, a, in[11] + K1, 19);
1356 /* Round 2 */
1357 ROUND(G, a, b, c, d, in[ 1] + K2, 3);
1358 ROUND(G, d, a, b, c, in[ 3] + K2, 5);
1359 ROUND(G, c, d, a, b, in[ 5] + K2, 9);
1360 ROUND(G, b, c, d, a, in[ 7] + K2, 13);
1361 ROUND(G, a, b, c, d, in[ 9] + K2, 3);
1362 ROUND(G, d, a, b, c, in[11] + K2, 5);
1363 ROUND(G, c, d, a, b, in[ 0] + K2, 9);
1364 ROUND(G, b, c, d, a, in[ 2] + K2, 13);
1365 ROUND(G, a, b, c, d, in[ 4] + K2, 3);
1366 ROUND(G, d, a, b, c, in[ 6] + K2, 5);
1367 ROUND(G, c, d, a, b, in[ 8] + K2, 9);
1368 ROUND(G, b, c, d, a, in[10] + K2, 13);
1370 /* Round 3 */
1371 ROUND(H, a, b, c, d, in[ 3] + K3, 3);
1372 ROUND(H, d, a, b, c, in[ 7] + K3, 9);
1373 ROUND(H, c, d, a, b, in[11] + K3, 11);
1374 ROUND(H, b, c, d, a, in[ 2] + K3, 15);
1375 ROUND(H, a, b, c, d, in[ 6] + K3, 3);
1376 ROUND(H, d, a, b, c, in[10] + K3, 9);
1377 ROUND(H, c, d, a, b, in[ 1] + K3, 11);
1378 ROUND(H, b, c, d, a, in[ 5] + K3, 15);
1379 ROUND(H, a, b, c, d, in[ 9] + K3, 3);
1380 ROUND(H, d, a, b, c, in[ 0] + K3, 9);
1381 ROUND(H, c, d, a, b, in[ 4] + K3, 11);
1382 ROUND(H, b, c, d, a, in[ 8] + K3, 15);
1384 return buf[1] + b; /* "most hashed" word */
1385 /* Alternative: return sum of all words? */
1387 #endif
1389 #undef ROUND
1390 #undef F
1391 #undef G
1392 #undef H
1393 #undef K1
1394 #undef K2
1395 #undef K3
1397 /* This should not be decreased so low that ISNs wrap too fast. */
1398 #define REKEY_INTERVAL (300 * HZ)
1400 * Bit layout of the tcp sequence numbers (before adding current time):
1401 * bit 24-31: increased after every key exchange
1402 * bit 0-23: hash(source,dest)
1404 * The implementation is similar to the algorithm described
1405 * in the Appendix of RFC 1185, except that
1406 * - it uses a 1 MHz clock instead of a 250 kHz clock
1407 * - it performs a rekey every 5 minutes, which is equivalent
1408 * to a (source,dest) tulple dependent forward jump of the
1409 * clock by 0..2^(HASH_BITS+1)
1411 * Thus the average ISN wraparound time is 68 minutes instead of
1412 * 4.55 hours.
1414 * SMP cleanup and lock avoidance with poor man's RCU.
1415 * Manfred Spraul <manfred@colorfullife.com>
1418 #define COUNT_BITS 8
1419 #define COUNT_MASK ((1 << COUNT_BITS) - 1)
1420 #define HASH_BITS 24
1421 #define HASH_MASK ((1 << HASH_BITS) - 1)
1423 static struct keydata {
1424 __u32 count; /* already shifted to the final position */
1425 __u32 secret[12];
1426 } ____cacheline_aligned ip_keydata[2];
1428 static unsigned int ip_cnt;
1430 static void rekey_seq_generator(struct work_struct *work);
1432 static DECLARE_DELAYED_WORK(rekey_work, rekey_seq_generator);
1435 * Lock avoidance:
1436 * The ISN generation runs lockless - it's just a hash over random data.
1437 * State changes happen every 5 minutes when the random key is replaced.
1438 * Synchronization is performed by having two copies of the hash function
1439 * state and rekey_seq_generator always updates the inactive copy.
1440 * The copy is then activated by updating ip_cnt.
1441 * The implementation breaks down if someone blocks the thread
1442 * that processes SYN requests for more than 5 minutes. Should never
1443 * happen, and even if that happens only a not perfectly compliant
1444 * ISN is generated, nothing fatal.
1446 static void rekey_seq_generator(struct work_struct *work)
1448 struct keydata *keyptr = &ip_keydata[1 ^ (ip_cnt & 1)];
1450 get_random_bytes(keyptr->secret, sizeof(keyptr->secret));
1451 keyptr->count = (ip_cnt & COUNT_MASK) << HASH_BITS;
1452 smp_wmb();
1453 ip_cnt++;
1454 schedule_delayed_work(&rekey_work, REKEY_INTERVAL);
1457 static inline struct keydata *get_keyptr(void)
1459 struct keydata *keyptr = &ip_keydata[ip_cnt & 1];
1461 smp_rmb();
1463 return keyptr;
1466 static __init int seqgen_init(void)
1468 rekey_seq_generator(NULL);
1469 return 0;
1471 late_initcall(seqgen_init);
1473 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1474 __u32 secure_tcpv6_sequence_number(__be32 *saddr, __be32 *daddr,
1475 __be16 sport, __be16 dport)
1477 __u32 seq;
1478 __u32 hash[12];
1479 struct keydata *keyptr = get_keyptr();
1481 /* The procedure is the same as for IPv4, but addresses are longer.
1482 * Thus we must use twothirdsMD4Transform.
1485 memcpy(hash, saddr, 16);
1486 hash[4]=((__force u16)sport << 16) + (__force u16)dport;
1487 memcpy(&hash[5],keyptr->secret,sizeof(__u32) * 7);
1489 seq = twothirdsMD4Transform((const __u32 *)daddr, hash) & HASH_MASK;
1490 seq += keyptr->count;
1492 seq += ktime_get_real().tv64;
1494 return seq;
1496 EXPORT_SYMBOL(secure_tcpv6_sequence_number);
1497 #endif
1499 /* The code below is shamelessly stolen from secure_tcp_sequence_number().
1500 * All blames to Andrey V. Savochkin <saw@msu.ru>.
1502 __u32 secure_ip_id(__be32 daddr)
1504 struct keydata *keyptr;
1505 __u32 hash[4];
1507 keyptr = get_keyptr();
1510 * Pick a unique starting offset for each IP destination.
1511 * The dest ip address is placed in the starting vector,
1512 * which is then hashed with random data.
1514 hash[0] = (__force __u32)daddr;
1515 hash[1] = keyptr->secret[9];
1516 hash[2] = keyptr->secret[10];
1517 hash[3] = keyptr->secret[11];
1519 return half_md4_transform(hash, keyptr->secret);
1522 #ifdef CONFIG_INET
1524 __u32 secure_tcp_sequence_number(__be32 saddr, __be32 daddr,
1525 __be16 sport, __be16 dport)
1527 __u32 seq;
1528 __u32 hash[4];
1529 struct keydata *keyptr = get_keyptr();
1532 * Pick a unique starting offset for each TCP connection endpoints
1533 * (saddr, daddr, sport, dport).
1534 * Note that the words are placed into the starting vector, which is
1535 * then mixed with a partial MD4 over random data.
1537 hash[0]=(__force u32)saddr;
1538 hash[1]=(__force u32)daddr;
1539 hash[2]=((__force u16)sport << 16) + (__force u16)dport;
1540 hash[3]=keyptr->secret[11];
1542 seq = half_md4_transform(hash, keyptr->secret) & HASH_MASK;
1543 seq += keyptr->count;
1545 * As close as possible to RFC 793, which
1546 * suggests using a 250 kHz clock.
1547 * Further reading shows this assumes 2 Mb/s networks.
1548 * For 10 Gb/s Ethernet, a 1 GHz clock is appropriate.
1549 * That's funny, Linux has one built in! Use it!
1550 * (Networks are faster now - should this be increased?)
1552 seq += ktime_get_real().tv64;
1553 #if 0
1554 printk("init_seq(%lx, %lx, %d, %d) = %d\n",
1555 saddr, daddr, sport, dport, seq);
1556 #endif
1557 return seq;
1560 /* Generate secure starting point for ephemeral IPV4 transport port search */
1561 u32 secure_ipv4_port_ephemeral(__be32 saddr, __be32 daddr, __be16 dport)
1563 struct keydata *keyptr = get_keyptr();
1564 u32 hash[4];
1567 * Pick a unique starting offset for each ephemeral port search
1568 * (saddr, daddr, dport) and 48bits of random data.
1570 hash[0] = (__force u32)saddr;
1571 hash[1] = (__force u32)daddr;
1572 hash[2] = (__force u32)dport ^ keyptr->secret[10];
1573 hash[3] = keyptr->secret[11];
1575 return half_md4_transform(hash, keyptr->secret);
1578 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1579 u32 secure_ipv6_port_ephemeral(const __be32 *saddr, const __be32 *daddr, __be16 dport)
1581 struct keydata *keyptr = get_keyptr();
1582 u32 hash[12];
1584 memcpy(hash, saddr, 16);
1585 hash[4] = (__force u32)dport;
1586 memcpy(&hash[5],keyptr->secret,sizeof(__u32) * 7);
1588 return twothirdsMD4Transform((const __u32 *)daddr, hash);
1590 #endif
1592 #if defined(CONFIG_IP_DCCP) || defined(CONFIG_IP_DCCP_MODULE)
1593 /* Similar to secure_tcp_sequence_number but generate a 48 bit value
1594 * bit's 32-47 increase every key exchange
1595 * 0-31 hash(source, dest)
1597 u64 secure_dccp_sequence_number(__be32 saddr, __be32 daddr,
1598 __be16 sport, __be16 dport)
1600 u64 seq;
1601 __u32 hash[4];
1602 struct keydata *keyptr = get_keyptr();
1604 hash[0] = (__force u32)saddr;
1605 hash[1] = (__force u32)daddr;
1606 hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
1607 hash[3] = keyptr->secret[11];
1609 seq = half_md4_transform(hash, keyptr->secret);
1610 seq |= ((u64)keyptr->count) << (32 - HASH_BITS);
1612 seq += ktime_get_real().tv64;
1613 seq &= (1ull << 48) - 1;
1614 #if 0
1615 printk("dccp init_seq(%lx, %lx, %d, %d) = %d\n",
1616 saddr, daddr, sport, dport, seq);
1617 #endif
1618 return seq;
1621 EXPORT_SYMBOL(secure_dccp_sequence_number);
1622 #endif
1624 #endif /* CONFIG_INET */
1628 * Get a random word for internal kernel use only. Similar to urandom but
1629 * with the goal of minimal entropy pool depletion. As a result, the random
1630 * value is not cryptographically secure but for several uses the cost of
1631 * depleting entropy is too high
1633 unsigned int get_random_int(void)
1636 * Use IP's RNG. It suits our purpose perfectly: it re-keys itself
1637 * every second, from the entropy pool (and thus creates a limited
1638 * drain on it), and uses halfMD4Transform within the second. We
1639 * also mix it with jiffies and the PID:
1641 return secure_ip_id((__force __be32)(current->pid + jiffies));
1645 * randomize_range() returns a start address such that
1647 * [...... <range> .....]
1648 * start end
1650 * a <range> with size "len" starting at the return value is inside in the
1651 * area defined by [start, end], but is otherwise randomized.
1653 unsigned long
1654 randomize_range(unsigned long start, unsigned long end, unsigned long len)
1656 unsigned long range = end - len - start;
1658 if (end <= start + len)
1659 return 0;
1660 return PAGE_ALIGN(get_random_int() % range + start);