[PATCH] aic7xxx_osm build fix
[cris-mirror.git] / drivers / char / random.c
blob460b5d475edd2324628cc95ed13347807a74904b
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/config.h>
228 #include <linux/module.h>
229 #include <linux/kernel.h>
230 #include <linux/major.h>
231 #include <linux/string.h>
232 #include <linux/fcntl.h>
233 #include <linux/slab.h>
234 #include <linux/random.h>
235 #include <linux/poll.h>
236 #include <linux/init.h>
237 #include <linux/fs.h>
238 #include <linux/genhd.h>
239 #include <linux/interrupt.h>
240 #include <linux/spinlock.h>
241 #include <linux/percpu.h>
242 #include <linux/cryptohash.h>
244 #include <asm/processor.h>
245 #include <asm/uaccess.h>
246 #include <asm/irq.h>
247 #include <asm/io.h>
250 * Configuration information
252 #define INPUT_POOL_WORDS 128
253 #define OUTPUT_POOL_WORDS 32
254 #define SEC_XFER_SIZE 512
257 * The minimum number of bits of entropy before we wake up a read on
258 * /dev/random. Should be enough to do a significant reseed.
260 static int random_read_wakeup_thresh = 64;
263 * If the entropy count falls under this number of bits, then we
264 * should wake up processes which are selecting or polling on write
265 * access to /dev/random.
267 static int random_write_wakeup_thresh = 128;
270 * When the input pool goes over trickle_thresh, start dropping most
271 * samples to avoid wasting CPU time and reduce lock contention.
274 static int trickle_thresh = INPUT_POOL_WORDS * 28;
276 static DEFINE_PER_CPU(int, trickle_count) = 0;
279 * A pool of size .poolwords is stirred with a primitive polynomial
280 * of degree .poolwords over GF(2). The taps for various sizes are
281 * defined below. They are chosen to be evenly spaced (minimum RMS
282 * distance from evenly spaced; the numbers in the comments are a
283 * scaled squared error sum) except for the last tap, which is 1 to
284 * get the twisting happening as fast as possible.
286 static struct poolinfo {
287 int poolwords;
288 int tap1, tap2, tap3, tap4, tap5;
289 } poolinfo_table[] = {
290 /* x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 -- 105 */
291 { 128, 103, 76, 51, 25, 1 },
292 /* x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 -- 15 */
293 { 32, 26, 20, 14, 7, 1 },
294 #if 0
295 /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1 -- 115 */
296 { 2048, 1638, 1231, 819, 411, 1 },
298 /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
299 { 1024, 817, 615, 412, 204, 1 },
301 /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
302 { 1024, 819, 616, 410, 207, 2 },
304 /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
305 { 512, 411, 308, 208, 104, 1 },
307 /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
308 { 512, 409, 307, 206, 102, 2 },
309 /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
310 { 512, 409, 309, 205, 103, 2 },
312 /* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
313 { 256, 205, 155, 101, 52, 1 },
315 /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
316 { 128, 103, 78, 51, 27, 2 },
318 /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
319 { 64, 52, 39, 26, 14, 1 },
320 #endif
323 #define POOLBITS poolwords*32
324 #define POOLBYTES poolwords*4
327 * For the purposes of better mixing, we use the CRC-32 polynomial as
328 * well to make a twisted Generalized Feedback Shift Reigster
330 * (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR generators. ACM
331 * Transactions on Modeling and Computer Simulation 2(3):179-194.
332 * Also see M. Matsumoto & Y. Kurita, 1994. Twisted GFSR generators
333 * II. ACM Transactions on Mdeling and Computer Simulation 4:254-266)
335 * Thanks to Colin Plumb for suggesting this.
337 * We have not analyzed the resultant polynomial to prove it primitive;
338 * in fact it almost certainly isn't. Nonetheless, the irreducible factors
339 * of a random large-degree polynomial over GF(2) are more than large enough
340 * that periodicity is not a concern.
342 * The input hash is much less sensitive than the output hash. All
343 * that we want of it is that it be a good non-cryptographic hash;
344 * i.e. it not produce collisions when fed "random" data of the sort
345 * we expect to see. As long as the pool state differs for different
346 * inputs, we have preserved the input entropy and done a good job.
347 * The fact that an intelligent attacker can construct inputs that
348 * will produce controlled alterations to the pool's state is not
349 * important because we don't consider such inputs to contribute any
350 * randomness. The only property we need with respect to them is that
351 * the attacker can't increase his/her knowledge of the pool's state.
352 * Since all additions are reversible (knowing the final state and the
353 * input, you can reconstruct the initial state), if an attacker has
354 * any uncertainty about the initial state, he/she can only shuffle
355 * that uncertainty about, but never cause any collisions (which would
356 * decrease the uncertainty).
358 * The chosen system lets the state of the pool be (essentially) the input
359 * modulo the generator polymnomial. Now, for random primitive polynomials,
360 * this is a universal class of hash functions, meaning that the chance
361 * of a collision is limited by the attacker's knowledge of the generator
362 * polynomail, so if it is chosen at random, an attacker can never force
363 * a collision. Here, we use a fixed polynomial, but we *can* assume that
364 * ###--> it is unknown to the processes generating the input entropy. <-###
365 * Because of this important property, this is a good, collision-resistant
366 * hash; hash collisions will occur no more often than chance.
370 * Static global variables
372 static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
373 static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
375 #if 0
376 static int debug = 0;
377 module_param(debug, bool, 0644);
378 #define DEBUG_ENT(fmt, arg...) do { if (debug) \
379 printk(KERN_DEBUG "random %04d %04d %04d: " \
380 fmt,\
381 input_pool.entropy_count,\
382 blocking_pool.entropy_count,\
383 nonblocking_pool.entropy_count,\
384 ## arg); } while (0)
385 #else
386 #define DEBUG_ENT(fmt, arg...) do {} while (0)
387 #endif
389 /**********************************************************************
391 * OS independent entropy store. Here are the functions which handle
392 * storing entropy in an entropy pool.
394 **********************************************************************/
396 struct entropy_store;
397 struct entropy_store {
398 /* mostly-read data: */
399 struct poolinfo *poolinfo;
400 __u32 *pool;
401 const char *name;
402 int limit;
403 struct entropy_store *pull;
405 /* read-write data: */
406 spinlock_t lock ____cacheline_aligned_in_smp;
407 unsigned add_ptr;
408 int entropy_count;
409 int input_rotate;
412 static __u32 input_pool_data[INPUT_POOL_WORDS];
413 static __u32 blocking_pool_data[OUTPUT_POOL_WORDS];
414 static __u32 nonblocking_pool_data[OUTPUT_POOL_WORDS];
416 static struct entropy_store input_pool = {
417 .poolinfo = &poolinfo_table[0],
418 .name = "input",
419 .limit = 1,
420 .lock = SPIN_LOCK_UNLOCKED,
421 .pool = input_pool_data
424 static struct entropy_store blocking_pool = {
425 .poolinfo = &poolinfo_table[1],
426 .name = "blocking",
427 .limit = 1,
428 .pull = &input_pool,
429 .lock = SPIN_LOCK_UNLOCKED,
430 .pool = blocking_pool_data
433 static struct entropy_store nonblocking_pool = {
434 .poolinfo = &poolinfo_table[1],
435 .name = "nonblocking",
436 .pull = &input_pool,
437 .lock = SPIN_LOCK_UNLOCKED,
438 .pool = nonblocking_pool_data
442 * This function adds a byte into the entropy "pool". It does not
443 * update the entropy estimate. The caller should call
444 * credit_entropy_store if this is appropriate.
446 * The pool is stirred with a primitive polynomial of the appropriate
447 * degree, and then twisted. We twist by three bits at a time because
448 * it's cheap to do so and helps slightly in the expected case where
449 * the entropy is concentrated in the low-order bits.
451 static void __add_entropy_words(struct entropy_store *r, const __u32 *in,
452 int nwords, __u32 out[16])
454 static __u32 const twist_table[8] = {
455 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
456 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
457 unsigned long i, add_ptr, tap1, tap2, tap3, tap4, tap5;
458 int new_rotate, input_rotate;
459 int wordmask = r->poolinfo->poolwords - 1;
460 __u32 w, next_w;
461 unsigned long flags;
463 /* Taps are constant, so we can load them without holding r->lock. */
464 tap1 = r->poolinfo->tap1;
465 tap2 = r->poolinfo->tap2;
466 tap3 = r->poolinfo->tap3;
467 tap4 = r->poolinfo->tap4;
468 tap5 = r->poolinfo->tap5;
469 next_w = *in++;
471 spin_lock_irqsave(&r->lock, flags);
472 prefetch_range(r->pool, wordmask);
473 input_rotate = r->input_rotate;
474 add_ptr = r->add_ptr;
476 while (nwords--) {
477 w = rol32(next_w, input_rotate);
478 if (nwords > 0)
479 next_w = *in++;
480 i = add_ptr = (add_ptr - 1) & wordmask;
482 * Normally, we add 7 bits of rotation to the pool.
483 * At the beginning of the pool, add an extra 7 bits
484 * rotation, so that successive passes spread the
485 * input bits across the pool evenly.
487 new_rotate = input_rotate + 14;
488 if (i)
489 new_rotate = input_rotate + 7;
490 input_rotate = new_rotate & 31;
492 /* XOR in the various taps */
493 w ^= r->pool[(i + tap1) & wordmask];
494 w ^= r->pool[(i + tap2) & wordmask];
495 w ^= r->pool[(i + tap3) & wordmask];
496 w ^= r->pool[(i + tap4) & wordmask];
497 w ^= r->pool[(i + tap5) & wordmask];
498 w ^= r->pool[i];
499 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
502 r->input_rotate = input_rotate;
503 r->add_ptr = add_ptr;
505 if (out) {
506 for (i = 0; i < 16; i++) {
507 out[i] = r->pool[add_ptr];
508 add_ptr = (add_ptr - 1) & wordmask;
512 spin_unlock_irqrestore(&r->lock, flags);
515 static inline void add_entropy_words(struct entropy_store *r, const __u32 *in,
516 int nwords)
518 __add_entropy_words(r, in, nwords, NULL);
522 * Credit (or debit) the entropy store with n bits of entropy
524 static void credit_entropy_store(struct entropy_store *r, int nbits)
526 unsigned long flags;
528 spin_lock_irqsave(&r->lock, flags);
530 if (r->entropy_count + nbits < 0) {
531 DEBUG_ENT("negative entropy/overflow (%d+%d)\n",
532 r->entropy_count, nbits);
533 r->entropy_count = 0;
534 } else if (r->entropy_count + nbits > r->poolinfo->POOLBITS) {
535 r->entropy_count = r->poolinfo->POOLBITS;
536 } else {
537 r->entropy_count += nbits;
538 if (nbits)
539 DEBUG_ENT("added %d entropy credits to %s\n",
540 nbits, r->name);
543 spin_unlock_irqrestore(&r->lock, flags);
546 /*********************************************************************
548 * Entropy input management
550 *********************************************************************/
552 /* There is one of these per entropy source */
553 struct timer_rand_state {
554 cycles_t last_time;
555 long last_delta,last_delta2;
556 unsigned dont_count_entropy:1;
559 static struct timer_rand_state input_timer_state;
560 static struct timer_rand_state *irq_timer_state[NR_IRQS];
563 * This function adds entropy to the entropy "pool" by using timing
564 * delays. It uses the timer_rand_state structure to make an estimate
565 * of how many bits of entropy this call has added to the pool.
567 * The number "num" is also added to the pool - it should somehow describe
568 * the type of event which just happened. This is currently 0-255 for
569 * keyboard scan codes, and 256 upwards for interrupts.
572 static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
574 struct {
575 cycles_t cycles;
576 long jiffies;
577 unsigned num;
578 } sample;
579 long delta, delta2, delta3;
581 preempt_disable();
582 /* if over the trickle threshold, use only 1 in 4096 samples */
583 if (input_pool.entropy_count > trickle_thresh &&
584 (__get_cpu_var(trickle_count)++ & 0xfff))
585 goto out;
587 sample.jiffies = jiffies;
588 sample.cycles = get_cycles();
589 sample.num = num;
590 add_entropy_words(&input_pool, (u32 *)&sample, sizeof(sample)/4);
593 * Calculate number of bits of randomness we probably added.
594 * We take into account the first, second and third-order deltas
595 * in order to make our estimate.
598 if (!state->dont_count_entropy) {
599 delta = sample.jiffies - state->last_time;
600 state->last_time = sample.jiffies;
602 delta2 = delta - state->last_delta;
603 state->last_delta = delta;
605 delta3 = delta2 - state->last_delta2;
606 state->last_delta2 = delta2;
608 if (delta < 0)
609 delta = -delta;
610 if (delta2 < 0)
611 delta2 = -delta2;
612 if (delta3 < 0)
613 delta3 = -delta3;
614 if (delta > delta2)
615 delta = delta2;
616 if (delta > delta3)
617 delta = delta3;
620 * delta is now minimum absolute delta.
621 * Round down by 1 bit on general principles,
622 * and limit entropy entimate to 12 bits.
624 credit_entropy_store(&input_pool,
625 min_t(int, fls(delta>>1), 11));
628 if(input_pool.entropy_count >= random_read_wakeup_thresh)
629 wake_up_interruptible(&random_read_wait);
631 out:
632 preempt_enable();
635 extern void add_input_randomness(unsigned int type, unsigned int code,
636 unsigned int value)
638 static unsigned char last_value;
640 /* ignore autorepeat and the like */
641 if (value == last_value)
642 return;
644 DEBUG_ENT("input event\n");
645 last_value = value;
646 add_timer_randomness(&input_timer_state,
647 (type << 4) ^ code ^ (code >> 4) ^ value);
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 void add_disk_randomness(struct gendisk *disk)
661 if (!disk || !disk->random)
662 return;
663 /* first major is 1, so we get >= 0x200 here */
664 DEBUG_ENT("disk event %d:%d\n", disk->major, disk->first_minor);
666 add_timer_randomness(disk->random,
667 0x100 + MKDEV(disk->major, disk->first_minor));
670 EXPORT_SYMBOL(add_disk_randomness);
672 #define EXTRACT_SIZE 10
674 /*********************************************************************
676 * Entropy extraction routines
678 *********************************************************************/
680 static ssize_t extract_entropy(struct entropy_store *r, void * buf,
681 size_t nbytes, int min, int rsvd);
684 * This utility inline function is responsible for transfering entropy
685 * from the primary pool to the secondary extraction pool. We make
686 * sure we pull enough for a 'catastrophic reseed'.
688 static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
690 __u32 tmp[OUTPUT_POOL_WORDS];
692 if (r->pull && r->entropy_count < nbytes * 8 &&
693 r->entropy_count < r->poolinfo->POOLBITS) {
694 int bytes = max_t(int, random_read_wakeup_thresh / 8,
695 min_t(int, nbytes, sizeof(tmp)));
696 int rsvd = r->limit ? 0 : random_read_wakeup_thresh/4;
698 DEBUG_ENT("going to reseed %s with %d bits "
699 "(%d of %d requested)\n",
700 r->name, bytes * 8, nbytes * 8, r->entropy_count);
702 bytes=extract_entropy(r->pull, tmp, bytes,
703 random_read_wakeup_thresh / 8, rsvd);
704 add_entropy_words(r, tmp, (bytes + 3) / 4);
705 credit_entropy_store(r, bytes*8);
710 * These functions extracts randomness from the "entropy pool", and
711 * returns it in a buffer.
713 * The min parameter specifies the minimum amount we can pull before
714 * failing to avoid races that defeat catastrophic reseeding while the
715 * reserved parameter indicates how much entropy we must leave in the
716 * pool after each pull to avoid starving other readers.
718 * Note: extract_entropy() assumes that .poolwords is a multiple of 16 words.
721 static size_t account(struct entropy_store *r, size_t nbytes, int min,
722 int reserved)
724 unsigned long flags;
726 BUG_ON(r->entropy_count > r->poolinfo->POOLBITS);
728 /* Hold lock while accounting */
729 spin_lock_irqsave(&r->lock, flags);
731 DEBUG_ENT("trying to extract %d bits from %s\n",
732 nbytes * 8, r->name);
734 /* Can we pull enough? */
735 if (r->entropy_count / 8 < min + reserved) {
736 nbytes = 0;
737 } else {
738 /* If limited, never pull more than available */
739 if (r->limit && nbytes + reserved >= r->entropy_count / 8)
740 nbytes = r->entropy_count/8 - reserved;
742 if(r->entropy_count / 8 >= nbytes + reserved)
743 r->entropy_count -= nbytes*8;
744 else
745 r->entropy_count = reserved;
747 if (r->entropy_count < random_write_wakeup_thresh)
748 wake_up_interruptible(&random_write_wait);
751 DEBUG_ENT("debiting %d entropy credits from %s%s\n",
752 nbytes * 8, r->name, r->limit ? "" : " (unlimited)");
754 spin_unlock_irqrestore(&r->lock, flags);
756 return nbytes;
759 static void extract_buf(struct entropy_store *r, __u8 *out)
761 int i, x;
762 __u32 data[16], buf[5 + SHA_WORKSPACE_WORDS];
764 sha_init(buf);
766 * As we hash the pool, we mix intermediate values of
767 * the hash back into the pool. This eliminates
768 * backtracking attacks (where the attacker knows
769 * the state of the pool plus the current outputs, and
770 * attempts to find previous ouputs), unless the hash
771 * function can be inverted.
773 for (i = 0, x = 0; i < r->poolinfo->poolwords; i += 16, x+=2) {
774 sha_transform(buf, (__u8 *)r->pool+i, buf + 5);
775 add_entropy_words(r, &buf[x % 5], 1);
779 * To avoid duplicates, we atomically extract a
780 * portion of the pool while mixing, and hash one
781 * final time.
783 __add_entropy_words(r, &buf[x % 5], 1, data);
784 sha_transform(buf, (__u8 *)data, buf + 5);
787 * In case the hash function has some recognizable
788 * output pattern, we fold it in half.
791 buf[0] ^= buf[3];
792 buf[1] ^= buf[4];
793 buf[0] ^= rol32(buf[3], 16);
794 memcpy(out, buf, EXTRACT_SIZE);
795 memset(buf, 0, sizeof(buf));
798 static ssize_t extract_entropy(struct entropy_store *r, void * buf,
799 size_t nbytes, int min, int reserved)
801 ssize_t ret = 0, i;
802 __u8 tmp[EXTRACT_SIZE];
804 xfer_secondary_pool(r, nbytes);
805 nbytes = account(r, nbytes, min, reserved);
807 while (nbytes) {
808 extract_buf(r, tmp);
809 i = min_t(int, nbytes, EXTRACT_SIZE);
810 memcpy(buf, tmp, i);
811 nbytes -= i;
812 buf += i;
813 ret += i;
816 /* Wipe data just returned from memory */
817 memset(tmp, 0, sizeof(tmp));
819 return ret;
822 static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
823 size_t nbytes)
825 ssize_t ret = 0, i;
826 __u8 tmp[EXTRACT_SIZE];
828 xfer_secondary_pool(r, nbytes);
829 nbytes = account(r, nbytes, 0, 0);
831 while (nbytes) {
832 if (need_resched()) {
833 if (signal_pending(current)) {
834 if (ret == 0)
835 ret = -ERESTARTSYS;
836 break;
838 schedule();
841 extract_buf(r, tmp);
842 i = min_t(int, nbytes, EXTRACT_SIZE);
843 if (copy_to_user(buf, tmp, i)) {
844 ret = -EFAULT;
845 break;
848 nbytes -= i;
849 buf += i;
850 ret += i;
853 /* Wipe data just returned from memory */
854 memset(tmp, 0, sizeof(tmp));
856 return ret;
860 * This function is the exported kernel interface. It returns some
861 * number of good random numbers, suitable for seeding TCP sequence
862 * numbers, etc.
864 void get_random_bytes(void *buf, int nbytes)
866 extract_entropy(&nonblocking_pool, buf, nbytes, 0, 0);
869 EXPORT_SYMBOL(get_random_bytes);
872 * init_std_data - initialize pool with system data
874 * @r: pool to initialize
876 * This function clears the pool's entropy count and mixes some system
877 * data into the pool to prepare it for use. The pool is not cleared
878 * as that can only decrease the entropy in the pool.
880 static void init_std_data(struct entropy_store *r)
882 struct timeval tv;
883 unsigned long flags;
885 spin_lock_irqsave(&r->lock, flags);
886 r->entropy_count = 0;
887 spin_unlock_irqrestore(&r->lock, flags);
889 do_gettimeofday(&tv);
890 add_entropy_words(r, (__u32 *)&tv, sizeof(tv)/4);
891 add_entropy_words(r, (__u32 *)&system_utsname,
892 sizeof(system_utsname)/4);
895 static int __init rand_initialize(void)
897 init_std_data(&input_pool);
898 init_std_data(&blocking_pool);
899 init_std_data(&nonblocking_pool);
900 return 0;
902 module_init(rand_initialize);
904 void rand_initialize_irq(int irq)
906 struct timer_rand_state *state;
908 if (irq >= NR_IRQS || irq_timer_state[irq])
909 return;
912 * If kmalloc returns null, we just won't use that entropy
913 * source.
915 state = kmalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
916 if (state) {
917 memset(state, 0, sizeof(struct timer_rand_state));
918 irq_timer_state[irq] = state;
922 void rand_initialize_disk(struct gendisk *disk)
924 struct timer_rand_state *state;
927 * If kmalloc returns null, we just won't use that entropy
928 * source.
930 state = kmalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
931 if (state) {
932 memset(state, 0, sizeof(struct timer_rand_state));
933 disk->random = state;
937 static ssize_t
938 random_read(struct file * file, char __user * buf, size_t nbytes, loff_t *ppos)
940 ssize_t n, retval = 0, count = 0;
942 if (nbytes == 0)
943 return 0;
945 while (nbytes > 0) {
946 n = nbytes;
947 if (n > SEC_XFER_SIZE)
948 n = SEC_XFER_SIZE;
950 DEBUG_ENT("reading %d bits\n", n*8);
952 n = extract_entropy_user(&blocking_pool, buf, n);
954 DEBUG_ENT("read got %d bits (%d still needed)\n",
955 n*8, (nbytes-n)*8);
957 if (n == 0) {
958 if (file->f_flags & O_NONBLOCK) {
959 retval = -EAGAIN;
960 break;
963 DEBUG_ENT("sleeping?\n");
965 wait_event_interruptible(random_read_wait,
966 input_pool.entropy_count >=
967 random_read_wakeup_thresh);
969 DEBUG_ENT("awake\n");
971 if (signal_pending(current)) {
972 retval = -ERESTARTSYS;
973 break;
976 continue;
979 if (n < 0) {
980 retval = n;
981 break;
983 count += n;
984 buf += n;
985 nbytes -= n;
986 break; /* This break makes the device work */
987 /* like a named pipe */
991 * If we gave the user some bytes, update the access time.
993 if (count)
994 file_accessed(file);
996 return (count ? count : retval);
999 static ssize_t
1000 urandom_read(struct file * file, char __user * buf,
1001 size_t nbytes, loff_t *ppos)
1003 return extract_entropy_user(&nonblocking_pool, buf, nbytes);
1006 static unsigned int
1007 random_poll(struct file *file, poll_table * wait)
1009 unsigned int mask;
1011 poll_wait(file, &random_read_wait, wait);
1012 poll_wait(file, &random_write_wait, wait);
1013 mask = 0;
1014 if (input_pool.entropy_count >= random_read_wakeup_thresh)
1015 mask |= POLLIN | POLLRDNORM;
1016 if (input_pool.entropy_count < random_write_wakeup_thresh)
1017 mask |= POLLOUT | POLLWRNORM;
1018 return mask;
1021 static ssize_t
1022 random_write(struct file * file, const char __user * buffer,
1023 size_t count, loff_t *ppos)
1025 int ret = 0;
1026 size_t bytes;
1027 __u32 buf[16];
1028 const char __user *p = buffer;
1029 size_t c = count;
1031 while (c > 0) {
1032 bytes = min(c, sizeof(buf));
1034 bytes -= copy_from_user(&buf, p, bytes);
1035 if (!bytes) {
1036 ret = -EFAULT;
1037 break;
1039 c -= bytes;
1040 p += bytes;
1042 add_entropy_words(&input_pool, buf, (bytes + 3) / 4);
1044 if (p == buffer) {
1045 return (ssize_t)ret;
1046 } else {
1047 struct inode *inode = file->f_dentry->d_inode;
1048 inode->i_mtime = current_fs_time(inode->i_sb);
1049 mark_inode_dirty(inode);
1050 return (ssize_t)(p - buffer);
1054 static int
1055 random_ioctl(struct inode * inode, struct file * file,
1056 unsigned int cmd, unsigned long arg)
1058 int size, ent_count;
1059 int __user *p = (int __user *)arg;
1060 int retval;
1062 switch (cmd) {
1063 case RNDGETENTCNT:
1064 ent_count = input_pool.entropy_count;
1065 if (put_user(ent_count, p))
1066 return -EFAULT;
1067 return 0;
1068 case RNDADDTOENTCNT:
1069 if (!capable(CAP_SYS_ADMIN))
1070 return -EPERM;
1071 if (get_user(ent_count, p))
1072 return -EFAULT;
1073 credit_entropy_store(&input_pool, ent_count);
1075 * Wake up waiting processes if we have enough
1076 * entropy.
1078 if (input_pool.entropy_count >= random_read_wakeup_thresh)
1079 wake_up_interruptible(&random_read_wait);
1080 return 0;
1081 case RNDADDENTROPY:
1082 if (!capable(CAP_SYS_ADMIN))
1083 return -EPERM;
1084 if (get_user(ent_count, p++))
1085 return -EFAULT;
1086 if (ent_count < 0)
1087 return -EINVAL;
1088 if (get_user(size, p++))
1089 return -EFAULT;
1090 retval = random_write(file, (const char __user *) p,
1091 size, &file->f_pos);
1092 if (retval < 0)
1093 return retval;
1094 credit_entropy_store(&input_pool, ent_count);
1096 * Wake up waiting processes if we have enough
1097 * entropy.
1099 if (input_pool.entropy_count >= random_read_wakeup_thresh)
1100 wake_up_interruptible(&random_read_wait);
1101 return 0;
1102 case RNDZAPENTCNT:
1103 case RNDCLEARPOOL:
1104 /* Clear the entropy pool counters. */
1105 if (!capable(CAP_SYS_ADMIN))
1106 return -EPERM;
1107 init_std_data(&input_pool);
1108 init_std_data(&blocking_pool);
1109 init_std_data(&nonblocking_pool);
1110 return 0;
1111 default:
1112 return -EINVAL;
1116 struct file_operations random_fops = {
1117 .read = random_read,
1118 .write = random_write,
1119 .poll = random_poll,
1120 .ioctl = random_ioctl,
1123 struct file_operations urandom_fops = {
1124 .read = urandom_read,
1125 .write = random_write,
1126 .ioctl = random_ioctl,
1129 /***************************************************************
1130 * Random UUID interface
1132 * Used here for a Boot ID, but can be useful for other kernel
1133 * drivers.
1134 ***************************************************************/
1137 * Generate random UUID
1139 void generate_random_uuid(unsigned char uuid_out[16])
1141 get_random_bytes(uuid_out, 16);
1142 /* Set UUID version to 4 --- truely random generation */
1143 uuid_out[6] = (uuid_out[6] & 0x0F) | 0x40;
1144 /* Set the UUID variant to DCE */
1145 uuid_out[8] = (uuid_out[8] & 0x3F) | 0x80;
1148 EXPORT_SYMBOL(generate_random_uuid);
1150 /********************************************************************
1152 * Sysctl interface
1154 ********************************************************************/
1156 #ifdef CONFIG_SYSCTL
1158 #include <linux/sysctl.h>
1160 static int min_read_thresh = 8, min_write_thresh;
1161 static int max_read_thresh = INPUT_POOL_WORDS * 32;
1162 static int max_write_thresh = INPUT_POOL_WORDS * 32;
1163 static char sysctl_bootid[16];
1166 * These functions is used to return both the bootid UUID, and random
1167 * UUID. The difference is in whether table->data is NULL; if it is,
1168 * then a new UUID is generated and returned to the user.
1170 * If the user accesses this via the proc interface, it will be returned
1171 * as an ASCII string in the standard UUID format. If accesses via the
1172 * sysctl system call, it is returned as 16 bytes of binary data.
1174 static int proc_do_uuid(ctl_table *table, int write, struct file *filp,
1175 void __user *buffer, size_t *lenp, loff_t *ppos)
1177 ctl_table fake_table;
1178 unsigned char buf[64], tmp_uuid[16], *uuid;
1180 uuid = table->data;
1181 if (!uuid) {
1182 uuid = tmp_uuid;
1183 uuid[8] = 0;
1185 if (uuid[8] == 0)
1186 generate_random_uuid(uuid);
1188 sprintf(buf, "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-"
1189 "%02x%02x%02x%02x%02x%02x",
1190 uuid[0], uuid[1], uuid[2], uuid[3],
1191 uuid[4], uuid[5], uuid[6], uuid[7],
1192 uuid[8], uuid[9], uuid[10], uuid[11],
1193 uuid[12], uuid[13], uuid[14], uuid[15]);
1194 fake_table.data = buf;
1195 fake_table.maxlen = sizeof(buf);
1197 return proc_dostring(&fake_table, write, filp, buffer, lenp, ppos);
1200 static int uuid_strategy(ctl_table *table, int __user *name, int nlen,
1201 void __user *oldval, size_t __user *oldlenp,
1202 void __user *newval, size_t newlen, void **context)
1204 unsigned char tmp_uuid[16], *uuid;
1205 unsigned int len;
1207 if (!oldval || !oldlenp)
1208 return 1;
1210 uuid = table->data;
1211 if (!uuid) {
1212 uuid = tmp_uuid;
1213 uuid[8] = 0;
1215 if (uuid[8] == 0)
1216 generate_random_uuid(uuid);
1218 if (get_user(len, oldlenp))
1219 return -EFAULT;
1220 if (len) {
1221 if (len > 16)
1222 len = 16;
1223 if (copy_to_user(oldval, uuid, len) ||
1224 put_user(len, oldlenp))
1225 return -EFAULT;
1227 return 1;
1230 static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
1231 ctl_table random_table[] = {
1233 .ctl_name = RANDOM_POOLSIZE,
1234 .procname = "poolsize",
1235 .data = &sysctl_poolsize,
1236 .maxlen = sizeof(int),
1237 .mode = 0444,
1238 .proc_handler = &proc_dointvec,
1241 .ctl_name = RANDOM_ENTROPY_COUNT,
1242 .procname = "entropy_avail",
1243 .maxlen = sizeof(int),
1244 .mode = 0444,
1245 .proc_handler = &proc_dointvec,
1246 .data = &input_pool.entropy_count,
1249 .ctl_name = RANDOM_READ_THRESH,
1250 .procname = "read_wakeup_threshold",
1251 .data = &random_read_wakeup_thresh,
1252 .maxlen = sizeof(int),
1253 .mode = 0644,
1254 .proc_handler = &proc_dointvec_minmax,
1255 .strategy = &sysctl_intvec,
1256 .extra1 = &min_read_thresh,
1257 .extra2 = &max_read_thresh,
1260 .ctl_name = RANDOM_WRITE_THRESH,
1261 .procname = "write_wakeup_threshold",
1262 .data = &random_write_wakeup_thresh,
1263 .maxlen = sizeof(int),
1264 .mode = 0644,
1265 .proc_handler = &proc_dointvec_minmax,
1266 .strategy = &sysctl_intvec,
1267 .extra1 = &min_write_thresh,
1268 .extra2 = &max_write_thresh,
1271 .ctl_name = RANDOM_BOOT_ID,
1272 .procname = "boot_id",
1273 .data = &sysctl_bootid,
1274 .maxlen = 16,
1275 .mode = 0444,
1276 .proc_handler = &proc_do_uuid,
1277 .strategy = &uuid_strategy,
1280 .ctl_name = RANDOM_UUID,
1281 .procname = "uuid",
1282 .maxlen = 16,
1283 .mode = 0444,
1284 .proc_handler = &proc_do_uuid,
1285 .strategy = &uuid_strategy,
1287 { .ctl_name = 0 }
1289 #endif /* CONFIG_SYSCTL */
1291 /********************************************************************
1293 * Random funtions for networking
1295 ********************************************************************/
1298 * TCP initial sequence number picking. This uses the random number
1299 * generator to pick an initial secret value. This value is hashed
1300 * along with the TCP endpoint information to provide a unique
1301 * starting point for each pair of TCP endpoints. This defeats
1302 * attacks which rely on guessing the initial TCP sequence number.
1303 * This algorithm was suggested by Steve Bellovin.
1305 * Using a very strong hash was taking an appreciable amount of the total
1306 * TCP connection establishment time, so this is a weaker hash,
1307 * compensated for by changing the secret periodically.
1310 /* F, G and H are basic MD4 functions: selection, majority, parity */
1311 #define F(x, y, z) ((z) ^ ((x) & ((y) ^ (z))))
1312 #define G(x, y, z) (((x) & (y)) + (((x) ^ (y)) & (z)))
1313 #define H(x, y, z) ((x) ^ (y) ^ (z))
1316 * The generic round function. The application is so specific that
1317 * we don't bother protecting all the arguments with parens, as is generally
1318 * good macro practice, in favor of extra legibility.
1319 * Rotation is separate from addition to prevent recomputation
1321 #define ROUND(f, a, b, c, d, x, s) \
1322 (a += f(b, c, d) + x, a = (a << s) | (a >> (32 - s)))
1323 #define K1 0
1324 #define K2 013240474631UL
1325 #define K3 015666365641UL
1327 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1329 static __u32 twothirdsMD4Transform (__u32 const buf[4], __u32 const in[12])
1331 __u32 a = buf[0], b = buf[1], c = buf[2], d = buf[3];
1333 /* Round 1 */
1334 ROUND(F, a, b, c, d, in[ 0] + K1, 3);
1335 ROUND(F, d, a, b, c, in[ 1] + K1, 7);
1336 ROUND(F, c, d, a, b, in[ 2] + K1, 11);
1337 ROUND(F, b, c, d, a, in[ 3] + K1, 19);
1338 ROUND(F, a, b, c, d, in[ 4] + K1, 3);
1339 ROUND(F, d, a, b, c, in[ 5] + K1, 7);
1340 ROUND(F, c, d, a, b, in[ 6] + K1, 11);
1341 ROUND(F, b, c, d, a, in[ 7] + K1, 19);
1342 ROUND(F, a, b, c, d, in[ 8] + K1, 3);
1343 ROUND(F, d, a, b, c, in[ 9] + K1, 7);
1344 ROUND(F, c, d, a, b, in[10] + K1, 11);
1345 ROUND(F, b, c, d, a, in[11] + K1, 19);
1347 /* Round 2 */
1348 ROUND(G, a, b, c, d, in[ 1] + K2, 3);
1349 ROUND(G, d, a, b, c, in[ 3] + K2, 5);
1350 ROUND(G, c, d, a, b, in[ 5] + K2, 9);
1351 ROUND(G, b, c, d, a, in[ 7] + K2, 13);
1352 ROUND(G, a, b, c, d, in[ 9] + K2, 3);
1353 ROUND(G, d, a, b, c, in[11] + K2, 5);
1354 ROUND(G, c, d, a, b, in[ 0] + K2, 9);
1355 ROUND(G, b, c, d, a, in[ 2] + K2, 13);
1356 ROUND(G, a, b, c, d, in[ 4] + K2, 3);
1357 ROUND(G, d, a, b, c, in[ 6] + K2, 5);
1358 ROUND(G, c, d, a, b, in[ 8] + K2, 9);
1359 ROUND(G, b, c, d, a, in[10] + K2, 13);
1361 /* Round 3 */
1362 ROUND(H, a, b, c, d, in[ 3] + K3, 3);
1363 ROUND(H, d, a, b, c, in[ 7] + K3, 9);
1364 ROUND(H, c, d, a, b, in[11] + K3, 11);
1365 ROUND(H, b, c, d, a, in[ 2] + K3, 15);
1366 ROUND(H, a, b, c, d, in[ 6] + K3, 3);
1367 ROUND(H, d, a, b, c, in[10] + K3, 9);
1368 ROUND(H, c, d, a, b, in[ 1] + K3, 11);
1369 ROUND(H, b, c, d, a, in[ 5] + K3, 15);
1370 ROUND(H, a, b, c, d, in[ 9] + K3, 3);
1371 ROUND(H, d, a, b, c, in[ 0] + K3, 9);
1372 ROUND(H, c, d, a, b, in[ 4] + K3, 11);
1373 ROUND(H, b, c, d, a, in[ 8] + K3, 15);
1375 return buf[1] + b; /* "most hashed" word */
1376 /* Alternative: return sum of all words? */
1378 #endif
1380 #undef ROUND
1381 #undef F
1382 #undef G
1383 #undef H
1384 #undef K1
1385 #undef K2
1386 #undef K3
1388 /* This should not be decreased so low that ISNs wrap too fast. */
1389 #define REKEY_INTERVAL (300 * HZ)
1391 * Bit layout of the tcp sequence numbers (before adding current time):
1392 * bit 24-31: increased after every key exchange
1393 * bit 0-23: hash(source,dest)
1395 * The implementation is similar to the algorithm described
1396 * in the Appendix of RFC 1185, except that
1397 * - it uses a 1 MHz clock instead of a 250 kHz clock
1398 * - it performs a rekey every 5 minutes, which is equivalent
1399 * to a (source,dest) tulple dependent forward jump of the
1400 * clock by 0..2^(HASH_BITS+1)
1402 * Thus the average ISN wraparound time is 68 minutes instead of
1403 * 4.55 hours.
1405 * SMP cleanup and lock avoidance with poor man's RCU.
1406 * Manfred Spraul <manfred@colorfullife.com>
1409 #define COUNT_BITS 8
1410 #define COUNT_MASK ((1 << COUNT_BITS) - 1)
1411 #define HASH_BITS 24
1412 #define HASH_MASK ((1 << HASH_BITS) - 1)
1414 static struct keydata {
1415 __u32 count; /* already shifted to the final position */
1416 __u32 secret[12];
1417 } ____cacheline_aligned ip_keydata[2];
1419 static unsigned int ip_cnt;
1421 static void rekey_seq_generator(void *private_);
1423 static DECLARE_WORK(rekey_work, rekey_seq_generator, NULL);
1426 * Lock avoidance:
1427 * The ISN generation runs lockless - it's just a hash over random data.
1428 * State changes happen every 5 minutes when the random key is replaced.
1429 * Synchronization is performed by having two copies of the hash function
1430 * state and rekey_seq_generator always updates the inactive copy.
1431 * The copy is then activated by updating ip_cnt.
1432 * The implementation breaks down if someone blocks the thread
1433 * that processes SYN requests for more than 5 minutes. Should never
1434 * happen, and even if that happens only a not perfectly compliant
1435 * ISN is generated, nothing fatal.
1437 static void rekey_seq_generator(void *private_)
1439 struct keydata *keyptr = &ip_keydata[1 ^ (ip_cnt & 1)];
1441 get_random_bytes(keyptr->secret, sizeof(keyptr->secret));
1442 keyptr->count = (ip_cnt & COUNT_MASK) << HASH_BITS;
1443 smp_wmb();
1444 ip_cnt++;
1445 schedule_delayed_work(&rekey_work, REKEY_INTERVAL);
1448 static inline struct keydata *get_keyptr(void)
1450 struct keydata *keyptr = &ip_keydata[ip_cnt & 1];
1452 smp_rmb();
1454 return keyptr;
1457 static __init int seqgen_init(void)
1459 rekey_seq_generator(NULL);
1460 return 0;
1462 late_initcall(seqgen_init);
1464 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1465 __u32 secure_tcpv6_sequence_number(__u32 *saddr, __u32 *daddr,
1466 __u16 sport, __u16 dport)
1468 struct timeval tv;
1469 __u32 seq;
1470 __u32 hash[12];
1471 struct keydata *keyptr = get_keyptr();
1473 /* The procedure is the same as for IPv4, but addresses are longer.
1474 * Thus we must use twothirdsMD4Transform.
1477 memcpy(hash, saddr, 16);
1478 hash[4]=(sport << 16) + dport;
1479 memcpy(&hash[5],keyptr->secret,sizeof(__u32) * 7);
1481 seq = twothirdsMD4Transform(daddr, hash) & HASH_MASK;
1482 seq += keyptr->count;
1484 do_gettimeofday(&tv);
1485 seq += tv.tv_usec + tv.tv_sec * 1000000;
1487 return seq;
1489 EXPORT_SYMBOL(secure_tcpv6_sequence_number);
1490 #endif
1492 /* The code below is shamelessly stolen from secure_tcp_sequence_number().
1493 * All blames to Andrey V. Savochkin <saw@msu.ru>.
1495 __u32 secure_ip_id(__u32 daddr)
1497 struct keydata *keyptr;
1498 __u32 hash[4];
1500 keyptr = get_keyptr();
1503 * Pick a unique starting offset for each IP destination.
1504 * The dest ip address is placed in the starting vector,
1505 * which is then hashed with random data.
1507 hash[0] = daddr;
1508 hash[1] = keyptr->secret[9];
1509 hash[2] = keyptr->secret[10];
1510 hash[3] = keyptr->secret[11];
1512 return half_md4_transform(hash, keyptr->secret);
1515 #ifdef CONFIG_INET
1517 __u32 secure_tcp_sequence_number(__u32 saddr, __u32 daddr,
1518 __u16 sport, __u16 dport)
1520 struct timeval tv;
1521 __u32 seq;
1522 __u32 hash[4];
1523 struct keydata *keyptr = get_keyptr();
1526 * Pick a unique starting offset for each TCP connection endpoints
1527 * (saddr, daddr, sport, dport).
1528 * Note that the words are placed into the starting vector, which is
1529 * then mixed with a partial MD4 over random data.
1531 hash[0]=saddr;
1532 hash[1]=daddr;
1533 hash[2]=(sport << 16) + dport;
1534 hash[3]=keyptr->secret[11];
1536 seq = half_md4_transform(hash, keyptr->secret) & HASH_MASK;
1537 seq += keyptr->count;
1539 * As close as possible to RFC 793, which
1540 * suggests using a 250 kHz clock.
1541 * Further reading shows this assumes 2 Mb/s networks.
1542 * For 10 Mb/s Ethernet, a 1 MHz clock is appropriate.
1543 * That's funny, Linux has one built in! Use it!
1544 * (Networks are faster now - should this be increased?)
1546 do_gettimeofday(&tv);
1547 seq += tv.tv_usec + tv.tv_sec * 1000000;
1548 #if 0
1549 printk("init_seq(%lx, %lx, %d, %d) = %d\n",
1550 saddr, daddr, sport, dport, seq);
1551 #endif
1552 return seq;
1555 EXPORT_SYMBOL(secure_tcp_sequence_number);
1559 /* Generate secure starting point for ephemeral TCP port search */
1560 u32 secure_tcp_port_ephemeral(__u32 saddr, __u32 daddr, __u16 dport)
1562 struct keydata *keyptr = get_keyptr();
1563 u32 hash[4];
1566 * Pick a unique starting offset for each ephemeral port search
1567 * (saddr, daddr, dport) and 48bits of random data.
1569 hash[0] = saddr;
1570 hash[1] = daddr;
1571 hash[2] = dport ^ keyptr->secret[10];
1572 hash[3] = keyptr->secret[11];
1574 return half_md4_transform(hash, keyptr->secret);
1577 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1578 u32 secure_tcpv6_port_ephemeral(const __u32 *saddr, const __u32 *daddr, __u16 dport)
1580 struct keydata *keyptr = get_keyptr();
1581 u32 hash[12];
1583 memcpy(hash, saddr, 16);
1584 hash[4] = dport;
1585 memcpy(&hash[5],keyptr->secret,sizeof(__u32) * 7);
1587 return twothirdsMD4Transform(daddr, hash);
1589 EXPORT_SYMBOL(secure_tcpv6_port_ephemeral);
1590 #endif
1592 #endif /* CONFIG_INET */
1596 * Get a random word for internal kernel use only. Similar to urandom but
1597 * with the goal of minimal entropy pool depletion. As a result, the random
1598 * value is not cryptographically secure but for several uses the cost of
1599 * depleting entropy is too high
1601 unsigned int get_random_int(void)
1604 * Use IP's RNG. It suits our purpose perfectly: it re-keys itself
1605 * every second, from the entropy pool (and thus creates a limited
1606 * drain on it), and uses halfMD4Transform within the second. We
1607 * also mix it with jiffies and the PID:
1609 return secure_ip_id(current->pid + jiffies);
1613 * randomize_range() returns a start address such that
1615 * [...... <range> .....]
1616 * start end
1618 * a <range> with size "len" starting at the return value is inside in the
1619 * area defined by [start, end], but is otherwise randomized.
1621 unsigned long
1622 randomize_range(unsigned long start, unsigned long end, unsigned long len)
1624 unsigned long range = end - len - start;
1626 if (end <= start + len)
1627 return 0;
1628 return PAGE_ALIGN(get_random_int() % range + start);