nilfs2: insert checkpoint number in segment summary header
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / char / random.c
blob2fd3d39995d52cb896b05a729ec8d3e04767f99d
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/mm.h>
240 #include <linux/spinlock.h>
241 #include <linux/percpu.h>
242 #include <linux/cryptohash.h>
243 #include <linux/fips.h>
245 #ifdef CONFIG_GENERIC_HARDIRQS
246 # include <linux/irq.h>
247 #endif
249 #include <asm/processor.h>
250 #include <asm/uaccess.h>
251 #include <asm/irq.h>
252 #include <asm/io.h>
255 * Configuration information
257 #define INPUT_POOL_WORDS 128
258 #define OUTPUT_POOL_WORDS 32
259 #define SEC_XFER_SIZE 512
262 * The minimum number of bits of entropy before we wake up a read on
263 * /dev/random. Should be enough to do a significant reseed.
265 static int random_read_wakeup_thresh = 64;
268 * If the entropy count falls under this number of bits, then we
269 * should wake up processes which are selecting or polling on write
270 * access to /dev/random.
272 static int random_write_wakeup_thresh = 128;
275 * When the input pool goes over trickle_thresh, start dropping most
276 * samples to avoid wasting CPU time and reduce lock contention.
279 static int trickle_thresh __read_mostly = INPUT_POOL_WORDS * 28;
281 static DEFINE_PER_CPU(int, trickle_count);
284 * A pool of size .poolwords is stirred with a primitive polynomial
285 * of degree .poolwords over GF(2). The taps for various sizes are
286 * defined below. They are chosen to be evenly spaced (minimum RMS
287 * distance from evenly spaced; the numbers in the comments are a
288 * scaled squared error sum) except for the last tap, which is 1 to
289 * get the twisting happening as fast as possible.
291 static struct poolinfo {
292 int poolwords;
293 int tap1, tap2, tap3, tap4, tap5;
294 } poolinfo_table[] = {
295 /* x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 -- 105 */
296 { 128, 103, 76, 51, 25, 1 },
297 /* x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 -- 15 */
298 { 32, 26, 20, 14, 7, 1 },
299 #if 0
300 /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1 -- 115 */
301 { 2048, 1638, 1231, 819, 411, 1 },
303 /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
304 { 1024, 817, 615, 412, 204, 1 },
306 /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
307 { 1024, 819, 616, 410, 207, 2 },
309 /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
310 { 512, 411, 308, 208, 104, 1 },
312 /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
313 { 512, 409, 307, 206, 102, 2 },
314 /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
315 { 512, 409, 309, 205, 103, 2 },
317 /* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
318 { 256, 205, 155, 101, 52, 1 },
320 /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
321 { 128, 103, 78, 51, 27, 2 },
323 /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
324 { 64, 52, 39, 26, 14, 1 },
325 #endif
328 #define POOLBITS poolwords*32
329 #define POOLBYTES poolwords*4
332 * For the purposes of better mixing, we use the CRC-32 polynomial as
333 * well to make a twisted Generalized Feedback Shift Reigster
335 * (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR generators. ACM
336 * Transactions on Modeling and Computer Simulation 2(3):179-194.
337 * Also see M. Matsumoto & Y. Kurita, 1994. Twisted GFSR generators
338 * II. ACM Transactions on Mdeling and Computer Simulation 4:254-266)
340 * Thanks to Colin Plumb for suggesting this.
342 * We have not analyzed the resultant polynomial to prove it primitive;
343 * in fact it almost certainly isn't. Nonetheless, the irreducible factors
344 * of a random large-degree polynomial over GF(2) are more than large enough
345 * that periodicity is not a concern.
347 * The input hash is much less sensitive than the output hash. All
348 * that we want of it is that it be a good non-cryptographic hash;
349 * i.e. it not produce collisions when fed "random" data of the sort
350 * we expect to see. As long as the pool state differs for different
351 * inputs, we have preserved the input entropy and done a good job.
352 * The fact that an intelligent attacker can construct inputs that
353 * will produce controlled alterations to the pool's state is not
354 * important because we don't consider such inputs to contribute any
355 * randomness. The only property we need with respect to them is that
356 * the attacker can't increase his/her knowledge of the pool's state.
357 * Since all additions are reversible (knowing the final state and the
358 * input, you can reconstruct the initial state), if an attacker has
359 * any uncertainty about the initial state, he/she can only shuffle
360 * that uncertainty about, but never cause any collisions (which would
361 * decrease the uncertainty).
363 * The chosen system lets the state of the pool be (essentially) the input
364 * modulo the generator polymnomial. Now, for random primitive polynomials,
365 * this is a universal class of hash functions, meaning that the chance
366 * of a collision is limited by the attacker's knowledge of the generator
367 * polynomail, so if it is chosen at random, an attacker can never force
368 * a collision. Here, we use a fixed polynomial, but we *can* assume that
369 * ###--> it is unknown to the processes generating the input entropy. <-###
370 * Because of this important property, this is a good, collision-resistant
371 * hash; hash collisions will occur no more often than chance.
375 * Static global variables
377 static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
378 static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
379 static struct fasync_struct *fasync;
381 #if 0
382 static int debug;
383 module_param(debug, bool, 0644);
384 #define DEBUG_ENT(fmt, arg...) do { \
385 if (debug) \
386 printk(KERN_DEBUG "random %04d %04d %04d: " \
387 fmt,\
388 input_pool.entropy_count,\
389 blocking_pool.entropy_count,\
390 nonblocking_pool.entropy_count,\
391 ## arg); } while (0)
392 #else
393 #define DEBUG_ENT(fmt, arg...) do {} while (0)
394 #endif
396 /**********************************************************************
398 * OS independent entropy store. Here are the functions which handle
399 * storing entropy in an entropy pool.
401 **********************************************************************/
403 struct entropy_store;
404 struct entropy_store {
405 /* read-only data: */
406 struct poolinfo *poolinfo;
407 __u32 *pool;
408 const char *name;
409 int limit;
410 struct entropy_store *pull;
412 /* read-write data: */
413 spinlock_t lock;
414 unsigned add_ptr;
415 int entropy_count;
416 int input_rotate;
417 __u8 *last_data;
420 static __u32 input_pool_data[INPUT_POOL_WORDS];
421 static __u32 blocking_pool_data[OUTPUT_POOL_WORDS];
422 static __u32 nonblocking_pool_data[OUTPUT_POOL_WORDS];
424 static struct entropy_store input_pool = {
425 .poolinfo = &poolinfo_table[0],
426 .name = "input",
427 .limit = 1,
428 .lock = __SPIN_LOCK_UNLOCKED(&input_pool.lock),
429 .pool = input_pool_data
432 static struct entropy_store blocking_pool = {
433 .poolinfo = &poolinfo_table[1],
434 .name = "blocking",
435 .limit = 1,
436 .pull = &input_pool,
437 .lock = __SPIN_LOCK_UNLOCKED(&blocking_pool.lock),
438 .pool = blocking_pool_data
441 static struct entropy_store nonblocking_pool = {
442 .poolinfo = &poolinfo_table[1],
443 .name = "nonblocking",
444 .pull = &input_pool,
445 .lock = __SPIN_LOCK_UNLOCKED(&nonblocking_pool.lock),
446 .pool = nonblocking_pool_data
450 * This function adds bytes into the entropy "pool". It does not
451 * update the entropy estimate. The caller should call
452 * credit_entropy_bits if this is appropriate.
454 * The pool is stirred with a primitive polynomial of the appropriate
455 * degree, and then twisted. We twist by three bits at a time because
456 * it's cheap to do so and helps slightly in the expected case where
457 * the entropy is concentrated in the low-order bits.
459 static void mix_pool_bytes_extract(struct entropy_store *r, const void *in,
460 int nbytes, __u8 out[64])
462 static __u32 const twist_table[8] = {
463 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
464 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
465 unsigned long i, j, tap1, tap2, tap3, tap4, tap5;
466 int input_rotate;
467 int wordmask = r->poolinfo->poolwords - 1;
468 const char *bytes = in;
469 __u32 w;
470 unsigned long flags;
472 /* Taps are constant, so we can load them without holding r->lock. */
473 tap1 = r->poolinfo->tap1;
474 tap2 = r->poolinfo->tap2;
475 tap3 = r->poolinfo->tap3;
476 tap4 = r->poolinfo->tap4;
477 tap5 = r->poolinfo->tap5;
479 spin_lock_irqsave(&r->lock, flags);
480 input_rotate = r->input_rotate;
481 i = r->add_ptr;
483 /* mix one byte at a time to simplify size handling and churn faster */
484 while (nbytes--) {
485 w = rol32(*bytes++, input_rotate & 31);
486 i = (i - 1) & wordmask;
488 /* XOR in the various taps */
489 w ^= r->pool[i];
490 w ^= r->pool[(i + tap1) & wordmask];
491 w ^= r->pool[(i + tap2) & wordmask];
492 w ^= r->pool[(i + tap3) & wordmask];
493 w ^= r->pool[(i + tap4) & wordmask];
494 w ^= r->pool[(i + tap5) & wordmask];
496 /* Mix the result back in with a twist */
497 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
500 * Normally, we add 7 bits of rotation to the pool.
501 * At the beginning of the pool, add an extra 7 bits
502 * rotation, so that successive passes spread the
503 * input bits across the pool evenly.
505 input_rotate += i ? 7 : 14;
508 r->input_rotate = input_rotate;
509 r->add_ptr = i;
511 if (out)
512 for (j = 0; j < 16; j++)
513 ((__u32 *)out)[j] = r->pool[(i - j) & wordmask];
515 spin_unlock_irqrestore(&r->lock, flags);
518 static void mix_pool_bytes(struct entropy_store *r, const void *in, int bytes)
520 mix_pool_bytes_extract(r, in, bytes, NULL);
524 * Credit (or debit) the entropy store with n bits of entropy
526 static void credit_entropy_bits(struct entropy_store *r, int nbits)
528 unsigned long flags;
529 int entropy_count;
531 if (!nbits)
532 return;
534 spin_lock_irqsave(&r->lock, flags);
536 DEBUG_ENT("added %d entropy credits to %s\n", nbits, r->name);
537 entropy_count = r->entropy_count;
538 entropy_count += nbits;
539 if (entropy_count < 0) {
540 DEBUG_ENT("negative entropy/overflow\n");
541 entropy_count = 0;
542 } else if (entropy_count > r->poolinfo->POOLBITS)
543 entropy_count = r->poolinfo->POOLBITS;
544 r->entropy_count = entropy_count;
546 /* should we wake readers? */
547 if (r == &input_pool && entropy_count >= random_read_wakeup_thresh) {
548 wake_up_interruptible(&random_read_wait);
549 kill_fasync(&fasync, SIGIO, POLL_IN);
551 spin_unlock_irqrestore(&r->lock, flags);
554 /*********************************************************************
556 * Entropy input management
558 *********************************************************************/
560 /* There is one of these per entropy source */
561 struct timer_rand_state {
562 cycles_t last_time;
563 long last_delta, last_delta2;
564 unsigned dont_count_entropy:1;
567 #ifndef CONFIG_GENERIC_HARDIRQS
569 static struct timer_rand_state *irq_timer_state[NR_IRQS];
571 static struct timer_rand_state *get_timer_rand_state(unsigned int irq)
573 return irq_timer_state[irq];
576 static void set_timer_rand_state(unsigned int irq,
577 struct timer_rand_state *state)
579 irq_timer_state[irq] = state;
582 #else
584 static struct timer_rand_state *get_timer_rand_state(unsigned int irq)
586 struct irq_desc *desc;
588 desc = irq_to_desc(irq);
590 return desc->timer_rand_state;
593 static void set_timer_rand_state(unsigned int irq,
594 struct timer_rand_state *state)
596 struct irq_desc *desc;
598 desc = irq_to_desc(irq);
600 desc->timer_rand_state = state;
602 #endif
604 static struct timer_rand_state input_timer_state;
607 * This function adds entropy to the entropy "pool" by using timing
608 * delays. It uses the timer_rand_state structure to make an estimate
609 * of how many bits of entropy this call has added to the pool.
611 * The number "num" is also added to the pool - it should somehow describe
612 * the type of event which just happened. This is currently 0-255 for
613 * keyboard scan codes, and 256 upwards for interrupts.
616 static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
618 struct {
619 cycles_t cycles;
620 long jiffies;
621 unsigned num;
622 } sample;
623 long delta, delta2, delta3;
625 preempt_disable();
626 /* if over the trickle threshold, use only 1 in 4096 samples */
627 if (input_pool.entropy_count > trickle_thresh &&
628 (__get_cpu_var(trickle_count)++ & 0xfff))
629 goto out;
631 sample.jiffies = jiffies;
632 sample.cycles = get_cycles();
633 sample.num = num;
634 mix_pool_bytes(&input_pool, &sample, sizeof(sample));
637 * Calculate number of bits of randomness we probably added.
638 * We take into account the first, second and third-order deltas
639 * in order to make our estimate.
642 if (!state->dont_count_entropy) {
643 delta = sample.jiffies - state->last_time;
644 state->last_time = sample.jiffies;
646 delta2 = delta - state->last_delta;
647 state->last_delta = delta;
649 delta3 = delta2 - state->last_delta2;
650 state->last_delta2 = delta2;
652 if (delta < 0)
653 delta = -delta;
654 if (delta2 < 0)
655 delta2 = -delta2;
656 if (delta3 < 0)
657 delta3 = -delta3;
658 if (delta > delta2)
659 delta = delta2;
660 if (delta > delta3)
661 delta = delta3;
664 * delta is now minimum absolute delta.
665 * Round down by 1 bit on general principles,
666 * and limit entropy entimate to 12 bits.
668 credit_entropy_bits(&input_pool,
669 min_t(int, fls(delta>>1), 11));
671 out:
672 preempt_enable();
675 void add_input_randomness(unsigned int type, unsigned int code,
676 unsigned int value)
678 static unsigned char last_value;
680 /* ignore autorepeat and the like */
681 if (value == last_value)
682 return;
684 DEBUG_ENT("input event\n");
685 last_value = value;
686 add_timer_randomness(&input_timer_state,
687 (type << 4) ^ code ^ (code >> 4) ^ value);
689 EXPORT_SYMBOL_GPL(add_input_randomness);
691 void add_interrupt_randomness(int irq)
693 struct timer_rand_state *state;
695 state = get_timer_rand_state(irq);
697 if (state == NULL)
698 return;
700 DEBUG_ENT("irq event %d\n", irq);
701 add_timer_randomness(state, 0x100 + irq);
704 #ifdef CONFIG_BLOCK
705 void add_disk_randomness(struct gendisk *disk)
707 if (!disk || !disk->random)
708 return;
709 /* first major is 1, so we get >= 0x200 here */
710 DEBUG_ENT("disk event %d:%d\n",
711 MAJOR(disk_devt(disk)), MINOR(disk_devt(disk)));
713 add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
715 #endif
717 #define EXTRACT_SIZE 10
719 /*********************************************************************
721 * Entropy extraction routines
723 *********************************************************************/
725 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
726 size_t nbytes, int min, int rsvd);
729 * This utility inline function is responsible for transfering entropy
730 * from the primary pool to the secondary extraction pool. We make
731 * sure we pull enough for a 'catastrophic reseed'.
733 static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
735 __u32 tmp[OUTPUT_POOL_WORDS];
737 if (r->pull && r->entropy_count < nbytes * 8 &&
738 r->entropy_count < r->poolinfo->POOLBITS) {
739 /* If we're limited, always leave two wakeup worth's BITS */
740 int rsvd = r->limit ? 0 : random_read_wakeup_thresh/4;
741 int bytes = nbytes;
743 /* pull at least as many as BYTES as wakeup BITS */
744 bytes = max_t(int, bytes, random_read_wakeup_thresh / 8);
745 /* but never more than the buffer size */
746 bytes = min_t(int, bytes, sizeof(tmp));
748 DEBUG_ENT("going to reseed %s with %d bits "
749 "(%d of %d requested)\n",
750 r->name, bytes * 8, nbytes * 8, r->entropy_count);
752 bytes = extract_entropy(r->pull, tmp, bytes,
753 random_read_wakeup_thresh / 8, rsvd);
754 mix_pool_bytes(r, tmp, bytes);
755 credit_entropy_bits(r, bytes*8);
760 * These functions extracts randomness from the "entropy pool", and
761 * returns it in a buffer.
763 * The min parameter specifies the minimum amount we can pull before
764 * failing to avoid races that defeat catastrophic reseeding while the
765 * reserved parameter indicates how much entropy we must leave in the
766 * pool after each pull to avoid starving other readers.
768 * Note: extract_entropy() assumes that .poolwords is a multiple of 16 words.
771 static size_t account(struct entropy_store *r, size_t nbytes, int min,
772 int reserved)
774 unsigned long flags;
776 /* Hold lock while accounting */
777 spin_lock_irqsave(&r->lock, flags);
779 BUG_ON(r->entropy_count > r->poolinfo->POOLBITS);
780 DEBUG_ENT("trying to extract %d bits from %s\n",
781 nbytes * 8, r->name);
783 /* Can we pull enough? */
784 if (r->entropy_count / 8 < min + reserved) {
785 nbytes = 0;
786 } else {
787 /* If limited, never pull more than available */
788 if (r->limit && nbytes + reserved >= r->entropy_count / 8)
789 nbytes = r->entropy_count/8 - reserved;
791 if (r->entropy_count / 8 >= nbytes + reserved)
792 r->entropy_count -= nbytes*8;
793 else
794 r->entropy_count = reserved;
796 if (r->entropy_count < random_write_wakeup_thresh) {
797 wake_up_interruptible(&random_write_wait);
798 kill_fasync(&fasync, SIGIO, POLL_OUT);
802 DEBUG_ENT("debiting %d entropy credits from %s%s\n",
803 nbytes * 8, r->name, r->limit ? "" : " (unlimited)");
805 spin_unlock_irqrestore(&r->lock, flags);
807 return nbytes;
810 static void extract_buf(struct entropy_store *r, __u8 *out)
812 int i;
813 __u32 hash[5], workspace[SHA_WORKSPACE_WORDS];
814 __u8 extract[64];
816 /* Generate a hash across the pool, 16 words (512 bits) at a time */
817 sha_init(hash);
818 for (i = 0; i < r->poolinfo->poolwords; i += 16)
819 sha_transform(hash, (__u8 *)(r->pool + i), workspace);
822 * We mix the hash back into the pool to prevent backtracking
823 * attacks (where the attacker knows the state of the pool
824 * plus the current outputs, and attempts to find previous
825 * ouputs), unless the hash function can be inverted. By
826 * mixing at least a SHA1 worth of hash data back, we make
827 * brute-forcing the feedback as hard as brute-forcing the
828 * hash.
830 mix_pool_bytes_extract(r, hash, sizeof(hash), extract);
833 * To avoid duplicates, we atomically extract a portion of the
834 * pool while mixing, and hash one final time.
836 sha_transform(hash, extract, workspace);
837 memset(extract, 0, sizeof(extract));
838 memset(workspace, 0, sizeof(workspace));
841 * In case the hash function has some recognizable output
842 * pattern, we fold it in half. Thus, we always feed back
843 * twice as much data as we output.
845 hash[0] ^= hash[3];
846 hash[1] ^= hash[4];
847 hash[2] ^= rol32(hash[2], 16);
848 memcpy(out, hash, EXTRACT_SIZE);
849 memset(hash, 0, sizeof(hash));
852 static ssize_t extract_entropy(struct entropy_store *r, void *buf,
853 size_t nbytes, int min, int reserved)
855 ssize_t ret = 0, i;
856 __u8 tmp[EXTRACT_SIZE];
857 unsigned long flags;
859 xfer_secondary_pool(r, nbytes);
860 nbytes = account(r, nbytes, min, reserved);
862 while (nbytes) {
863 extract_buf(r, tmp);
865 if (r->last_data) {
866 spin_lock_irqsave(&r->lock, flags);
867 if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
868 panic("Hardware RNG duplicated output!\n");
869 memcpy(r->last_data, tmp, EXTRACT_SIZE);
870 spin_unlock_irqrestore(&r->lock, flags);
872 i = min_t(int, nbytes, EXTRACT_SIZE);
873 memcpy(buf, tmp, i);
874 nbytes -= i;
875 buf += i;
876 ret += i;
879 /* Wipe data just returned from memory */
880 memset(tmp, 0, sizeof(tmp));
882 return ret;
885 static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
886 size_t nbytes)
888 ssize_t ret = 0, i;
889 __u8 tmp[EXTRACT_SIZE];
891 xfer_secondary_pool(r, nbytes);
892 nbytes = account(r, nbytes, 0, 0);
894 while (nbytes) {
895 if (need_resched()) {
896 if (signal_pending(current)) {
897 if (ret == 0)
898 ret = -ERESTARTSYS;
899 break;
901 schedule();
904 extract_buf(r, tmp);
905 i = min_t(int, nbytes, EXTRACT_SIZE);
906 if (copy_to_user(buf, tmp, i)) {
907 ret = -EFAULT;
908 break;
911 nbytes -= i;
912 buf += i;
913 ret += i;
916 /* Wipe data just returned from memory */
917 memset(tmp, 0, sizeof(tmp));
919 return ret;
923 * This function is the exported kernel interface. It returns some
924 * number of good random numbers, suitable for seeding TCP sequence
925 * numbers, etc.
927 void get_random_bytes(void *buf, int nbytes)
929 extract_entropy(&nonblocking_pool, buf, nbytes, 0, 0);
931 EXPORT_SYMBOL(get_random_bytes);
934 * init_std_data - initialize pool with system data
936 * @r: pool to initialize
938 * This function clears the pool's entropy count and mixes some system
939 * data into the pool to prepare it for use. The pool is not cleared
940 * as that can only decrease the entropy in the pool.
942 static void init_std_data(struct entropy_store *r)
944 ktime_t now;
945 unsigned long flags;
947 spin_lock_irqsave(&r->lock, flags);
948 r->entropy_count = 0;
949 spin_unlock_irqrestore(&r->lock, flags);
951 now = ktime_get_real();
952 mix_pool_bytes(r, &now, sizeof(now));
953 mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
954 /* Enable continuous test in fips mode */
955 if (fips_enabled)
956 r->last_data = kmalloc(EXTRACT_SIZE, GFP_KERNEL);
959 static int rand_initialize(void)
961 init_std_data(&input_pool);
962 init_std_data(&blocking_pool);
963 init_std_data(&nonblocking_pool);
964 return 0;
966 module_init(rand_initialize);
968 void rand_initialize_irq(int irq)
970 struct timer_rand_state *state;
972 state = get_timer_rand_state(irq);
974 if (state)
975 return;
978 * If kzalloc returns null, we just won't use that entropy
979 * source.
981 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
982 if (state)
983 set_timer_rand_state(irq, state);
986 #ifdef CONFIG_BLOCK
987 void rand_initialize_disk(struct gendisk *disk)
989 struct timer_rand_state *state;
992 * If kzalloc returns null, we just won't use that entropy
993 * source.
995 state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
996 if (state)
997 disk->random = state;
999 #endif
1001 static ssize_t
1002 random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1004 ssize_t n, retval = 0, count = 0;
1006 if (nbytes == 0)
1007 return 0;
1009 while (nbytes > 0) {
1010 n = nbytes;
1011 if (n > SEC_XFER_SIZE)
1012 n = SEC_XFER_SIZE;
1014 DEBUG_ENT("reading %d bits\n", n*8);
1016 n = extract_entropy_user(&blocking_pool, buf, n);
1018 DEBUG_ENT("read got %d bits (%d still needed)\n",
1019 n*8, (nbytes-n)*8);
1021 if (n == 0) {
1022 if (file->f_flags & O_NONBLOCK) {
1023 retval = -EAGAIN;
1024 break;
1027 DEBUG_ENT("sleeping?\n");
1029 wait_event_interruptible(random_read_wait,
1030 input_pool.entropy_count >=
1031 random_read_wakeup_thresh);
1033 DEBUG_ENT("awake\n");
1035 if (signal_pending(current)) {
1036 retval = -ERESTARTSYS;
1037 break;
1040 continue;
1043 if (n < 0) {
1044 retval = n;
1045 break;
1047 count += n;
1048 buf += n;
1049 nbytes -= n;
1050 break; /* This break makes the device work */
1051 /* like a named pipe */
1054 return (count ? count : retval);
1057 static ssize_t
1058 urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1060 return extract_entropy_user(&nonblocking_pool, buf, nbytes);
1063 static unsigned int
1064 random_poll(struct file *file, poll_table * wait)
1066 unsigned int mask;
1068 poll_wait(file, &random_read_wait, wait);
1069 poll_wait(file, &random_write_wait, wait);
1070 mask = 0;
1071 if (input_pool.entropy_count >= random_read_wakeup_thresh)
1072 mask |= POLLIN | POLLRDNORM;
1073 if (input_pool.entropy_count < random_write_wakeup_thresh)
1074 mask |= POLLOUT | POLLWRNORM;
1075 return mask;
1078 static int
1079 write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1081 size_t bytes;
1082 __u32 buf[16];
1083 const char __user *p = buffer;
1085 while (count > 0) {
1086 bytes = min(count, sizeof(buf));
1087 if (copy_from_user(&buf, p, bytes))
1088 return -EFAULT;
1090 count -= bytes;
1091 p += bytes;
1093 mix_pool_bytes(r, buf, bytes);
1094 cond_resched();
1097 return 0;
1100 static ssize_t random_write(struct file *file, const char __user *buffer,
1101 size_t count, loff_t *ppos)
1103 size_t ret;
1105 ret = write_pool(&blocking_pool, buffer, count);
1106 if (ret)
1107 return ret;
1108 ret = write_pool(&nonblocking_pool, buffer, count);
1109 if (ret)
1110 return ret;
1112 return (ssize_t)count;
1115 static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1117 int size, ent_count;
1118 int __user *p = (int __user *)arg;
1119 int retval;
1121 switch (cmd) {
1122 case RNDGETENTCNT:
1123 /* inherently racy, no point locking */
1124 if (put_user(input_pool.entropy_count, p))
1125 return -EFAULT;
1126 return 0;
1127 case RNDADDTOENTCNT:
1128 if (!capable(CAP_SYS_ADMIN))
1129 return -EPERM;
1130 if (get_user(ent_count, p))
1131 return -EFAULT;
1132 credit_entropy_bits(&input_pool, ent_count);
1133 return 0;
1134 case RNDADDENTROPY:
1135 if (!capable(CAP_SYS_ADMIN))
1136 return -EPERM;
1137 if (get_user(ent_count, p++))
1138 return -EFAULT;
1139 if (ent_count < 0)
1140 return -EINVAL;
1141 if (get_user(size, p++))
1142 return -EFAULT;
1143 retval = write_pool(&input_pool, (const char __user *)p,
1144 size);
1145 if (retval < 0)
1146 return retval;
1147 credit_entropy_bits(&input_pool, ent_count);
1148 return 0;
1149 case RNDZAPENTCNT:
1150 case RNDCLEARPOOL:
1151 /* Clear the entropy pool counters. */
1152 if (!capable(CAP_SYS_ADMIN))
1153 return -EPERM;
1154 rand_initialize();
1155 return 0;
1156 default:
1157 return -EINVAL;
1161 static int random_fasync(int fd, struct file *filp, int on)
1163 return fasync_helper(fd, filp, on, &fasync);
1166 const struct file_operations random_fops = {
1167 .read = random_read,
1168 .write = random_write,
1169 .poll = random_poll,
1170 .unlocked_ioctl = random_ioctl,
1171 .fasync = random_fasync,
1174 const struct file_operations urandom_fops = {
1175 .read = urandom_read,
1176 .write = random_write,
1177 .unlocked_ioctl = random_ioctl,
1178 .fasync = random_fasync,
1181 /***************************************************************
1182 * Random UUID interface
1184 * Used here for a Boot ID, but can be useful for other kernel
1185 * drivers.
1186 ***************************************************************/
1189 * Generate random UUID
1191 void generate_random_uuid(unsigned char uuid_out[16])
1193 get_random_bytes(uuid_out, 16);
1194 /* Set UUID version to 4 --- truly random generation */
1195 uuid_out[6] = (uuid_out[6] & 0x0F) | 0x40;
1196 /* Set the UUID variant to DCE */
1197 uuid_out[8] = (uuid_out[8] & 0x3F) | 0x80;
1199 EXPORT_SYMBOL(generate_random_uuid);
1201 /********************************************************************
1203 * Sysctl interface
1205 ********************************************************************/
1207 #ifdef CONFIG_SYSCTL
1209 #include <linux/sysctl.h>
1211 static int min_read_thresh = 8, min_write_thresh;
1212 static int max_read_thresh = INPUT_POOL_WORDS * 32;
1213 static int max_write_thresh = INPUT_POOL_WORDS * 32;
1214 static char sysctl_bootid[16];
1217 * These functions is used to return both the bootid UUID, and random
1218 * UUID. The difference is in whether table->data is NULL; if it is,
1219 * then a new UUID is generated and returned to the user.
1221 * If the user accesses this via the proc interface, it will be returned
1222 * as an ASCII string in the standard UUID format. If accesses via the
1223 * sysctl system call, it is returned as 16 bytes of binary data.
1225 static int proc_do_uuid(ctl_table *table, int write,
1226 void __user *buffer, size_t *lenp, loff_t *ppos)
1228 ctl_table fake_table;
1229 unsigned char buf[64], tmp_uuid[16], *uuid;
1231 uuid = table->data;
1232 if (!uuid) {
1233 uuid = tmp_uuid;
1234 uuid[8] = 0;
1236 if (uuid[8] == 0)
1237 generate_random_uuid(uuid);
1239 sprintf(buf, "%pU", uuid);
1241 fake_table.data = buf;
1242 fake_table.maxlen = sizeof(buf);
1244 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
1247 static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
1248 ctl_table random_table[] = {
1250 .procname = "poolsize",
1251 .data = &sysctl_poolsize,
1252 .maxlen = sizeof(int),
1253 .mode = 0444,
1254 .proc_handler = proc_dointvec,
1257 .procname = "entropy_avail",
1258 .maxlen = sizeof(int),
1259 .mode = 0444,
1260 .proc_handler = proc_dointvec,
1261 .data = &input_pool.entropy_count,
1264 .procname = "read_wakeup_threshold",
1265 .data = &random_read_wakeup_thresh,
1266 .maxlen = sizeof(int),
1267 .mode = 0644,
1268 .proc_handler = proc_dointvec_minmax,
1269 .extra1 = &min_read_thresh,
1270 .extra2 = &max_read_thresh,
1273 .procname = "write_wakeup_threshold",
1274 .data = &random_write_wakeup_thresh,
1275 .maxlen = sizeof(int),
1276 .mode = 0644,
1277 .proc_handler = proc_dointvec_minmax,
1278 .extra1 = &min_write_thresh,
1279 .extra2 = &max_write_thresh,
1282 .procname = "boot_id",
1283 .data = &sysctl_bootid,
1284 .maxlen = 16,
1285 .mode = 0444,
1286 .proc_handler = proc_do_uuid,
1289 .procname = "uuid",
1290 .maxlen = 16,
1291 .mode = 0444,
1292 .proc_handler = proc_do_uuid,
1296 #endif /* CONFIG_SYSCTL */
1298 /********************************************************************
1300 * Random functions for networking
1302 ********************************************************************/
1305 * TCP initial sequence number picking. This uses the random number
1306 * generator to pick an initial secret value. This value is hashed
1307 * along with the TCP endpoint information to provide a unique
1308 * starting point for each pair of TCP endpoints. This defeats
1309 * attacks which rely on guessing the initial TCP sequence number.
1310 * This algorithm was suggested by Steve Bellovin.
1312 * Using a very strong hash was taking an appreciable amount of the total
1313 * TCP connection establishment time, so this is a weaker hash,
1314 * compensated for by changing the secret periodically.
1317 /* F, G and H are basic MD4 functions: selection, majority, parity */
1318 #define F(x, y, z) ((z) ^ ((x) & ((y) ^ (z))))
1319 #define G(x, y, z) (((x) & (y)) + (((x) ^ (y)) & (z)))
1320 #define H(x, y, z) ((x) ^ (y) ^ (z))
1323 * The generic round function. The application is so specific that
1324 * we don't bother protecting all the arguments with parens, as is generally
1325 * good macro practice, in favor of extra legibility.
1326 * Rotation is separate from addition to prevent recomputation
1328 #define ROUND(f, a, b, c, d, x, s) \
1329 (a += f(b, c, d) + x, a = (a << s) | (a >> (32 - s)))
1330 #define K1 0
1331 #define K2 013240474631UL
1332 #define K3 015666365641UL
1334 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1336 static __u32 twothirdsMD4Transform(__u32 const buf[4], __u32 const in[12])
1338 __u32 a = buf[0], b = buf[1], c = buf[2], d = buf[3];
1340 /* Round 1 */
1341 ROUND(F, a, b, c, d, in[ 0] + K1, 3);
1342 ROUND(F, d, a, b, c, in[ 1] + K1, 7);
1343 ROUND(F, c, d, a, b, in[ 2] + K1, 11);
1344 ROUND(F, b, c, d, a, in[ 3] + K1, 19);
1345 ROUND(F, a, b, c, d, in[ 4] + K1, 3);
1346 ROUND(F, d, a, b, c, in[ 5] + K1, 7);
1347 ROUND(F, c, d, a, b, in[ 6] + K1, 11);
1348 ROUND(F, b, c, d, a, in[ 7] + K1, 19);
1349 ROUND(F, a, b, c, d, in[ 8] + K1, 3);
1350 ROUND(F, d, a, b, c, in[ 9] + K1, 7);
1351 ROUND(F, c, d, a, b, in[10] + K1, 11);
1352 ROUND(F, b, c, d, a, in[11] + K1, 19);
1354 /* Round 2 */
1355 ROUND(G, a, b, c, d, in[ 1] + K2, 3);
1356 ROUND(G, d, a, b, c, in[ 3] + K2, 5);
1357 ROUND(G, c, d, a, b, in[ 5] + K2, 9);
1358 ROUND(G, b, c, d, a, in[ 7] + K2, 13);
1359 ROUND(G, a, b, c, d, in[ 9] + K2, 3);
1360 ROUND(G, d, a, b, c, in[11] + K2, 5);
1361 ROUND(G, c, d, a, b, in[ 0] + K2, 9);
1362 ROUND(G, b, c, d, a, in[ 2] + K2, 13);
1363 ROUND(G, a, b, c, d, in[ 4] + K2, 3);
1364 ROUND(G, d, a, b, c, in[ 6] + K2, 5);
1365 ROUND(G, c, d, a, b, in[ 8] + K2, 9);
1366 ROUND(G, b, c, d, a, in[10] + K2, 13);
1368 /* Round 3 */
1369 ROUND(H, a, b, c, d, in[ 3] + K3, 3);
1370 ROUND(H, d, a, b, c, in[ 7] + K3, 9);
1371 ROUND(H, c, d, a, b, in[11] + K3, 11);
1372 ROUND(H, b, c, d, a, in[ 2] + K3, 15);
1373 ROUND(H, a, b, c, d, in[ 6] + K3, 3);
1374 ROUND(H, d, a, b, c, in[10] + K3, 9);
1375 ROUND(H, c, d, a, b, in[ 1] + K3, 11);
1376 ROUND(H, b, c, d, a, in[ 5] + K3, 15);
1377 ROUND(H, a, b, c, d, in[ 9] + K3, 3);
1378 ROUND(H, d, a, b, c, in[ 0] + K3, 9);
1379 ROUND(H, c, d, a, b, in[ 4] + K3, 11);
1380 ROUND(H, b, c, d, a, in[ 8] + K3, 15);
1382 return buf[1] + b; /* "most hashed" word */
1383 /* Alternative: return sum of all words? */
1385 #endif
1387 #undef ROUND
1388 #undef F
1389 #undef G
1390 #undef H
1391 #undef K1
1392 #undef K2
1393 #undef K3
1395 /* This should not be decreased so low that ISNs wrap too fast. */
1396 #define REKEY_INTERVAL (300 * HZ)
1398 * Bit layout of the tcp sequence numbers (before adding current time):
1399 * bit 24-31: increased after every key exchange
1400 * bit 0-23: hash(source,dest)
1402 * The implementation is similar to the algorithm described
1403 * in the Appendix of RFC 1185, except that
1404 * - it uses a 1 MHz clock instead of a 250 kHz clock
1405 * - it performs a rekey every 5 minutes, which is equivalent
1406 * to a (source,dest) tulple dependent forward jump of the
1407 * clock by 0..2^(HASH_BITS+1)
1409 * Thus the average ISN wraparound time is 68 minutes instead of
1410 * 4.55 hours.
1412 * SMP cleanup and lock avoidance with poor man's RCU.
1413 * Manfred Spraul <manfred@colorfullife.com>
1416 #define COUNT_BITS 8
1417 #define COUNT_MASK ((1 << COUNT_BITS) - 1)
1418 #define HASH_BITS 24
1419 #define HASH_MASK ((1 << HASH_BITS) - 1)
1421 static struct keydata {
1422 __u32 count; /* already shifted to the final position */
1423 __u32 secret[12];
1424 } ____cacheline_aligned ip_keydata[2];
1426 static unsigned int ip_cnt;
1428 static void rekey_seq_generator(struct work_struct *work);
1430 static DECLARE_DELAYED_WORK(rekey_work, rekey_seq_generator);
1433 * Lock avoidance:
1434 * The ISN generation runs lockless - it's just a hash over random data.
1435 * State changes happen every 5 minutes when the random key is replaced.
1436 * Synchronization is performed by having two copies of the hash function
1437 * state and rekey_seq_generator always updates the inactive copy.
1438 * The copy is then activated by updating ip_cnt.
1439 * The implementation breaks down if someone blocks the thread
1440 * that processes SYN requests for more than 5 minutes. Should never
1441 * happen, and even if that happens only a not perfectly compliant
1442 * ISN is generated, nothing fatal.
1444 static void rekey_seq_generator(struct work_struct *work)
1446 struct keydata *keyptr = &ip_keydata[1 ^ (ip_cnt & 1)];
1448 get_random_bytes(keyptr->secret, sizeof(keyptr->secret));
1449 keyptr->count = (ip_cnt & COUNT_MASK) << HASH_BITS;
1450 smp_wmb();
1451 ip_cnt++;
1452 schedule_delayed_work(&rekey_work,
1453 round_jiffies_relative(REKEY_INTERVAL));
1456 static inline struct keydata *get_keyptr(void)
1458 struct keydata *keyptr = &ip_keydata[ip_cnt & 1];
1460 smp_rmb();
1462 return keyptr;
1465 static __init int seqgen_init(void)
1467 rekey_seq_generator(NULL);
1468 return 0;
1470 late_initcall(seqgen_init);
1472 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1473 __u32 secure_tcpv6_sequence_number(__be32 *saddr, __be32 *daddr,
1474 __be16 sport, __be16 dport)
1476 __u32 seq;
1477 __u32 hash[12];
1478 struct keydata *keyptr = get_keyptr();
1480 /* The procedure is the same as for IPv4, but addresses are longer.
1481 * Thus we must use twothirdsMD4Transform.
1484 memcpy(hash, saddr, 16);
1485 hash[4] = ((__force u16)sport << 16) + (__force u16)dport;
1486 memcpy(&hash[5], keyptr->secret, sizeof(__u32) * 7);
1488 seq = twothirdsMD4Transform((const __u32 *)daddr, hash) & HASH_MASK;
1489 seq += keyptr->count;
1491 seq += ktime_to_ns(ktime_get_real());
1493 return seq;
1495 EXPORT_SYMBOL(secure_tcpv6_sequence_number);
1496 #endif
1498 /* The code below is shamelessly stolen from secure_tcp_sequence_number().
1499 * All blames to Andrey V. Savochkin <saw@msu.ru>.
1501 __u32 secure_ip_id(__be32 daddr)
1503 struct keydata *keyptr;
1504 __u32 hash[4];
1506 keyptr = get_keyptr();
1509 * Pick a unique starting offset for each IP destination.
1510 * The dest ip address is placed in the starting vector,
1511 * which is then hashed with random data.
1513 hash[0] = (__force __u32)daddr;
1514 hash[1] = keyptr->secret[9];
1515 hash[2] = keyptr->secret[10];
1516 hash[3] = keyptr->secret[11];
1518 return half_md4_transform(hash, keyptr->secret);
1521 #ifdef CONFIG_INET
1523 __u32 secure_tcp_sequence_number(__be32 saddr, __be32 daddr,
1524 __be16 sport, __be16 dport)
1526 __u32 seq;
1527 __u32 hash[4];
1528 struct keydata *keyptr = get_keyptr();
1531 * Pick a unique starting offset for each TCP connection endpoints
1532 * (saddr, daddr, sport, dport).
1533 * Note that the words are placed into the starting vector, which is
1534 * then mixed with a partial MD4 over random data.
1536 hash[0] = (__force u32)saddr;
1537 hash[1] = (__force u32)daddr;
1538 hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
1539 hash[3] = keyptr->secret[11];
1541 seq = half_md4_transform(hash, keyptr->secret) & HASH_MASK;
1542 seq += keyptr->count;
1544 * As close as possible to RFC 793, which
1545 * suggests using a 250 kHz clock.
1546 * Further reading shows this assumes 2 Mb/s networks.
1547 * For 10 Mb/s Ethernet, a 1 MHz clock is appropriate.
1548 * For 10 Gb/s Ethernet, a 1 GHz clock should be ok, but
1549 * we also need to limit the resolution so that the u32 seq
1550 * overlaps less than one time per MSL (2 minutes).
1551 * Choosing a clock of 64 ns period is OK. (period of 274 s)
1553 seq += ktime_to_ns(ktime_get_real()) >> 6;
1555 return seq;
1558 /* Generate secure starting point for ephemeral IPV4 transport port search */
1559 u32 secure_ipv4_port_ephemeral(__be32 saddr, __be32 daddr, __be16 dport)
1561 struct keydata *keyptr = get_keyptr();
1562 u32 hash[4];
1565 * Pick a unique starting offset for each ephemeral port search
1566 * (saddr, daddr, dport) and 48bits of random data.
1568 hash[0] = (__force u32)saddr;
1569 hash[1] = (__force u32)daddr;
1570 hash[2] = (__force u32)dport ^ keyptr->secret[10];
1571 hash[3] = keyptr->secret[11];
1573 return half_md4_transform(hash, keyptr->secret);
1575 EXPORT_SYMBOL_GPL(secure_ipv4_port_ephemeral);
1577 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1578 u32 secure_ipv6_port_ephemeral(const __be32 *saddr, const __be32 *daddr,
1579 __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_to_ns(ktime_get_real());
1613 seq &= (1ull << 48) - 1;
1615 return seq;
1617 EXPORT_SYMBOL(secure_dccp_sequence_number);
1618 #endif
1620 #endif /* CONFIG_INET */
1624 * Get a random word for internal kernel use only. Similar to urandom but
1625 * with the goal of minimal entropy pool depletion. As a result, the random
1626 * value is not cryptographically secure but for several uses the cost of
1627 * depleting entropy is too high
1629 DEFINE_PER_CPU(__u32 [4], get_random_int_hash);
1630 unsigned int get_random_int(void)
1632 struct keydata *keyptr;
1633 __u32 *hash = get_cpu_var(get_random_int_hash);
1634 int ret;
1636 keyptr = get_keyptr();
1637 hash[0] += current->pid + jiffies + get_cycles();
1639 ret = half_md4_transform(hash, keyptr->secret);
1640 put_cpu_var(get_random_int_hash);
1642 return ret;
1646 * randomize_range() returns a start address such that
1648 * [...... <range> .....]
1649 * start end
1651 * a <range> with size "len" starting at the return value is inside in the
1652 * area defined by [start, end], but is otherwise randomized.
1654 unsigned long
1655 randomize_range(unsigned long start, unsigned long end, unsigned long len)
1657 unsigned long range = end - len - start;
1659 if (end <= start + len)
1660 return 0;
1661 return PAGE_ALIGN(get_random_int() % range + start);