2 * random.c -- A strong random number generator
4 * Version 1.89, last modified 19-Sep-99
6 * Copyright Theodore Ts'o, 1994, 1995, 1996, 1997, 1998, 1999. All
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
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
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
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.
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
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
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
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
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
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
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
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 * The code for SHA transform was taken from Peter Gutmann's
222 * implementation, which has been placed in the public domain.
223 * The code for MD5 transform was taken from Colin Plumb's
224 * implementation, which has been placed in the public domain.
225 * The MD5 cryptographic checksum was devised by Ronald Rivest, and is
226 * documented in RFC 1321, "The MD5 Message Digest Algorithm".
228 * Further background information on this topic may be obtained from
229 * RFC 1750, "Randomness Recommendations for Security", by Donald
230 * Eastlake, Steve Crocker, and Jeff Schiller.
233 #include <linux/utsname.h>
234 #include <linux/config.h>
235 #include <linux/module.h>
236 #include <linux/kernel.h>
237 #include <linux/major.h>
238 #include <linux/string.h>
239 #include <linux/fcntl.h>
240 #include <linux/slab.h>
241 #include <linux/random.h>
242 #include <linux/poll.h>
243 #include <linux/init.h>
244 #include <linux/fs.h>
245 #include <linux/workqueue.h>
246 #include <linux/genhd.h>
247 #include <linux/interrupt.h>
248 #include <linux/spinlock.h>
249 #include <linux/percpu.h>
251 #include <asm/processor.h>
252 #include <asm/uaccess.h>
257 * Configuration information
259 #define DEFAULT_POOL_SIZE 512
260 #define SECONDARY_POOL_SIZE 128
261 #define BATCH_ENTROPY_SIZE 256
265 * The minimum number of bits of entropy before we wake up a read on
266 * /dev/random. Should be enough to do a significant reseed.
268 static int random_read_wakeup_thresh = 64;
271 * If the entropy count falls under this number of bits, then we
272 * should wake up processes which are selecting or polling on write
273 * access to /dev/random.
275 static int random_write_wakeup_thresh = 128;
278 * When the input pool goes over trickle_thresh, start dropping most
279 * samples to avoid wasting CPU time and reduce lock contention.
282 static int trickle_thresh = DEFAULT_POOL_SIZE * 7;
284 static DEFINE_PER_CPU(int, trickle_count) = 0;
287 * A pool of size .poolwords is stirred with a primitive polynomial
288 * of degree .poolwords over GF(2). The taps for various sizes are
289 * defined below. They are chosen to be evenly spaced (minimum RMS
290 * distance from evenly spaced; the numbers in the comments are a
291 * scaled squared error sum) except for the last tap, which is 1 to
292 * get the twisting happening as fast as possible.
294 static struct poolinfo {
296 int tap1, tap2, tap3, tap4, tap5;
297 } poolinfo_table[] = {
298 /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1 -- 115 */
299 { 2048, 1638, 1231, 819, 411, 1 },
301 /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
302 { 1024, 817, 615, 412, 204, 1 },
303 #if 0 /* Alternate polynomial */
304 /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
305 { 1024, 819, 616, 410, 207, 2 },
308 /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
309 { 512, 411, 308, 208, 104, 1 },
310 #if 0 /* Alternates */
311 /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
312 { 512, 409, 307, 206, 102, 2 },
313 /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
314 { 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^76 + x^51 +x^25 + x + 1 -- 105 */
321 { 128, 103, 76, 51, 25, 1 },
322 #if 0 /* Alternate polynomial */
323 /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
324 { 128, 103, 78, 51, 27, 2 },
327 /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
328 { 64, 52, 39, 26, 14, 1 },
330 /* x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 -- 15 */
331 { 32, 26, 20, 14, 7, 1 },
333 { 0, 0, 0, 0, 0, 0 },
336 #define POOLBITS poolwords*32
337 #define POOLBYTES poolwords*4
340 * For the purposes of better mixing, we use the CRC-32 polynomial as
341 * well to make a twisted Generalized Feedback Shift Reigster
343 * (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR generators. ACM
344 * Transactions on Modeling and Computer Simulation 2(3):179-194.
345 * Also see M. Matsumoto & Y. Kurita, 1994. Twisted GFSR generators
346 * II. ACM Transactions on Mdeling and Computer Simulation 4:254-266)
348 * Thanks to Colin Plumb for suggesting this.
350 * We have not analyzed the resultant polynomial to prove it primitive;
351 * in fact it almost certainly isn't. Nonetheless, the irreducible factors
352 * of a random large-degree polynomial over GF(2) are more than large enough
353 * that periodicity is not a concern.
355 * The input hash is much less sensitive than the output hash. All
356 * that we want of it is that it be a good non-cryptographic hash;
357 * i.e. it not produce collisions when fed "random" data of the sort
358 * we expect to see. As long as the pool state differs for different
359 * inputs, we have preserved the input entropy and done a good job.
360 * The fact that an intelligent attacker can construct inputs that
361 * will produce controlled alterations to the pool's state is not
362 * important because we don't consider such inputs to contribute any
363 * randomness. The only property we need with respect to them is that
364 * the attacker can't increase his/her knowledge of the pool's state.
365 * Since all additions are reversible (knowing the final state and the
366 * input, you can reconstruct the initial state), if an attacker has
367 * any uncertainty about the initial state, he/she can only shuffle
368 * that uncertainty about, but never cause any collisions (which would
369 * decrease the uncertainty).
371 * The chosen system lets the state of the pool be (essentially) the input
372 * modulo the generator polymnomial. Now, for random primitive polynomials,
373 * this is a universal class of hash functions, meaning that the chance
374 * of a collision is limited by the attacker's knowledge of the generator
375 * polynomail, so if it is chosen at random, an attacker can never force
376 * a collision. Here, we use a fixed polynomial, but we *can* assume that
377 * ###--> it is unknown to the processes generating the input entropy. <-###
378 * Because of this important property, this is a good, collision-resistant
379 * hash; hash collisions will occur no more often than chance.
383 * Linux 2.2 compatibility
385 #ifndef DECLARE_WAITQUEUE
386 #define DECLARE_WAITQUEUE(WAIT, PTR) struct wait_queue WAIT = { PTR, NULL }
388 #ifndef DECLARE_WAIT_QUEUE_HEAD
389 #define DECLARE_WAIT_QUEUE_HEAD(WAIT) struct wait_queue *WAIT
393 * Static global variables
395 static struct entropy_store *random_state; /* The default global store */
396 static struct entropy_store *sec_random_state; /* secondary store */
397 static struct entropy_store *urandom_state; /* For urandom */
398 static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
399 static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
402 * Forward procedure declarations
405 static void sysctl_init_random(struct entropy_store *random_state);
408 /*****************************************************************
410 * Utility functions, with some ASM defined functions for speed
413 *****************************************************************/
416 * Unfortunately, while the GCC optimizer for the i386 understands how
417 * to optimize a static rotate left of x bits, it doesn't know how to
418 * deal with a variable rotate of x bits. So we use a bit of asm magic.
420 #if (!defined (__i386__))
421 static inline __u32 rotate_left(int i, __u32 word)
423 return (word << i) | (word >> (32 - i));
426 static inline __u32 rotate_left(int i, __u32 word)
428 __asm__("roll %%cl,%0"
430 :"0" (word),"c" (i));
438 * For entropy estimation, we need to do an integral base 2
441 * Note the "12bits" suffix - this is used for numbers between
442 * 0 and 4095 only. This allows a few shortcuts.
444 #if 0 /* Slow but clear version */
445 static inline __u32 int_ln_12bits(__u32 word)
453 #else /* Faster (more clever) version, courtesy Colin Plumb */
454 static inline __u32 int_ln_12bits(__u32 word)
456 /* Smear msbit right to make an n-bit mask */
461 /* Remove one bit to make this a logarithm */
463 /* Count the bits set in the word */
464 word -= (word >> 1) & 0x555;
465 word = (word & 0x333) + ((word >> 2) & 0x333);
473 static int debug = 0;
474 module_param(debug, bool, 0644);
475 #define DEBUG_ENT(fmt, arg...) do { if (debug) \
476 printk(KERN_DEBUG "random %04d %04d %04d: " \
478 random_state->entropy_count,\
479 sec_random_state->entropy_count,\
480 urandom_state->entropy_count,\
483 #define DEBUG_ENT(fmt, arg...) do {} while (0)
486 /**********************************************************************
488 * OS independent entropy store. Here are the functions which handle
489 * storing entropy in an entropy pool.
491 **********************************************************************/
493 struct entropy_store {
494 /* mostly-read data: */
495 struct poolinfo poolinfo;
499 /* read-write data: */
500 spinlock_t lock ____cacheline_aligned_in_smp;
507 * Initialize the entropy store. The input argument is the size of
510 * Returns an negative error if there is a problem.
512 static int create_entropy_store(int size, const char *name,
513 struct entropy_store **ret_bucket)
515 struct entropy_store *r;
519 poolwords = (size + 3) / 4; /* Convert bytes->words */
520 /* The pool size must be a multiple of 16 32-bit words */
521 poolwords = ((poolwords + 15) / 16) * 16;
523 for (p = poolinfo_table; p->poolwords; p++) {
524 if (poolwords == p->poolwords)
527 if (p->poolwords == 0)
530 r = kmalloc(sizeof(struct entropy_store), GFP_KERNEL);
534 memset (r, 0, sizeof(struct entropy_store));
537 r->pool = kmalloc(POOLBYTES, GFP_KERNEL);
542 memset(r->pool, 0, POOLBYTES);
543 spin_lock_init(&r->lock);
549 /* Clear the entropy pool and associated counters. */
550 static void clear_entropy_store(struct entropy_store *r)
553 r->entropy_count = 0;
555 memset(r->pool, 0, r->poolinfo.POOLBYTES);
559 * This function adds a byte into the entropy "pool". It does not
560 * update the entropy estimate. The caller should call
561 * credit_entropy_store if this is appropriate.
563 * The pool is stirred with a primitive polynomial of the appropriate
564 * degree, and then twisted. We twist by three bits at a time because
565 * it's cheap to do so and helps slightly in the expected case where
566 * the entropy is concentrated in the low-order bits.
568 static void __add_entropy_words(struct entropy_store *r, const __u32 *in,
569 int nwords, __u32 out[16])
571 static __u32 const twist_table[8] = {
572 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
573 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
574 unsigned long i, add_ptr, tap1, tap2, tap3, tap4, tap5;
575 int new_rotate, input_rotate;
576 int wordmask = r->poolinfo.poolwords - 1;
580 /* Taps are constant, so we can load them without holding r->lock. */
581 tap1 = r->poolinfo.tap1;
582 tap2 = r->poolinfo.tap2;
583 tap3 = r->poolinfo.tap3;
584 tap4 = r->poolinfo.tap4;
585 tap5 = r->poolinfo.tap5;
588 spin_lock_irqsave(&r->lock, flags);
589 prefetch_range(r->pool, wordmask);
590 input_rotate = r->input_rotate;
591 add_ptr = r->add_ptr;
594 w = rotate_left(input_rotate, next_w);
597 i = add_ptr = (add_ptr - 1) & wordmask;
599 * Normally, we add 7 bits of rotation to the pool.
600 * At the beginning of the pool, add an extra 7 bits
601 * rotation, so that successive passes spread the
602 * input bits across the pool evenly.
604 new_rotate = input_rotate + 14;
606 new_rotate = input_rotate + 7;
607 input_rotate = new_rotate & 31;
609 /* XOR in the various taps */
610 w ^= r->pool[(i + tap1) & wordmask];
611 w ^= r->pool[(i + tap2) & wordmask];
612 w ^= r->pool[(i + tap3) & wordmask];
613 w ^= r->pool[(i + tap4) & wordmask];
614 w ^= r->pool[(i + tap5) & wordmask];
616 r->pool[i] = (w >> 3) ^ twist_table[w & 7];
619 r->input_rotate = input_rotate;
620 r->add_ptr = add_ptr;
623 for (i = 0; i < 16; i++) {
624 out[i] = r->pool[add_ptr];
625 add_ptr = (add_ptr - 1) & wordmask;
629 spin_unlock_irqrestore(&r->lock, flags);
632 static inline void add_entropy_words(struct entropy_store *r, const __u32 *in,
635 __add_entropy_words(r, in, nwords, NULL);
639 * Credit (or debit) the entropy store with n bits of entropy
641 static void credit_entropy_store(struct entropy_store *r, int nbits)
645 spin_lock_irqsave(&r->lock, flags);
647 if (r->entropy_count + nbits < 0) {
648 DEBUG_ENT("negative entropy/overflow (%d+%d)\n",
649 r->entropy_count, nbits);
650 r->entropy_count = 0;
651 } else if (r->entropy_count + nbits > r->poolinfo.POOLBITS) {
652 r->entropy_count = r->poolinfo.POOLBITS;
654 r->entropy_count += nbits;
656 DEBUG_ENT("added %d entropy credits to %s\n",
660 spin_unlock_irqrestore(&r->lock, flags);
663 /**********************************************************************
665 * Entropy batch input management
667 * We batch entropy to be added to avoid increasing interrupt latency
669 **********************************************************************/
676 static struct sample *batch_entropy_pool, *batch_entropy_copy;
677 static int batch_head, batch_tail;
678 static DEFINE_SPINLOCK(batch_lock);
680 static int batch_max;
681 static void batch_entropy_process(void *private_);
682 static DECLARE_WORK(batch_work, batch_entropy_process, NULL);
684 /* note: the size must be a power of 2 */
685 static int __init batch_entropy_init(int size, struct entropy_store *r)
687 batch_entropy_pool = kmalloc(size*sizeof(struct sample), GFP_KERNEL);
688 if (!batch_entropy_pool)
690 batch_entropy_copy = kmalloc(size*sizeof(struct sample), GFP_KERNEL);
691 if (!batch_entropy_copy) {
692 kfree(batch_entropy_pool);
695 batch_head = batch_tail = 0;
702 * Changes to the entropy data is put into a queue rather than being added to
703 * the entropy counts directly. This is presumably to avoid doing heavy
704 * hashing calculations during an interrupt in add_timer_randomness().
705 * Instead, the entropy is only added to the pool by keventd.
707 static void batch_entropy_store(u32 a, u32 b, int num)
715 spin_lock_irqsave(&batch_lock, flags);
717 batch_entropy_pool[batch_head].data[0] = a;
718 batch_entropy_pool[batch_head].data[1] = b;
719 batch_entropy_pool[batch_head].credit = num;
721 if (((batch_head - batch_tail) & (batch_max - 1)) >= (batch_max / 2))
722 schedule_delayed_work(&batch_work, 1);
724 new = (batch_head + 1) & (batch_max - 1);
725 if (new == batch_tail)
726 DEBUG_ENT("batch entropy buffer full\n");
730 spin_unlock_irqrestore(&batch_lock, flags);
734 * Flush out the accumulated entropy operations, adding entropy to the passed
735 * store (normally random_state). If that store has enough entropy, alternate
736 * between randomizing the data of the primary and secondary stores.
738 static void batch_entropy_process(void *private_)
740 struct entropy_store *r = (struct entropy_store *) private_, *p;
741 int max_entropy = r->poolinfo.POOLBITS;
744 /* Mixing into the pool is expensive, so copy over the batch
745 * data and release the batch lock. The pool is at least half
746 * full, so don't worry too much about copying only the used
749 spin_lock_irq(&batch_lock);
751 memcpy(batch_entropy_copy, batch_entropy_pool,
752 batch_max * sizeof(struct sample));
756 batch_tail = batch_head;
758 spin_unlock_irq(&batch_lock);
761 while (head != tail) {
762 if (r->entropy_count >= max_entropy) {
763 r = (r == sec_random_state) ? random_state :
765 max_entropy = r->poolinfo.POOLBITS;
767 add_entropy_words(r, batch_entropy_copy[tail].data, 2);
768 credit_entropy_store(r, batch_entropy_copy[tail].credit);
769 tail = (tail + 1) & (batch_max - 1);
771 if (p->entropy_count >= random_read_wakeup_thresh)
772 wake_up_interruptible(&random_read_wait);
775 /*********************************************************************
777 * Entropy input management
779 *********************************************************************/
781 /* There is one of these per entropy source */
782 struct timer_rand_state {
784 long last_delta,last_delta2;
785 unsigned dont_count_entropy:1;
788 static struct timer_rand_state input_timer_state;
789 static struct timer_rand_state extract_timer_state;
790 static struct timer_rand_state *irq_timer_state[NR_IRQS];
793 * This function adds entropy to the entropy "pool" by using timing
794 * delays. It uses the timer_rand_state structure to make an estimate
795 * of how many bits of entropy this call has added to the pool.
797 * The number "num" is also added to the pool - it should somehow describe
798 * the type of event which just happened. This is currently 0-255 for
799 * keyboard scan codes, and 256 upwards for interrupts.
802 static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
805 long delta, delta2, delta3, time;
809 /* if over the trickle threshold, use only 1 in 4096 samples */
810 if (random_state->entropy_count > trickle_thresh &&
811 (__get_cpu_var(trickle_count)++ & 0xfff))
815 * Calculate number of bits of randomness we probably added.
816 * We take into account the first, second and third-order deltas
817 * in order to make our estimate.
821 if (!state->dont_count_entropy) {
822 delta = time - state->last_time;
823 state->last_time = time;
825 delta2 = delta - state->last_delta;
826 state->last_delta = delta;
828 delta3 = delta2 - state->last_delta2;
829 state->last_delta2 = delta2;
843 * delta is now minimum absolute delta.
844 * Round down by 1 bit on general principles,
845 * and limit entropy entimate to 12 bits.
848 delta &= (1 << 12) - 1;
850 entropy = int_ln_12bits(delta);
854 * Use get_cycles() if implemented, otherwise fall back to
859 num ^= (u32)((data >> 31) >> 1);
863 batch_entropy_store(num, data, entropy);
868 extern void add_input_randomness(unsigned int type, unsigned int code,
871 static unsigned char last_value;
873 /* ignore autorepeat and the like */
874 if (value == last_value)
877 DEBUG_ENT("input event\n");
879 add_timer_randomness(&input_timer_state,
880 (type << 4) ^ code ^ (code >> 4) ^ value);
883 void add_interrupt_randomness(int irq)
885 if (irq >= NR_IRQS || irq_timer_state[irq] == 0)
888 DEBUG_ENT("irq event %d\n", irq);
889 add_timer_randomness(irq_timer_state[irq], 0x100 + irq);
892 void add_disk_randomness(struct gendisk *disk)
894 if (!disk || !disk->random)
896 /* first major is 1, so we get >= 0x200 here */
897 DEBUG_ENT("disk event %d:%d\n", disk->major, disk->first_minor);
899 add_timer_randomness(disk->random,
900 0x100 + MKDEV(disk->major, disk->first_minor));
903 EXPORT_SYMBOL(add_disk_randomness);
905 /******************************************************************
907 * Hash function definition
909 *******************************************************************/
912 * This chunk of code defines a function
913 * void HASH_TRANSFORM(__u32 digest[HASH_BUFFER_SIZE + HASH_EXTRA_SIZE],
914 * __u32 const data[16])
916 * The function hashes the input data to produce a digest in the first
917 * HASH_BUFFER_SIZE words of the digest[] array, and uses HASH_EXTRA_SIZE
918 * more words for internal purposes. (This buffer is exported so the
919 * caller can wipe it once rather than this code doing it each call,
920 * and tacking it onto the end of the digest[] array is the quick and
921 * dirty way of doing it.)
923 * It so happens that MD5 and SHA share most of the initial vector
924 * used to initialize the digest[] array before the first call:
929 * 5) 0xc3d2e1f0 (SHA only)
931 * For /dev/random purposes, the length of the data being hashed is
932 * fixed in length, so appending a bit count in the usual way is not
933 * cryptographically necessary.
938 #define HASH_BUFFER_SIZE 5
939 #define HASH_EXTRA_SIZE 80
940 #define HASH_TRANSFORM SHATransform
942 /* Various size/speed tradeoffs are available. Choose 0..3. */
943 #define SHA_CODE_SIZE 0
946 * SHA transform algorithm, taken from code written by Peter Gutmann,
947 * and placed in the public domain.
950 /* The SHA f()-functions. */
952 #define f1(x,y,z) (z ^ (x & (y ^ z))) /* Rounds 0-19: x ? y : z */
953 #define f2(x,y,z) (x ^ y ^ z) /* Rounds 20-39: XOR */
954 #define f3(x,y,z) ((x & y) + (z & (x ^ y))) /* Rounds 40-59: majority */
955 #define f4(x,y,z) (x ^ y ^ z) /* Rounds 60-79: XOR */
957 /* The SHA Mysterious Constants */
959 #define K1 0x5A827999L /* Rounds 0-19: sqrt(2) * 2^30 */
960 #define K2 0x6ED9EBA1L /* Rounds 20-39: sqrt(3) * 2^30 */
961 #define K3 0x8F1BBCDCL /* Rounds 40-59: sqrt(5) * 2^30 */
962 #define K4 0xCA62C1D6L /* Rounds 60-79: sqrt(10) * 2^30 */
964 #define ROTL(n,X) (((X) << n ) | ((X) >> (32 - n)))
966 #define subRound(a, b, c, d, e, f, k, data) \
967 (e += ROTL(5, a) + f(b, c, d) + k + data, b = ROTL(30, b))
969 static void SHATransform(__u32 digest[85], __u32 const data[16])
971 __u32 A, B, C, D, E; /* Local vars */
974 #define W (digest + HASH_BUFFER_SIZE) /* Expanded data array */
977 * Do the preliminary expansion of 16 to 80 words. Doing it
978 * out-of-line line this is faster than doing it in-line on
979 * register-starved machines like the x86, and not really any
980 * slower on real processors.
982 memcpy(W, data, 16*sizeof(__u32));
983 for (i = 0; i < 64; i++) {
984 TEMP = W[i] ^ W[i+2] ^ W[i+8] ^ W[i+13];
985 W[i+16] = ROTL(1, TEMP);
988 /* Set up first buffer and local data buffer */
995 /* Heavy mangling, in 4 sub-rounds of 20 iterations each. */
996 #if SHA_CODE_SIZE == 0
998 * Approximately 50% of the speed of the largest version, but
999 * takes up 1/16 the space. Saves about 6k on an i386 kernel.
1001 for (i = 0; i < 80; i++) {
1004 TEMP = f1(B, C, D) + K1;
1006 TEMP = f2(B, C, D) + K2;
1009 TEMP = f3(B, C, D) + K3;
1011 TEMP = f4(B, C, D) + K4;
1013 TEMP += ROTL(5, A) + E + W[i];
1014 E = D; D = C; C = ROTL(30, B); B = A; A = TEMP;
1016 #elif SHA_CODE_SIZE == 1
1017 for (i = 0; i < 20; i++) {
1018 TEMP = f1(B, C, D) + K1 + ROTL(5, A) + E + W[i];
1019 E = D; D = C; C = ROTL(30, B); B = A; A = TEMP;
1021 for (; i < 40; i++) {
1022 TEMP = f2(B, C, D) + K2 + ROTL(5, A) + E + W[i];
1023 E = D; D = C; C = ROTL(30, B); B = A; A = TEMP;
1025 for (; i < 60; i++) {
1026 TEMP = f3(B, C, D) + K3 + ROTL(5, A) + E + W[i];
1027 E = D; D = C; C = ROTL(30, B); B = A; A = TEMP;
1029 for (; i < 80; i++) {
1030 TEMP = f4(B, C, D) + K4 + ROTL(5, A) + E + W[i];
1031 E = D; D = C; C = ROTL(30, B); B = A; A = TEMP;
1033 #elif SHA_CODE_SIZE == 2
1034 for (i = 0; i < 20; i += 5) {
1035 subRound(A, B, C, D, E, f1, K1, W[i ]);
1036 subRound(E, A, B, C, D, f1, K1, W[i+1]);
1037 subRound(D, E, A, B, C, f1, K1, W[i+2]);
1038 subRound(C, D, E, A, B, f1, K1, W[i+3]);
1039 subRound(B, C, D, E, A, f1, K1, W[i+4]);
1041 for (; i < 40; i += 5) {
1042 subRound(A, B, C, D, E, f2, K2, W[i ]);
1043 subRound(E, A, B, C, D, f2, K2, W[i+1]);
1044 subRound(D, E, A, B, C, f2, K2, W[i+2]);
1045 subRound(C, D, E, A, B, f2, K2, W[i+3]);
1046 subRound(B, C, D, E, A, f2, K2, W[i+4]);
1048 for (; i < 60; i += 5) {
1049 subRound(A, B, C, D, E, f3, K3, W[i ]);
1050 subRound(E, A, B, C, D, f3, K3, W[i+1]);
1051 subRound(D, E, A, B, C, f3, K3, W[i+2]);
1052 subRound(C, D, E, A, B, f3, K3, W[i+3]);
1053 subRound(B, C, D, E, A, f3, K3, W[i+4]);
1055 for (; i < 80; i += 5) {
1056 subRound(A, B, C, D, E, f4, K4, W[i ]);
1057 subRound(E, A, B, C, D, f4, K4, W[i+1]);
1058 subRound(D, E, A, B, C, f4, K4, W[i+2]);
1059 subRound(C, D, E, A, B, f4, K4, W[i+3]);
1060 subRound(B, C, D, E, A, f4, K4, W[i+4]);
1062 #elif SHA_CODE_SIZE == 3 /* Really large version */
1063 subRound(A, B, C, D, E, f1, K1, W[ 0]);
1064 subRound(E, A, B, C, D, f1, K1, W[ 1]);
1065 subRound(D, E, A, B, C, f1, K1, W[ 2]);
1066 subRound(C, D, E, A, B, f1, K1, W[ 3]);
1067 subRound(B, C, D, E, A, f1, K1, W[ 4]);
1068 subRound(A, B, C, D, E, f1, K1, W[ 5]);
1069 subRound(E, A, B, C, D, f1, K1, W[ 6]);
1070 subRound(D, E, A, B, C, f1, K1, W[ 7]);
1071 subRound(C, D, E, A, B, f1, K1, W[ 8]);
1072 subRound(B, C, D, E, A, f1, K1, W[ 9]);
1073 subRound(A, B, C, D, E, f1, K1, W[10]);
1074 subRound(E, A, B, C, D, f1, K1, W[11]);
1075 subRound(D, E, A, B, C, f1, K1, W[12]);
1076 subRound(C, D, E, A, B, f1, K1, W[13]);
1077 subRound(B, C, D, E, A, f1, K1, W[14]);
1078 subRound(A, B, C, D, E, f1, K1, W[15]);
1079 subRound(E, A, B, C, D, f1, K1, W[16]);
1080 subRound(D, E, A, B, C, f1, K1, W[17]);
1081 subRound(C, D, E, A, B, f1, K1, W[18]);
1082 subRound(B, C, D, E, A, f1, K1, W[19]);
1084 subRound(A, B, C, D, E, f2, K2, W[20]);
1085 subRound(E, A, B, C, D, f2, K2, W[21]);
1086 subRound(D, E, A, B, C, f2, K2, W[22]);
1087 subRound(C, D, E, A, B, f2, K2, W[23]);
1088 subRound(B, C, D, E, A, f2, K2, W[24]);
1089 subRound(A, B, C, D, E, f2, K2, W[25]);
1090 subRound(E, A, B, C, D, f2, K2, W[26]);
1091 subRound(D, E, A, B, C, f2, K2, W[27]);
1092 subRound(C, D, E, A, B, f2, K2, W[28]);
1093 subRound(B, C, D, E, A, f2, K2, W[29]);
1094 subRound(A, B, C, D, E, f2, K2, W[30]);
1095 subRound(E, A, B, C, D, f2, K2, W[31]);
1096 subRound(D, E, A, B, C, f2, K2, W[32]);
1097 subRound(C, D, E, A, B, f2, K2, W[33]);
1098 subRound(B, C, D, E, A, f2, K2, W[34]);
1099 subRound(A, B, C, D, E, f2, K2, W[35]);
1100 subRound(E, A, B, C, D, f2, K2, W[36]);
1101 subRound(D, E, A, B, C, f2, K2, W[37]);
1102 subRound(C, D, E, A, B, f2, K2, W[38]);
1103 subRound(B, C, D, E, A, f2, K2, W[39]);
1105 subRound(A, B, C, D, E, f3, K3, W[40]);
1106 subRound(E, A, B, C, D, f3, K3, W[41]);
1107 subRound(D, E, A, B, C, f3, K3, W[42]);
1108 subRound(C, D, E, A, B, f3, K3, W[43]);
1109 subRound(B, C, D, E, A, f3, K3, W[44]);
1110 subRound(A, B, C, D, E, f3, K3, W[45]);
1111 subRound(E, A, B, C, D, f3, K3, W[46]);
1112 subRound(D, E, A, B, C, f3, K3, W[47]);
1113 subRound(C, D, E, A, B, f3, K3, W[48]);
1114 subRound(B, C, D, E, A, f3, K3, W[49]);
1115 subRound(A, B, C, D, E, f3, K3, W[50]);
1116 subRound(E, A, B, C, D, f3, K3, W[51]);
1117 subRound(D, E, A, B, C, f3, K3, W[52]);
1118 subRound(C, D, E, A, B, f3, K3, W[53]);
1119 subRound(B, C, D, E, A, f3, K3, W[54]);
1120 subRound(A, B, C, D, E, f3, K3, W[55]);
1121 subRound(E, A, B, C, D, f3, K3, W[56]);
1122 subRound(D, E, A, B, C, f3, K3, W[57]);
1123 subRound(C, D, E, A, B, f3, K3, W[58]);
1124 subRound(B, C, D, E, A, f3, K3, W[59]);
1126 subRound(A, B, C, D, E, f4, K4, W[60]);
1127 subRound(E, A, B, C, D, f4, K4, W[61]);
1128 subRound(D, E, A, B, C, f4, K4, W[62]);
1129 subRound(C, D, E, A, B, f4, K4, W[63]);
1130 subRound(B, C, D, E, A, f4, K4, W[64]);
1131 subRound(A, B, C, D, E, f4, K4, W[65]);
1132 subRound(E, A, B, C, D, f4, K4, W[66]);
1133 subRound(D, E, A, B, C, f4, K4, W[67]);
1134 subRound(C, D, E, A, B, f4, K4, W[68]);
1135 subRound(B, C, D, E, A, f4, K4, W[69]);
1136 subRound(A, B, C, D, E, f4, K4, W[70]);
1137 subRound(E, A, B, C, D, f4, K4, W[71]);
1138 subRound(D, E, A, B, C, f4, K4, W[72]);
1139 subRound(C, D, E, A, B, f4, K4, W[73]);
1140 subRound(B, C, D, E, A, f4, K4, W[74]);
1141 subRound(A, B, C, D, E, f4, K4, W[75]);
1142 subRound(E, A, B, C, D, f4, K4, W[76]);
1143 subRound(D, E, A, B, C, f4, K4, W[77]);
1144 subRound(C, D, E, A, B, f4, K4, W[78]);
1145 subRound(B, C, D, E, A, f4, K4, W[79]);
1147 #error Illegal SHA_CODE_SIZE
1150 /* Build message digest */
1157 /* W is wiped by the caller */
1172 #else /* !USE_SHA - Use MD5 */
1174 #define HASH_BUFFER_SIZE 4
1175 #define HASH_EXTRA_SIZE 0
1176 #define HASH_TRANSFORM MD5Transform
1179 * MD5 transform algorithm, taken from code written by Colin Plumb,
1180 * and put into the public domain
1183 /* The four core functions - F1 is optimized somewhat */
1185 /* #define F1(x, y, z) (x & y | ~x & z) */
1186 #define F1(x, y, z) (z ^ (x & (y ^ z)))
1187 #define F2(x, y, z) F1(z, x, y)
1188 #define F3(x, y, z) (x ^ y ^ z)
1189 #define F4(x, y, z) (y ^ (x | ~z))
1191 /* This is the central step in the MD5 algorithm. */
1192 #define MD5STEP(f, w, x, y, z, data, s) \
1193 (w += f(x, y, z) + data, w = w << s | w >> (32 - s), w += x )
1196 * The core of the MD5 algorithm, this alters an existing MD5 hash to
1197 * reflect the addition of 16 longwords of new data. MD5Update blocks
1198 * the data and converts bytes into longwords for this routine.
1200 static void MD5Transform(__u32 buf[HASH_BUFFER_SIZE], __u32 const in[16])
1209 MD5STEP(F1, a, b, c, d, in[ 0]+0xd76aa478, 7);
1210 MD5STEP(F1, d, a, b, c, in[ 1]+0xe8c7b756, 12);
1211 MD5STEP(F1, c, d, a, b, in[ 2]+0x242070db, 17);
1212 MD5STEP(F1, b, c, d, a, in[ 3]+0xc1bdceee, 22);
1213 MD5STEP(F1, a, b, c, d, in[ 4]+0xf57c0faf, 7);
1214 MD5STEP(F1, d, a, b, c, in[ 5]+0x4787c62a, 12);
1215 MD5STEP(F1, c, d, a, b, in[ 6]+0xa8304613, 17);
1216 MD5STEP(F1, b, c, d, a, in[ 7]+0xfd469501, 22);
1217 MD5STEP(F1, a, b, c, d, in[ 8]+0x698098d8, 7);
1218 MD5STEP(F1, d, a, b, c, in[ 9]+0x8b44f7af, 12);
1219 MD5STEP(F1, c, d, a, b, in[10]+0xffff5bb1, 17);
1220 MD5STEP(F1, b, c, d, a, in[11]+0x895cd7be, 22);
1221 MD5STEP(F1, a, b, c, d, in[12]+0x6b901122, 7);
1222 MD5STEP(F1, d, a, b, c, in[13]+0xfd987193, 12);
1223 MD5STEP(F1, c, d, a, b, in[14]+0xa679438e, 17);
1224 MD5STEP(F1, b, c, d, a, in[15]+0x49b40821, 22);
1226 MD5STEP(F2, a, b, c, d, in[ 1]+0xf61e2562, 5);
1227 MD5STEP(F2, d, a, b, c, in[ 6]+0xc040b340, 9);
1228 MD5STEP(F2, c, d, a, b, in[11]+0x265e5a51, 14);
1229 MD5STEP(F2, b, c, d, a, in[ 0]+0xe9b6c7aa, 20);
1230 MD5STEP(F2, a, b, c, d, in[ 5]+0xd62f105d, 5);
1231 MD5STEP(F2, d, a, b, c, in[10]+0x02441453, 9);
1232 MD5STEP(F2, c, d, a, b, in[15]+0xd8a1e681, 14);
1233 MD5STEP(F2, b, c, d, a, in[ 4]+0xe7d3fbc8, 20);
1234 MD5STEP(F2, a, b, c, d, in[ 9]+0x21e1cde6, 5);
1235 MD5STEP(F2, d, a, b, c, in[14]+0xc33707d6, 9);
1236 MD5STEP(F2, c, d, a, b, in[ 3]+0xf4d50d87, 14);
1237 MD5STEP(F2, b, c, d, a, in[ 8]+0x455a14ed, 20);
1238 MD5STEP(F2, a, b, c, d, in[13]+0xa9e3e905, 5);
1239 MD5STEP(F2, d, a, b, c, in[ 2]+0xfcefa3f8, 9);
1240 MD5STEP(F2, c, d, a, b, in[ 7]+0x676f02d9, 14);
1241 MD5STEP(F2, b, c, d, a, in[12]+0x8d2a4c8a, 20);
1243 MD5STEP(F3, a, b, c, d, in[ 5]+0xfffa3942, 4);
1244 MD5STEP(F3, d, a, b, c, in[ 8]+0x8771f681, 11);
1245 MD5STEP(F3, c, d, a, b, in[11]+0x6d9d6122, 16);
1246 MD5STEP(F3, b, c, d, a, in[14]+0xfde5380c, 23);
1247 MD5STEP(F3, a, b, c, d, in[ 1]+0xa4beea44, 4);
1248 MD5STEP(F3, d, a, b, c, in[ 4]+0x4bdecfa9, 11);
1249 MD5STEP(F3, c, d, a, b, in[ 7]+0xf6bb4b60, 16);
1250 MD5STEP(F3, b, c, d, a, in[10]+0xbebfbc70, 23);
1251 MD5STEP(F3, a, b, c, d, in[13]+0x289b7ec6, 4);
1252 MD5STEP(F3, d, a, b, c, in[ 0]+0xeaa127fa, 11);
1253 MD5STEP(F3, c, d, a, b, in[ 3]+0xd4ef3085, 16);
1254 MD5STEP(F3, b, c, d, a, in[ 6]+0x04881d05, 23);
1255 MD5STEP(F3, a, b, c, d, in[ 9]+0xd9d4d039, 4);
1256 MD5STEP(F3, d, a, b, c, in[12]+0xe6db99e5, 11);
1257 MD5STEP(F3, c, d, a, b, in[15]+0x1fa27cf8, 16);
1258 MD5STEP(F3, b, c, d, a, in[ 2]+0xc4ac5665, 23);
1260 MD5STEP(F4, a, b, c, d, in[ 0]+0xf4292244, 6);
1261 MD5STEP(F4, d, a, b, c, in[ 7]+0x432aff97, 10);
1262 MD5STEP(F4, c, d, a, b, in[14]+0xab9423a7, 15);
1263 MD5STEP(F4, b, c, d, a, in[ 5]+0xfc93a039, 21);
1264 MD5STEP(F4, a, b, c, d, in[12]+0x655b59c3, 6);
1265 MD5STEP(F4, d, a, b, c, in[ 3]+0x8f0ccc92, 10);
1266 MD5STEP(F4, c, d, a, b, in[10]+0xffeff47d, 15);
1267 MD5STEP(F4, b, c, d, a, in[ 1]+0x85845dd1, 21);
1268 MD5STEP(F4, a, b, c, d, in[ 8]+0x6fa87e4f, 6);
1269 MD5STEP(F4, d, a, b, c, in[15]+0xfe2ce6e0, 10);
1270 MD5STEP(F4, c, d, a, b, in[ 6]+0xa3014314, 15);
1271 MD5STEP(F4, b, c, d, a, in[13]+0x4e0811a1, 21);
1272 MD5STEP(F4, a, b, c, d, in[ 4]+0xf7537e82, 6);
1273 MD5STEP(F4, d, a, b, c, in[11]+0xbd3af235, 10);
1274 MD5STEP(F4, c, d, a, b, in[ 2]+0x2ad7d2bb, 15);
1275 MD5STEP(F4, b, c, d, a, in[ 9]+0xeb86d391, 21);
1289 #endif /* !USE_SHA */
1291 /*********************************************************************
1293 * Entropy extraction routines
1295 *********************************************************************/
1297 #define EXTRACT_ENTROPY_USER 1
1298 #define EXTRACT_ENTROPY_SECONDARY 2
1299 #define EXTRACT_ENTROPY_LIMIT 4
1300 #define TMP_BUF_SIZE (HASH_BUFFER_SIZE + HASH_EXTRA_SIZE)
1301 #define SEC_XFER_SIZE (TMP_BUF_SIZE*4)
1303 static ssize_t extract_entropy(struct entropy_store *r, void * buf,
1304 size_t nbytes, int flags);
1307 * This utility inline function is responsible for transfering entropy
1308 * from the primary pool to the secondary extraction pool. We make
1309 * sure we pull enough for a 'catastrophic reseed'.
1311 static inline void xfer_secondary_pool(struct entropy_store *r,
1312 size_t nbytes, __u32 *tmp)
1314 if (r->entropy_count < nbytes * 8 &&
1315 r->entropy_count < r->poolinfo.POOLBITS) {
1316 int bytes = max_t(int, random_read_wakeup_thresh / 8,
1317 min_t(int, nbytes, TMP_BUF_SIZE));
1319 DEBUG_ENT("going to reseed %s with %d bits "
1320 "(%d of %d requested)\n",
1321 r->name, bytes * 8, nbytes * 8, r->entropy_count);
1323 bytes=extract_entropy(random_state, tmp, bytes,
1324 EXTRACT_ENTROPY_LIMIT);
1325 add_entropy_words(r, tmp, (bytes + 3) / 4);
1326 credit_entropy_store(r, bytes*8);
1331 * This function extracts randomness from the "entropy pool", and
1332 * returns it in a buffer. This function computes how many remaining
1333 * bits of entropy are left in the pool, but it does not restrict the
1334 * number of bytes that are actually obtained. If the EXTRACT_ENTROPY_USER
1335 * flag is given, then the buf pointer is assumed to be in user space.
1337 * If the EXTRACT_ENTROPY_SECONDARY flag is given, then we are actually
1338 * extracting entropy from the secondary pool, and can refill from the
1339 * primary pool if needed.
1341 * Note: extract_entropy() assumes that .poolwords is a multiple of 16 words.
1343 static ssize_t extract_entropy(struct entropy_store *r, void * buf,
1344 size_t nbytes, int flags)
1347 __u32 tmp[TMP_BUF_SIZE], data[16];
1349 unsigned long cpuflags;
1351 /* Redundant, but just in case... */
1352 if (r->entropy_count > r->poolinfo.POOLBITS)
1353 r->entropy_count = r->poolinfo.POOLBITS;
1355 if (flags & EXTRACT_ENTROPY_SECONDARY)
1356 xfer_secondary_pool(r, nbytes, tmp);
1358 /* Hold lock while accounting */
1359 spin_lock_irqsave(&r->lock, cpuflags);
1361 DEBUG_ENT("trying to extract %d bits from %s\n",
1362 nbytes * 8, r->name);
1364 if (flags & EXTRACT_ENTROPY_LIMIT && nbytes >= r->entropy_count / 8)
1365 nbytes = r->entropy_count / 8;
1367 if (r->entropy_count / 8 >= nbytes)
1368 r->entropy_count -= nbytes*8;
1370 r->entropy_count = 0;
1372 if (r->entropy_count < random_write_wakeup_thresh)
1373 wake_up_interruptible(&random_write_wait);
1375 DEBUG_ENT("debiting %d entropy credits from %s%s\n",
1376 nbytes * 8, r->name,
1377 flags & EXTRACT_ENTROPY_LIMIT ? "" : " (unlimited)");
1379 spin_unlock_irqrestore(&r->lock, cpuflags);
1384 * Check if we need to break out or reschedule....
1386 if ((flags & EXTRACT_ENTROPY_USER) && need_resched()) {
1387 if (signal_pending(current)) {
1396 /* Hash the pool to get the output */
1397 tmp[0] = 0x67452301;
1398 tmp[1] = 0xefcdab89;
1399 tmp[2] = 0x98badcfe;
1400 tmp[3] = 0x10325476;
1402 tmp[4] = 0xc3d2e1f0;
1405 * As we hash the pool, we mix intermediate values of
1406 * the hash back into the pool. This eliminates
1407 * backtracking attacks (where the attacker knows
1408 * the state of the pool plus the current outputs, and
1409 * attempts to find previous ouputs), unless the hash
1410 * function can be inverted.
1412 for (i = 0, x = 0; i < r->poolinfo.poolwords; i += 16, x+=2) {
1413 HASH_TRANSFORM(tmp, r->pool+i);
1414 add_entropy_words(r, &tmp[x%HASH_BUFFER_SIZE], 1);
1418 * To avoid duplicates, we atomically extract a
1419 * portion of the pool while mixing, and hash one
1422 __add_entropy_words(r, &tmp[x%HASH_BUFFER_SIZE], 1, data);
1423 HASH_TRANSFORM(tmp, data);
1426 * In case the hash function has some recognizable
1427 * output pattern, we fold it in half.
1429 for (i = 0; i < HASH_BUFFER_SIZE/2; i++)
1430 tmp[i] ^= tmp[i + (HASH_BUFFER_SIZE+1)/2];
1431 #if HASH_BUFFER_SIZE & 1 /* There's a middle word to deal with */
1432 x = tmp[HASH_BUFFER_SIZE/2];
1433 x ^= (x >> 16); /* Fold it in half */
1434 ((__u16 *)tmp)[HASH_BUFFER_SIZE-1] = (__u16)x;
1437 /* Copy data to destination buffer */
1438 i = min(nbytes, HASH_BUFFER_SIZE*sizeof(__u32)/2);
1439 if (flags & EXTRACT_ENTROPY_USER) {
1440 i -= copy_to_user(buf, (__u8 const *)tmp, i);
1446 memcpy(buf, (__u8 const *)tmp, i);
1453 /* Wipe data just returned from memory */
1454 memset(tmp, 0, sizeof(tmp));
1460 * This function is the exported kernel interface. It returns some
1461 * number of good random numbers, suitable for seeding TCP sequence
1464 void get_random_bytes(void *buf, int nbytes)
1466 struct entropy_store *r = urandom_state;
1467 int flags = EXTRACT_ENTROPY_SECONDARY;
1470 r = sec_random_state;
1476 printk(KERN_NOTICE "get_random_bytes called before "
1477 "random driver initialization\n");
1480 extract_entropy(r, (char *) buf, nbytes, flags);
1483 EXPORT_SYMBOL(get_random_bytes);
1485 /*********************************************************************
1487 * Functions to interface with Linux
1489 *********************************************************************/
1492 * Initialize the random pool with standard stuff.
1494 * NOTE: This is an OS-dependent function.
1496 static void init_std_data(struct entropy_store *r)
1503 do_gettimeofday(&tv);
1504 words[0] = tv.tv_sec;
1505 words[1] = tv.tv_usec;
1506 add_entropy_words(r, words, 2);
1509 * This doesn't lock system.utsname. However, we are generating
1510 * entropy so a race with a name set here is fine.
1512 p = (char *) &system_utsname;
1513 for (i = sizeof(system_utsname) / sizeof(words); i; i--) {
1514 memcpy(words, p, sizeof(words));
1515 add_entropy_words(r, words, sizeof(words)/4);
1520 static int __init rand_initialize(void)
1524 if (create_entropy_store(DEFAULT_POOL_SIZE, "primary", &random_state))
1526 if (batch_entropy_init(BATCH_ENTROPY_SIZE, random_state))
1528 if (create_entropy_store(SECONDARY_POOL_SIZE, "secondary",
1531 if (create_entropy_store(SECONDARY_POOL_SIZE, "urandom",
1534 clear_entropy_store(random_state);
1535 clear_entropy_store(sec_random_state);
1536 clear_entropy_store(urandom_state);
1537 init_std_data(random_state);
1538 init_std_data(sec_random_state);
1539 init_std_data(urandom_state);
1540 #ifdef CONFIG_SYSCTL
1541 sysctl_init_random(random_state);
1543 for (i = 0; i < NR_IRQS; i++)
1544 irq_timer_state[i] = NULL;
1545 memset(&input_timer_state, 0, sizeof(struct timer_rand_state));
1546 memset(&extract_timer_state, 0, sizeof(struct timer_rand_state));
1547 extract_timer_state.dont_count_entropy = 1;
1552 module_init(rand_initialize);
1554 void rand_initialize_irq(int irq)
1556 struct timer_rand_state *state;
1558 if (irq >= NR_IRQS || irq_timer_state[irq])
1562 * If kmalloc returns null, we just won't use that entropy
1565 state = kmalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1567 memset(state, 0, sizeof(struct timer_rand_state));
1568 irq_timer_state[irq] = state;
1572 void rand_initialize_disk(struct gendisk *disk)
1574 struct timer_rand_state *state;
1577 * If kmalloc returns null, we just won't use that entropy
1580 state = kmalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1582 memset(state, 0, sizeof(struct timer_rand_state));
1583 disk->random = state;
1588 random_read(struct file * file, char __user * buf, size_t nbytes, loff_t *ppos)
1590 DECLARE_WAITQUEUE(wait, current);
1591 ssize_t n, retval = 0, count = 0;
1596 while (nbytes > 0) {
1598 if (n > SEC_XFER_SIZE)
1601 DEBUG_ENT("reading %d bits\n", n*8);
1603 n = extract_entropy(sec_random_state, buf, n,
1604 EXTRACT_ENTROPY_USER |
1605 EXTRACT_ENTROPY_LIMIT |
1606 EXTRACT_ENTROPY_SECONDARY);
1608 DEBUG_ENT("read got %d bits (%d still needed)\n",
1612 if (file->f_flags & O_NONBLOCK) {
1616 if (signal_pending(current)) {
1617 retval = -ERESTARTSYS;
1621 set_current_state(TASK_INTERRUPTIBLE);
1622 add_wait_queue(&random_read_wait, &wait);
1624 if (sec_random_state->entropy_count / 8 == 0)
1627 set_current_state(TASK_RUNNING);
1628 remove_wait_queue(&random_read_wait, &wait);
1640 break; /* This break makes the device work */
1641 /* like a named pipe */
1645 * If we gave the user some bytes, update the access time.
1648 file_accessed(file);
1650 return (count ? count : retval);
1654 urandom_read(struct file * file, char __user * buf,
1655 size_t nbytes, loff_t *ppos)
1657 int flags = EXTRACT_ENTROPY_USER;
1658 unsigned long cpuflags;
1660 spin_lock_irqsave(&random_state->lock, cpuflags);
1661 if (random_state->entropy_count > random_state->poolinfo.POOLBITS)
1662 flags |= EXTRACT_ENTROPY_SECONDARY;
1663 spin_unlock_irqrestore(&random_state->lock, cpuflags);
1665 return extract_entropy(urandom_state, buf, nbytes, flags);
1669 random_poll(struct file *file, poll_table * wait)
1673 poll_wait(file, &random_read_wait, wait);
1674 poll_wait(file, &random_write_wait, wait);
1676 if (random_state->entropy_count >= random_read_wakeup_thresh)
1677 mask |= POLLIN | POLLRDNORM;
1678 if (random_state->entropy_count < random_write_wakeup_thresh)
1679 mask |= POLLOUT | POLLWRNORM;
1684 random_write(struct file * file, const char __user * buffer,
1685 size_t count, loff_t *ppos)
1690 const char __user *p = buffer;
1694 bytes = min(c, sizeof(buf));
1696 bytes -= copy_from_user(&buf, p, bytes);
1704 add_entropy_words(random_state, buf, (bytes + 3) / 4);
1707 return (ssize_t)ret;
1709 struct inode *inode = file->f_dentry->d_inode;
1710 inode->i_mtime = current_fs_time(inode->i_sb);
1711 mark_inode_dirty(inode);
1712 return (ssize_t)(p - buffer);
1717 random_ioctl(struct inode * inode, struct file * file,
1718 unsigned int cmd, unsigned long arg)
1720 int size, ent_count;
1721 int __user *p = (int __user *)arg;
1726 ent_count = random_state->entropy_count;
1727 if (put_user(ent_count, p))
1730 case RNDADDTOENTCNT:
1731 if (!capable(CAP_SYS_ADMIN))
1733 if (get_user(ent_count, p))
1735 credit_entropy_store(random_state, ent_count);
1737 * Wake up waiting processes if we have enough
1740 if (random_state->entropy_count >= random_read_wakeup_thresh)
1741 wake_up_interruptible(&random_read_wait);
1744 if (!capable(CAP_SYS_ADMIN))
1746 if (get_user(ent_count, p++))
1750 if (get_user(size, p++))
1752 retval = random_write(file, (const char __user *) p,
1753 size, &file->f_pos);
1756 credit_entropy_store(random_state, ent_count);
1758 * Wake up waiting processes if we have enough
1761 if (random_state->entropy_count >= random_read_wakeup_thresh)
1762 wake_up_interruptible(&random_read_wait);
1765 if (!capable(CAP_SYS_ADMIN))
1767 random_state->entropy_count = 0;
1770 /* Clear the entropy pool and associated counters. */
1771 if (!capable(CAP_SYS_ADMIN))
1773 clear_entropy_store(random_state);
1774 init_std_data(random_state);
1781 struct file_operations random_fops = {
1782 .read = random_read,
1783 .write = random_write,
1784 .poll = random_poll,
1785 .ioctl = random_ioctl,
1788 struct file_operations urandom_fops = {
1789 .read = urandom_read,
1790 .write = random_write,
1791 .ioctl = random_ioctl,
1794 /***************************************************************
1795 * Random UUID interface
1797 * Used here for a Boot ID, but can be useful for other kernel
1799 ***************************************************************/
1802 * Generate random UUID
1804 void generate_random_uuid(unsigned char uuid_out[16])
1806 get_random_bytes(uuid_out, 16);
1807 /* Set UUID version to 4 --- truely random generation */
1808 uuid_out[6] = (uuid_out[6] & 0x0F) | 0x40;
1809 /* Set the UUID variant to DCE */
1810 uuid_out[8] = (uuid_out[8] & 0x3F) | 0x80;
1813 EXPORT_SYMBOL(generate_random_uuid);
1815 /********************************************************************
1819 ********************************************************************/
1821 #ifdef CONFIG_SYSCTL
1823 #include <linux/sysctl.h>
1825 static int min_read_thresh, max_read_thresh;
1826 static int min_write_thresh, max_write_thresh;
1827 static char sysctl_bootid[16];
1830 * These functions is used to return both the bootid UUID, and random
1831 * UUID. The difference is in whether table->data is NULL; if it is,
1832 * then a new UUID is generated and returned to the user.
1834 * If the user accesses this via the proc interface, it will be returned
1835 * as an ASCII string in the standard UUID format. If accesses via the
1836 * sysctl system call, it is returned as 16 bytes of binary data.
1838 static int proc_do_uuid(ctl_table *table, int write, struct file *filp,
1839 void __user *buffer, size_t *lenp, loff_t *ppos)
1841 ctl_table fake_table;
1842 unsigned char buf[64], tmp_uuid[16], *uuid;
1850 generate_random_uuid(uuid);
1852 sprintf(buf, "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-"
1853 "%02x%02x%02x%02x%02x%02x",
1854 uuid[0], uuid[1], uuid[2], uuid[3],
1855 uuid[4], uuid[5], uuid[6], uuid[7],
1856 uuid[8], uuid[9], uuid[10], uuid[11],
1857 uuid[12], uuid[13], uuid[14], uuid[15]);
1858 fake_table.data = buf;
1859 fake_table.maxlen = sizeof(buf);
1861 return proc_dostring(&fake_table, write, filp, buffer, lenp, ppos);
1864 static int uuid_strategy(ctl_table *table, int __user *name, int nlen,
1865 void __user *oldval, size_t __user *oldlenp,
1866 void __user *newval, size_t newlen, void **context)
1868 unsigned char tmp_uuid[16], *uuid;
1871 if (!oldval || !oldlenp)
1880 generate_random_uuid(uuid);
1882 if (get_user(len, oldlenp))
1887 if (copy_to_user(oldval, uuid, len) ||
1888 put_user(len, oldlenp))
1894 static int sysctl_poolsize = DEFAULT_POOL_SIZE;
1895 ctl_table random_table[] = {
1897 .ctl_name = RANDOM_POOLSIZE,
1898 .procname = "poolsize",
1899 .data = &sysctl_poolsize,
1900 .maxlen = sizeof(int),
1902 .proc_handler = &proc_dointvec,
1905 .ctl_name = RANDOM_ENTROPY_COUNT,
1906 .procname = "entropy_avail",
1907 .maxlen = sizeof(int),
1909 .proc_handler = &proc_dointvec,
1912 .ctl_name = RANDOM_READ_THRESH,
1913 .procname = "read_wakeup_threshold",
1914 .data = &random_read_wakeup_thresh,
1915 .maxlen = sizeof(int),
1917 .proc_handler = &proc_dointvec_minmax,
1918 .strategy = &sysctl_intvec,
1919 .extra1 = &min_read_thresh,
1920 .extra2 = &max_read_thresh,
1923 .ctl_name = RANDOM_WRITE_THRESH,
1924 .procname = "write_wakeup_threshold",
1925 .data = &random_write_wakeup_thresh,
1926 .maxlen = sizeof(int),
1928 .proc_handler = &proc_dointvec_minmax,
1929 .strategy = &sysctl_intvec,
1930 .extra1 = &min_write_thresh,
1931 .extra2 = &max_write_thresh,
1934 .ctl_name = RANDOM_BOOT_ID,
1935 .procname = "boot_id",
1936 .data = &sysctl_bootid,
1939 .proc_handler = &proc_do_uuid,
1940 .strategy = &uuid_strategy,
1943 .ctl_name = RANDOM_UUID,
1947 .proc_handler = &proc_do_uuid,
1948 .strategy = &uuid_strategy,
1953 static void sysctl_init_random(struct entropy_store *random_state)
1955 min_read_thresh = 8;
1956 min_write_thresh = 0;
1957 max_read_thresh = max_write_thresh = random_state->poolinfo.POOLBITS;
1958 random_table[1].data = &random_state->entropy_count;
1960 #endif /* CONFIG_SYSCTL */
1962 /********************************************************************
1964 * Random funtions for networking
1966 ********************************************************************/
1970 * TCP initial sequence number picking. This uses the random number
1971 * generator to pick an initial secret value. This value is hashed
1972 * along with the TCP endpoint information to provide a unique
1973 * starting point for each pair of TCP endpoints. This defeats
1974 * attacks which rely on guessing the initial TCP sequence number.
1975 * This algorithm was suggested by Steve Bellovin.
1977 * Using a very strong hash was taking an appreciable amount of the total
1978 * TCP connection establishment time, so this is a weaker hash,
1979 * compensated for by changing the secret periodically.
1982 /* F, G and H are basic MD4 functions: selection, majority, parity */
1983 #define F(x, y, z) ((z) ^ ((x) & ((y) ^ (z))))
1984 #define G(x, y, z) (((x) & (y)) + (((x) ^ (y)) & (z)))
1985 #define H(x, y, z) ((x) ^ (y) ^ (z))
1988 * The generic round function. The application is so specific that
1989 * we don't bother protecting all the arguments with parens, as is generally
1990 * good macro practice, in favor of extra legibility.
1991 * Rotation is separate from addition to prevent recomputation
1993 #define ROUND(f, a, b, c, d, x, s) \
1994 (a += f(b, c, d) + x, a = (a << s) | (a >> (32 - s)))
1996 #define K2 013240474631UL
1997 #define K3 015666365641UL
2000 * Basic cut-down MD4 transform. Returns only 32 bits of result.
2002 static __u32 halfMD4Transform (__u32 const buf[4], __u32 const in[8])
2004 __u32 a = buf[0], b = buf[1], c = buf[2], d = buf[3];
2007 ROUND(F, a, b, c, d, in[0] + K1, 3);
2008 ROUND(F, d, a, b, c, in[1] + K1, 7);
2009 ROUND(F, c, d, a, b, in[2] + K1, 11);
2010 ROUND(F, b, c, d, a, in[3] + K1, 19);
2011 ROUND(F, a, b, c, d, in[4] + K1, 3);
2012 ROUND(F, d, a, b, c, in[5] + K1, 7);
2013 ROUND(F, c, d, a, b, in[6] + K1, 11);
2014 ROUND(F, b, c, d, a, in[7] + K1, 19);
2017 ROUND(G, a, b, c, d, in[1] + K2, 3);
2018 ROUND(G, d, a, b, c, in[3] + K2, 5);
2019 ROUND(G, c, d, a, b, in[5] + K2, 9);
2020 ROUND(G, b, c, d, a, in[7] + K2, 13);
2021 ROUND(G, a, b, c, d, in[0] + K2, 3);
2022 ROUND(G, d, a, b, c, in[2] + K2, 5);
2023 ROUND(G, c, d, a, b, in[4] + K2, 9);
2024 ROUND(G, b, c, d, a, in[6] + K2, 13);
2027 ROUND(H, a, b, c, d, in[3] + K3, 3);
2028 ROUND(H, d, a, b, c, in[7] + K3, 9);
2029 ROUND(H, c, d, a, b, in[2] + K3, 11);
2030 ROUND(H, b, c, d, a, in[6] + K3, 15);
2031 ROUND(H, a, b, c, d, in[1] + K3, 3);
2032 ROUND(H, d, a, b, c, in[5] + K3, 9);
2033 ROUND(H, c, d, a, b, in[0] + K3, 11);
2034 ROUND(H, b, c, d, a, in[4] + K3, 15);
2036 return buf[1] + b; /* "most hashed" word */
2037 /* Alternative: return sum of all words? */
2040 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2042 static __u32 twothirdsMD4Transform (__u32 const buf[4], __u32 const in[12])
2044 __u32 a = buf[0], b = buf[1], c = buf[2], d = buf[3];
2047 ROUND(F, a, b, c, d, in[ 0] + K1, 3);
2048 ROUND(F, d, a, b, c, in[ 1] + K1, 7);
2049 ROUND(F, c, d, a, b, in[ 2] + K1, 11);
2050 ROUND(F, b, c, d, a, in[ 3] + K1, 19);
2051 ROUND(F, a, b, c, d, in[ 4] + K1, 3);
2052 ROUND(F, d, a, b, c, in[ 5] + K1, 7);
2053 ROUND(F, c, d, a, b, in[ 6] + K1, 11);
2054 ROUND(F, b, c, d, a, in[ 7] + K1, 19);
2055 ROUND(F, a, b, c, d, in[ 8] + K1, 3);
2056 ROUND(F, d, a, b, c, in[ 9] + K1, 7);
2057 ROUND(F, c, d, a, b, in[10] + K1, 11);
2058 ROUND(F, b, c, d, a, in[11] + K1, 19);
2061 ROUND(G, a, b, c, d, in[ 1] + K2, 3);
2062 ROUND(G, d, a, b, c, in[ 3] + K2, 5);
2063 ROUND(G, c, d, a, b, in[ 5] + K2, 9);
2064 ROUND(G, b, c, d, a, in[ 7] + K2, 13);
2065 ROUND(G, a, b, c, d, in[ 9] + K2, 3);
2066 ROUND(G, d, a, b, c, in[11] + K2, 5);
2067 ROUND(G, c, d, a, b, in[ 0] + K2, 9);
2068 ROUND(G, b, c, d, a, in[ 2] + K2, 13);
2069 ROUND(G, a, b, c, d, in[ 4] + K2, 3);
2070 ROUND(G, d, a, b, c, in[ 6] + K2, 5);
2071 ROUND(G, c, d, a, b, in[ 8] + K2, 9);
2072 ROUND(G, b, c, d, a, in[10] + K2, 13);
2075 ROUND(H, a, b, c, d, in[ 3] + K3, 3);
2076 ROUND(H, d, a, b, c, in[ 7] + K3, 9);
2077 ROUND(H, c, d, a, b, in[11] + K3, 11);
2078 ROUND(H, b, c, d, a, in[ 2] + K3, 15);
2079 ROUND(H, a, b, c, d, in[ 6] + K3, 3);
2080 ROUND(H, d, a, b, c, in[10] + K3, 9);
2081 ROUND(H, c, d, a, b, in[ 1] + K3, 11);
2082 ROUND(H, b, c, d, a, in[ 5] + K3, 15);
2083 ROUND(H, a, b, c, d, in[ 9] + K3, 3);
2084 ROUND(H, d, a, b, c, in[ 0] + K3, 9);
2085 ROUND(H, c, d, a, b, in[ 4] + K3, 11);
2086 ROUND(H, b, c, d, a, in[ 8] + K3, 15);
2088 return buf[1] + b; /* "most hashed" word */
2089 /* Alternative: return sum of all words? */
2101 /* This should not be decreased so low that ISNs wrap too fast. */
2102 #define REKEY_INTERVAL (300 * HZ)
2104 * Bit layout of the tcp sequence numbers (before adding current time):
2105 * bit 24-31: increased after every key exchange
2106 * bit 0-23: hash(source,dest)
2108 * The implementation is similar to the algorithm described
2109 * in the Appendix of RFC 1185, except that
2110 * - it uses a 1 MHz clock instead of a 250 kHz clock
2111 * - it performs a rekey every 5 minutes, which is equivalent
2112 * to a (source,dest) tulple dependent forward jump of the
2113 * clock by 0..2^(HASH_BITS+1)
2115 * Thus the average ISN wraparound time is 68 minutes instead of
2118 * SMP cleanup and lock avoidance with poor man's RCU.
2119 * Manfred Spraul <manfred@colorfullife.com>
2122 #define COUNT_BITS 8
2123 #define COUNT_MASK ((1 << COUNT_BITS) - 1)
2124 #define HASH_BITS 24
2125 #define HASH_MASK ((1 << HASH_BITS) - 1)
2127 static struct keydata {
2128 __u32 count; /* already shifted to the final position */
2130 } ____cacheline_aligned ip_keydata[2];
2132 static unsigned int ip_cnt;
2134 static void rekey_seq_generator(void *private_);
2136 static DECLARE_WORK(rekey_work, rekey_seq_generator, NULL);
2140 * The ISN generation runs lockless - it's just a hash over random data.
2141 * State changes happen every 5 minutes when the random key is replaced.
2142 * Synchronization is performed by having two copies of the hash function
2143 * state and rekey_seq_generator always updates the inactive copy.
2144 * The copy is then activated by updating ip_cnt.
2145 * The implementation breaks down if someone blocks the thread
2146 * that processes SYN requests for more than 5 minutes. Should never
2147 * happen, and even if that happens only a not perfectly compliant
2148 * ISN is generated, nothing fatal.
2150 static void rekey_seq_generator(void *private_)
2152 struct keydata *keyptr = &ip_keydata[1 ^ (ip_cnt & 1)];
2154 get_random_bytes(keyptr->secret, sizeof(keyptr->secret));
2155 keyptr->count = (ip_cnt & COUNT_MASK) << HASH_BITS;
2158 schedule_delayed_work(&rekey_work, REKEY_INTERVAL);
2161 static inline struct keydata *get_keyptr(void)
2163 struct keydata *keyptr = &ip_keydata[ip_cnt & 1];
2170 static __init int seqgen_init(void)
2172 rekey_seq_generator(NULL);
2175 late_initcall(seqgen_init);
2177 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2178 __u32 secure_tcpv6_sequence_number(__u32 *saddr, __u32 *daddr,
2179 __u16 sport, __u16 dport)
2184 struct keydata *keyptr = get_keyptr();
2186 /* The procedure is the same as for IPv4, but addresses are longer.
2187 * Thus we must use twothirdsMD4Transform.
2190 memcpy(hash, saddr, 16);
2191 hash[4]=(sport << 16) + dport;
2192 memcpy(&hash[5],keyptr->secret,sizeof(__u32) * 7);
2194 seq = twothirdsMD4Transform(daddr, hash) & HASH_MASK;
2195 seq += keyptr->count;
2197 do_gettimeofday(&tv);
2198 seq += tv.tv_usec + tv.tv_sec * 1000000;
2202 EXPORT_SYMBOL(secure_tcpv6_sequence_number);
2205 __u32 secure_tcp_sequence_number(__u32 saddr, __u32 daddr,
2206 __u16 sport, __u16 dport)
2211 struct keydata *keyptr = get_keyptr();
2214 * Pick a unique starting offset for each TCP connection endpoints
2215 * (saddr, daddr, sport, dport).
2216 * Note that the words are placed into the starting vector, which is
2217 * then mixed with a partial MD4 over random data.
2221 hash[2]=(sport << 16) + dport;
2222 hash[3]=keyptr->secret[11];
2224 seq = halfMD4Transform(hash, keyptr->secret) & HASH_MASK;
2225 seq += keyptr->count;
2227 * As close as possible to RFC 793, which
2228 * suggests using a 250 kHz clock.
2229 * Further reading shows this assumes 2 Mb/s networks.
2230 * For 10 Mb/s Ethernet, a 1 MHz clock is appropriate.
2231 * That's funny, Linux has one built in! Use it!
2232 * (Networks are faster now - should this be increased?)
2234 do_gettimeofday(&tv);
2235 seq += tv.tv_usec + tv.tv_sec * 1000000;
2237 printk("init_seq(%lx, %lx, %d, %d) = %d\n",
2238 saddr, daddr, sport, dport, seq);
2243 EXPORT_SYMBOL(secure_tcp_sequence_number);
2245 /* The code below is shamelessly stolen from secure_tcp_sequence_number().
2246 * All blames to Andrey V. Savochkin <saw@msu.ru>.
2248 __u32 secure_ip_id(__u32 daddr)
2250 struct keydata *keyptr;
2253 keyptr = get_keyptr();
2256 * Pick a unique starting offset for each IP destination.
2257 * The dest ip address is placed in the starting vector,
2258 * which is then hashed with random data.
2261 hash[1] = keyptr->secret[9];
2262 hash[2] = keyptr->secret[10];
2263 hash[3] = keyptr->secret[11];
2265 return halfMD4Transform(hash, keyptr->secret);
2268 /* Generate secure starting point for ephemeral TCP port search */
2269 u32 secure_tcp_port_ephemeral(__u32 saddr, __u32 daddr, __u16 dport)
2271 struct keydata *keyptr = get_keyptr();
2275 * Pick a unique starting offset for each ephemeral port search
2276 * (saddr, daddr, dport) and 48bits of random data.
2280 hash[2] = dport ^ keyptr->secret[10];
2281 hash[3] = keyptr->secret[11];
2283 return halfMD4Transform(hash, keyptr->secret);
2286 #ifdef CONFIG_SYN_COOKIES
2288 * Secure SYN cookie computation. This is the algorithm worked out by
2289 * Dan Bernstein and Eric Schenk.
2291 * For linux I implement the 1 minute counter by looking at the jiffies clock.
2292 * The count is passed in as a parameter, so this code doesn't much care.
2295 #define COOKIEBITS 24 /* Upper bits store count */
2296 #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
2298 static int syncookie_init;
2299 static __u32 syncookie_secret[2][16-3+HASH_BUFFER_SIZE];
2301 __u32 secure_tcp_syn_cookie(__u32 saddr, __u32 daddr, __u16 sport,
2302 __u16 dport, __u32 sseq, __u32 count, __u32 data)
2304 __u32 tmp[16 + HASH_BUFFER_SIZE + HASH_EXTRA_SIZE];
2308 * Pick two random secrets the first time we need a cookie.
2310 if (syncookie_init == 0) {
2311 get_random_bytes(syncookie_secret, sizeof(syncookie_secret));
2316 * Compute the secure sequence number.
2317 * The output should be:
2318 * HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24)
2319 * + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24).
2320 * Where sseq is their sequence number and count increases every
2322 * As an extra hack, we add a small "data" value that encodes the
2323 * MSS into the second hash value.
2326 memcpy(tmp + 3, syncookie_secret[0], sizeof(syncookie_secret[0]));
2329 tmp[2]=(sport << 16) + dport;
2330 HASH_TRANSFORM(tmp+16, tmp);
2331 seq = tmp[17] + sseq + (count << COOKIEBITS);
2333 memcpy(tmp + 3, syncookie_secret[1], sizeof(syncookie_secret[1]));
2336 tmp[2]=(sport << 16) + dport;
2337 tmp[3] = count; /* minute counter */
2338 HASH_TRANSFORM(tmp + 16, tmp);
2340 /* Add in the second hash and the data */
2341 return seq + ((tmp[17] + data) & COOKIEMASK);
2345 * This retrieves the small "data" value from the syncookie.
2346 * If the syncookie is bad, the data returned will be out of
2347 * range. This must be checked by the caller.
2349 * The count value used to generate the cookie must be within
2350 * "maxdiff" if the current (passed-in) "count". The return value
2351 * is (__u32)-1 if this test fails.
2353 __u32 check_tcp_syn_cookie(__u32 cookie, __u32 saddr, __u32 daddr, __u16 sport,
2354 __u16 dport, __u32 sseq, __u32 count, __u32 maxdiff)
2356 __u32 tmp[16 + HASH_BUFFER_SIZE + HASH_EXTRA_SIZE];
2359 if (syncookie_init == 0)
2360 return (__u32)-1; /* Well, duh! */
2362 /* Strip away the layers from the cookie */
2363 memcpy(tmp + 3, syncookie_secret[0], sizeof(syncookie_secret[0]));
2366 tmp[2]=(sport << 16) + dport;
2367 HASH_TRANSFORM(tmp + 16, tmp);
2368 cookie -= tmp[17] + sseq;
2369 /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */
2371 diff = (count - (cookie >> COOKIEBITS)) & ((__u32)-1 >> COOKIEBITS);
2372 if (diff >= maxdiff)
2375 memcpy(tmp+3, syncookie_secret[1], sizeof(syncookie_secret[1]));
2378 tmp[2] = (sport << 16) + dport;
2379 tmp[3] = count - diff; /* minute counter */
2380 HASH_TRANSFORM(tmp + 16, tmp);
2382 return (cookie - tmp[17]) & COOKIEMASK; /* Leaving the data behind */
2385 #endif /* CONFIG_INET */