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
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 * 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>
244 #include <asm/processor.h>
245 #include <asm/uaccess.h>
250 * Configuration information
252 #define INPUT_POOL_WORDS 128
253 #define OUTPUT_POOL_WORDS 32
254 #define SEC_XFER_SIZE 512
257 * The minimum number of bits of entropy before we wake up a read on
258 * /dev/random. Should be enough to do a significant reseed.
260 static int random_read_wakeup_thresh
= 64;
263 * If the entropy count falls under this number of bits, then we
264 * should wake up processes which are selecting or polling on write
265 * access to /dev/random.
267 static int random_write_wakeup_thresh
= 128;
270 * When the input pool goes over trickle_thresh, start dropping most
271 * samples to avoid wasting CPU time and reduce lock contention.
274 static int trickle_thresh __read_mostly
= INPUT_POOL_WORDS
* 28;
276 static DEFINE_PER_CPU(int, trickle_count
);
279 * A pool of size .poolwords is stirred with a primitive polynomial
280 * of degree .poolwords over GF(2). The taps for various sizes are
281 * defined below. They are chosen to be evenly spaced (minimum RMS
282 * distance from evenly spaced; the numbers in the comments are a
283 * scaled squared error sum) except for the last tap, which is 1 to
284 * get the twisting happening as fast as possible.
286 static struct poolinfo
{
288 int tap1
, tap2
, tap3
, tap4
, tap5
;
289 } poolinfo_table
[] = {
290 /* x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 -- 105 */
291 { 128, 103, 76, 51, 25, 1 },
292 /* x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 -- 15 */
293 { 32, 26, 20, 14, 7, 1 },
295 /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1 -- 115 */
296 { 2048, 1638, 1231, 819, 411, 1 },
298 /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
299 { 1024, 817, 615, 412, 204, 1 },
301 /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
302 { 1024, 819, 616, 410, 207, 2 },
304 /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
305 { 512, 411, 308, 208, 104, 1 },
307 /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
308 { 512, 409, 307, 206, 102, 2 },
309 /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
310 { 512, 409, 309, 205, 103, 2 },
312 /* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
313 { 256, 205, 155, 101, 52, 1 },
315 /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
316 { 128, 103, 78, 51, 27, 2 },
318 /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
319 { 64, 52, 39, 26, 14, 1 },
323 #define POOLBITS poolwords*32
324 #define POOLBYTES poolwords*4
327 * For the purposes of better mixing, we use the CRC-32 polynomial as
328 * well to make a twisted Generalized Feedback Shift Reigster
330 * (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR generators. ACM
331 * Transactions on Modeling and Computer Simulation 2(3):179-194.
332 * Also see M. Matsumoto & Y. Kurita, 1994. Twisted GFSR generators
333 * II. ACM Transactions on Mdeling and Computer Simulation 4:254-266)
335 * Thanks to Colin Plumb for suggesting this.
337 * We have not analyzed the resultant polynomial to prove it primitive;
338 * in fact it almost certainly isn't. Nonetheless, the irreducible factors
339 * of a random large-degree polynomial over GF(2) are more than large enough
340 * that periodicity is not a concern.
342 * The input hash is much less sensitive than the output hash. All
343 * that we want of it is that it be a good non-cryptographic hash;
344 * i.e. it not produce collisions when fed "random" data of the sort
345 * we expect to see. As long as the pool state differs for different
346 * inputs, we have preserved the input entropy and done a good job.
347 * The fact that an intelligent attacker can construct inputs that
348 * will produce controlled alterations to the pool's state is not
349 * important because we don't consider such inputs to contribute any
350 * randomness. The only property we need with respect to them is that
351 * the attacker can't increase his/her knowledge of the pool's state.
352 * Since all additions are reversible (knowing the final state and the
353 * input, you can reconstruct the initial state), if an attacker has
354 * any uncertainty about the initial state, he/she can only shuffle
355 * that uncertainty about, but never cause any collisions (which would
356 * decrease the uncertainty).
358 * The chosen system lets the state of the pool be (essentially) the input
359 * modulo the generator polymnomial. Now, for random primitive polynomials,
360 * this is a universal class of hash functions, meaning that the chance
361 * of a collision is limited by the attacker's knowledge of the generator
362 * polynomail, so if it is chosen at random, an attacker can never force
363 * a collision. Here, we use a fixed polynomial, but we *can* assume that
364 * ###--> it is unknown to the processes generating the input entropy. <-###
365 * Because of this important property, this is a good, collision-resistant
366 * hash; hash collisions will occur no more often than chance.
370 * Static global variables
372 static DECLARE_WAIT_QUEUE_HEAD(random_read_wait
);
373 static DECLARE_WAIT_QUEUE_HEAD(random_write_wait
);
374 static struct fasync_struct
*fasync
;
378 module_param(debug
, bool, 0644);
379 #define DEBUG_ENT(fmt, arg...) do { \
381 printk(KERN_DEBUG "random %04d %04d %04d: " \
383 input_pool.entropy_count,\
384 blocking_pool.entropy_count,\
385 nonblocking_pool.entropy_count,\
388 #define DEBUG_ENT(fmt, arg...) do {} while (0)
391 /**********************************************************************
393 * OS independent entropy store. Here are the functions which handle
394 * storing entropy in an entropy pool.
396 **********************************************************************/
398 struct entropy_store
;
399 struct entropy_store
{
400 /* read-only data: */
401 struct poolinfo
*poolinfo
;
405 struct entropy_store
*pull
;
407 /* read-write data: */
414 static __u32 input_pool_data
[INPUT_POOL_WORDS
];
415 static __u32 blocking_pool_data
[OUTPUT_POOL_WORDS
];
416 static __u32 nonblocking_pool_data
[OUTPUT_POOL_WORDS
];
418 static struct entropy_store input_pool
= {
419 .poolinfo
= &poolinfo_table
[0],
422 .lock
= __SPIN_LOCK_UNLOCKED(&input_pool
.lock
),
423 .pool
= input_pool_data
426 static struct entropy_store blocking_pool
= {
427 .poolinfo
= &poolinfo_table
[1],
431 .lock
= __SPIN_LOCK_UNLOCKED(&blocking_pool
.lock
),
432 .pool
= blocking_pool_data
435 static struct entropy_store nonblocking_pool
= {
436 .poolinfo
= &poolinfo_table
[1],
437 .name
= "nonblocking",
439 .lock
= __SPIN_LOCK_UNLOCKED(&nonblocking_pool
.lock
),
440 .pool
= nonblocking_pool_data
444 * This function adds bytes into the entropy "pool". It does not
445 * update the entropy estimate. The caller should call
446 * credit_entropy_bits if this is appropriate.
448 * The pool is stirred with a primitive polynomial of the appropriate
449 * degree, and then twisted. We twist by three bits at a time because
450 * it's cheap to do so and helps slightly in the expected case where
451 * the entropy is concentrated in the low-order bits.
453 static void mix_pool_bytes_extract(struct entropy_store
*r
, const void *in
,
454 int nbytes
, __u8 out
[64])
456 static __u32
const twist_table
[8] = {
457 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
458 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
459 unsigned long i
, j
, tap1
, tap2
, tap3
, tap4
, tap5
;
461 int wordmask
= r
->poolinfo
->poolwords
- 1;
462 const char *bytes
= in
;
466 /* Taps are constant, so we can load them without holding r->lock. */
467 tap1
= r
->poolinfo
->tap1
;
468 tap2
= r
->poolinfo
->tap2
;
469 tap3
= r
->poolinfo
->tap3
;
470 tap4
= r
->poolinfo
->tap4
;
471 tap5
= r
->poolinfo
->tap5
;
473 spin_lock_irqsave(&r
->lock
, flags
);
474 input_rotate
= r
->input_rotate
;
477 /* mix one byte at a time to simplify size handling and churn faster */
479 w
= rol32(*bytes
++, input_rotate
& 31);
480 i
= (i
- 1) & wordmask
;
482 /* XOR in the various taps */
484 w
^= r
->pool
[(i
+ tap1
) & wordmask
];
485 w
^= r
->pool
[(i
+ tap2
) & wordmask
];
486 w
^= r
->pool
[(i
+ tap3
) & wordmask
];
487 w
^= r
->pool
[(i
+ tap4
) & wordmask
];
488 w
^= r
->pool
[(i
+ tap5
) & wordmask
];
490 /* Mix the result back in with a twist */
491 r
->pool
[i
] = (w
>> 3) ^ twist_table
[w
& 7];
494 * Normally, we add 7 bits of rotation to the pool.
495 * At the beginning of the pool, add an extra 7 bits
496 * rotation, so that successive passes spread the
497 * input bits across the pool evenly.
499 input_rotate
+= i
? 7 : 14;
502 r
->input_rotate
= input_rotate
;
506 for (j
= 0; j
< 16; j
++)
507 ((__u32
*)out
)[j
] = r
->pool
[(i
- j
) & wordmask
];
509 spin_unlock_irqrestore(&r
->lock
, flags
);
512 static void mix_pool_bytes(struct entropy_store
*r
, const void *in
, int bytes
)
514 mix_pool_bytes_extract(r
, in
, bytes
, NULL
);
518 * Credit (or debit) the entropy store with n bits of entropy
520 static void credit_entropy_bits(struct entropy_store
*r
, int nbits
)
528 spin_lock_irqsave(&r
->lock
, flags
);
530 DEBUG_ENT("added %d entropy credits to %s\n", nbits
, r
->name
);
531 entropy_count
= r
->entropy_count
;
532 entropy_count
+= nbits
;
533 if (entropy_count
< 0) {
534 DEBUG_ENT("negative entropy/overflow\n");
536 } else if (entropy_count
> r
->poolinfo
->POOLBITS
)
537 entropy_count
= r
->poolinfo
->POOLBITS
;
538 r
->entropy_count
= entropy_count
;
540 /* should we wake readers? */
541 if (r
== &input_pool
&& entropy_count
>= random_read_wakeup_thresh
) {
542 wake_up_interruptible(&random_read_wait
);
543 kill_fasync(&fasync
, SIGIO
, POLL_IN
);
545 spin_unlock_irqrestore(&r
->lock
, flags
);
548 /*********************************************************************
550 * Entropy input management
552 *********************************************************************/
554 /* There is one of these per entropy source */
555 struct timer_rand_state
{
557 long last_delta
, last_delta2
;
558 unsigned dont_count_entropy
:1;
561 #ifndef CONFIG_SPARSE_IRQ
563 static struct timer_rand_state
*irq_timer_state
[NR_IRQS
];
565 static struct timer_rand_state
*get_timer_rand_state(unsigned int irq
)
567 return irq_timer_state
[irq
];
570 static void set_timer_rand_state(unsigned int irq
,
571 struct timer_rand_state
*state
)
573 irq_timer_state
[irq
] = state
;
578 static struct timer_rand_state
*get_timer_rand_state(unsigned int irq
)
580 struct irq_desc
*desc
;
582 desc
= irq_to_desc(irq
);
584 return desc
->timer_rand_state
;
587 static void set_timer_rand_state(unsigned int irq
,
588 struct timer_rand_state
*state
)
590 struct irq_desc
*desc
;
592 desc
= irq_to_desc(irq
);
594 desc
->timer_rand_state
= state
;
598 static struct timer_rand_state input_timer_state
;
601 * This function adds entropy to the entropy "pool" by using timing
602 * delays. It uses the timer_rand_state structure to make an estimate
603 * of how many bits of entropy this call has added to the pool.
605 * The number "num" is also added to the pool - it should somehow describe
606 * the type of event which just happened. This is currently 0-255 for
607 * keyboard scan codes, and 256 upwards for interrupts.
610 static void add_timer_randomness(struct timer_rand_state
*state
, unsigned num
)
617 long delta
, delta2
, delta3
;
620 /* if over the trickle threshold, use only 1 in 4096 samples */
621 if (input_pool
.entropy_count
> trickle_thresh
&&
622 (__get_cpu_var(trickle_count
)++ & 0xfff))
625 sample
.jiffies
= jiffies
;
626 sample
.cycles
= get_cycles();
628 mix_pool_bytes(&input_pool
, &sample
, sizeof(sample
));
631 * Calculate number of bits of randomness we probably added.
632 * We take into account the first, second and third-order deltas
633 * in order to make our estimate.
636 if (!state
->dont_count_entropy
) {
637 delta
= sample
.jiffies
- state
->last_time
;
638 state
->last_time
= sample
.jiffies
;
640 delta2
= delta
- state
->last_delta
;
641 state
->last_delta
= delta
;
643 delta3
= delta2
- state
->last_delta2
;
644 state
->last_delta2
= delta2
;
658 * delta is now minimum absolute delta.
659 * Round down by 1 bit on general principles,
660 * and limit entropy entimate to 12 bits.
662 credit_entropy_bits(&input_pool
,
663 min_t(int, fls(delta
>>1), 11));
669 void add_input_randomness(unsigned int type
, unsigned int code
,
672 static unsigned char last_value
;
674 /* ignore autorepeat and the like */
675 if (value
== last_value
)
678 DEBUG_ENT("input event\n");
680 add_timer_randomness(&input_timer_state
,
681 (type
<< 4) ^ code
^ (code
>> 4) ^ value
);
683 EXPORT_SYMBOL_GPL(add_input_randomness
);
685 void add_interrupt_randomness(int irq
)
687 struct timer_rand_state
*state
;
689 state
= get_timer_rand_state(irq
);
694 DEBUG_ENT("irq event %d\n", irq
);
695 add_timer_randomness(state
, 0x100 + irq
);
699 void add_disk_randomness(struct gendisk
*disk
)
701 if (!disk
|| !disk
->random
)
703 /* first major is 1, so we get >= 0x200 here */
704 DEBUG_ENT("disk event %d:%d\n",
705 MAJOR(disk_devt(disk
)), MINOR(disk_devt(disk
)));
707 add_timer_randomness(disk
->random
, 0x100 + disk_devt(disk
));
711 #define EXTRACT_SIZE 10
713 /*********************************************************************
715 * Entropy extraction routines
717 *********************************************************************/
719 static ssize_t
extract_entropy(struct entropy_store
*r
, void *buf
,
720 size_t nbytes
, int min
, int rsvd
);
723 * This utility inline function is responsible for transfering entropy
724 * from the primary pool to the secondary extraction pool. We make
725 * sure we pull enough for a 'catastrophic reseed'.
727 static void xfer_secondary_pool(struct entropy_store
*r
, size_t nbytes
)
729 __u32 tmp
[OUTPUT_POOL_WORDS
];
731 if (r
->pull
&& r
->entropy_count
< nbytes
* 8 &&
732 r
->entropy_count
< r
->poolinfo
->POOLBITS
) {
733 /* If we're limited, always leave two wakeup worth's BITS */
734 int rsvd
= r
->limit
? 0 : random_read_wakeup_thresh
/4;
737 /* pull at least as many as BYTES as wakeup BITS */
738 bytes
= max_t(int, bytes
, random_read_wakeup_thresh
/ 8);
739 /* but never more than the buffer size */
740 bytes
= min_t(int, bytes
, sizeof(tmp
));
742 DEBUG_ENT("going to reseed %s with %d bits "
743 "(%d of %d requested)\n",
744 r
->name
, bytes
* 8, nbytes
* 8, r
->entropy_count
);
746 bytes
= extract_entropy(r
->pull
, tmp
, bytes
,
747 random_read_wakeup_thresh
/ 8, rsvd
);
748 mix_pool_bytes(r
, tmp
, bytes
);
749 credit_entropy_bits(r
, bytes
*8);
754 * These functions extracts randomness from the "entropy pool", and
755 * returns it in a buffer.
757 * The min parameter specifies the minimum amount we can pull before
758 * failing to avoid races that defeat catastrophic reseeding while the
759 * reserved parameter indicates how much entropy we must leave in the
760 * pool after each pull to avoid starving other readers.
762 * Note: extract_entropy() assumes that .poolwords is a multiple of 16 words.
765 static size_t account(struct entropy_store
*r
, size_t nbytes
, int min
,
770 /* Hold lock while accounting */
771 spin_lock_irqsave(&r
->lock
, flags
);
773 BUG_ON(r
->entropy_count
> r
->poolinfo
->POOLBITS
);
774 DEBUG_ENT("trying to extract %d bits from %s\n",
775 nbytes
* 8, r
->name
);
777 /* Can we pull enough? */
778 if (r
->entropy_count
/ 8 < min
+ reserved
) {
781 /* If limited, never pull more than available */
782 if (r
->limit
&& nbytes
+ reserved
>= r
->entropy_count
/ 8)
783 nbytes
= r
->entropy_count
/8 - reserved
;
785 if (r
->entropy_count
/ 8 >= nbytes
+ reserved
)
786 r
->entropy_count
-= nbytes
*8;
788 r
->entropy_count
= reserved
;
790 if (r
->entropy_count
< random_write_wakeup_thresh
) {
791 wake_up_interruptible(&random_write_wait
);
792 kill_fasync(&fasync
, SIGIO
, POLL_OUT
);
796 DEBUG_ENT("debiting %d entropy credits from %s%s\n",
797 nbytes
* 8, r
->name
, r
->limit
? "" : " (unlimited)");
799 spin_unlock_irqrestore(&r
->lock
, flags
);
804 static void extract_buf(struct entropy_store
*r
, __u8
*out
)
807 __u32 hash
[5], workspace
[SHA_WORKSPACE_WORDS
];
810 /* Generate a hash across the pool, 16 words (512 bits) at a time */
812 for (i
= 0; i
< r
->poolinfo
->poolwords
; i
+= 16)
813 sha_transform(hash
, (__u8
*)(r
->pool
+ i
), workspace
);
816 * We mix the hash back into the pool to prevent backtracking
817 * attacks (where the attacker knows the state of the pool
818 * plus the current outputs, and attempts to find previous
819 * ouputs), unless the hash function can be inverted. By
820 * mixing at least a SHA1 worth of hash data back, we make
821 * brute-forcing the feedback as hard as brute-forcing the
824 mix_pool_bytes_extract(r
, hash
, sizeof(hash
), extract
);
827 * To avoid duplicates, we atomically extract a portion of the
828 * pool while mixing, and hash one final time.
830 sha_transform(hash
, extract
, workspace
);
831 memset(extract
, 0, sizeof(extract
));
832 memset(workspace
, 0, sizeof(workspace
));
835 * In case the hash function has some recognizable output
836 * pattern, we fold it in half. Thus, we always feed back
837 * twice as much data as we output.
841 hash
[2] ^= rol32(hash
[2], 16);
842 memcpy(out
, hash
, EXTRACT_SIZE
);
843 memset(hash
, 0, sizeof(hash
));
846 static ssize_t
extract_entropy(struct entropy_store
*r
, void *buf
,
847 size_t nbytes
, int min
, int reserved
)
850 __u8 tmp
[EXTRACT_SIZE
];
852 xfer_secondary_pool(r
, nbytes
);
853 nbytes
= account(r
, nbytes
, min
, reserved
);
857 i
= min_t(int, nbytes
, EXTRACT_SIZE
);
864 /* Wipe data just returned from memory */
865 memset(tmp
, 0, sizeof(tmp
));
870 static ssize_t
extract_entropy_user(struct entropy_store
*r
, void __user
*buf
,
874 __u8 tmp
[EXTRACT_SIZE
];
876 xfer_secondary_pool(r
, nbytes
);
877 nbytes
= account(r
, nbytes
, 0, 0);
880 if (need_resched()) {
881 if (signal_pending(current
)) {
890 i
= min_t(int, nbytes
, EXTRACT_SIZE
);
891 if (copy_to_user(buf
, tmp
, i
)) {
901 /* Wipe data just returned from memory */
902 memset(tmp
, 0, sizeof(tmp
));
908 * This function is the exported kernel interface. It returns some
909 * number of good random numbers, suitable for seeding TCP sequence
912 void get_random_bytes(void *buf
, int nbytes
)
914 extract_entropy(&nonblocking_pool
, buf
, nbytes
, 0, 0);
916 EXPORT_SYMBOL(get_random_bytes
);
919 * init_std_data - initialize pool with system data
921 * @r: pool to initialize
923 * This function clears the pool's entropy count and mixes some system
924 * data into the pool to prepare it for use. The pool is not cleared
925 * as that can only decrease the entropy in the pool.
927 static void init_std_data(struct entropy_store
*r
)
932 spin_lock_irqsave(&r
->lock
, flags
);
933 r
->entropy_count
= 0;
934 spin_unlock_irqrestore(&r
->lock
, flags
);
936 now
= ktime_get_real();
937 mix_pool_bytes(r
, &now
, sizeof(now
));
938 mix_pool_bytes(r
, utsname(), sizeof(*(utsname())));
941 static int rand_initialize(void)
943 init_std_data(&input_pool
);
944 init_std_data(&blocking_pool
);
945 init_std_data(&nonblocking_pool
);
948 module_init(rand_initialize
);
950 void rand_initialize_irq(int irq
)
952 struct timer_rand_state
*state
;
954 state
= get_timer_rand_state(irq
);
960 * If kzalloc returns null, we just won't use that entropy
963 state
= kzalloc(sizeof(struct timer_rand_state
), GFP_KERNEL
);
965 set_timer_rand_state(irq
, state
);
969 void rand_initialize_disk(struct gendisk
*disk
)
971 struct timer_rand_state
*state
;
974 * If kzalloc returns null, we just won't use that entropy
977 state
= kzalloc(sizeof(struct timer_rand_state
), GFP_KERNEL
);
979 disk
->random
= state
;
984 random_read(struct file
*file
, char __user
*buf
, size_t nbytes
, loff_t
*ppos
)
986 ssize_t n
, retval
= 0, count
= 0;
993 if (n
> SEC_XFER_SIZE
)
996 DEBUG_ENT("reading %d bits\n", n
*8);
998 n
= extract_entropy_user(&blocking_pool
, buf
, n
);
1000 DEBUG_ENT("read got %d bits (%d still needed)\n",
1004 if (file
->f_flags
& O_NONBLOCK
) {
1009 DEBUG_ENT("sleeping?\n");
1011 wait_event_interruptible(random_read_wait
,
1012 input_pool
.entropy_count
>=
1013 random_read_wakeup_thresh
);
1015 DEBUG_ENT("awake\n");
1017 if (signal_pending(current
)) {
1018 retval
= -ERESTARTSYS
;
1032 break; /* This break makes the device work */
1033 /* like a named pipe */
1037 * If we gave the user some bytes, update the access time.
1040 file_accessed(file
);
1042 return (count
? count
: retval
);
1046 urandom_read(struct file
*file
, char __user
*buf
, size_t nbytes
, loff_t
*ppos
)
1048 return extract_entropy_user(&nonblocking_pool
, buf
, nbytes
);
1052 random_poll(struct file
*file
, poll_table
* wait
)
1056 poll_wait(file
, &random_read_wait
, wait
);
1057 poll_wait(file
, &random_write_wait
, wait
);
1059 if (input_pool
.entropy_count
>= random_read_wakeup_thresh
)
1060 mask
|= POLLIN
| POLLRDNORM
;
1061 if (input_pool
.entropy_count
< random_write_wakeup_thresh
)
1062 mask
|= POLLOUT
| POLLWRNORM
;
1067 write_pool(struct entropy_store
*r
, const char __user
*buffer
, size_t count
)
1071 const char __user
*p
= buffer
;
1074 bytes
= min(count
, sizeof(buf
));
1075 if (copy_from_user(&buf
, p
, bytes
))
1081 mix_pool_bytes(r
, buf
, bytes
);
1088 static ssize_t
random_write(struct file
*file
, const char __user
*buffer
,
1089 size_t count
, loff_t
*ppos
)
1092 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1094 ret
= write_pool(&blocking_pool
, buffer
, count
);
1097 ret
= write_pool(&nonblocking_pool
, buffer
, count
);
1101 inode
->i_mtime
= current_fs_time(inode
->i_sb
);
1102 mark_inode_dirty(inode
);
1103 return (ssize_t
)count
;
1106 static long random_ioctl(struct file
*f
, unsigned int cmd
, unsigned long arg
)
1108 int size
, ent_count
;
1109 int __user
*p
= (int __user
*)arg
;
1114 /* inherently racy, no point locking */
1115 if (put_user(input_pool
.entropy_count
, p
))
1118 case RNDADDTOENTCNT
:
1119 if (!capable(CAP_SYS_ADMIN
))
1121 if (get_user(ent_count
, p
))
1123 credit_entropy_bits(&input_pool
, ent_count
);
1126 if (!capable(CAP_SYS_ADMIN
))
1128 if (get_user(ent_count
, p
++))
1132 if (get_user(size
, p
++))
1134 retval
= write_pool(&input_pool
, (const char __user
*)p
,
1138 credit_entropy_bits(&input_pool
, ent_count
);
1142 /* Clear the entropy pool counters. */
1143 if (!capable(CAP_SYS_ADMIN
))
1152 static int random_fasync(int fd
, struct file
*filp
, int on
)
1154 return fasync_helper(fd
, filp
, on
, &fasync
);
1157 const struct file_operations random_fops
= {
1158 .read
= random_read
,
1159 .write
= random_write
,
1160 .poll
= random_poll
,
1161 .unlocked_ioctl
= random_ioctl
,
1162 .fasync
= random_fasync
,
1165 const struct file_operations urandom_fops
= {
1166 .read
= urandom_read
,
1167 .write
= random_write
,
1168 .unlocked_ioctl
= random_ioctl
,
1169 .fasync
= random_fasync
,
1172 /***************************************************************
1173 * Random UUID interface
1175 * Used here for a Boot ID, but can be useful for other kernel
1177 ***************************************************************/
1180 * Generate random UUID
1182 void generate_random_uuid(unsigned char uuid_out
[16])
1184 get_random_bytes(uuid_out
, 16);
1185 /* Set UUID version to 4 --- truely random generation */
1186 uuid_out
[6] = (uuid_out
[6] & 0x0F) | 0x40;
1187 /* Set the UUID variant to DCE */
1188 uuid_out
[8] = (uuid_out
[8] & 0x3F) | 0x80;
1190 EXPORT_SYMBOL(generate_random_uuid
);
1192 /********************************************************************
1196 ********************************************************************/
1198 #ifdef CONFIG_SYSCTL
1200 #include <linux/sysctl.h>
1202 static int min_read_thresh
= 8, min_write_thresh
;
1203 static int max_read_thresh
= INPUT_POOL_WORDS
* 32;
1204 static int max_write_thresh
= INPUT_POOL_WORDS
* 32;
1205 static char sysctl_bootid
[16];
1208 * These functions is used to return both the bootid UUID, and random
1209 * UUID. The difference is in whether table->data is NULL; if it is,
1210 * then a new UUID is generated and returned to the user.
1212 * If the user accesses this via the proc interface, it will be returned
1213 * as an ASCII string in the standard UUID format. If accesses via the
1214 * sysctl system call, it is returned as 16 bytes of binary data.
1216 static int proc_do_uuid(ctl_table
*table
, int write
, struct file
*filp
,
1217 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
1219 ctl_table fake_table
;
1220 unsigned char buf
[64], tmp_uuid
[16], *uuid
;
1228 generate_random_uuid(uuid
);
1230 sprintf(buf
, "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-"
1231 "%02x%02x%02x%02x%02x%02x",
1232 uuid
[0], uuid
[1], uuid
[2], uuid
[3],
1233 uuid
[4], uuid
[5], uuid
[6], uuid
[7],
1234 uuid
[8], uuid
[9], uuid
[10], uuid
[11],
1235 uuid
[12], uuid
[13], uuid
[14], uuid
[15]);
1236 fake_table
.data
= buf
;
1237 fake_table
.maxlen
= sizeof(buf
);
1239 return proc_dostring(&fake_table
, write
, filp
, buffer
, lenp
, ppos
);
1242 static int uuid_strategy(ctl_table
*table
,
1243 void __user
*oldval
, size_t __user
*oldlenp
,
1244 void __user
*newval
, size_t newlen
)
1246 unsigned char tmp_uuid
[16], *uuid
;
1249 if (!oldval
|| !oldlenp
)
1258 generate_random_uuid(uuid
);
1260 if (get_user(len
, oldlenp
))
1265 if (copy_to_user(oldval
, uuid
, len
) ||
1266 put_user(len
, oldlenp
))
1272 static int sysctl_poolsize
= INPUT_POOL_WORDS
* 32;
1273 ctl_table random_table
[] = {
1275 .ctl_name
= RANDOM_POOLSIZE
,
1276 .procname
= "poolsize",
1277 .data
= &sysctl_poolsize
,
1278 .maxlen
= sizeof(int),
1280 .proc_handler
= &proc_dointvec
,
1283 .ctl_name
= RANDOM_ENTROPY_COUNT
,
1284 .procname
= "entropy_avail",
1285 .maxlen
= sizeof(int),
1287 .proc_handler
= &proc_dointvec
,
1288 .data
= &input_pool
.entropy_count
,
1291 .ctl_name
= RANDOM_READ_THRESH
,
1292 .procname
= "read_wakeup_threshold",
1293 .data
= &random_read_wakeup_thresh
,
1294 .maxlen
= sizeof(int),
1296 .proc_handler
= &proc_dointvec_minmax
,
1297 .strategy
= &sysctl_intvec
,
1298 .extra1
= &min_read_thresh
,
1299 .extra2
= &max_read_thresh
,
1302 .ctl_name
= RANDOM_WRITE_THRESH
,
1303 .procname
= "write_wakeup_threshold",
1304 .data
= &random_write_wakeup_thresh
,
1305 .maxlen
= sizeof(int),
1307 .proc_handler
= &proc_dointvec_minmax
,
1308 .strategy
= &sysctl_intvec
,
1309 .extra1
= &min_write_thresh
,
1310 .extra2
= &max_write_thresh
,
1313 .ctl_name
= RANDOM_BOOT_ID
,
1314 .procname
= "boot_id",
1315 .data
= &sysctl_bootid
,
1318 .proc_handler
= &proc_do_uuid
,
1319 .strategy
= &uuid_strategy
,
1322 .ctl_name
= RANDOM_UUID
,
1326 .proc_handler
= &proc_do_uuid
,
1327 .strategy
= &uuid_strategy
,
1331 #endif /* CONFIG_SYSCTL */
1333 /********************************************************************
1335 * Random funtions for networking
1337 ********************************************************************/
1340 * TCP initial sequence number picking. This uses the random number
1341 * generator to pick an initial secret value. This value is hashed
1342 * along with the TCP endpoint information to provide a unique
1343 * starting point for each pair of TCP endpoints. This defeats
1344 * attacks which rely on guessing the initial TCP sequence number.
1345 * This algorithm was suggested by Steve Bellovin.
1347 * Using a very strong hash was taking an appreciable amount of the total
1348 * TCP connection establishment time, so this is a weaker hash,
1349 * compensated for by changing the secret periodically.
1352 /* F, G and H are basic MD4 functions: selection, majority, parity */
1353 #define F(x, y, z) ((z) ^ ((x) & ((y) ^ (z))))
1354 #define G(x, y, z) (((x) & (y)) + (((x) ^ (y)) & (z)))
1355 #define H(x, y, z) ((x) ^ (y) ^ (z))
1358 * The generic round function. The application is so specific that
1359 * we don't bother protecting all the arguments with parens, as is generally
1360 * good macro practice, in favor of extra legibility.
1361 * Rotation is separate from addition to prevent recomputation
1363 #define ROUND(f, a, b, c, d, x, s) \
1364 (a += f(b, c, d) + x, a = (a << s) | (a >> (32 - s)))
1366 #define K2 013240474631UL
1367 #define K3 015666365641UL
1369 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1371 static __u32
twothirdsMD4Transform(__u32
const buf
[4], __u32
const in
[12])
1373 __u32 a
= buf
[0], b
= buf
[1], c
= buf
[2], d
= buf
[3];
1376 ROUND(F
, a
, b
, c
, d
, in
[ 0] + K1
, 3);
1377 ROUND(F
, d
, a
, b
, c
, in
[ 1] + K1
, 7);
1378 ROUND(F
, c
, d
, a
, b
, in
[ 2] + K1
, 11);
1379 ROUND(F
, b
, c
, d
, a
, in
[ 3] + K1
, 19);
1380 ROUND(F
, a
, b
, c
, d
, in
[ 4] + K1
, 3);
1381 ROUND(F
, d
, a
, b
, c
, in
[ 5] + K1
, 7);
1382 ROUND(F
, c
, d
, a
, b
, in
[ 6] + K1
, 11);
1383 ROUND(F
, b
, c
, d
, a
, in
[ 7] + K1
, 19);
1384 ROUND(F
, a
, b
, c
, d
, in
[ 8] + K1
, 3);
1385 ROUND(F
, d
, a
, b
, c
, in
[ 9] + K1
, 7);
1386 ROUND(F
, c
, d
, a
, b
, in
[10] + K1
, 11);
1387 ROUND(F
, b
, c
, d
, a
, in
[11] + K1
, 19);
1390 ROUND(G
, a
, b
, c
, d
, in
[ 1] + K2
, 3);
1391 ROUND(G
, d
, a
, b
, c
, in
[ 3] + K2
, 5);
1392 ROUND(G
, c
, d
, a
, b
, in
[ 5] + K2
, 9);
1393 ROUND(G
, b
, c
, d
, a
, in
[ 7] + K2
, 13);
1394 ROUND(G
, a
, b
, c
, d
, in
[ 9] + K2
, 3);
1395 ROUND(G
, d
, a
, b
, c
, in
[11] + K2
, 5);
1396 ROUND(G
, c
, d
, a
, b
, in
[ 0] + K2
, 9);
1397 ROUND(G
, b
, c
, d
, a
, in
[ 2] + K2
, 13);
1398 ROUND(G
, a
, b
, c
, d
, in
[ 4] + K2
, 3);
1399 ROUND(G
, d
, a
, b
, c
, in
[ 6] + K2
, 5);
1400 ROUND(G
, c
, d
, a
, b
, in
[ 8] + K2
, 9);
1401 ROUND(G
, b
, c
, d
, a
, in
[10] + K2
, 13);
1404 ROUND(H
, a
, b
, c
, d
, in
[ 3] + K3
, 3);
1405 ROUND(H
, d
, a
, b
, c
, in
[ 7] + K3
, 9);
1406 ROUND(H
, c
, d
, a
, b
, in
[11] + K3
, 11);
1407 ROUND(H
, b
, c
, d
, a
, in
[ 2] + K3
, 15);
1408 ROUND(H
, a
, b
, c
, d
, in
[ 6] + K3
, 3);
1409 ROUND(H
, d
, a
, b
, c
, in
[10] + K3
, 9);
1410 ROUND(H
, c
, d
, a
, b
, in
[ 1] + K3
, 11);
1411 ROUND(H
, b
, c
, d
, a
, in
[ 5] + K3
, 15);
1412 ROUND(H
, a
, b
, c
, d
, in
[ 9] + K3
, 3);
1413 ROUND(H
, d
, a
, b
, c
, in
[ 0] + K3
, 9);
1414 ROUND(H
, c
, d
, a
, b
, in
[ 4] + K3
, 11);
1415 ROUND(H
, b
, c
, d
, a
, in
[ 8] + K3
, 15);
1417 return buf
[1] + b
; /* "most hashed" word */
1418 /* Alternative: return sum of all words? */
1430 /* This should not be decreased so low that ISNs wrap too fast. */
1431 #define REKEY_INTERVAL (300 * HZ)
1433 * Bit layout of the tcp sequence numbers (before adding current time):
1434 * bit 24-31: increased after every key exchange
1435 * bit 0-23: hash(source,dest)
1437 * The implementation is similar to the algorithm described
1438 * in the Appendix of RFC 1185, except that
1439 * - it uses a 1 MHz clock instead of a 250 kHz clock
1440 * - it performs a rekey every 5 minutes, which is equivalent
1441 * to a (source,dest) tulple dependent forward jump of the
1442 * clock by 0..2^(HASH_BITS+1)
1444 * Thus the average ISN wraparound time is 68 minutes instead of
1447 * SMP cleanup and lock avoidance with poor man's RCU.
1448 * Manfred Spraul <manfred@colorfullife.com>
1451 #define COUNT_BITS 8
1452 #define COUNT_MASK ((1 << COUNT_BITS) - 1)
1453 #define HASH_BITS 24
1454 #define HASH_MASK ((1 << HASH_BITS) - 1)
1456 static struct keydata
{
1457 __u32 count
; /* already shifted to the final position */
1459 } ____cacheline_aligned ip_keydata
[2];
1461 static unsigned int ip_cnt
;
1463 static void rekey_seq_generator(struct work_struct
*work
);
1465 static DECLARE_DELAYED_WORK(rekey_work
, rekey_seq_generator
);
1469 * The ISN generation runs lockless - it's just a hash over random data.
1470 * State changes happen every 5 minutes when the random key is replaced.
1471 * Synchronization is performed by having two copies of the hash function
1472 * state and rekey_seq_generator always updates the inactive copy.
1473 * The copy is then activated by updating ip_cnt.
1474 * The implementation breaks down if someone blocks the thread
1475 * that processes SYN requests for more than 5 minutes. Should never
1476 * happen, and even if that happens only a not perfectly compliant
1477 * ISN is generated, nothing fatal.
1479 static void rekey_seq_generator(struct work_struct
*work
)
1481 struct keydata
*keyptr
= &ip_keydata
[1 ^ (ip_cnt
& 1)];
1483 get_random_bytes(keyptr
->secret
, sizeof(keyptr
->secret
));
1484 keyptr
->count
= (ip_cnt
& COUNT_MASK
) << HASH_BITS
;
1487 schedule_delayed_work(&rekey_work
, REKEY_INTERVAL
);
1490 static inline struct keydata
*get_keyptr(void)
1492 struct keydata
*keyptr
= &ip_keydata
[ip_cnt
& 1];
1499 static __init
int seqgen_init(void)
1501 rekey_seq_generator(NULL
);
1504 late_initcall(seqgen_init
);
1506 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1507 __u32
secure_tcpv6_sequence_number(__be32
*saddr
, __be32
*daddr
,
1508 __be16 sport
, __be16 dport
)
1512 struct keydata
*keyptr
= get_keyptr();
1514 /* The procedure is the same as for IPv4, but addresses are longer.
1515 * Thus we must use twothirdsMD4Transform.
1518 memcpy(hash
, saddr
, 16);
1519 hash
[4] = ((__force u16
)sport
<< 16) + (__force u16
)dport
;
1520 memcpy(&hash
[5], keyptr
->secret
, sizeof(__u32
) * 7);
1522 seq
= twothirdsMD4Transform((const __u32
*)daddr
, hash
) & HASH_MASK
;
1523 seq
+= keyptr
->count
;
1525 seq
+= ktime_to_ns(ktime_get_real());
1529 EXPORT_SYMBOL(secure_tcpv6_sequence_number
);
1532 /* The code below is shamelessly stolen from secure_tcp_sequence_number().
1533 * All blames to Andrey V. Savochkin <saw@msu.ru>.
1535 __u32
secure_ip_id(__be32 daddr
)
1537 struct keydata
*keyptr
;
1540 keyptr
= get_keyptr();
1543 * Pick a unique starting offset for each IP destination.
1544 * The dest ip address is placed in the starting vector,
1545 * which is then hashed with random data.
1547 hash
[0] = (__force __u32
)daddr
;
1548 hash
[1] = keyptr
->secret
[9];
1549 hash
[2] = keyptr
->secret
[10];
1550 hash
[3] = keyptr
->secret
[11];
1552 return half_md4_transform(hash
, keyptr
->secret
);
1557 __u32
secure_tcp_sequence_number(__be32 saddr
, __be32 daddr
,
1558 __be16 sport
, __be16 dport
)
1562 struct keydata
*keyptr
= get_keyptr();
1565 * Pick a unique starting offset for each TCP connection endpoints
1566 * (saddr, daddr, sport, dport).
1567 * Note that the words are placed into the starting vector, which is
1568 * then mixed with a partial MD4 over random data.
1570 hash
[0] = (__force u32
)saddr
;
1571 hash
[1] = (__force u32
)daddr
;
1572 hash
[2] = ((__force u16
)sport
<< 16) + (__force u16
)dport
;
1573 hash
[3] = keyptr
->secret
[11];
1575 seq
= half_md4_transform(hash
, keyptr
->secret
) & HASH_MASK
;
1576 seq
+= keyptr
->count
;
1578 * As close as possible to RFC 793, which
1579 * suggests using a 250 kHz clock.
1580 * Further reading shows this assumes 2 Mb/s networks.
1581 * For 10 Mb/s Ethernet, a 1 MHz clock is appropriate.
1582 * For 10 Gb/s Ethernet, a 1 GHz clock should be ok, but
1583 * we also need to limit the resolution so that the u32 seq
1584 * overlaps less than one time per MSL (2 minutes).
1585 * Choosing a clock of 64 ns period is OK. (period of 274 s)
1587 seq
+= ktime_to_ns(ktime_get_real()) >> 6;
1592 /* Generate secure starting point for ephemeral IPV4 transport port search */
1593 u32
secure_ipv4_port_ephemeral(__be32 saddr
, __be32 daddr
, __be16 dport
)
1595 struct keydata
*keyptr
= get_keyptr();
1599 * Pick a unique starting offset for each ephemeral port search
1600 * (saddr, daddr, dport) and 48bits of random data.
1602 hash
[0] = (__force u32
)saddr
;
1603 hash
[1] = (__force u32
)daddr
;
1604 hash
[2] = (__force u32
)dport
^ keyptr
->secret
[10];
1605 hash
[3] = keyptr
->secret
[11];
1607 return half_md4_transform(hash
, keyptr
->secret
);
1609 EXPORT_SYMBOL_GPL(secure_ipv4_port_ephemeral
);
1611 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1612 u32
secure_ipv6_port_ephemeral(const __be32
*saddr
, const __be32
*daddr
,
1615 struct keydata
*keyptr
= get_keyptr();
1618 memcpy(hash
, saddr
, 16);
1619 hash
[4] = (__force u32
)dport
;
1620 memcpy(&hash
[5], keyptr
->secret
, sizeof(__u32
) * 7);
1622 return twothirdsMD4Transform((const __u32
*)daddr
, hash
);
1626 #if defined(CONFIG_IP_DCCP) || defined(CONFIG_IP_DCCP_MODULE)
1627 /* Similar to secure_tcp_sequence_number but generate a 48 bit value
1628 * bit's 32-47 increase every key exchange
1629 * 0-31 hash(source, dest)
1631 u64
secure_dccp_sequence_number(__be32 saddr
, __be32 daddr
,
1632 __be16 sport
, __be16 dport
)
1636 struct keydata
*keyptr
= get_keyptr();
1638 hash
[0] = (__force u32
)saddr
;
1639 hash
[1] = (__force u32
)daddr
;
1640 hash
[2] = ((__force u16
)sport
<< 16) + (__force u16
)dport
;
1641 hash
[3] = keyptr
->secret
[11];
1643 seq
= half_md4_transform(hash
, keyptr
->secret
);
1644 seq
|= ((u64
)keyptr
->count
) << (32 - HASH_BITS
);
1646 seq
+= ktime_to_ns(ktime_get_real());
1647 seq
&= (1ull << 48) - 1;
1651 EXPORT_SYMBOL(secure_dccp_sequence_number
);
1654 #endif /* CONFIG_INET */
1658 * Get a random word for internal kernel use only. Similar to urandom but
1659 * with the goal of minimal entropy pool depletion. As a result, the random
1660 * value is not cryptographically secure but for several uses the cost of
1661 * depleting entropy is too high
1663 unsigned int get_random_int(void)
1666 * Use IP's RNG. It suits our purpose perfectly: it re-keys itself
1667 * every second, from the entropy pool (and thus creates a limited
1668 * drain on it), and uses halfMD4Transform within the second. We
1669 * also mix it with jiffies and the PID:
1671 return secure_ip_id((__force __be32
)(current
->pid
+ jiffies
));
1675 * randomize_range() returns a start address such that
1677 * [...... <range> .....]
1680 * a <range> with size "len" starting at the return value is inside in the
1681 * area defined by [start, end], but is otherwise randomized.
1684 randomize_range(unsigned long start
, unsigned long end
, unsigned long len
)
1686 unsigned long range
= end
- len
- start
;
1688 if (end
<= start
+ len
)
1690 return PAGE_ALIGN(get_random_int() % range
+ start
);