Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * random.c -- A strong random number generator |
| 3 | * |
Matt Mackall | 9e95ce2 | 2005-04-16 15:25:56 -0700 | [diff] [blame] | 4 | * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5 | * |
| 6 | * Copyright Theodore Ts'o, 1994, 1995, 1996, 1997, 1998, 1999. All |
| 7 | * rights reserved. |
| 8 | * |
| 9 | * Redistribution and use in source and binary forms, with or without |
| 10 | * modification, are permitted provided that the following conditions |
| 11 | * are met: |
| 12 | * 1. Redistributions of source code must retain the above copyright |
| 13 | * notice, and the entire permission notice in its entirety, |
| 14 | * including the disclaimer of warranties. |
| 15 | * 2. Redistributions in binary form must reproduce the above copyright |
| 16 | * notice, this list of conditions and the following disclaimer in the |
| 17 | * documentation and/or other materials provided with the distribution. |
| 18 | * 3. The name of the author may not be used to endorse or promote |
| 19 | * products derived from this software without specific prior |
| 20 | * written permission. |
| 21 | * |
| 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.) |
| 27 | * |
| 28 | * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED |
| 29 | * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES |
| 30 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ALL OF |
| 31 | * WHICH ARE HEREBY DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE |
| 32 | * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
| 33 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT |
| 34 | * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR |
| 35 | * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF |
| 36 | * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 37 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE |
| 38 | * USE OF THIS SOFTWARE, EVEN IF NOT ADVISED OF THE POSSIBILITY OF SUCH |
| 39 | * DAMAGE. |
| 40 | */ |
| 41 | |
| 42 | /* |
| 43 | * (now, with legal B.S. out of the way.....) |
| 44 | * |
| 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. |
| 51 | * |
| 52 | * Theory of operation |
| 53 | * =================== |
| 54 | * |
| 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. |
| 65 | * |
| 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. |
| 77 | * |
| 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. |
| 89 | * |
| 90 | * If this estimate goes to zero, the routine can still generate |
| 91 | * random numbers; however, an attacker may (at least in theory) be |
| 92 | * able to infer the future output of the generator from prior |
| 93 | * outputs. This requires successful cryptanalysis of SHA, which is |
| 94 | * not believed to be feasible, but there is a remote possibility. |
| 95 | * Nonetheless, these numbers should be useful for the vast majority |
| 96 | * of purposes. |
| 97 | * |
| 98 | * Exported interfaces ---- output |
| 99 | * =============================== |
| 100 | * |
| 101 | * There are three exported interfaces; the first is one designed to |
| 102 | * be used from within the kernel: |
| 103 | * |
| 104 | * void get_random_bytes(void *buf, int nbytes); |
| 105 | * |
| 106 | * This interface will return the requested number of random bytes, |
| 107 | * and place it in the requested buffer. |
| 108 | * |
| 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. |
| 115 | * |
| 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. |
| 121 | * |
| 122 | * Exported interfaces ---- input |
| 123 | * ============================== |
| 124 | * |
| 125 | * The current exported interfaces for gathering environmental noise |
| 126 | * from the devices are: |
| 127 | * |
| 128 | * void add_input_randomness(unsigned int type, unsigned int code, |
| 129 | * unsigned int value); |
| 130 | * void add_interrupt_randomness(int irq); |
| 131 | * |
| 132 | * add_input_randomness() uses the input layer interrupt timing, as well as |
| 133 | * the event type information from the hardware. |
| 134 | * |
| 135 | * add_interrupt_randomness() uses the inter-interrupt timing as random |
| 136 | * inputs to the entropy pool. Note that not all interrupts are good |
| 137 | * sources of randomness! For example, the timer interrupts is not a |
| 138 | * good choice, because the periodicity of the interrupts is too |
| 139 | * regular, and hence predictable to an attacker. Disk interrupts are |
| 140 | * a better measure, since the timing of the disk interrupts are more |
| 141 | * unpredictable. |
| 142 | * |
| 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. |
| 146 | * |
| 147 | * Ensuring unpredictability at system startup |
| 148 | * ============================================ |
| 149 | * |
| 150 | * When any operating system starts up, it will go through a sequence |
| 151 | * of actions that are fairly predictable by an adversary, especially |
| 152 | * if the start-up does not involve interaction with a human operator. |
| 153 | * This reduces the actual number of bits of unpredictability in the |
| 154 | * entropy pool below the value in entropy_count. In order to |
| 155 | * counteract this effect, it helps to carry information in the |
| 156 | * entropy pool across shut-downs and start-ups. To do this, put the |
| 157 | * following lines an appropriate script which is run during the boot |
| 158 | * sequence: |
| 159 | * |
| 160 | * echo "Initializing random number generator..." |
| 161 | * random_seed=/var/run/random-seed |
| 162 | * # Carry a random seed from start-up to start-up |
| 163 | * # Load and then save the whole entropy pool |
| 164 | * if [ -f $random_seed ]; then |
| 165 | * cat $random_seed >/dev/urandom |
| 166 | * else |
| 167 | * touch $random_seed |
| 168 | * fi |
| 169 | * chmod 600 $random_seed |
| 170 | * dd if=/dev/urandom of=$random_seed count=1 bs=512 |
| 171 | * |
| 172 | * and the following lines in an appropriate script which is run as |
| 173 | * the system is shutdown: |
| 174 | * |
| 175 | * # Carry a random seed from shut-down to start-up |
| 176 | * # Save the whole entropy pool |
| 177 | * echo "Saving random seed..." |
| 178 | * random_seed=/var/run/random-seed |
| 179 | * touch $random_seed |
| 180 | * chmod 600 $random_seed |
| 181 | * dd if=/dev/urandom of=$random_seed count=1 bs=512 |
| 182 | * |
| 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. |
| 187 | * |
| 188 | * Effectively, these commands cause the contents of the entropy pool |
| 189 | * to be saved at shut-down time and reloaded into the entropy pool at |
| 190 | * start-up. (The 'dd' in the addition to the bootup script is to |
| 191 | * make sure that /etc/random-seed is different for every start-up, |
| 192 | * even if the system crashes without executing rc.0.) Even with |
| 193 | * complete knowledge of the start-up activities, predicting the state |
| 194 | * of the entropy pool requires knowledge of the previous history of |
| 195 | * the system. |
| 196 | * |
| 197 | * Configuring the /dev/random driver under Linux |
| 198 | * ============================================== |
| 199 | * |
| 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: |
| 204 | * |
| 205 | * mknod /dev/random c 1 8 |
| 206 | * mknod /dev/urandom c 1 9 |
| 207 | * |
| 208 | * Acknowledgements: |
| 209 | * ================= |
| 210 | * |
| 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. |
| 217 | * |
| 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. |
| 220 | * |
| 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. |
| 224 | */ |
| 225 | |
| 226 | #include <linux/utsname.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | #include <linux/module.h> |
| 228 | #include <linux/kernel.h> |
| 229 | #include <linux/major.h> |
| 230 | #include <linux/string.h> |
| 231 | #include <linux/fcntl.h> |
| 232 | #include <linux/slab.h> |
| 233 | #include <linux/random.h> |
| 234 | #include <linux/poll.h> |
| 235 | #include <linux/init.h> |
| 236 | #include <linux/fs.h> |
| 237 | #include <linux/genhd.h> |
| 238 | #include <linux/interrupt.h> |
| 239 | #include <linux/spinlock.h> |
| 240 | #include <linux/percpu.h> |
| 241 | #include <linux/cryptohash.h> |
| 242 | |
| 243 | #include <asm/processor.h> |
| 244 | #include <asm/uaccess.h> |
| 245 | #include <asm/irq.h> |
| 246 | #include <asm/io.h> |
| 247 | |
| 248 | /* |
| 249 | * Configuration information |
| 250 | */ |
| 251 | #define INPUT_POOL_WORDS 128 |
| 252 | #define OUTPUT_POOL_WORDS 32 |
| 253 | #define SEC_XFER_SIZE 512 |
| 254 | |
| 255 | /* |
| 256 | * The minimum number of bits of entropy before we wake up a read on |
| 257 | * /dev/random. Should be enough to do a significant reseed. |
| 258 | */ |
| 259 | static int random_read_wakeup_thresh = 64; |
| 260 | |
| 261 | /* |
| 262 | * If the entropy count falls under this number of bits, then we |
| 263 | * should wake up processes which are selecting or polling on write |
| 264 | * access to /dev/random. |
| 265 | */ |
| 266 | static int random_write_wakeup_thresh = 128; |
| 267 | |
| 268 | /* |
| 269 | * When the input pool goes over trickle_thresh, start dropping most |
| 270 | * samples to avoid wasting CPU time and reduce lock contention. |
| 271 | */ |
| 272 | |
Christoph Lameter | 6c03652 | 2005-07-07 17:56:59 -0700 | [diff] [blame] | 273 | static int trickle_thresh __read_mostly = INPUT_POOL_WORDS * 28; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 274 | |
| 275 | static DEFINE_PER_CPU(int, trickle_count) = 0; |
| 276 | |
| 277 | /* |
| 278 | * A pool of size .poolwords is stirred with a primitive polynomial |
| 279 | * of degree .poolwords over GF(2). The taps for various sizes are |
| 280 | * defined below. They are chosen to be evenly spaced (minimum RMS |
| 281 | * distance from evenly spaced; the numbers in the comments are a |
| 282 | * scaled squared error sum) except for the last tap, which is 1 to |
| 283 | * get the twisting happening as fast as possible. |
| 284 | */ |
| 285 | static struct poolinfo { |
| 286 | int poolwords; |
| 287 | int tap1, tap2, tap3, tap4, tap5; |
| 288 | } poolinfo_table[] = { |
| 289 | /* x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 -- 105 */ |
| 290 | { 128, 103, 76, 51, 25, 1 }, |
| 291 | /* x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 -- 15 */ |
| 292 | { 32, 26, 20, 14, 7, 1 }, |
| 293 | #if 0 |
| 294 | /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1 -- 115 */ |
| 295 | { 2048, 1638, 1231, 819, 411, 1 }, |
| 296 | |
| 297 | /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */ |
| 298 | { 1024, 817, 615, 412, 204, 1 }, |
| 299 | |
| 300 | /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */ |
| 301 | { 1024, 819, 616, 410, 207, 2 }, |
| 302 | |
| 303 | /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */ |
| 304 | { 512, 411, 308, 208, 104, 1 }, |
| 305 | |
| 306 | /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */ |
| 307 | { 512, 409, 307, 206, 102, 2 }, |
| 308 | /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */ |
| 309 | { 512, 409, 309, 205, 103, 2 }, |
| 310 | |
| 311 | /* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */ |
| 312 | { 256, 205, 155, 101, 52, 1 }, |
| 313 | |
| 314 | /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */ |
| 315 | { 128, 103, 78, 51, 27, 2 }, |
| 316 | |
| 317 | /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */ |
| 318 | { 64, 52, 39, 26, 14, 1 }, |
| 319 | #endif |
| 320 | }; |
| 321 | |
| 322 | #define POOLBITS poolwords*32 |
| 323 | #define POOLBYTES poolwords*4 |
| 324 | |
| 325 | /* |
| 326 | * For the purposes of better mixing, we use the CRC-32 polynomial as |
| 327 | * well to make a twisted Generalized Feedback Shift Reigster |
| 328 | * |
| 329 | * (See M. Matsumoto & Y. Kurita, 1992. Twisted GFSR generators. ACM |
| 330 | * Transactions on Modeling and Computer Simulation 2(3):179-194. |
| 331 | * Also see M. Matsumoto & Y. Kurita, 1994. Twisted GFSR generators |
| 332 | * II. ACM Transactions on Mdeling and Computer Simulation 4:254-266) |
| 333 | * |
| 334 | * Thanks to Colin Plumb for suggesting this. |
| 335 | * |
| 336 | * We have not analyzed the resultant polynomial to prove it primitive; |
| 337 | * in fact it almost certainly isn't. Nonetheless, the irreducible factors |
| 338 | * of a random large-degree polynomial over GF(2) are more than large enough |
| 339 | * that periodicity is not a concern. |
| 340 | * |
| 341 | * The input hash is much less sensitive than the output hash. All |
| 342 | * that we want of it is that it be a good non-cryptographic hash; |
| 343 | * i.e. it not produce collisions when fed "random" data of the sort |
| 344 | * we expect to see. As long as the pool state differs for different |
| 345 | * inputs, we have preserved the input entropy and done a good job. |
| 346 | * The fact that an intelligent attacker can construct inputs that |
| 347 | * will produce controlled alterations to the pool's state is not |
| 348 | * important because we don't consider such inputs to contribute any |
| 349 | * randomness. The only property we need with respect to them is that |
| 350 | * the attacker can't increase his/her knowledge of the pool's state. |
| 351 | * Since all additions are reversible (knowing the final state and the |
| 352 | * input, you can reconstruct the initial state), if an attacker has |
| 353 | * any uncertainty about the initial state, he/she can only shuffle |
| 354 | * that uncertainty about, but never cause any collisions (which would |
| 355 | * decrease the uncertainty). |
| 356 | * |
| 357 | * The chosen system lets the state of the pool be (essentially) the input |
| 358 | * modulo the generator polymnomial. Now, for random primitive polynomials, |
| 359 | * this is a universal class of hash functions, meaning that the chance |
| 360 | * of a collision is limited by the attacker's knowledge of the generator |
| 361 | * polynomail, so if it is chosen at random, an attacker can never force |
| 362 | * a collision. Here, we use a fixed polynomial, but we *can* assume that |
| 363 | * ###--> it is unknown to the processes generating the input entropy. <-### |
| 364 | * Because of this important property, this is a good, collision-resistant |
| 365 | * hash; hash collisions will occur no more often than chance. |
| 366 | */ |
| 367 | |
| 368 | /* |
| 369 | * Static global variables |
| 370 | */ |
| 371 | static DECLARE_WAIT_QUEUE_HEAD(random_read_wait); |
| 372 | static DECLARE_WAIT_QUEUE_HEAD(random_write_wait); |
| 373 | |
| 374 | #if 0 |
| 375 | static int debug = 0; |
| 376 | module_param(debug, bool, 0644); |
| 377 | #define DEBUG_ENT(fmt, arg...) do { if (debug) \ |
| 378 | printk(KERN_DEBUG "random %04d %04d %04d: " \ |
| 379 | fmt,\ |
| 380 | input_pool.entropy_count,\ |
| 381 | blocking_pool.entropy_count,\ |
| 382 | nonblocking_pool.entropy_count,\ |
| 383 | ## arg); } while (0) |
| 384 | #else |
| 385 | #define DEBUG_ENT(fmt, arg...) do {} while (0) |
| 386 | #endif |
| 387 | |
| 388 | /********************************************************************** |
| 389 | * |
| 390 | * OS independent entropy store. Here are the functions which handle |
| 391 | * storing entropy in an entropy pool. |
| 392 | * |
| 393 | **********************************************************************/ |
| 394 | |
| 395 | struct entropy_store; |
| 396 | struct entropy_store { |
| 397 | /* mostly-read data: */ |
| 398 | struct poolinfo *poolinfo; |
| 399 | __u32 *pool; |
| 400 | const char *name; |
| 401 | int limit; |
| 402 | struct entropy_store *pull; |
| 403 | |
| 404 | /* read-write data: */ |
| 405 | spinlock_t lock ____cacheline_aligned_in_smp; |
| 406 | unsigned add_ptr; |
| 407 | int entropy_count; |
| 408 | int input_rotate; |
| 409 | }; |
| 410 | |
| 411 | static __u32 input_pool_data[INPUT_POOL_WORDS]; |
| 412 | static __u32 blocking_pool_data[OUTPUT_POOL_WORDS]; |
| 413 | static __u32 nonblocking_pool_data[OUTPUT_POOL_WORDS]; |
| 414 | |
| 415 | static struct entropy_store input_pool = { |
| 416 | .poolinfo = &poolinfo_table[0], |
| 417 | .name = "input", |
| 418 | .limit = 1, |
Ingo Molnar | e4d9191 | 2006-07-03 00:24:34 -0700 | [diff] [blame] | 419 | .lock = __SPIN_LOCK_UNLOCKED(&input_pool.lock), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 420 | .pool = input_pool_data |
| 421 | }; |
| 422 | |
| 423 | static struct entropy_store blocking_pool = { |
| 424 | .poolinfo = &poolinfo_table[1], |
| 425 | .name = "blocking", |
| 426 | .limit = 1, |
| 427 | .pull = &input_pool, |
Ingo Molnar | e4d9191 | 2006-07-03 00:24:34 -0700 | [diff] [blame] | 428 | .lock = __SPIN_LOCK_UNLOCKED(&blocking_pool.lock), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 429 | .pool = blocking_pool_data |
| 430 | }; |
| 431 | |
| 432 | static struct entropy_store nonblocking_pool = { |
| 433 | .poolinfo = &poolinfo_table[1], |
| 434 | .name = "nonblocking", |
| 435 | .pull = &input_pool, |
Ingo Molnar | e4d9191 | 2006-07-03 00:24:34 -0700 | [diff] [blame] | 436 | .lock = __SPIN_LOCK_UNLOCKED(&nonblocking_pool.lock), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 437 | .pool = nonblocking_pool_data |
| 438 | }; |
| 439 | |
| 440 | /* |
| 441 | * This function adds a byte into the entropy "pool". It does not |
| 442 | * update the entropy estimate. The caller should call |
| 443 | * credit_entropy_store if this is appropriate. |
| 444 | * |
| 445 | * The pool is stirred with a primitive polynomial of the appropriate |
| 446 | * degree, and then twisted. We twist by three bits at a time because |
| 447 | * it's cheap to do so and helps slightly in the expected case where |
| 448 | * the entropy is concentrated in the low-order bits. |
| 449 | */ |
| 450 | static void __add_entropy_words(struct entropy_store *r, const __u32 *in, |
| 451 | int nwords, __u32 out[16]) |
| 452 | { |
| 453 | static __u32 const twist_table[8] = { |
| 454 | 0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158, |
| 455 | 0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 }; |
| 456 | unsigned long i, add_ptr, tap1, tap2, tap3, tap4, tap5; |
| 457 | int new_rotate, input_rotate; |
| 458 | int wordmask = r->poolinfo->poolwords - 1; |
| 459 | __u32 w, next_w; |
| 460 | unsigned long flags; |
| 461 | |
| 462 | /* Taps are constant, so we can load them without holding r->lock. */ |
| 463 | tap1 = r->poolinfo->tap1; |
| 464 | tap2 = r->poolinfo->tap2; |
| 465 | tap3 = r->poolinfo->tap3; |
| 466 | tap4 = r->poolinfo->tap4; |
| 467 | tap5 = r->poolinfo->tap5; |
| 468 | next_w = *in++; |
| 469 | |
| 470 | spin_lock_irqsave(&r->lock, flags); |
| 471 | prefetch_range(r->pool, wordmask); |
| 472 | input_rotate = r->input_rotate; |
| 473 | add_ptr = r->add_ptr; |
| 474 | |
| 475 | while (nwords--) { |
| 476 | w = rol32(next_w, input_rotate); |
| 477 | if (nwords > 0) |
| 478 | next_w = *in++; |
| 479 | i = add_ptr = (add_ptr - 1) & wordmask; |
| 480 | /* |
| 481 | * Normally, we add 7 bits of rotation to the pool. |
| 482 | * At the beginning of the pool, add an extra 7 bits |
| 483 | * rotation, so that successive passes spread the |
| 484 | * input bits across the pool evenly. |
| 485 | */ |
| 486 | new_rotate = input_rotate + 14; |
| 487 | if (i) |
| 488 | new_rotate = input_rotate + 7; |
| 489 | input_rotate = new_rotate & 31; |
| 490 | |
| 491 | /* XOR in the various taps */ |
| 492 | w ^= r->pool[(i + tap1) & wordmask]; |
| 493 | w ^= r->pool[(i + tap2) & wordmask]; |
| 494 | w ^= r->pool[(i + tap3) & wordmask]; |
| 495 | w ^= r->pool[(i + tap4) & wordmask]; |
| 496 | w ^= r->pool[(i + tap5) & wordmask]; |
| 497 | w ^= r->pool[i]; |
| 498 | r->pool[i] = (w >> 3) ^ twist_table[w & 7]; |
| 499 | } |
| 500 | |
| 501 | r->input_rotate = input_rotate; |
| 502 | r->add_ptr = add_ptr; |
| 503 | |
| 504 | if (out) { |
| 505 | for (i = 0; i < 16; i++) { |
| 506 | out[i] = r->pool[add_ptr]; |
| 507 | add_ptr = (add_ptr - 1) & wordmask; |
| 508 | } |
| 509 | } |
| 510 | |
| 511 | spin_unlock_irqrestore(&r->lock, flags); |
| 512 | } |
| 513 | |
| 514 | static inline void add_entropy_words(struct entropy_store *r, const __u32 *in, |
| 515 | int nwords) |
| 516 | { |
| 517 | __add_entropy_words(r, in, nwords, NULL); |
| 518 | } |
| 519 | |
| 520 | /* |
| 521 | * Credit (or debit) the entropy store with n bits of entropy |
| 522 | */ |
| 523 | static void credit_entropy_store(struct entropy_store *r, int nbits) |
| 524 | { |
| 525 | unsigned long flags; |
| 526 | |
| 527 | spin_lock_irqsave(&r->lock, flags); |
| 528 | |
| 529 | if (r->entropy_count + nbits < 0) { |
| 530 | DEBUG_ENT("negative entropy/overflow (%d+%d)\n", |
| 531 | r->entropy_count, nbits); |
| 532 | r->entropy_count = 0; |
| 533 | } else if (r->entropy_count + nbits > r->poolinfo->POOLBITS) { |
| 534 | r->entropy_count = r->poolinfo->POOLBITS; |
| 535 | } else { |
| 536 | r->entropy_count += nbits; |
| 537 | if (nbits) |
| 538 | DEBUG_ENT("added %d entropy credits to %s\n", |
| 539 | nbits, r->name); |
| 540 | } |
| 541 | |
| 542 | spin_unlock_irqrestore(&r->lock, flags); |
| 543 | } |
| 544 | |
| 545 | /********************************************************************* |
| 546 | * |
| 547 | * Entropy input management |
| 548 | * |
| 549 | *********************************************************************/ |
| 550 | |
| 551 | /* There is one of these per entropy source */ |
| 552 | struct timer_rand_state { |
| 553 | cycles_t last_time; |
| 554 | long last_delta,last_delta2; |
| 555 | unsigned dont_count_entropy:1; |
| 556 | }; |
| 557 | |
| 558 | static struct timer_rand_state input_timer_state; |
| 559 | static struct timer_rand_state *irq_timer_state[NR_IRQS]; |
| 560 | |
| 561 | /* |
| 562 | * This function adds entropy to the entropy "pool" by using timing |
| 563 | * delays. It uses the timer_rand_state structure to make an estimate |
| 564 | * of how many bits of entropy this call has added to the pool. |
| 565 | * |
| 566 | * The number "num" is also added to the pool - it should somehow describe |
| 567 | * the type of event which just happened. This is currently 0-255 for |
| 568 | * keyboard scan codes, and 256 upwards for interrupts. |
| 569 | * |
| 570 | */ |
| 571 | static void add_timer_randomness(struct timer_rand_state *state, unsigned num) |
| 572 | { |
| 573 | struct { |
| 574 | cycles_t cycles; |
| 575 | long jiffies; |
| 576 | unsigned num; |
| 577 | } sample; |
| 578 | long delta, delta2, delta3; |
| 579 | |
| 580 | preempt_disable(); |
| 581 | /* if over the trickle threshold, use only 1 in 4096 samples */ |
| 582 | if (input_pool.entropy_count > trickle_thresh && |
| 583 | (__get_cpu_var(trickle_count)++ & 0xfff)) |
| 584 | goto out; |
| 585 | |
| 586 | sample.jiffies = jiffies; |
| 587 | sample.cycles = get_cycles(); |
| 588 | sample.num = num; |
| 589 | add_entropy_words(&input_pool, (u32 *)&sample, sizeof(sample)/4); |
| 590 | |
| 591 | /* |
| 592 | * Calculate number of bits of randomness we probably added. |
| 593 | * We take into account the first, second and third-order deltas |
| 594 | * in order to make our estimate. |
| 595 | */ |
| 596 | |
| 597 | if (!state->dont_count_entropy) { |
| 598 | delta = sample.jiffies - state->last_time; |
| 599 | state->last_time = sample.jiffies; |
| 600 | |
| 601 | delta2 = delta - state->last_delta; |
| 602 | state->last_delta = delta; |
| 603 | |
| 604 | delta3 = delta2 - state->last_delta2; |
| 605 | state->last_delta2 = delta2; |
| 606 | |
| 607 | if (delta < 0) |
| 608 | delta = -delta; |
| 609 | if (delta2 < 0) |
| 610 | delta2 = -delta2; |
| 611 | if (delta3 < 0) |
| 612 | delta3 = -delta3; |
| 613 | if (delta > delta2) |
| 614 | delta = delta2; |
| 615 | if (delta > delta3) |
| 616 | delta = delta3; |
| 617 | |
| 618 | /* |
| 619 | * delta is now minimum absolute delta. |
| 620 | * Round down by 1 bit on general principles, |
| 621 | * and limit entropy entimate to 12 bits. |
| 622 | */ |
| 623 | credit_entropy_store(&input_pool, |
| 624 | min_t(int, fls(delta>>1), 11)); |
| 625 | } |
| 626 | |
| 627 | if(input_pool.entropy_count >= random_read_wakeup_thresh) |
| 628 | wake_up_interruptible(&random_read_wait); |
| 629 | |
| 630 | out: |
| 631 | preempt_enable(); |
| 632 | } |
| 633 | |
Stephen Hemminger | d251575 | 2006-01-11 12:17:38 -0800 | [diff] [blame] | 634 | void add_input_randomness(unsigned int type, unsigned int code, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 635 | unsigned int value) |
| 636 | { |
| 637 | static unsigned char last_value; |
| 638 | |
| 639 | /* ignore autorepeat and the like */ |
| 640 | if (value == last_value) |
| 641 | return; |
| 642 | |
| 643 | DEBUG_ENT("input event\n"); |
| 644 | last_value = value; |
| 645 | add_timer_randomness(&input_timer_state, |
| 646 | (type << 4) ^ code ^ (code >> 4) ^ value); |
| 647 | } |
Dmitry Torokhov | 80fc9f5 | 2006-10-11 01:43:58 -0400 | [diff] [blame] | 648 | EXPORT_SYMBOL_GPL(add_input_randomness); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 649 | |
| 650 | void add_interrupt_randomness(int irq) |
| 651 | { |
Stephen Hemminger | c80544d | 2007-10-18 03:07:05 -0700 | [diff] [blame] | 652 | if (irq >= NR_IRQS || irq_timer_state[irq] == NULL) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 653 | return; |
| 654 | |
| 655 | DEBUG_ENT("irq event %d\n", irq); |
| 656 | add_timer_randomness(irq_timer_state[irq], 0x100 + irq); |
| 657 | } |
| 658 | |
David Howells | 9361401 | 2006-09-30 20:45:40 +0200 | [diff] [blame] | 659 | #ifdef CONFIG_BLOCK |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 660 | void add_disk_randomness(struct gendisk *disk) |
| 661 | { |
| 662 | if (!disk || !disk->random) |
| 663 | return; |
| 664 | /* first major is 1, so we get >= 0x200 here */ |
| 665 | DEBUG_ENT("disk event %d:%d\n", disk->major, disk->first_minor); |
| 666 | |
| 667 | add_timer_randomness(disk->random, |
| 668 | 0x100 + MKDEV(disk->major, disk->first_minor)); |
| 669 | } |
David Howells | 9361401 | 2006-09-30 20:45:40 +0200 | [diff] [blame] | 670 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 671 | |
| 672 | #define EXTRACT_SIZE 10 |
| 673 | |
| 674 | /********************************************************************* |
| 675 | * |
| 676 | * Entropy extraction routines |
| 677 | * |
| 678 | *********************************************************************/ |
| 679 | |
| 680 | static ssize_t extract_entropy(struct entropy_store *r, void * buf, |
| 681 | size_t nbytes, int min, int rsvd); |
| 682 | |
| 683 | /* |
| 684 | * This utility inline function is responsible for transfering entropy |
| 685 | * from the primary pool to the secondary extraction pool. We make |
| 686 | * sure we pull enough for a 'catastrophic reseed'. |
| 687 | */ |
| 688 | static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes) |
| 689 | { |
| 690 | __u32 tmp[OUTPUT_POOL_WORDS]; |
| 691 | |
| 692 | if (r->pull && r->entropy_count < nbytes * 8 && |
| 693 | r->entropy_count < r->poolinfo->POOLBITS) { |
Matt Mackall | 5a021e9 | 2007-07-19 11:30:14 -0700 | [diff] [blame] | 694 | /* If we're limited, always leave two wakeup worth's BITS */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 695 | int rsvd = r->limit ? 0 : random_read_wakeup_thresh/4; |
Matt Mackall | 5a021e9 | 2007-07-19 11:30:14 -0700 | [diff] [blame] | 696 | int bytes = nbytes; |
| 697 | |
| 698 | /* pull at least as many as BYTES as wakeup BITS */ |
| 699 | bytes = max_t(int, bytes, random_read_wakeup_thresh / 8); |
| 700 | /* but never more than the buffer size */ |
| 701 | bytes = min_t(int, bytes, sizeof(tmp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 702 | |
| 703 | DEBUG_ENT("going to reseed %s with %d bits " |
| 704 | "(%d of %d requested)\n", |
| 705 | r->name, bytes * 8, nbytes * 8, r->entropy_count); |
| 706 | |
| 707 | bytes=extract_entropy(r->pull, tmp, bytes, |
| 708 | random_read_wakeup_thresh / 8, rsvd); |
| 709 | add_entropy_words(r, tmp, (bytes + 3) / 4); |
| 710 | credit_entropy_store(r, bytes*8); |
| 711 | } |
| 712 | } |
| 713 | |
| 714 | /* |
| 715 | * These functions extracts randomness from the "entropy pool", and |
| 716 | * returns it in a buffer. |
| 717 | * |
| 718 | * The min parameter specifies the minimum amount we can pull before |
| 719 | * failing to avoid races that defeat catastrophic reseeding while the |
| 720 | * reserved parameter indicates how much entropy we must leave in the |
| 721 | * pool after each pull to avoid starving other readers. |
| 722 | * |
| 723 | * Note: extract_entropy() assumes that .poolwords is a multiple of 16 words. |
| 724 | */ |
| 725 | |
| 726 | static size_t account(struct entropy_store *r, size_t nbytes, int min, |
| 727 | int reserved) |
| 728 | { |
| 729 | unsigned long flags; |
| 730 | |
| 731 | BUG_ON(r->entropy_count > r->poolinfo->POOLBITS); |
| 732 | |
| 733 | /* Hold lock while accounting */ |
| 734 | spin_lock_irqsave(&r->lock, flags); |
| 735 | |
| 736 | DEBUG_ENT("trying to extract %d bits from %s\n", |
| 737 | nbytes * 8, r->name); |
| 738 | |
| 739 | /* Can we pull enough? */ |
| 740 | if (r->entropy_count / 8 < min + reserved) { |
| 741 | nbytes = 0; |
| 742 | } else { |
| 743 | /* If limited, never pull more than available */ |
| 744 | if (r->limit && nbytes + reserved >= r->entropy_count / 8) |
| 745 | nbytes = r->entropy_count/8 - reserved; |
| 746 | |
| 747 | if(r->entropy_count / 8 >= nbytes + reserved) |
| 748 | r->entropy_count -= nbytes*8; |
| 749 | else |
| 750 | r->entropy_count = reserved; |
| 751 | |
| 752 | if (r->entropy_count < random_write_wakeup_thresh) |
| 753 | wake_up_interruptible(&random_write_wait); |
| 754 | } |
| 755 | |
| 756 | DEBUG_ENT("debiting %d entropy credits from %s%s\n", |
| 757 | nbytes * 8, r->name, r->limit ? "" : " (unlimited)"); |
| 758 | |
| 759 | spin_unlock_irqrestore(&r->lock, flags); |
| 760 | |
| 761 | return nbytes; |
| 762 | } |
| 763 | |
| 764 | static void extract_buf(struct entropy_store *r, __u8 *out) |
| 765 | { |
Matt Mackall | 602b6ae | 2007-05-29 21:54:27 -0500 | [diff] [blame] | 766 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 767 | __u32 data[16], buf[5 + SHA_WORKSPACE_WORDS]; |
| 768 | |
| 769 | sha_init(buf); |
| 770 | /* |
| 771 | * As we hash the pool, we mix intermediate values of |
| 772 | * the hash back into the pool. This eliminates |
| 773 | * backtracking attacks (where the attacker knows |
| 774 | * the state of the pool plus the current outputs, and |
| 775 | * attempts to find previous ouputs), unless the hash |
| 776 | * function can be inverted. |
| 777 | */ |
Matt Mackall | 602b6ae | 2007-05-29 21:54:27 -0500 | [diff] [blame] | 778 | for (i = 0; i < r->poolinfo->poolwords; i += 16) { |
| 779 | /* hash blocks of 16 words = 512 bits */ |
| 780 | sha_transform(buf, (__u8 *)(r->pool + i), buf + 5); |
| 781 | /* feed back portion of the resulting hash */ |
| 782 | add_entropy_words(r, &buf[i % 5], 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 783 | } |
| 784 | |
| 785 | /* |
| 786 | * To avoid duplicates, we atomically extract a |
| 787 | * portion of the pool while mixing, and hash one |
| 788 | * final time. |
| 789 | */ |
Matt Mackall | 602b6ae | 2007-05-29 21:54:27 -0500 | [diff] [blame] | 790 | __add_entropy_words(r, &buf[i % 5], 1, data); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 791 | sha_transform(buf, (__u8 *)data, buf + 5); |
| 792 | |
| 793 | /* |
| 794 | * In case the hash function has some recognizable |
| 795 | * output pattern, we fold it in half. |
| 796 | */ |
| 797 | |
| 798 | buf[0] ^= buf[3]; |
| 799 | buf[1] ^= buf[4]; |
Matt Mackall | 679ce0a | 2007-06-16 10:16:11 -0700 | [diff] [blame] | 800 | buf[2] ^= rol32(buf[2], 16); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 801 | memcpy(out, buf, EXTRACT_SIZE); |
| 802 | memset(buf, 0, sizeof(buf)); |
| 803 | } |
| 804 | |
| 805 | static ssize_t extract_entropy(struct entropy_store *r, void * buf, |
| 806 | size_t nbytes, int min, int reserved) |
| 807 | { |
| 808 | ssize_t ret = 0, i; |
| 809 | __u8 tmp[EXTRACT_SIZE]; |
| 810 | |
| 811 | xfer_secondary_pool(r, nbytes); |
| 812 | nbytes = account(r, nbytes, min, reserved); |
| 813 | |
| 814 | while (nbytes) { |
| 815 | extract_buf(r, tmp); |
| 816 | i = min_t(int, nbytes, EXTRACT_SIZE); |
| 817 | memcpy(buf, tmp, i); |
| 818 | nbytes -= i; |
| 819 | buf += i; |
| 820 | ret += i; |
| 821 | } |
| 822 | |
| 823 | /* Wipe data just returned from memory */ |
| 824 | memset(tmp, 0, sizeof(tmp)); |
| 825 | |
| 826 | return ret; |
| 827 | } |
| 828 | |
| 829 | static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf, |
| 830 | size_t nbytes) |
| 831 | { |
| 832 | ssize_t ret = 0, i; |
| 833 | __u8 tmp[EXTRACT_SIZE]; |
| 834 | |
| 835 | xfer_secondary_pool(r, nbytes); |
| 836 | nbytes = account(r, nbytes, 0, 0); |
| 837 | |
| 838 | while (nbytes) { |
| 839 | if (need_resched()) { |
| 840 | if (signal_pending(current)) { |
| 841 | if (ret == 0) |
| 842 | ret = -ERESTARTSYS; |
| 843 | break; |
| 844 | } |
| 845 | schedule(); |
| 846 | } |
| 847 | |
| 848 | extract_buf(r, tmp); |
| 849 | i = min_t(int, nbytes, EXTRACT_SIZE); |
| 850 | if (copy_to_user(buf, tmp, i)) { |
| 851 | ret = -EFAULT; |
| 852 | break; |
| 853 | } |
| 854 | |
| 855 | nbytes -= i; |
| 856 | buf += i; |
| 857 | ret += i; |
| 858 | } |
| 859 | |
| 860 | /* Wipe data just returned from memory */ |
| 861 | memset(tmp, 0, sizeof(tmp)); |
| 862 | |
| 863 | return ret; |
| 864 | } |
| 865 | |
| 866 | /* |
| 867 | * This function is the exported kernel interface. It returns some |
| 868 | * number of good random numbers, suitable for seeding TCP sequence |
| 869 | * numbers, etc. |
| 870 | */ |
| 871 | void get_random_bytes(void *buf, int nbytes) |
| 872 | { |
| 873 | extract_entropy(&nonblocking_pool, buf, nbytes, 0, 0); |
| 874 | } |
| 875 | |
| 876 | EXPORT_SYMBOL(get_random_bytes); |
| 877 | |
| 878 | /* |
| 879 | * init_std_data - initialize pool with system data |
| 880 | * |
| 881 | * @r: pool to initialize |
| 882 | * |
| 883 | * This function clears the pool's entropy count and mixes some system |
| 884 | * data into the pool to prepare it for use. The pool is not cleared |
| 885 | * as that can only decrease the entropy in the pool. |
| 886 | */ |
| 887 | static void init_std_data(struct entropy_store *r) |
| 888 | { |
Eric Dumazet | f859581 | 2007-03-28 14:22:33 -0700 | [diff] [blame] | 889 | ktime_t now; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 890 | unsigned long flags; |
| 891 | |
| 892 | spin_lock_irqsave(&r->lock, flags); |
| 893 | r->entropy_count = 0; |
| 894 | spin_unlock_irqrestore(&r->lock, flags); |
| 895 | |
Eric Dumazet | f859581 | 2007-03-28 14:22:33 -0700 | [diff] [blame] | 896 | now = ktime_get_real(); |
| 897 | add_entropy_words(r, (__u32 *)&now, sizeof(now)/4); |
Serge E. Hallyn | e9ff399 | 2006-10-02 02:18:11 -0700 | [diff] [blame] | 898 | add_entropy_words(r, (__u32 *)utsname(), |
| 899 | sizeof(*(utsname()))/4); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 900 | } |
| 901 | |
| 902 | static int __init rand_initialize(void) |
| 903 | { |
| 904 | init_std_data(&input_pool); |
| 905 | init_std_data(&blocking_pool); |
| 906 | init_std_data(&nonblocking_pool); |
| 907 | return 0; |
| 908 | } |
| 909 | module_init(rand_initialize); |
| 910 | |
| 911 | void rand_initialize_irq(int irq) |
| 912 | { |
| 913 | struct timer_rand_state *state; |
| 914 | |
| 915 | if (irq >= NR_IRQS || irq_timer_state[irq]) |
| 916 | return; |
| 917 | |
| 918 | /* |
Eric Dumazet | f859581 | 2007-03-28 14:22:33 -0700 | [diff] [blame] | 919 | * If kzalloc returns null, we just won't use that entropy |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 920 | * source. |
| 921 | */ |
Eric Dumazet | f859581 | 2007-03-28 14:22:33 -0700 | [diff] [blame] | 922 | state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL); |
| 923 | if (state) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 924 | irq_timer_state[irq] = state; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 925 | } |
| 926 | |
David Howells | 9361401 | 2006-09-30 20:45:40 +0200 | [diff] [blame] | 927 | #ifdef CONFIG_BLOCK |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 928 | void rand_initialize_disk(struct gendisk *disk) |
| 929 | { |
| 930 | struct timer_rand_state *state; |
| 931 | |
| 932 | /* |
Eric Dumazet | f859581 | 2007-03-28 14:22:33 -0700 | [diff] [blame] | 933 | * If kzalloc returns null, we just won't use that entropy |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 934 | * source. |
| 935 | */ |
Eric Dumazet | f859581 | 2007-03-28 14:22:33 -0700 | [diff] [blame] | 936 | state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL); |
| 937 | if (state) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 938 | disk->random = state; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 939 | } |
David Howells | 9361401 | 2006-09-30 20:45:40 +0200 | [diff] [blame] | 940 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 941 | |
| 942 | static ssize_t |
| 943 | random_read(struct file * file, char __user * buf, size_t nbytes, loff_t *ppos) |
| 944 | { |
| 945 | ssize_t n, retval = 0, count = 0; |
| 946 | |
| 947 | if (nbytes == 0) |
| 948 | return 0; |
| 949 | |
| 950 | while (nbytes > 0) { |
| 951 | n = nbytes; |
| 952 | if (n > SEC_XFER_SIZE) |
| 953 | n = SEC_XFER_SIZE; |
| 954 | |
| 955 | DEBUG_ENT("reading %d bits\n", n*8); |
| 956 | |
| 957 | n = extract_entropy_user(&blocking_pool, buf, n); |
| 958 | |
| 959 | DEBUG_ENT("read got %d bits (%d still needed)\n", |
| 960 | n*8, (nbytes-n)*8); |
| 961 | |
| 962 | if (n == 0) { |
| 963 | if (file->f_flags & O_NONBLOCK) { |
| 964 | retval = -EAGAIN; |
| 965 | break; |
| 966 | } |
| 967 | |
| 968 | DEBUG_ENT("sleeping?\n"); |
| 969 | |
| 970 | wait_event_interruptible(random_read_wait, |
| 971 | input_pool.entropy_count >= |
| 972 | random_read_wakeup_thresh); |
| 973 | |
| 974 | DEBUG_ENT("awake\n"); |
| 975 | |
| 976 | if (signal_pending(current)) { |
| 977 | retval = -ERESTARTSYS; |
| 978 | break; |
| 979 | } |
| 980 | |
| 981 | continue; |
| 982 | } |
| 983 | |
| 984 | if (n < 0) { |
| 985 | retval = n; |
| 986 | break; |
| 987 | } |
| 988 | count += n; |
| 989 | buf += n; |
| 990 | nbytes -= n; |
| 991 | break; /* This break makes the device work */ |
| 992 | /* like a named pipe */ |
| 993 | } |
| 994 | |
| 995 | /* |
| 996 | * If we gave the user some bytes, update the access time. |
| 997 | */ |
| 998 | if (count) |
| 999 | file_accessed(file); |
| 1000 | |
| 1001 | return (count ? count : retval); |
| 1002 | } |
| 1003 | |
| 1004 | static ssize_t |
| 1005 | urandom_read(struct file * file, char __user * buf, |
| 1006 | size_t nbytes, loff_t *ppos) |
| 1007 | { |
| 1008 | return extract_entropy_user(&nonblocking_pool, buf, nbytes); |
| 1009 | } |
| 1010 | |
| 1011 | static unsigned int |
| 1012 | random_poll(struct file *file, poll_table * wait) |
| 1013 | { |
| 1014 | unsigned int mask; |
| 1015 | |
| 1016 | poll_wait(file, &random_read_wait, wait); |
| 1017 | poll_wait(file, &random_write_wait, wait); |
| 1018 | mask = 0; |
| 1019 | if (input_pool.entropy_count >= random_read_wakeup_thresh) |
| 1020 | mask |= POLLIN | POLLRDNORM; |
| 1021 | if (input_pool.entropy_count < random_write_wakeup_thresh) |
| 1022 | mask |= POLLOUT | POLLWRNORM; |
| 1023 | return mask; |
| 1024 | } |
| 1025 | |
Matt Mackall | 7f397dc | 2007-05-29 21:58:10 -0500 | [diff] [blame] | 1026 | static int |
| 1027 | write_pool(struct entropy_store *r, const char __user *buffer, size_t count) |
| 1028 | { |
| 1029 | size_t bytes; |
| 1030 | __u32 buf[16]; |
| 1031 | const char __user *p = buffer; |
| 1032 | |
| 1033 | while (count > 0) { |
| 1034 | bytes = min(count, sizeof(buf)); |
| 1035 | if (copy_from_user(&buf, p, bytes)) |
| 1036 | return -EFAULT; |
| 1037 | |
| 1038 | count -= bytes; |
| 1039 | p += bytes; |
| 1040 | |
| 1041 | add_entropy_words(r, buf, (bytes + 3) / 4); |
Matt Mackall | 91f3f1e | 2008-02-06 01:37:20 -0800 | [diff] [blame] | 1042 | cond_resched(); |
Matt Mackall | 7f397dc | 2007-05-29 21:58:10 -0500 | [diff] [blame] | 1043 | } |
| 1044 | |
| 1045 | return 0; |
| 1046 | } |
| 1047 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1048 | static ssize_t |
| 1049 | random_write(struct file * file, const char __user * buffer, |
| 1050 | size_t count, loff_t *ppos) |
| 1051 | { |
Matt Mackall | 7f397dc | 2007-05-29 21:58:10 -0500 | [diff] [blame] | 1052 | size_t ret; |
| 1053 | struct inode *inode = file->f_path.dentry->d_inode; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1054 | |
Matt Mackall | 7f397dc | 2007-05-29 21:58:10 -0500 | [diff] [blame] | 1055 | ret = write_pool(&blocking_pool, buffer, count); |
| 1056 | if (ret) |
| 1057 | return ret; |
| 1058 | ret = write_pool(&nonblocking_pool, buffer, count); |
| 1059 | if (ret) |
| 1060 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1061 | |
Matt Mackall | 7f397dc | 2007-05-29 21:58:10 -0500 | [diff] [blame] | 1062 | inode->i_mtime = current_fs_time(inode->i_sb); |
| 1063 | mark_inode_dirty(inode); |
| 1064 | return (ssize_t)count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1065 | } |
| 1066 | |
| 1067 | static int |
| 1068 | random_ioctl(struct inode * inode, struct file * file, |
| 1069 | unsigned int cmd, unsigned long arg) |
| 1070 | { |
| 1071 | int size, ent_count; |
| 1072 | int __user *p = (int __user *)arg; |
| 1073 | int retval; |
| 1074 | |
| 1075 | switch (cmd) { |
| 1076 | case RNDGETENTCNT: |
| 1077 | ent_count = input_pool.entropy_count; |
| 1078 | if (put_user(ent_count, p)) |
| 1079 | return -EFAULT; |
| 1080 | return 0; |
| 1081 | case RNDADDTOENTCNT: |
| 1082 | if (!capable(CAP_SYS_ADMIN)) |
| 1083 | return -EPERM; |
| 1084 | if (get_user(ent_count, p)) |
| 1085 | return -EFAULT; |
| 1086 | credit_entropy_store(&input_pool, ent_count); |
| 1087 | /* |
| 1088 | * Wake up waiting processes if we have enough |
| 1089 | * entropy. |
| 1090 | */ |
| 1091 | if (input_pool.entropy_count >= random_read_wakeup_thresh) |
| 1092 | wake_up_interruptible(&random_read_wait); |
| 1093 | return 0; |
| 1094 | case RNDADDENTROPY: |
| 1095 | if (!capable(CAP_SYS_ADMIN)) |
| 1096 | return -EPERM; |
| 1097 | if (get_user(ent_count, p++)) |
| 1098 | return -EFAULT; |
| 1099 | if (ent_count < 0) |
| 1100 | return -EINVAL; |
| 1101 | if (get_user(size, p++)) |
| 1102 | return -EFAULT; |
Matt Mackall | 7f397dc | 2007-05-29 21:58:10 -0500 | [diff] [blame] | 1103 | retval = write_pool(&input_pool, (const char __user *)p, |
| 1104 | size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1105 | if (retval < 0) |
| 1106 | return retval; |
| 1107 | credit_entropy_store(&input_pool, ent_count); |
| 1108 | /* |
| 1109 | * Wake up waiting processes if we have enough |
| 1110 | * entropy. |
| 1111 | */ |
| 1112 | if (input_pool.entropy_count >= random_read_wakeup_thresh) |
| 1113 | wake_up_interruptible(&random_read_wait); |
| 1114 | return 0; |
| 1115 | case RNDZAPENTCNT: |
| 1116 | case RNDCLEARPOOL: |
| 1117 | /* Clear the entropy pool counters. */ |
| 1118 | if (!capable(CAP_SYS_ADMIN)) |
| 1119 | return -EPERM; |
| 1120 | init_std_data(&input_pool); |
| 1121 | init_std_data(&blocking_pool); |
| 1122 | init_std_data(&nonblocking_pool); |
| 1123 | return 0; |
| 1124 | default: |
| 1125 | return -EINVAL; |
| 1126 | } |
| 1127 | } |
| 1128 | |
Arjan van de Ven | 2b8693c | 2007-02-12 00:55:32 -0800 | [diff] [blame] | 1129 | const struct file_operations random_fops = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1130 | .read = random_read, |
| 1131 | .write = random_write, |
| 1132 | .poll = random_poll, |
| 1133 | .ioctl = random_ioctl, |
| 1134 | }; |
| 1135 | |
Arjan van de Ven | 2b8693c | 2007-02-12 00:55:32 -0800 | [diff] [blame] | 1136 | const struct file_operations urandom_fops = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1137 | .read = urandom_read, |
| 1138 | .write = random_write, |
| 1139 | .ioctl = random_ioctl, |
| 1140 | }; |
| 1141 | |
| 1142 | /*************************************************************** |
| 1143 | * Random UUID interface |
| 1144 | * |
| 1145 | * Used here for a Boot ID, but can be useful for other kernel |
| 1146 | * drivers. |
| 1147 | ***************************************************************/ |
| 1148 | |
| 1149 | /* |
| 1150 | * Generate random UUID |
| 1151 | */ |
| 1152 | void generate_random_uuid(unsigned char uuid_out[16]) |
| 1153 | { |
| 1154 | get_random_bytes(uuid_out, 16); |
| 1155 | /* Set UUID version to 4 --- truely random generation */ |
| 1156 | uuid_out[6] = (uuid_out[6] & 0x0F) | 0x40; |
| 1157 | /* Set the UUID variant to DCE */ |
| 1158 | uuid_out[8] = (uuid_out[8] & 0x3F) | 0x80; |
| 1159 | } |
| 1160 | |
| 1161 | EXPORT_SYMBOL(generate_random_uuid); |
| 1162 | |
| 1163 | /******************************************************************** |
| 1164 | * |
| 1165 | * Sysctl interface |
| 1166 | * |
| 1167 | ********************************************************************/ |
| 1168 | |
| 1169 | #ifdef CONFIG_SYSCTL |
| 1170 | |
| 1171 | #include <linux/sysctl.h> |
| 1172 | |
| 1173 | static int min_read_thresh = 8, min_write_thresh; |
| 1174 | static int max_read_thresh = INPUT_POOL_WORDS * 32; |
| 1175 | static int max_write_thresh = INPUT_POOL_WORDS * 32; |
| 1176 | static char sysctl_bootid[16]; |
| 1177 | |
| 1178 | /* |
| 1179 | * These functions is used to return both the bootid UUID, and random |
| 1180 | * UUID. The difference is in whether table->data is NULL; if it is, |
| 1181 | * then a new UUID is generated and returned to the user. |
| 1182 | * |
| 1183 | * If the user accesses this via the proc interface, it will be returned |
| 1184 | * as an ASCII string in the standard UUID format. If accesses via the |
| 1185 | * sysctl system call, it is returned as 16 bytes of binary data. |
| 1186 | */ |
| 1187 | static int proc_do_uuid(ctl_table *table, int write, struct file *filp, |
| 1188 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 1189 | { |
| 1190 | ctl_table fake_table; |
| 1191 | unsigned char buf[64], tmp_uuid[16], *uuid; |
| 1192 | |
| 1193 | uuid = table->data; |
| 1194 | if (!uuid) { |
| 1195 | uuid = tmp_uuid; |
| 1196 | uuid[8] = 0; |
| 1197 | } |
| 1198 | if (uuid[8] == 0) |
| 1199 | generate_random_uuid(uuid); |
| 1200 | |
| 1201 | sprintf(buf, "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-" |
| 1202 | "%02x%02x%02x%02x%02x%02x", |
| 1203 | uuid[0], uuid[1], uuid[2], uuid[3], |
| 1204 | uuid[4], uuid[5], uuid[6], uuid[7], |
| 1205 | uuid[8], uuid[9], uuid[10], uuid[11], |
| 1206 | uuid[12], uuid[13], uuid[14], uuid[15]); |
| 1207 | fake_table.data = buf; |
| 1208 | fake_table.maxlen = sizeof(buf); |
| 1209 | |
| 1210 | return proc_dostring(&fake_table, write, filp, buffer, lenp, ppos); |
| 1211 | } |
| 1212 | |
| 1213 | static int uuid_strategy(ctl_table *table, int __user *name, int nlen, |
| 1214 | void __user *oldval, size_t __user *oldlenp, |
Alexey Dobriyan | 1f29bcd | 2006-12-10 02:19:10 -0800 | [diff] [blame] | 1215 | void __user *newval, size_t newlen) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1216 | { |
| 1217 | unsigned char tmp_uuid[16], *uuid; |
| 1218 | unsigned int len; |
| 1219 | |
| 1220 | if (!oldval || !oldlenp) |
| 1221 | return 1; |
| 1222 | |
| 1223 | uuid = table->data; |
| 1224 | if (!uuid) { |
| 1225 | uuid = tmp_uuid; |
| 1226 | uuid[8] = 0; |
| 1227 | } |
| 1228 | if (uuid[8] == 0) |
| 1229 | generate_random_uuid(uuid); |
| 1230 | |
| 1231 | if (get_user(len, oldlenp)) |
| 1232 | return -EFAULT; |
| 1233 | if (len) { |
| 1234 | if (len > 16) |
| 1235 | len = 16; |
| 1236 | if (copy_to_user(oldval, uuid, len) || |
| 1237 | put_user(len, oldlenp)) |
| 1238 | return -EFAULT; |
| 1239 | } |
| 1240 | return 1; |
| 1241 | } |
| 1242 | |
| 1243 | static int sysctl_poolsize = INPUT_POOL_WORDS * 32; |
| 1244 | ctl_table random_table[] = { |
| 1245 | { |
| 1246 | .ctl_name = RANDOM_POOLSIZE, |
| 1247 | .procname = "poolsize", |
| 1248 | .data = &sysctl_poolsize, |
| 1249 | .maxlen = sizeof(int), |
| 1250 | .mode = 0444, |
| 1251 | .proc_handler = &proc_dointvec, |
| 1252 | }, |
| 1253 | { |
| 1254 | .ctl_name = RANDOM_ENTROPY_COUNT, |
| 1255 | .procname = "entropy_avail", |
| 1256 | .maxlen = sizeof(int), |
| 1257 | .mode = 0444, |
| 1258 | .proc_handler = &proc_dointvec, |
| 1259 | .data = &input_pool.entropy_count, |
| 1260 | }, |
| 1261 | { |
| 1262 | .ctl_name = RANDOM_READ_THRESH, |
| 1263 | .procname = "read_wakeup_threshold", |
| 1264 | .data = &random_read_wakeup_thresh, |
| 1265 | .maxlen = sizeof(int), |
| 1266 | .mode = 0644, |
| 1267 | .proc_handler = &proc_dointvec_minmax, |
| 1268 | .strategy = &sysctl_intvec, |
| 1269 | .extra1 = &min_read_thresh, |
| 1270 | .extra2 = &max_read_thresh, |
| 1271 | }, |
| 1272 | { |
| 1273 | .ctl_name = RANDOM_WRITE_THRESH, |
| 1274 | .procname = "write_wakeup_threshold", |
| 1275 | .data = &random_write_wakeup_thresh, |
| 1276 | .maxlen = sizeof(int), |
| 1277 | .mode = 0644, |
| 1278 | .proc_handler = &proc_dointvec_minmax, |
| 1279 | .strategy = &sysctl_intvec, |
| 1280 | .extra1 = &min_write_thresh, |
| 1281 | .extra2 = &max_write_thresh, |
| 1282 | }, |
| 1283 | { |
| 1284 | .ctl_name = RANDOM_BOOT_ID, |
| 1285 | .procname = "boot_id", |
| 1286 | .data = &sysctl_bootid, |
| 1287 | .maxlen = 16, |
| 1288 | .mode = 0444, |
| 1289 | .proc_handler = &proc_do_uuid, |
| 1290 | .strategy = &uuid_strategy, |
| 1291 | }, |
| 1292 | { |
| 1293 | .ctl_name = RANDOM_UUID, |
| 1294 | .procname = "uuid", |
| 1295 | .maxlen = 16, |
| 1296 | .mode = 0444, |
| 1297 | .proc_handler = &proc_do_uuid, |
| 1298 | .strategy = &uuid_strategy, |
| 1299 | }, |
| 1300 | { .ctl_name = 0 } |
| 1301 | }; |
| 1302 | #endif /* CONFIG_SYSCTL */ |
| 1303 | |
| 1304 | /******************************************************************** |
| 1305 | * |
| 1306 | * Random funtions for networking |
| 1307 | * |
| 1308 | ********************************************************************/ |
| 1309 | |
| 1310 | /* |
| 1311 | * TCP initial sequence number picking. This uses the random number |
| 1312 | * generator to pick an initial secret value. This value is hashed |
| 1313 | * along with the TCP endpoint information to provide a unique |
| 1314 | * starting point for each pair of TCP endpoints. This defeats |
| 1315 | * attacks which rely on guessing the initial TCP sequence number. |
| 1316 | * This algorithm was suggested by Steve Bellovin. |
| 1317 | * |
| 1318 | * Using a very strong hash was taking an appreciable amount of the total |
| 1319 | * TCP connection establishment time, so this is a weaker hash, |
| 1320 | * compensated for by changing the secret periodically. |
| 1321 | */ |
| 1322 | |
| 1323 | /* F, G and H are basic MD4 functions: selection, majority, parity */ |
| 1324 | #define F(x, y, z) ((z) ^ ((x) & ((y) ^ (z)))) |
| 1325 | #define G(x, y, z) (((x) & (y)) + (((x) ^ (y)) & (z))) |
| 1326 | #define H(x, y, z) ((x) ^ (y) ^ (z)) |
| 1327 | |
| 1328 | /* |
| 1329 | * The generic round function. The application is so specific that |
| 1330 | * we don't bother protecting all the arguments with parens, as is generally |
| 1331 | * good macro practice, in favor of extra legibility. |
| 1332 | * Rotation is separate from addition to prevent recomputation |
| 1333 | */ |
| 1334 | #define ROUND(f, a, b, c, d, x, s) \ |
| 1335 | (a += f(b, c, d) + x, a = (a << s) | (a >> (32 - s))) |
| 1336 | #define K1 0 |
| 1337 | #define K2 013240474631UL |
| 1338 | #define K3 015666365641UL |
| 1339 | |
| 1340 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
| 1341 | |
| 1342 | static __u32 twothirdsMD4Transform (__u32 const buf[4], __u32 const in[12]) |
| 1343 | { |
| 1344 | __u32 a = buf[0], b = buf[1], c = buf[2], d = buf[3]; |
| 1345 | |
| 1346 | /* Round 1 */ |
| 1347 | ROUND(F, a, b, c, d, in[ 0] + K1, 3); |
| 1348 | ROUND(F, d, a, b, c, in[ 1] + K1, 7); |
| 1349 | ROUND(F, c, d, a, b, in[ 2] + K1, 11); |
| 1350 | ROUND(F, b, c, d, a, in[ 3] + K1, 19); |
| 1351 | ROUND(F, a, b, c, d, in[ 4] + K1, 3); |
| 1352 | ROUND(F, d, a, b, c, in[ 5] + K1, 7); |
| 1353 | ROUND(F, c, d, a, b, in[ 6] + K1, 11); |
| 1354 | ROUND(F, b, c, d, a, in[ 7] + K1, 19); |
| 1355 | ROUND(F, a, b, c, d, in[ 8] + K1, 3); |
| 1356 | ROUND(F, d, a, b, c, in[ 9] + K1, 7); |
| 1357 | ROUND(F, c, d, a, b, in[10] + K1, 11); |
| 1358 | ROUND(F, b, c, d, a, in[11] + K1, 19); |
| 1359 | |
| 1360 | /* Round 2 */ |
| 1361 | ROUND(G, a, b, c, d, in[ 1] + K2, 3); |
| 1362 | ROUND(G, d, a, b, c, in[ 3] + K2, 5); |
| 1363 | ROUND(G, c, d, a, b, in[ 5] + K2, 9); |
| 1364 | ROUND(G, b, c, d, a, in[ 7] + K2, 13); |
| 1365 | ROUND(G, a, b, c, d, in[ 9] + K2, 3); |
| 1366 | ROUND(G, d, a, b, c, in[11] + K2, 5); |
| 1367 | ROUND(G, c, d, a, b, in[ 0] + K2, 9); |
| 1368 | ROUND(G, b, c, d, a, in[ 2] + K2, 13); |
| 1369 | ROUND(G, a, b, c, d, in[ 4] + K2, 3); |
| 1370 | ROUND(G, d, a, b, c, in[ 6] + K2, 5); |
| 1371 | ROUND(G, c, d, a, b, in[ 8] + K2, 9); |
| 1372 | ROUND(G, b, c, d, a, in[10] + K2, 13); |
| 1373 | |
| 1374 | /* Round 3 */ |
| 1375 | ROUND(H, a, b, c, d, in[ 3] + K3, 3); |
| 1376 | ROUND(H, d, a, b, c, in[ 7] + K3, 9); |
| 1377 | ROUND(H, c, d, a, b, in[11] + K3, 11); |
| 1378 | ROUND(H, b, c, d, a, in[ 2] + K3, 15); |
| 1379 | ROUND(H, a, b, c, d, in[ 6] + K3, 3); |
| 1380 | ROUND(H, d, a, b, c, in[10] + K3, 9); |
| 1381 | ROUND(H, c, d, a, b, in[ 1] + K3, 11); |
| 1382 | ROUND(H, b, c, d, a, in[ 5] + K3, 15); |
| 1383 | ROUND(H, a, b, c, d, in[ 9] + K3, 3); |
| 1384 | ROUND(H, d, a, b, c, in[ 0] + K3, 9); |
| 1385 | ROUND(H, c, d, a, b, in[ 4] + K3, 11); |
| 1386 | ROUND(H, b, c, d, a, in[ 8] + K3, 15); |
| 1387 | |
| 1388 | return buf[1] + b; /* "most hashed" word */ |
| 1389 | /* Alternative: return sum of all words? */ |
| 1390 | } |
| 1391 | #endif |
| 1392 | |
| 1393 | #undef ROUND |
| 1394 | #undef F |
| 1395 | #undef G |
| 1396 | #undef H |
| 1397 | #undef K1 |
| 1398 | #undef K2 |
| 1399 | #undef K3 |
| 1400 | |
| 1401 | /* This should not be decreased so low that ISNs wrap too fast. */ |
| 1402 | #define REKEY_INTERVAL (300 * HZ) |
| 1403 | /* |
| 1404 | * Bit layout of the tcp sequence numbers (before adding current time): |
| 1405 | * bit 24-31: increased after every key exchange |
| 1406 | * bit 0-23: hash(source,dest) |
| 1407 | * |
| 1408 | * The implementation is similar to the algorithm described |
| 1409 | * in the Appendix of RFC 1185, except that |
| 1410 | * - it uses a 1 MHz clock instead of a 250 kHz clock |
| 1411 | * - it performs a rekey every 5 minutes, which is equivalent |
| 1412 | * to a (source,dest) tulple dependent forward jump of the |
| 1413 | * clock by 0..2^(HASH_BITS+1) |
| 1414 | * |
| 1415 | * Thus the average ISN wraparound time is 68 minutes instead of |
| 1416 | * 4.55 hours. |
| 1417 | * |
| 1418 | * SMP cleanup and lock avoidance with poor man's RCU. |
| 1419 | * Manfred Spraul <manfred@colorfullife.com> |
| 1420 | * |
| 1421 | */ |
| 1422 | #define COUNT_BITS 8 |
| 1423 | #define COUNT_MASK ((1 << COUNT_BITS) - 1) |
| 1424 | #define HASH_BITS 24 |
| 1425 | #define HASH_MASK ((1 << HASH_BITS) - 1) |
| 1426 | |
| 1427 | static struct keydata { |
| 1428 | __u32 count; /* already shifted to the final position */ |
| 1429 | __u32 secret[12]; |
| 1430 | } ____cacheline_aligned ip_keydata[2]; |
| 1431 | |
| 1432 | static unsigned int ip_cnt; |
| 1433 | |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 1434 | static void rekey_seq_generator(struct work_struct *work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1435 | |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 1436 | static DECLARE_DELAYED_WORK(rekey_work, rekey_seq_generator); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1437 | |
| 1438 | /* |
| 1439 | * Lock avoidance: |
| 1440 | * The ISN generation runs lockless - it's just a hash over random data. |
| 1441 | * State changes happen every 5 minutes when the random key is replaced. |
| 1442 | * Synchronization is performed by having two copies of the hash function |
| 1443 | * state and rekey_seq_generator always updates the inactive copy. |
| 1444 | * The copy is then activated by updating ip_cnt. |
| 1445 | * The implementation breaks down if someone blocks the thread |
| 1446 | * that processes SYN requests for more than 5 minutes. Should never |
| 1447 | * happen, and even if that happens only a not perfectly compliant |
| 1448 | * ISN is generated, nothing fatal. |
| 1449 | */ |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 1450 | static void rekey_seq_generator(struct work_struct *work) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1451 | { |
| 1452 | struct keydata *keyptr = &ip_keydata[1 ^ (ip_cnt & 1)]; |
| 1453 | |
| 1454 | get_random_bytes(keyptr->secret, sizeof(keyptr->secret)); |
| 1455 | keyptr->count = (ip_cnt & COUNT_MASK) << HASH_BITS; |
| 1456 | smp_wmb(); |
| 1457 | ip_cnt++; |
| 1458 | schedule_delayed_work(&rekey_work, REKEY_INTERVAL); |
| 1459 | } |
| 1460 | |
| 1461 | static inline struct keydata *get_keyptr(void) |
| 1462 | { |
| 1463 | struct keydata *keyptr = &ip_keydata[ip_cnt & 1]; |
| 1464 | |
| 1465 | smp_rmb(); |
| 1466 | |
| 1467 | return keyptr; |
| 1468 | } |
| 1469 | |
| 1470 | static __init int seqgen_init(void) |
| 1471 | { |
| 1472 | rekey_seq_generator(NULL); |
| 1473 | return 0; |
| 1474 | } |
| 1475 | late_initcall(seqgen_init); |
| 1476 | |
| 1477 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
Al Viro | b09b845c | 2006-11-14 20:52:19 -0800 | [diff] [blame] | 1478 | __u32 secure_tcpv6_sequence_number(__be32 *saddr, __be32 *daddr, |
| 1479 | __be16 sport, __be16 dport) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1480 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1481 | __u32 seq; |
| 1482 | __u32 hash[12]; |
| 1483 | struct keydata *keyptr = get_keyptr(); |
| 1484 | |
| 1485 | /* The procedure is the same as for IPv4, but addresses are longer. |
| 1486 | * Thus we must use twothirdsMD4Transform. |
| 1487 | */ |
| 1488 | |
| 1489 | memcpy(hash, saddr, 16); |
Al Viro | b09b845c | 2006-11-14 20:52:19 -0800 | [diff] [blame] | 1490 | hash[4]=((__force u16)sport << 16) + (__force u16)dport; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1491 | memcpy(&hash[5],keyptr->secret,sizeof(__u32) * 7); |
| 1492 | |
Al Viro | b09b845c | 2006-11-14 20:52:19 -0800 | [diff] [blame] | 1493 | seq = twothirdsMD4Transform((const __u32 *)daddr, hash) & HASH_MASK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1494 | seq += keyptr->count; |
| 1495 | |
Eric Dumazet | 6dd10a6 | 2007-11-13 21:12:14 -0800 | [diff] [blame] | 1496 | seq += ktime_to_ns(ktime_get_real()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1497 | |
| 1498 | return seq; |
| 1499 | } |
| 1500 | EXPORT_SYMBOL(secure_tcpv6_sequence_number); |
| 1501 | #endif |
| 1502 | |
| 1503 | /* The code below is shamelessly stolen from secure_tcp_sequence_number(). |
| 1504 | * All blames to Andrey V. Savochkin <saw@msu.ru>. |
| 1505 | */ |
Al Viro | b09b845c | 2006-11-14 20:52:19 -0800 | [diff] [blame] | 1506 | __u32 secure_ip_id(__be32 daddr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1507 | { |
| 1508 | struct keydata *keyptr; |
| 1509 | __u32 hash[4]; |
| 1510 | |
| 1511 | keyptr = get_keyptr(); |
| 1512 | |
| 1513 | /* |
| 1514 | * Pick a unique starting offset for each IP destination. |
| 1515 | * The dest ip address is placed in the starting vector, |
| 1516 | * which is then hashed with random data. |
| 1517 | */ |
Al Viro | b09b845c | 2006-11-14 20:52:19 -0800 | [diff] [blame] | 1518 | hash[0] = (__force __u32)daddr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1519 | hash[1] = keyptr->secret[9]; |
| 1520 | hash[2] = keyptr->secret[10]; |
| 1521 | hash[3] = keyptr->secret[11]; |
| 1522 | |
| 1523 | return half_md4_transform(hash, keyptr->secret); |
| 1524 | } |
| 1525 | |
| 1526 | #ifdef CONFIG_INET |
| 1527 | |
Al Viro | b09b845c | 2006-11-14 20:52:19 -0800 | [diff] [blame] | 1528 | __u32 secure_tcp_sequence_number(__be32 saddr, __be32 daddr, |
| 1529 | __be16 sport, __be16 dport) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1530 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1531 | __u32 seq; |
| 1532 | __u32 hash[4]; |
| 1533 | struct keydata *keyptr = get_keyptr(); |
| 1534 | |
| 1535 | /* |
| 1536 | * Pick a unique starting offset for each TCP connection endpoints |
| 1537 | * (saddr, daddr, sport, dport). |
| 1538 | * Note that the words are placed into the starting vector, which is |
| 1539 | * then mixed with a partial MD4 over random data. |
| 1540 | */ |
Al Viro | b09b845c | 2006-11-14 20:52:19 -0800 | [diff] [blame] | 1541 | hash[0]=(__force u32)saddr; |
| 1542 | hash[1]=(__force u32)daddr; |
| 1543 | hash[2]=((__force u16)sport << 16) + (__force u16)dport; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1544 | hash[3]=keyptr->secret[11]; |
| 1545 | |
| 1546 | seq = half_md4_transform(hash, keyptr->secret) & HASH_MASK; |
| 1547 | seq += keyptr->count; |
| 1548 | /* |
| 1549 | * As close as possible to RFC 793, which |
| 1550 | * suggests using a 250 kHz clock. |
| 1551 | * Further reading shows this assumes 2 Mb/s networks. |
Eric Dumazet | 9b42c33 | 2007-10-01 13:58:36 -0700 | [diff] [blame] | 1552 | * For 10 Mb/s Ethernet, a 1 MHz clock is appropriate. |
| 1553 | * For 10 Gb/s Ethernet, a 1 GHz clock should be ok, but |
| 1554 | * we also need to limit the resolution so that the u32 seq |
| 1555 | * overlaps less than one time per MSL (2 minutes). |
| 1556 | * Choosing a clock of 64 ns period is OK. (period of 274 s) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1557 | */ |
Eric Dumazet | 6dd10a6 | 2007-11-13 21:12:14 -0800 | [diff] [blame] | 1558 | seq += ktime_to_ns(ktime_get_real()) >> 6; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1559 | #if 0 |
| 1560 | printk("init_seq(%lx, %lx, %d, %d) = %d\n", |
| 1561 | saddr, daddr, sport, dport, seq); |
| 1562 | #endif |
| 1563 | return seq; |
| 1564 | } |
| 1565 | |
Arnaldo Carvalho de Melo | a7f5e7f | 2005-12-13 23:25:31 -0800 | [diff] [blame] | 1566 | /* Generate secure starting point for ephemeral IPV4 transport port search */ |
Al Viro | b09b845c | 2006-11-14 20:52:19 -0800 | [diff] [blame] | 1567 | u32 secure_ipv4_port_ephemeral(__be32 saddr, __be32 daddr, __be16 dport) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1568 | { |
| 1569 | struct keydata *keyptr = get_keyptr(); |
| 1570 | u32 hash[4]; |
| 1571 | |
| 1572 | /* |
| 1573 | * Pick a unique starting offset for each ephemeral port search |
| 1574 | * (saddr, daddr, dport) and 48bits of random data. |
| 1575 | */ |
Al Viro | b09b845c | 2006-11-14 20:52:19 -0800 | [diff] [blame] | 1576 | hash[0] = (__force u32)saddr; |
| 1577 | hash[1] = (__force u32)daddr; |
| 1578 | hash[2] = (__force u32)dport ^ keyptr->secret[10]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1579 | hash[3] = keyptr->secret[11]; |
| 1580 | |
| 1581 | return half_md4_transform(hash, keyptr->secret); |
| 1582 | } |
| 1583 | |
| 1584 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
Al Viro | b09b845c | 2006-11-14 20:52:19 -0800 | [diff] [blame] | 1585 | u32 secure_ipv6_port_ephemeral(const __be32 *saddr, const __be32 *daddr, __be16 dport) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1586 | { |
| 1587 | struct keydata *keyptr = get_keyptr(); |
| 1588 | u32 hash[12]; |
| 1589 | |
| 1590 | memcpy(hash, saddr, 16); |
Al Viro | b09b845c | 2006-11-14 20:52:19 -0800 | [diff] [blame] | 1591 | hash[4] = (__force u32)dport; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1592 | memcpy(&hash[5],keyptr->secret,sizeof(__u32) * 7); |
| 1593 | |
Al Viro | b09b845c | 2006-11-14 20:52:19 -0800 | [diff] [blame] | 1594 | return twothirdsMD4Transform((const __u32 *)daddr, hash); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1595 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1596 | #endif |
| 1597 | |
Arnaldo Carvalho de Melo | c4365c9 | 2005-08-09 20:12:30 -0700 | [diff] [blame] | 1598 | #if defined(CONFIG_IP_DCCP) || defined(CONFIG_IP_DCCP_MODULE) |
| 1599 | /* Similar to secure_tcp_sequence_number but generate a 48 bit value |
| 1600 | * bit's 32-47 increase every key exchange |
| 1601 | * 0-31 hash(source, dest) |
| 1602 | */ |
Al Viro | b09b845c | 2006-11-14 20:52:19 -0800 | [diff] [blame] | 1603 | u64 secure_dccp_sequence_number(__be32 saddr, __be32 daddr, |
| 1604 | __be16 sport, __be16 dport) |
Arnaldo Carvalho de Melo | c4365c9 | 2005-08-09 20:12:30 -0700 | [diff] [blame] | 1605 | { |
Arnaldo Carvalho de Melo | c4365c9 | 2005-08-09 20:12:30 -0700 | [diff] [blame] | 1606 | u64 seq; |
| 1607 | __u32 hash[4]; |
| 1608 | struct keydata *keyptr = get_keyptr(); |
| 1609 | |
Al Viro | b09b845c | 2006-11-14 20:52:19 -0800 | [diff] [blame] | 1610 | hash[0] = (__force u32)saddr; |
| 1611 | hash[1] = (__force u32)daddr; |
| 1612 | hash[2] = ((__force u16)sport << 16) + (__force u16)dport; |
Arnaldo Carvalho de Melo | c4365c9 | 2005-08-09 20:12:30 -0700 | [diff] [blame] | 1613 | hash[3] = keyptr->secret[11]; |
| 1614 | |
| 1615 | seq = half_md4_transform(hash, keyptr->secret); |
| 1616 | seq |= ((u64)keyptr->count) << (32 - HASH_BITS); |
| 1617 | |
Eric Dumazet | 6dd10a6 | 2007-11-13 21:12:14 -0800 | [diff] [blame] | 1618 | seq += ktime_to_ns(ktime_get_real()); |
Arnaldo Carvalho de Melo | c4365c9 | 2005-08-09 20:12:30 -0700 | [diff] [blame] | 1619 | seq &= (1ull << 48) - 1; |
| 1620 | #if 0 |
| 1621 | printk("dccp init_seq(%lx, %lx, %d, %d) = %d\n", |
| 1622 | saddr, daddr, sport, dport, seq); |
| 1623 | #endif |
| 1624 | return seq; |
| 1625 | } |
| 1626 | |
| 1627 | EXPORT_SYMBOL(secure_dccp_sequence_number); |
| 1628 | #endif |
| 1629 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1630 | #endif /* CONFIG_INET */ |
| 1631 | |
| 1632 | |
| 1633 | /* |
| 1634 | * Get a random word for internal kernel use only. Similar to urandom but |
| 1635 | * with the goal of minimal entropy pool depletion. As a result, the random |
| 1636 | * value is not cryptographically secure but for several uses the cost of |
| 1637 | * depleting entropy is too high |
| 1638 | */ |
| 1639 | unsigned int get_random_int(void) |
| 1640 | { |
| 1641 | /* |
| 1642 | * Use IP's RNG. It suits our purpose perfectly: it re-keys itself |
| 1643 | * every second, from the entropy pool (and thus creates a limited |
| 1644 | * drain on it), and uses halfMD4Transform within the second. We |
| 1645 | * also mix it with jiffies and the PID: |
| 1646 | */ |
Al Viro | b09b845c | 2006-11-14 20:52:19 -0800 | [diff] [blame] | 1647 | return secure_ip_id((__force __be32)(current->pid + jiffies)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1648 | } |
| 1649 | |
| 1650 | /* |
| 1651 | * randomize_range() returns a start address such that |
| 1652 | * |
| 1653 | * [...... <range> .....] |
| 1654 | * start end |
| 1655 | * |
| 1656 | * a <range> with size "len" starting at the return value is inside in the |
| 1657 | * area defined by [start, end], but is otherwise randomized. |
| 1658 | */ |
| 1659 | unsigned long |
| 1660 | randomize_range(unsigned long start, unsigned long end, unsigned long len) |
| 1661 | { |
| 1662 | unsigned long range = end - len - start; |
| 1663 | |
| 1664 | if (end <= start + len) |
| 1665 | return 0; |
| 1666 | return PAGE_ALIGN(get_random_int() % range + start); |
| 1667 | } |