1 /* 2 * Copyright (c) 2004, 2005, 2006 Robin J Carey. All rights reserved. 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions 6 * are met: 7 * 1. Redistributions of source code must retain the above copyright 8 * notice, this list of conditions, and the following disclaimer, 9 * without modification, immediately at the beginning of the file. 10 * 2. The name of the author may not be used to endorse or promote products 11 * derived from this software without specific prior written permission. 12 * 13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR 17 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 23 * SUCH DAMAGE. 24 * 25 * $DragonFly: src/sys/kern/kern_nrandom.c,v 1.1 2006/06/18 01:34:59 dillon Exp $ 26 */ 27 /* --- NOTES --- 28 * 29 * Note: The word "entropy" is often incorrectly used to describe 30 * random data. The word "entropy" originates from the science of 31 * Physics. The correct descriptive definition would be something 32 * along the lines of "seed", "unpredictable numbers" or 33 * "unpredictable data". 34 * 35 * Note: Some /dev/[u]random implementations save "seed" between 36 * boots which represents a security hazard since an adversary 37 * could acquire this data (since it is stored in a file). If 38 * the unpredictable data used in the above routines is only 39 * generated during Kernel operation, then an adversary can only 40 * acquire that data through a Kernel security compromise and/or 41 * a cryptographic algorithm failure/cryptanalysis. 42 * 43 * Note: On FreeBSD-4.11, interrupts have to be manually enabled 44 * using the rndcontrol(8) command. 45 * 46 * --- DESIGN (FreeBSD-4.11 based) --- 47 * 48 * The rnddev module automatically initializes itself the first time 49 * it is used (client calls any public rnddev_*() interface routine). 50 * Both CSPRNGs are initially seeded from the precise nano[up]time() routines. 51 * Tests show this method produces good enough results, suitable for intended 52 * use. It is necessary for both CSPRNGs to be completely seeded, initially. 53 * 54 * After initialization and during Kernel operation the only suitable 55 * unpredictable data available is: 56 * 57 * (1) Keyboard scan-codes. 58 * (2) Nanouptime acquired by a Keyboard/Read-Event. 59 * (3) Suitable interrupt source; hard-disk/ATA-device. 60 * 61 * (X) Mouse-event (xyz-data unsuitable); NOT IMPLEMENTED. 62 * 63 * This data is added to both CSPRNGs in real-time as it happens/ 64 * becomes-available. Additionally, unpredictable (?) data may be 65 * acquired from a true-random number generator if such a device is 66 * available to the system (not advisable !). 67 * Nanouptime() acquired by a Read-Event is a very important aspect of 68 * this design, since it ensures that unpredictable data is added to 69 * the CSPRNGs even if there are no other sources. 70 * The nanouptime() Kernel routine is used since time relative to 71 * boot is less adversary-known than time itself. 72 * 73 * This design has been thoroughly tested with debug logging 74 * and the output from both /dev/random and /dev/urandom has 75 * been tested with the DIEHARD test-suite; both pass. 76 * 77 * MODIFICATIONS MADE TO ORIGINAL "kern_random.c": 78 * 79 * 6th July 2005: 80 * 81 * o Changed ReadSeed() function to schedule future read-seed-events 82 * by at least one second. Previous implementation used a randomised 83 * scheduling { 0, 1, 2, 3 seconds }. 84 * o Changed SEED_NANOUP() function to use a "previous" accumulator 85 * algorithm similar to ReadSeed(). This ensures that there is no 86 * way that an adversary can tell what number is being added to the 87 * CSPRNGs, since the number added to the CSPRNGs at Event-Time is 88 * the sum of nanouptime()@Event and an unknown/secret number. 89 * o Changed rnddev_add_interrupt() function to schedule future 90 * interrupt-events by at least one second. Previous implementation 91 * had no scheduling algorithm which allowed an "interrupt storm" 92 * to occur resulting in skewed data entering into the CSPRNGs. 93 * 94 * 95 * 9th July 2005: 96 * 97 * o Some small cleanups and change all internal functions to be 98 * static/private. 99 * o Removed ReadSeed() since its functionality is already performed 100 * by another function { rnddev_add_interrupt_OR_read() } and remove 101 * the silly rndByte accumulator/feedback-thing (since multipying by 102 * rndByte could yield a value of 0). 103 * o Made IBAA/L14 public interface become static/private; 104 * Local to this file (not changed to that in the original C modules). 105 * 106 * 16th July 2005: 107 * 108 * o SEED_NANOUP() -> NANOUP_EVENT() function rename. 109 * o Make NANOUP_EVENT() handle the time-buffering directly so that all 110 * time-stamp-events use this single time-buffer (including keyboard). 111 * This removes dependancy on "time_second" Kernel variable. 112 * o Removed second-time-buffer code in rnddev_add_interrupt_OR_read (void). 113 * o Rewrote the time-buffering algorithm in NANOUP_EVENT() to use a 114 * randomised time-delay range. 115 * 116 * 12th Dec 2005: 117 * 118 * o Updated to (hopefully final) L15 algorithm. 119 * 120 * 12th June 2006: 121 * 122 * o Added missing (u_char *) cast in RnddevRead() function. 123 * o Changed copyright to 3-clause BSD license and cleaned up the layout 124 * of this file. 125 */ 126 127 #include <sys/types.h> 128 #include <sys/kernel.h> 129 #include <sys/systm.h> 130 #include <sys/poll.h> 131 #include <sys/random.h> 132 #include <sys/systimer.h> 133 #include <sys/time.h> 134 #include <sys/proc.h> 135 #include <sys/lock.h> 136 #include <sys/sysctl.h> 137 #include <sys/spinlock.h> 138 #include <machine/clock.h> 139 140 #include <sys/thread2.h> 141 #include <sys/spinlock2.h> 142 143 /* 144 * Portability note: The u_char/unsigned char type is used where 145 * uint8_t from <stdint.h> or u_int8_t from <sys/types.h> should really 146 * be being used. On FreeBSD, it is safe to make the assumption that these 147 * different types are equivalent (on all architectures). 148 * The FreeBSD <sys/crypto/rc4> module also makes this assumption. 149 */ 150 151 /*------------------------------ IBAA ----------------------------------*/ 152 153 /*-------------------------- IBAA CSPRNG -------------------------------*/ 154 155 /* 156 * NOTE: The original source code from which this source code (IBAA) 157 * was taken has no copyright/license. The algorithm has no patent 158 * and is freely/publicly available from: 159 * 160 * http://www.burtleburtle.net/bob/rand/isaac.html 161 */ 162 163 /* 164 * ^ means XOR, & means bitwise AND, a<<b means shift a by b. 165 * barrel(a) shifts a 19 bits to the left, and bits wrap around 166 * ind(x) is (x AND 255), or (x mod 256) 167 */ 168 typedef u_int32_t u4; /* unsigned four bytes, 32 bits */ 169 170 #define ALPHA (8) 171 #define SIZE (1 << ALPHA) 172 #define ind(x) ((x) & (SIZE - 1)) 173 #define barrel(a) (((a) << 19) ^ ((a) >> 13)) /* beta=32,shift=19 */ 174 175 static void IBAA 176 ( 177 u4 *m, /* Memory: array of SIZE ALPHA-bit terms */ 178 u4 *r, /* Results: the sequence, same size as m */ 179 u4 *aa, /* Accumulator: a single value */ 180 u4 *bb /* the previous result */ 181 ) 182 { 183 u4 a, b, x, y, i; 184 185 a = *aa; b = *bb; 186 for (i = 0; i < SIZE; ++i) { 187 x = m[i]; 188 a = barrel(a) + m[ind(i + (SIZE / 2))]; /* set a */ 189 m[i] = y = m[ind(x)] + a + b; /* set m */ 190 r[i] = b = m[ind(y >> ALPHA)] + x; /* set r */ 191 } 192 *bb = b; *aa = a; 193 } 194 195 /*-------------------------- IBAA CSPRNG -------------------------------*/ 196 197 198 static u4 IBAA_memory[SIZE]; 199 static u4 IBAA_results[SIZE]; 200 static u4 IBAA_aa; 201 static u4 IBAA_bb; 202 203 static volatile int IBAA_byte_index; 204 205 206 static void IBAA_Init(void); 207 static void IBAA_Call(void); 208 static void IBAA_Seed(const u_int32_t val); 209 static u_char IBAA_Byte(void); 210 211 /* 212 * Initialize IBAA. 213 */ 214 static void 215 IBAA_Init(void) 216 { 217 size_t i; 218 219 for (i = 0; i < SIZE; ++i) { 220 IBAA_memory[i] = i; 221 } 222 IBAA_aa = IBAA_bb = 0; 223 IBAA_byte_index = sizeof(IBAA_results); /* force IBAA_Call() */ 224 } 225 226 /* 227 * PRIVATE: Call IBAA to produce 256 32-bit u4 results. 228 */ 229 static void 230 IBAA_Call (void) 231 { 232 IBAA(IBAA_memory, IBAA_results, &IBAA_aa, &IBAA_bb); 233 IBAA_byte_index = 0; 234 } 235 236 /* 237 * Add a 32-bit u4 seed value into IBAAs memory. 238 */ 239 static void 240 IBAA_Seed (const u_int32_t val) 241 { 242 static int memIndex = 0; 243 244 IBAA_memory[memIndex % SIZE] = val; 245 ++memIndex; 246 } 247 248 /* 249 * Extract a byte from IBAAs 256 32-bit u4 results array. 250 * 251 * NOTE: This code is designed to prevent MP races from taking 252 * IBAA_byte_index out of bounds. 253 */ 254 static u_char 255 IBAA_Byte(void) 256 { 257 u_char result; 258 int index; 259 260 index = IBAA_byte_index; 261 if (index == sizeof(IBAA_results)) { 262 IBAA_Call(); 263 index = 0; 264 } 265 result = ((u_char *)IBAA_results)[index]; 266 IBAA_byte_index = index + 1; 267 return result; 268 } 269 270 /*------------------------------ IBAA ----------------------------------*/ 271 272 273 /*------------------------------- L15 ----------------------------------*/ 274 275 /* 276 * IMPORTANT NOTE: LByteType must be exactly 8-bits in size or this software 277 * will not function correctly. 278 */ 279 typedef unsigned char LByteType; 280 281 #define L15_STATE_SIZE 256 282 283 static LByteType L15_x, L15_y; 284 static LByteType L15_start_x; 285 static LByteType L15_state[L15_STATE_SIZE]; 286 287 /* 288 * PRIVATE FUNCS: 289 */ 290 291 static void L15_Swap(const LByteType pos1, const LByteType pos2); 292 static void L15_InitState(void); 293 static void L15_KSA(const LByteType * const key, 294 const size_t keyLen); 295 static void L15_Discard(const LByteType numCalls); 296 297 /* 298 * PUBLIC INTERFACE: 299 */ 300 static void L15(const LByteType * const key, const size_t keyLen); 301 static LByteType L15_Byte(void); 302 static void L15_Vector(const LByteType * const key, 303 const size_t keyLen); 304 305 static __inline void 306 L15_Swap(const LByteType pos1, const LByteType pos2) 307 { 308 const LByteType save1 = L15_state[pos1]; 309 310 L15_state[pos1] = L15_state[pos2]; 311 L15_state[pos2] = save1; 312 } 313 314 static void 315 L15_InitState (void) 316 { 317 size_t i; 318 for (i = 0; i < L15_STATE_SIZE; ++i) 319 L15_state[i] = i; 320 } 321 322 #define L_SCHEDULE(xx) \ 323 \ 324 for (i = 0; i < L15_STATE_SIZE; ++i) { \ 325 L15_Swap(i, (stateIndex += (L15_state[i] + (xx)))); \ 326 } 327 328 static void 329 L15_KSA (const LByteType * const key, const size_t keyLen) 330 { 331 size_t i, keyIndex; 332 LByteType stateIndex = 0; 333 334 L_SCHEDULE(keyLen); 335 for (keyIndex = 0; keyIndex < keyLen; ++keyIndex) { 336 L_SCHEDULE(key[keyIndex]); 337 } 338 } 339 340 static void 341 L15_Discard(const LByteType numCalls) 342 { 343 LByteType i; 344 for (i = 0; i < numCalls; ++i) { 345 (void)L15_Byte(); 346 } 347 } 348 349 350 /* 351 * PUBLIC INTERFACE: 352 */ 353 static void 354 L15(const LByteType * const key, const size_t keyLen) 355 { 356 L15_x = L15_y = L15_start_x = 0; 357 L15_InitState(); 358 L15_KSA(key, keyLen); 359 L15_Discard(L15_Byte()); 360 } 361 362 static LByteType 363 L15_Byte(void) 364 { 365 LByteType z; 366 367 L15_Swap(L15_state[L15_x], L15_y); 368 z = (L15_state [L15_x++] + L15_state[L15_y--]); 369 if (L15_x == L15_start_x) { 370 --L15_y; 371 } 372 return (L15_state[z]); 373 } 374 375 static void 376 L15_Vector (const LByteType * const key, const size_t keyLen) 377 { 378 L15_KSA(key, keyLen); 379 } 380 381 /*------------------------------- L15 ----------------------------------*/ 382 383 /************************************************************************ 384 * KERNEL INTERFACE * 385 ************************************************************************ 386 * 387 * By Robin J Carey and Matthew Dillon. 388 */ 389 390 static int rand_thread_signal = 1; 391 static void NANOUP_EVENT(void); 392 static thread_t rand_td; 393 static struct spinlock rand_spin; 394 395 static int nrandevents; 396 SYSCTL_INT(_kern, OID_AUTO, nrandevents, CTLFLAG_RD, &nrandevents, 0, ""); 397 398 399 /* 400 * Called from early boot 401 */ 402 void 403 rand_initialize(void) 404 { 405 struct timespec now; 406 int i; 407 408 spin_init(&rand_spin); 409 410 /* Initialize IBAA. */ 411 IBAA_Init(); 412 413 /* Initialize L15. */ 414 nanouptime(&now); 415 L15((const LByteType *)&now.tv_nsec, sizeof(now.tv_nsec)); 416 for (i = 0; i < (SIZE / 2); ++i) { 417 nanotime(&now); 418 IBAA_Seed(now.tv_nsec); 419 L15_Vector((const LByteType *)&now.tv_nsec, 420 sizeof(now.tv_nsec)); 421 nanouptime(&now); 422 IBAA_Seed(now.tv_nsec); 423 L15_Vector((const LByteType *)&now.tv_nsec, 424 sizeof(now.tv_nsec)); 425 } 426 } 427 428 /* 429 * Keyboard events 430 */ 431 void 432 add_keyboard_randomness(u_char scancode) 433 { 434 spin_lock_wr(&rand_spin); 435 L15_Vector((const LByteType *) &scancode, sizeof (scancode)); 436 spin_unlock_wr(&rand_spin); 437 add_interrupt_randomness(0); 438 } 439 440 /* 441 * Interrupt events 442 */ 443 void 444 add_interrupt_randomness(int intr) 445 { 446 if (rand_thread_signal == 0) { 447 rand_thread_signal = 1; 448 lwkt_schedule(rand_td); 449 } 450 } 451 452 /* 453 * True random number source 454 */ 455 void 456 add_true_randomness(int val) 457 { 458 spin_lock_wr(&rand_spin); 459 IBAA_Seed(val); 460 L15_Vector((const LByteType *) &val, sizeof (val)); 461 ++nrandevents; 462 spin_unlock_wr(&rand_spin); 463 } 464 465 /* 466 * Poll (always succeeds) 467 */ 468 int 469 random_poll(dev_t dev, int events, struct thread *td) 470 { 471 int revents = 0; 472 473 if (events & (POLLIN | POLLRDNORM)) 474 revents |= events & (POLLIN | POLLRDNORM); 475 if (events & (POLLOUT | POLLWRNORM)) 476 revents |= events & (POLLOUT | POLLWRNORM); 477 478 return (revents); 479 } 480 481 /* 482 * Heavy weight random number generator. May return less then the 483 * requested number of bytes. 484 */ 485 u_int 486 read_random(void *buf, u_int nbytes) 487 { 488 u_int i; 489 490 spin_lock_wr(&rand_spin); 491 for (i = 0; i < nbytes; ++i) 492 ((u_char *)buf)[i] = IBAA_Byte(); 493 spin_unlock_wr(&rand_spin); 494 add_interrupt_randomness(0); 495 return(i); 496 } 497 498 /* 499 * Lightweight random number generator. Must return requested number of 500 * bytes. 501 */ 502 u_int 503 read_random_unlimited(void *buf, u_int nbytes) 504 { 505 u_int i; 506 507 spin_lock_wr(&rand_spin); 508 for (i = 0; i < nbytes; ++i) 509 ((u_char *)buf)[i] = L15_Byte(); 510 spin_unlock_wr(&rand_spin); 511 add_interrupt_randomness(0); 512 return (i); 513 } 514 515 /* 516 * Random number generator helper thread. This limits code overhead from 517 * high frequency events by delaying the clearing of rand_thread_signal. 518 */ 519 static 520 void 521 rand_thread_loop(void *dummy) 522 { 523 int count; 524 525 for (;;) { 526 NANOUP_EVENT (); 527 spin_lock_wr(&rand_spin); 528 count = (int)(L15_Byte() * hz / (256 * 10) + hz / 10); 529 spin_unlock_wr(&rand_spin); 530 tsleep(rand_td, 0, "rwait", count); 531 rand_thread_signal = 0; 532 lwkt_deschedule_self(rand_td); 533 lwkt_switch(); 534 } 535 } 536 537 static 538 void 539 rand_thread_init(void) 540 { 541 lwkt_create(rand_thread_loop, NULL, &rand_td, NULL, 0, 0, "random"); 542 } 543 544 SYSINIT(rand, SI_SUB_HELPER_THREADS, SI_ORDER_ANY, rand_thread_init, 0); 545 546 /* 547 * Time-buffered event time-stamping. This is necessary to cutoff higher 548 * event frequencies, e.g. an interrupt occuring at 25Hz. In such cases 549 * the CPU is being chewed and the timestamps are skewed (minimal variation). 550 * Use a nano-second time-delay to limit how many times an Event can occur 551 * in one second; <= 5Hz. Note that this doesn't prevent time-stamp skewing. 552 * This implementation randmoises the time-delay between events, which adds 553 * a layer of security/unpredictability with regard to read-events (a user 554 * controlled input). 555 * 556 * Note: now.tv_nsec should range [ 0 - 1000,000,000 ]. 557 * Note: "ACCUM" is a security measure (result = capped-unknown + unknown), 558 * and also produces an uncapped (>=32-bit) value. 559 */ 560 static void 561 NANOUP_EVENT(void) 562 { 563 static struct timespec ACCUM = { 0, 0 }; 564 static struct timespec NEXT = { 0, 0 }; 565 struct timespec now; 566 567 nanouptime(&now); 568 spin_lock_wr(&rand_spin); 569 if ((now.tv_nsec > NEXT.tv_nsec) || (now.tv_sec != NEXT.tv_sec)) { 570 /* 571 * Randomised time-delay: 200e6 - 350e6 ns; 5 - 2.86 Hz. 572 */ 573 unsigned long one_mil; 574 unsigned long timeDelay; 575 576 one_mil = 1000000UL; /* 0.001 s */ 577 timeDelay = (one_mil * 200) + (((unsigned)ACCUM.tv_nsec % 151) * one_mil); 578 NEXT.tv_nsec = now.tv_nsec + timeDelay; 579 NEXT.tv_sec = now.tv_sec; 580 ACCUM.tv_nsec += now.tv_nsec; 581 582 /* 583 * The TSC, if present, generally has an even higher 584 * resolution. Integrate a portion of it into our seed. 585 */ 586 if (tsc_present) 587 ACCUM.tv_nsec ^= rdtsc() & 255; 588 589 IBAA_Seed(ACCUM.tv_nsec); 590 L15_Vector((const LByteType *)&ACCUM.tv_nsec, 591 sizeof(ACCUM.tv_nsec)); 592 ++nrandevents; 593 } 594 spin_unlock_wr(&rand_spin); 595 } 596 597