1 /*- 2 * Copyright (c) 2017 W. Dean Freeman 3 * Copyright (c) 2013-2015 Mark R V Murray 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer 11 * in this position and unchanged. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 19 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 21 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 22 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 23 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 24 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 25 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 * 27 */ 28 29 /* 30 * This implementation of Fortuna is based on the descriptions found in 31 * ISBN 978-0-470-47424-2 "Cryptography Engineering" by Ferguson, Schneier 32 * and Kohno ("FS&K"). 33 */ 34 35 #include <sys/param.h> 36 #include <sys/limits.h> 37 38 #ifdef _KERNEL 39 #include <sys/fail.h> 40 #include <sys/kernel.h> 41 #include <sys/lock.h> 42 #include <sys/malloc.h> 43 #include <sys/mutex.h> 44 #include <sys/random.h> 45 #include <sys/sdt.h> 46 #include <sys/sysctl.h> 47 #include <sys/systm.h> 48 49 #include <machine/cpu.h> 50 #else /* !_KERNEL */ 51 #include <inttypes.h> 52 #include <stdbool.h> 53 #include <stdio.h> 54 #include <stdlib.h> 55 #include <string.h> 56 #include <threads.h> 57 58 #include "unit_test.h" 59 #endif /* _KERNEL */ 60 61 #include <crypto/chacha20/chacha.h> 62 #include <crypto/rijndael/rijndael-api-fst.h> 63 #include <crypto/sha2/sha256.h> 64 65 #include <dev/random/hash.h> 66 #include <dev/random/randomdev.h> 67 #ifdef _KERNEL 68 #include <dev/random/random_harvestq.h> 69 #endif 70 #include <dev/random/uint128.h> 71 #include <dev/random/fortuna.h> 72 73 /* Defined in FS&K */ 74 #define RANDOM_FORTUNA_NPOOLS 32 /* The number of accumulation pools */ 75 #define RANDOM_FORTUNA_DEFPOOLSIZE 64 /* The default pool size/length for a (re)seed */ 76 #define RANDOM_FORTUNA_MAX_READ (1 << 20) /* Max bytes from AES before rekeying */ 77 #define RANDOM_FORTUNA_BLOCKS_PER_KEY (1 << 16) /* Max blocks from AES before rekeying */ 78 CTASSERT(RANDOM_FORTUNA_BLOCKS_PER_KEY * RANDOM_BLOCKSIZE == 79 RANDOM_FORTUNA_MAX_READ); 80 81 /* 82 * The allowable range of RANDOM_FORTUNA_DEFPOOLSIZE. The default value is above. 83 * Making RANDOM_FORTUNA_DEFPOOLSIZE too large will mean a long time between reseeds, 84 * and too small may compromise initial security but get faster reseeds. 85 */ 86 #define RANDOM_FORTUNA_MINPOOLSIZE 16 87 #define RANDOM_FORTUNA_MAXPOOLSIZE INT_MAX 88 CTASSERT(RANDOM_FORTUNA_MINPOOLSIZE <= RANDOM_FORTUNA_DEFPOOLSIZE); 89 CTASSERT(RANDOM_FORTUNA_DEFPOOLSIZE <= RANDOM_FORTUNA_MAXPOOLSIZE); 90 91 /* This algorithm (and code) presumes that RANDOM_KEYSIZE is twice as large as RANDOM_BLOCKSIZE */ 92 CTASSERT(RANDOM_BLOCKSIZE == sizeof(uint128_t)); 93 CTASSERT(RANDOM_KEYSIZE == 2*RANDOM_BLOCKSIZE); 94 95 /* Probes for dtrace(1) */ 96 #ifdef _KERNEL 97 SDT_PROVIDER_DECLARE(random); 98 SDT_PROVIDER_DEFINE(random); 99 SDT_PROBE_DEFINE2(random, fortuna, event_processor, debug, "u_int", "struct fs_pool *"); 100 #endif /* _KERNEL */ 101 102 /* 103 * This is the beastie that needs protecting. It contains all of the 104 * state that we are excited about. Exactly one is instantiated. 105 */ 106 static struct fortuna_state { 107 struct fs_pool { /* P_i */ 108 u_int fsp_length; /* Only the first one is used by Fortuna */ 109 struct randomdev_hash fsp_hash; 110 } fs_pool[RANDOM_FORTUNA_NPOOLS]; 111 u_int fs_reseedcount; /* ReseedCnt */ 112 uint128_t fs_counter; /* C */ 113 union randomdev_key fs_key; /* K */ 114 u_int fs_minpoolsize; /* Extras */ 115 /* Extras for the OS */ 116 #ifdef _KERNEL 117 /* For use when 'pacing' the reseeds */ 118 sbintime_t fs_lasttime; 119 #endif 120 /* Reseed lock */ 121 mtx_t fs_mtx; 122 } fortuna_state; 123 124 /* 125 * This knob enables or disables the "Concurrent Reads" Fortuna feature. 126 * 127 * The benefit of Concurrent Reads is improved concurrency in Fortuna. That is 128 * reflected in two related aspects: 129 * 130 * 1. Concurrent full-rate devrandom readers can achieve similar throughput to 131 * a single reader thread (at least up to a modest number of cores; the 132 * non-concurrent design falls over at 2 readers). 133 * 134 * 2. The rand_harvestq process spends much less time spinning when one or more 135 * readers is processing a large request. Partially this is due to 136 * rand_harvestq / ra_event_processor design, which only passes one event at 137 * a time to the underlying algorithm. Each time, Fortuna must take its 138 * global state mutex, potentially blocking on a reader. Our adaptive 139 * mutexes assume that a lock holder currently on CPU will release the lock 140 * quickly, and spin if the owning thread is currently running. 141 * 142 * (There is no reason rand_harvestq necessarily has to use the same lock as 143 * the generator, or that it must necessarily drop and retake locks 144 * repeatedly, but that is the current status quo.) 145 * 146 * The concern is that the reduced lock scope might results in a less safe 147 * random(4) design. However, the reduced-lock scope design is still 148 * fundamentally Fortuna. This is discussed below. 149 * 150 * Fortuna Read() only needs mutual exclusion between readers to correctly 151 * update the shared read-side state: C, the 128-bit counter; and K, the 152 * current cipher/PRF key. 153 * 154 * In the Fortuna design, the global counter C should provide an independent 155 * range of values per request. 156 * 157 * Under lock, we can save a copy of C on the stack, and increment the global C 158 * by the number of blocks a Read request will require. 159 * 160 * Still under lock, we can save a copy of the key K on the stack, and then 161 * perform the usual key erasure K' <- Keystream(C, K, ...). This does require 162 * generating 256 bits (32 bytes) of cryptographic keystream output with the 163 * global lock held, but that's all; none of the API keystream generation must 164 * be performed under lock. 165 * 166 * At this point, we may unlock. 167 * 168 * Some example timelines below (to oversimplify, all requests are in units of 169 * native blocks, and the keysize happens to be equal or less to the native 170 * blocksize of the underlying cipher, and the same sequence of two requests 171 * arrive in the same order). The possibly expensive consumer keystream 172 * generation portion is marked with '**'. 173 * 174 * Status Quo fortuna_read() Reduced-scope locking 175 * ------------------------- --------------------- 176 * C=C_0, K=K_0 C=C_0, K=K_0 177 * <Thr 1 requests N blocks> <Thr 1 requests N blocks> 178 * 1:Lock() 1:Lock() 179 * <Thr 2 requests M blocks> <Thr 2 requests M blocks> 180 * 1:GenBytes() 1:stack_C := C_0 181 * 1: Keystream(C_0, K_0, N) 1:stack_K := K_0 182 * 1: <N blocks generated>** 1:C' := C_0 + N 183 * 1: C' := C_0 + N 1:K' := Keystream(C', K_0, 1) 184 * 1: <- Keystream 1: <1 block generated> 185 * 1: K' := Keystream(C', K_0, 1) 1: C'' := C' + 1 186 * 1: <1 block generated> 1: <- Keystream 187 * 1: C'' := C' + 1 1:Unlock() 188 * 1: <- Keystream 189 * 1: <- GenBytes() 190 * 1:Unlock() 191 * 192 * Just prior to unlock, shared state is identical: 193 * ------------------------------------------------ 194 * C'' == C_0 + N + 1 C'' == C_0 + N + 1 195 * K' == keystream generated from K' == keystream generated from 196 * C_0 + N, K_0. C_0 + N, K_0. 197 * K_0 has been erased. K_0 has been erased. 198 * 199 * After both designs unlock, the 2nd reader is unblocked. 200 * 201 * 2:Lock() 2:Lock() 202 * 2:GenBytes() 2:stack_C' := C'' 203 * 2: Keystream(C'', K', M) 2:stack_K' := K' 204 * 2: <M blocks generated>** 2:C''' := C'' + M 205 * 2: C''' := C'' + M 2:K'' := Keystream(C''', K', 1) 206 * 2: <- Keystream 2: <1 block generated> 207 * 2: K'' := Keystream(C''', K', 1) 2: C'''' := C''' + 1 208 * 2: <1 block generated> 2: <- Keystream 209 * 2: C'''' := C''' + 1 2:Unlock() 210 * 2: <- Keystream 211 * 2: <- GenBytes() 212 * 2:Unlock() 213 * 214 * Just prior to unlock, global state is identical: 215 * ------------------------------------------------------ 216 * 217 * C'''' == (C_0 + N + 1) + M + 1 C'''' == (C_0 + N + 1) + M + 1 218 * K'' == keystream generated from K'' == keystream generated from 219 * C_0 + N + 1 + M, K'. C_0 + N + 1 + M, K'. 220 * K' has been erased. K' has been erased. 221 * 222 * Finally, in the new design, the two consumer threads can finish the 223 * remainder of the generation at any time (including simultaneously): 224 * 225 * 1: GenBytes() 226 * 1: Keystream(stack_C, stack_K, N) 227 * 1: <N blocks generated>** 228 * 1: <- Keystream 229 * 1: <- GenBytes 230 * 1:ExplicitBzero(stack_C, stack_K) 231 * 232 * 2: GenBytes() 233 * 2: Keystream(stack_C', stack_K', M) 234 * 2: <M blocks generated>** 235 * 2: <- Keystream 236 * 2: <- GenBytes 237 * 2:ExplicitBzero(stack_C', stack_K') 238 * 239 * The generated user keystream for both threads is identical between the two 240 * implementations: 241 * 242 * 1: Keystream(C_0, K_0, N) 1: Keystream(stack_C, stack_K, N) 243 * 2: Keystream(C'', K', M) 2: Keystream(stack_C', stack_K', M) 244 * 245 * (stack_C == C_0; stack_K == K_0; stack_C' == C''; stack_K' == K'.) 246 */ 247 static bool fortuna_concurrent_read __read_frequently = true; 248 249 #ifdef _KERNEL 250 static struct sysctl_ctx_list random_clist; 251 RANDOM_CHECK_UINT(fs_minpoolsize, RANDOM_FORTUNA_MINPOOLSIZE, RANDOM_FORTUNA_MAXPOOLSIZE); 252 #else 253 static uint8_t zero_region[RANDOM_ZERO_BLOCKSIZE]; 254 #endif 255 256 static void random_fortuna_pre_read(void); 257 static void random_fortuna_read(uint8_t *, size_t); 258 static bool random_fortuna_seeded(void); 259 static bool random_fortuna_seeded_internal(void); 260 static void random_fortuna_process_event(struct harvest_event *); 261 262 static void random_fortuna_reseed_internal(uint32_t *entropy_data, u_int blockcount); 263 264 #ifdef RANDOM_LOADABLE 265 static 266 #endif 267 const struct random_algorithm random_alg_context = { 268 .ra_ident = "Fortuna", 269 .ra_pre_read = random_fortuna_pre_read, 270 .ra_read = random_fortuna_read, 271 .ra_seeded = random_fortuna_seeded, 272 .ra_event_processor = random_fortuna_process_event, 273 .ra_poolcount = RANDOM_FORTUNA_NPOOLS, 274 }; 275 276 /* ARGSUSED */ 277 static void 278 random_fortuna_init_alg(void *unused __unused) 279 { 280 int i; 281 #ifdef _KERNEL 282 struct sysctl_oid *random_fortuna_o; 283 #endif 284 285 #ifdef RANDOM_LOADABLE 286 p_random_alg_context = &random_alg_context; 287 #endif 288 289 RANDOM_RESEED_INIT_LOCK(); 290 /* 291 * Fortuna parameters. Do not adjust these unless you have 292 * have a very good clue about what they do! 293 */ 294 fortuna_state.fs_minpoolsize = RANDOM_FORTUNA_DEFPOOLSIZE; 295 #ifdef _KERNEL 296 fortuna_state.fs_lasttime = 0; 297 random_fortuna_o = SYSCTL_ADD_NODE(&random_clist, 298 SYSCTL_STATIC_CHILDREN(_kern_random), 299 OID_AUTO, "fortuna", CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 300 "Fortuna Parameters"); 301 SYSCTL_ADD_PROC(&random_clist, 302 SYSCTL_CHILDREN(random_fortuna_o), OID_AUTO, "minpoolsize", 303 CTLTYPE_UINT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, 304 &fortuna_state.fs_minpoolsize, RANDOM_FORTUNA_DEFPOOLSIZE, 305 random_check_uint_fs_minpoolsize, "IU", 306 "Minimum pool size necessary to cause a reseed"); 307 KASSERT(fortuna_state.fs_minpoolsize > 0, ("random: Fortuna threshold must be > 0 at startup")); 308 309 SYSCTL_ADD_BOOL(&random_clist, SYSCTL_CHILDREN(random_fortuna_o), 310 OID_AUTO, "concurrent_read", CTLFLAG_RDTUN, 311 &fortuna_concurrent_read, 0, "If non-zero, enable " 312 "feature to improve concurrent Fortuna performance."); 313 #endif 314 315 /*- 316 * FS&K - InitializePRNG() 317 * - P_i = \epsilon 318 * - ReseedCNT = 0 319 */ 320 for (i = 0; i < RANDOM_FORTUNA_NPOOLS; i++) { 321 randomdev_hash_init(&fortuna_state.fs_pool[i].fsp_hash); 322 fortuna_state.fs_pool[i].fsp_length = 0; 323 } 324 fortuna_state.fs_reseedcount = 0; 325 /*- 326 * FS&K - InitializeGenerator() 327 * - C = 0 328 * - K = 0 329 */ 330 fortuna_state.fs_counter = UINT128_ZERO; 331 explicit_bzero(&fortuna_state.fs_key, sizeof(fortuna_state.fs_key)); 332 } 333 SYSINIT(random_alg, SI_SUB_RANDOM, SI_ORDER_SECOND, random_fortuna_init_alg, 334 NULL); 335 336 /*- 337 * FS&K - AddRandomEvent() 338 * Process a single stochastic event off the harvest queue 339 */ 340 static void 341 random_fortuna_process_event(struct harvest_event *event) 342 { 343 u_int pl; 344 345 RANDOM_RESEED_LOCK(); 346 /*- 347 * FS&K - P_i = P_i|<harvested stuff> 348 * Accumulate the event into the appropriate pool 349 * where each event carries the destination information. 350 * 351 * The hash_init() and hash_finish() calls are done in 352 * random_fortuna_pre_read(). 353 * 354 * We must be locked against pool state modification which can happen 355 * during accumulation/reseeding and reading/regating. 356 */ 357 pl = event->he_destination % RANDOM_FORTUNA_NPOOLS; 358 /* 359 * If a VM generation ID changes (clone and play or VM rewind), we want 360 * to incorporate that as soon as possible. Override destingation pool 361 * for immediate next use. 362 */ 363 if (event->he_source == RANDOM_PURE_VMGENID) 364 pl = 0; 365 /* 366 * We ignore low entropy static/counter fields towards the end of the 367 * he_event structure in order to increase measurable entropy when 368 * conducting SP800-90B entropy analysis measurements of seed material 369 * fed into PRNG. 370 * -- wdf 371 */ 372 KASSERT(event->he_size <= sizeof(event->he_entropy), 373 ("%s: event->he_size: %hhu > sizeof(event->he_entropy): %zu\n", 374 __func__, event->he_size, sizeof(event->he_entropy))); 375 randomdev_hash_iterate(&fortuna_state.fs_pool[pl].fsp_hash, 376 &event->he_somecounter, sizeof(event->he_somecounter)); 377 randomdev_hash_iterate(&fortuna_state.fs_pool[pl].fsp_hash, 378 event->he_entropy, event->he_size); 379 380 /*- 381 * Don't wrap the length. This is a "saturating" add. 382 * XXX: FIX!!: We don't actually need lengths for anything but fs_pool[0], 383 * but it's been useful debugging to see them all. 384 */ 385 fortuna_state.fs_pool[pl].fsp_length = MIN(RANDOM_FORTUNA_MAXPOOLSIZE, 386 fortuna_state.fs_pool[pl].fsp_length + 387 sizeof(event->he_somecounter) + event->he_size); 388 RANDOM_RESEED_UNLOCK(); 389 } 390 391 /*- 392 * FS&K - Reseed() 393 * This introduces new key material into the output generator. 394 * Additionally it increments the output generator's counter 395 * variable C. When C > 0, the output generator is seeded and 396 * will deliver output. 397 * The entropy_data buffer passed is a very specific size; the 398 * product of RANDOM_FORTUNA_NPOOLS and RANDOM_KEYSIZE. 399 */ 400 static void 401 random_fortuna_reseed_internal(uint32_t *entropy_data, u_int blockcount) 402 { 403 struct randomdev_hash context; 404 uint8_t hash[RANDOM_KEYSIZE]; 405 const void *keymaterial; 406 size_t keysz; 407 bool seeded; 408 409 RANDOM_RESEED_ASSERT_LOCK_OWNED(); 410 411 seeded = random_fortuna_seeded_internal(); 412 if (seeded) { 413 randomdev_getkey(&fortuna_state.fs_key, &keymaterial, &keysz); 414 KASSERT(keysz == RANDOM_KEYSIZE, ("%s: key size %zu not %u", 415 __func__, keysz, (unsigned)RANDOM_KEYSIZE)); 416 } 417 418 /*- 419 * FS&K - K = Hd(K|s) where Hd(m) is H(H(0^512|m)) 420 * - C = C + 1 421 */ 422 randomdev_hash_init(&context); 423 randomdev_hash_iterate(&context, zero_region, RANDOM_ZERO_BLOCKSIZE); 424 if (seeded) 425 randomdev_hash_iterate(&context, keymaterial, keysz); 426 randomdev_hash_iterate(&context, entropy_data, RANDOM_KEYSIZE*blockcount); 427 randomdev_hash_finish(&context, hash); 428 randomdev_hash_init(&context); 429 randomdev_hash_iterate(&context, hash, RANDOM_KEYSIZE); 430 randomdev_hash_finish(&context, hash); 431 randomdev_encrypt_init(&fortuna_state.fs_key, hash); 432 explicit_bzero(hash, sizeof(hash)); 433 /* Unblock the device if this is the first time we are reseeding. */ 434 if (uint128_is_zero(fortuna_state.fs_counter)) 435 randomdev_unblock(); 436 uint128_increment(&fortuna_state.fs_counter); 437 } 438 439 /*- 440 * FS&K - RandomData() (Part 1) 441 * Used to return processed entropy from the PRNG. There is a pre_read 442 * required to be present (but it can be a stub) in order to allow 443 * specific actions at the begin of the read. 444 */ 445 void 446 random_fortuna_pre_read(void) 447 { 448 #ifdef _KERNEL 449 sbintime_t now; 450 #endif 451 struct randomdev_hash context; 452 uint32_t s[RANDOM_FORTUNA_NPOOLS*RANDOM_KEYSIZE_WORDS]; 453 uint8_t temp[RANDOM_KEYSIZE]; 454 u_int i; 455 456 KASSERT(fortuna_state.fs_minpoolsize > 0, ("random: Fortuna threshold must be > 0")); 457 RANDOM_RESEED_LOCK(); 458 #ifdef _KERNEL 459 /* FS&K - Use 'getsbinuptime()' to prevent reseed-spamming. */ 460 now = getsbinuptime(); 461 #endif 462 463 if (fortuna_state.fs_pool[0].fsp_length < fortuna_state.fs_minpoolsize 464 #ifdef _KERNEL 465 /* 466 * FS&K - Use 'getsbinuptime()' to prevent reseed-spamming, but do 467 * not block initial seeding (fs_lasttime == 0). 468 */ 469 || (__predict_true(fortuna_state.fs_lasttime != 0) && 470 now - fortuna_state.fs_lasttime <= SBT_1S/10) 471 #endif 472 ) { 473 RANDOM_RESEED_UNLOCK(); 474 return; 475 } 476 477 #ifdef _KERNEL 478 /* 479 * When set, pretend we do not have enough entropy to reseed yet. 480 */ 481 KFAIL_POINT_CODE(DEBUG_FP, random_fortuna_pre_read, { 482 if (RETURN_VALUE != 0) { 483 RANDOM_RESEED_UNLOCK(); 484 return; 485 } 486 }); 487 #endif 488 489 #ifdef _KERNEL 490 fortuna_state.fs_lasttime = now; 491 #endif 492 493 /* FS&K - ReseedCNT = ReseedCNT + 1 */ 494 fortuna_state.fs_reseedcount++; 495 /* s = \epsilon at start */ 496 for (i = 0; i < RANDOM_FORTUNA_NPOOLS; i++) { 497 /* FS&K - if Divides(ReseedCnt, 2^i) ... */ 498 if ((fortuna_state.fs_reseedcount % (1 << i)) == 0) { 499 /*- 500 * FS&K - temp = (P_i) 501 * - P_i = \epsilon 502 * - s = s|H(temp) 503 */ 504 randomdev_hash_finish(&fortuna_state.fs_pool[i].fsp_hash, temp); 505 randomdev_hash_init(&fortuna_state.fs_pool[i].fsp_hash); 506 fortuna_state.fs_pool[i].fsp_length = 0; 507 randomdev_hash_init(&context); 508 randomdev_hash_iterate(&context, temp, RANDOM_KEYSIZE); 509 randomdev_hash_finish(&context, s + i*RANDOM_KEYSIZE_WORDS); 510 } else 511 break; 512 } 513 #ifdef _KERNEL 514 SDT_PROBE2(random, fortuna, event_processor, debug, fortuna_state.fs_reseedcount, fortuna_state.fs_pool); 515 #endif 516 /* FS&K */ 517 random_fortuna_reseed_internal(s, i); 518 RANDOM_RESEED_UNLOCK(); 519 520 /* Clean up and secure */ 521 explicit_bzero(s, sizeof(s)); 522 explicit_bzero(temp, sizeof(temp)); 523 } 524 525 /* 526 * This is basically GenerateBlocks() from FS&K. 527 * 528 * It differs in two ways: 529 * 530 * 1. Chacha20 is tolerant of non-block-multiple request sizes, so we do not 531 * need to handle any remainder bytes specially and can just pass the length 532 * directly to the PRF construction; and 533 * 534 * 2. Chacha20 is a 512-bit block size cipher (whereas AES has 128-bit block 535 * size, regardless of key size). This means Chacha does not require re-keying 536 * every 1MiB. This is implied by the math in FS&K 9.4 and mentioned 537 * explicitly in the conclusion, "If we had a block cipher with a 256-bit [or 538 * greater] block size, then the collisions would not have been an issue at 539 * all" (p. 144). 540 * 541 * 3. In conventional ("locked") mode, we produce a maximum of PAGE_SIZE output 542 * at a time before dropping the lock, to not bully the lock especially. This 543 * has been the status quo since 2015 (r284959). 544 * 545 * The upstream caller random_fortuna_read is responsible for zeroing out 546 * sensitive buffers provided as parameters to this routine. 547 */ 548 enum { 549 FORTUNA_UNLOCKED = false, 550 FORTUNA_LOCKED = true 551 }; 552 static void 553 random_fortuna_genbytes(uint8_t *buf, size_t bytecount, 554 uint8_t newkey[static RANDOM_KEYSIZE], uint128_t *p_counter, 555 union randomdev_key *p_key, bool locked) 556 { 557 uint8_t remainder_buf[RANDOM_BLOCKSIZE]; 558 size_t chunk_size; 559 560 if (locked) 561 RANDOM_RESEED_ASSERT_LOCK_OWNED(); 562 else 563 RANDOM_RESEED_ASSERT_LOCK_NOT_OWNED(); 564 565 /* 566 * Easy case: don't have to worry about bullying the global mutex, 567 * don't have to worry about rekeying Chacha; API is byte-oriented. 568 */ 569 if (!locked && random_chachamode) { 570 randomdev_keystream(p_key, p_counter, buf, bytecount); 571 return; 572 } 573 574 if (locked) { 575 /* 576 * While holding the global lock, limit PRF generation to 577 * mitigate, but not eliminate, bullying symptoms. 578 */ 579 chunk_size = PAGE_SIZE; 580 } else { 581 /* 582 * 128-bit block ciphers like AES must be re-keyed at 1MB 583 * intervals to avoid unacceptable statistical differentiation 584 * from true random data (FS&K 9.4, p. 143-144). 585 */ 586 MPASS(!random_chachamode); 587 chunk_size = RANDOM_FORTUNA_MAX_READ; 588 } 589 590 chunk_size = MIN(bytecount, chunk_size); 591 if (!random_chachamode) 592 chunk_size = rounddown(chunk_size, RANDOM_BLOCKSIZE); 593 594 while (bytecount >= chunk_size && chunk_size > 0) { 595 randomdev_keystream(p_key, p_counter, buf, chunk_size); 596 597 buf += chunk_size; 598 bytecount -= chunk_size; 599 600 /* We have to rekey if there is any data remaining to be 601 * generated, in two scenarios: 602 * 603 * locked: we need to rekey before we unlock and release the 604 * global state to another consumer; or 605 * 606 * unlocked: we need to rekey because we're in AES mode and are 607 * required to rekey at chunk_size==1MB. But we do not need to 608 * rekey during the last trailing <1MB chunk. 609 */ 610 if (bytecount > 0) { 611 if (locked || chunk_size == RANDOM_FORTUNA_MAX_READ) { 612 randomdev_keystream(p_key, p_counter, newkey, 613 RANDOM_KEYSIZE); 614 randomdev_encrypt_init(p_key, newkey); 615 } 616 617 /* 618 * If we're holding the global lock, yield it briefly 619 * now. 620 */ 621 if (locked) { 622 RANDOM_RESEED_UNLOCK(); 623 RANDOM_RESEED_LOCK(); 624 } 625 626 /* 627 * At the trailing end, scale down chunk_size from 1MB or 628 * PAGE_SIZE to all remaining full blocks (AES) or all 629 * remaining bytes (Chacha). 630 */ 631 if (bytecount < chunk_size) { 632 if (random_chachamode) 633 chunk_size = bytecount; 634 else if (bytecount >= RANDOM_BLOCKSIZE) 635 chunk_size = rounddown(bytecount, 636 RANDOM_BLOCKSIZE); 637 else 638 break; 639 } 640 } 641 } 642 643 /* 644 * Generate any partial AES block remaining into a temporary buffer and 645 * copy the desired substring out. 646 */ 647 if (bytecount > 0) { 648 MPASS(!random_chachamode); 649 650 randomdev_keystream(p_key, p_counter, remainder_buf, 651 sizeof(remainder_buf)); 652 } 653 654 /* 655 * In locked mode, re-key global K before dropping the lock, which we 656 * don't need for memcpy/bzero below. 657 */ 658 if (locked) { 659 randomdev_keystream(p_key, p_counter, newkey, RANDOM_KEYSIZE); 660 randomdev_encrypt_init(p_key, newkey); 661 RANDOM_RESEED_UNLOCK(); 662 } 663 664 if (bytecount > 0) { 665 memcpy(buf, remainder_buf, bytecount); 666 explicit_bzero(remainder_buf, sizeof(remainder_buf)); 667 } 668 } 669 670 671 /* 672 * Handle only "concurrency-enabled" Fortuna reads to simplify logic. 673 * 674 * Caller (random_fortuna_read) is responsible for zeroing out sensitive 675 * buffers provided as parameters to this routine. 676 */ 677 static void 678 random_fortuna_read_concurrent(uint8_t *buf, size_t bytecount, 679 uint8_t newkey[static RANDOM_KEYSIZE]) 680 { 681 union randomdev_key key_copy; 682 uint128_t counter_copy; 683 size_t blockcount; 684 685 MPASS(fortuna_concurrent_read); 686 687 /* 688 * Compute number of blocks required for the PRF request ('delta C'). 689 * We will step the global counter 'C' by this number under lock, and 690 * then actually consume the counter values outside the lock. 691 * 692 * This ensures that contemporaneous but independent requests for 693 * randomness receive distinct 'C' values and thus independent PRF 694 * results. 695 */ 696 if (random_chachamode) { 697 blockcount = howmany(bytecount, CHACHA_BLOCKLEN); 698 } else { 699 blockcount = howmany(bytecount, RANDOM_BLOCKSIZE); 700 701 /* 702 * Need to account for the additional blocks generated by 703 * rekeying when updating the global fs_counter. 704 */ 705 blockcount += RANDOM_KEYS_PER_BLOCK * 706 (blockcount / RANDOM_FORTUNA_BLOCKS_PER_KEY); 707 } 708 709 RANDOM_RESEED_LOCK(); 710 KASSERT(!uint128_is_zero(fortuna_state.fs_counter), ("FS&K: C != 0")); 711 712 /* 713 * Save the original counter and key values that will be used as the 714 * PRF for this particular consumer. 715 */ 716 memcpy(&counter_copy, &fortuna_state.fs_counter, sizeof(counter_copy)); 717 memcpy(&key_copy, &fortuna_state.fs_key, sizeof(key_copy)); 718 719 /* 720 * Step the counter as if we had generated 'bytecount' blocks for this 721 * consumer. I.e., ensure that the next consumer gets an independent 722 * range of counter values once we drop the global lock. 723 */ 724 uint128_add64(&fortuna_state.fs_counter, blockcount); 725 726 /* 727 * We still need to Rekey the global 'K' between independent calls; 728 * this is no different from conventional Fortuna. Note that 729 * 'randomdev_keystream()' will step the fs_counter 'C' appropriately 730 * for the blocks needed for the 'newkey'. 731 * 732 * (This is part of PseudoRandomData() in FS&K, 9.4.4.) 733 */ 734 randomdev_keystream(&fortuna_state.fs_key, &fortuna_state.fs_counter, 735 newkey, RANDOM_KEYSIZE); 736 randomdev_encrypt_init(&fortuna_state.fs_key, newkey); 737 738 /* 739 * We have everything we need to generate a unique PRF for this 740 * consumer without touching global state. 741 */ 742 RANDOM_RESEED_UNLOCK(); 743 744 random_fortuna_genbytes(buf, bytecount, newkey, &counter_copy, 745 &key_copy, FORTUNA_UNLOCKED); 746 RANDOM_RESEED_ASSERT_LOCK_NOT_OWNED(); 747 748 explicit_bzero(&counter_copy, sizeof(counter_copy)); 749 explicit_bzero(&key_copy, sizeof(key_copy)); 750 } 751 752 /*- 753 * FS&K - RandomData() (Part 2) 754 * Main read from Fortuna, continued. May be called multiple times after 755 * the random_fortuna_pre_read() above. 756 * 757 * The supplied buf MAY not be a multiple of RANDOM_BLOCKSIZE in size; it is 758 * the responsibility of the algorithm to accommodate partial block reads, if a 759 * block output mode is used. 760 */ 761 void 762 random_fortuna_read(uint8_t *buf, size_t bytecount) 763 { 764 uint8_t newkey[RANDOM_KEYSIZE]; 765 766 if (fortuna_concurrent_read) { 767 random_fortuna_read_concurrent(buf, bytecount, newkey); 768 goto out; 769 } 770 771 RANDOM_RESEED_LOCK(); 772 KASSERT(!uint128_is_zero(fortuna_state.fs_counter), ("FS&K: C != 0")); 773 774 random_fortuna_genbytes(buf, bytecount, newkey, 775 &fortuna_state.fs_counter, &fortuna_state.fs_key, FORTUNA_LOCKED); 776 /* Returns unlocked */ 777 RANDOM_RESEED_ASSERT_LOCK_NOT_OWNED(); 778 779 out: 780 explicit_bzero(newkey, sizeof(newkey)); 781 } 782 783 #ifdef _KERNEL 784 static bool block_seeded_status = false; 785 SYSCTL_BOOL(_kern_random, OID_AUTO, block_seeded_status, CTLFLAG_RWTUN, 786 &block_seeded_status, 0, 787 "If non-zero, pretend Fortuna is in an unseeded state. By setting " 788 "this as a tunable, boot can be tested as if the random device is " 789 "unavailable."); 790 #endif 791 792 static bool 793 random_fortuna_seeded_internal(void) 794 { 795 return (!uint128_is_zero(fortuna_state.fs_counter)); 796 } 797 798 static bool 799 random_fortuna_seeded(void) 800 { 801 802 #ifdef _KERNEL 803 if (block_seeded_status) 804 return (false); 805 #endif 806 807 if (__predict_true(random_fortuna_seeded_internal())) 808 return (true); 809 810 /* 811 * Maybe we have enough entropy in the zeroth pool but just haven't 812 * kicked the initial seed step. Do so now. 813 */ 814 random_fortuna_pre_read(); 815 816 return (random_fortuna_seeded_internal()); 817 } 818