1.\" $NetBSD: rnd.4,v 1.35 2020/05/06 18:38:20 riastradh Exp $ 2.\" 3.\" Copyright (c) 2014-2020 The NetBSD Foundation, Inc. 4.\" All rights reserved. 5.\" 6.\" This code is derived from software contributed to The NetBSD Foundation 7.\" by Taylor R. Campbell. 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, this list of conditions and the following disclaimer. 14.\" 2. Redistributions in binary form must reproduce the above copyright 15.\" notice, this list of conditions and the following disclaimer in the 16.\" documentation and/or other materials provided with the distribution. 17.\" 18.\" THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 19.\" ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 20.\" TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 21.\" PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 22.\" BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 23.\" CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 24.\" SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 25.\" INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 26.\" CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 27.\" ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 28.\" POSSIBILITY OF SUCH DAMAGE. 29.\" 30.Dd May 1, 2020 31.Dt RND 4 32.Os 33.\""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" 34.Sh NAME 35.Nm rnd 36.Nd random number generator 37.\""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" 38.Sh DESCRIPTION 39The 40.Pa /dev/random 41and 42.Pa /dev/urandom 43devices generate bytes randomly with uniform distribution. 44Every read from them is independent. 45.Bl -tag -width /dev/urandom 46.It Pa /dev/urandom 47Never blocks. 48.It Pa /dev/random 49Sometimes blocks. 50Will block early at boot if the system's state is known to be 51predictable. 52.El 53.Pp 54Applications should read from 55.Pa /dev/urandom , 56or the 57.Xr sysctl 7 58variable 59.Li kern.arandom , 60when they need randomly generated data, e.g. key material for 61cryptography or seeds for simulations. 62(The 63.Xr sysctl 7 64variable 65.Li kern.arandom 66is limited to 256 bytes per read, but is otherwise equivalent to 67reading from 68.Pa /dev/urandom 69and always works even in a 70.Xr chroot 8 71environment without requiring a populated 72.Pa /dev 73tree and without opening a file descriptor, so 74.Li kern.arandom 75may be preferable to use in libraries.) 76.Pp 77Systems should be engineered to judiciously read at least once from 78.Pa /dev/random 79at boot before running any services that talk to the internet or 80otherwise require cryptography, in order to avoid generating keys 81predictably. 82.Pa /dev/random 83may block at any time, so programs that read from it must be prepared 84to handle blocking. 85Interactive programs that block due to reads from 86.Pa /dev/random 87can be especially frustrating. 88.Pp 89If interrupted by a signal, reads from either 90.Pa /dev/random 91or 92.Pa /dev/urandom 93may return short, so programs that handle signals must be prepared to 94retry reads. 95.Pp 96Writing to either 97.Pa /dev/random 98or 99.Pa /dev/urandom 100influences subsequent output of both devices, guaranteed to take 101effect at next open. 102If you have a coin in your pocket, you can flip it 256 times and feed 103the outputs to 104.Pa /dev/random 105to guarantee your system is in a state that nobody but you and the 106bored security guard watching the surveillance camera in your office 107can guess: 108.Bd -literal -offset abcd 109% echo tthhhhhthhhththtthhhhthtththttth... > /dev/random 110.Ed 111.Pp 112(Sequence generated from a genuine US quarter dollar, guaranteed 113random.) 114.\""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" 115.Sh SECURITY MODEL 116The 117.Nm 118subsystem provides the following security properties against two 119different classes of attackers, provided that there is enough entropy 120from entropy sources not seen by attackers: 121.Bl -bullet -offset abcd 122.It 123An attacker who has seen some outputs and can supply some entropy 124sources' inputs to the operating system cannot predict past or future 125unseen outputs. 126.It 127An attacker who has seen the entire state of the machine cannot predict 128past outputs. 129.El 130.Pp 131One 132.Sq output 133means a single read, no matter how short it is. 134.Pp 135.Sq Cannot predict 136means it is conjectured of the cryptography in 137.Fa /dev/random 138that any computationally bounded attacker who tries to distinguish 139outputs from uniform random cannot do more than negligibly better than 140uniform random guessing. 141.\""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" 142.Sh ENTROPY 143The operating system continuously makes observations of hardware 144devices, such as network packet timings, disk seek delays, and 145keystrokes. 146The observations are combined into a seed for a cryptographic 147pseudorandom number generator (PRNG) which is used to generate the 148outputs of both 149.Pa /dev/random 150and 151.Pa /dev/urandom . 152.Pp 153An attacker may be able to guess with nonnegligible chance of success 154what your last keystroke was, but guessing every observation the 155operating system may have made is more difficult. 156The difficulty of the best strategy at guessing a random variable is 157analyzed as the -log_2 of the highest probability of any outcome, 158measured in bits, and called its 159.Em min-entropy , 160or 161.Em entropy 162for short in cryptography. 163For example: 164.Bl -bullet -offset abcd -compact 165.It 166A fair coin toss has one bit of entropy. 167.It 168A fair (six-sided) die roll has a little over 2.5 bits of entropy. 169.It 170A string of two independent fair coin tosses has two bits of entropy. 171.It 172The toss of a pair of fair coins that are glued together has one bit of 173entropy. 174.It 175A uniform random distribution with 176.Fa n 177possibilities has log_2 178.Fa n 179bits of entropy. 180.It 181An utterance from an accounting troll who always says 182.Sq nine 183has zero bits of entropy. 184.El 185.Pp 186Note that entropy is a property of an observable physical process, like 187a coin toss, or of a state of knowledge about that physical process; it 188is not a property of a specific sample obtained by observing it, like 189the string 190.Sq tthhhhht . 191There are also kinds of entropy in information theory other than 192min-entropy, including the more well-known Shannon entropy, but they 193are not relevant here. 194.Pp 195Hardware devices that the operating system monitors for observations 196are called 197.Em "entropy sources" , 198and the observations are combined into an 199.Em "entropy pool" . 200The 201.Xr rndctl 8 202command queries information about entropy sources and the entropy pool, 203and can control which entropy sources the operating system uses or 204ignores. 205.Pp 206256 bits of entropy is typically considered intractable to guess with 207classical computers and with current models of the capabilities of 208quantum computers. 209.Pp 210Systems with nonvolatile storage should store a secret from 211.Pa /dev/urandom 212on disk during installation or shutdown, and feed it back during boot, 213so that the work the operating system has done to gather entropy \(em 214including the work its operator may have done to flip a coin! \(em can be 215saved from one boot to the next, and so that newly installed systems 216are not vulnerable to generating cryptographic keys predictably. 217.Pp 218The boot loaders in some 219.Nx 220ports support a command to load a seed from disk before the 221kernel has started. 222For those that don't, the 223.Xr rndctl 8 224command can do it once userland has started, for example by setting 225.Dq Li random_seed=YES 226in 227.Pa /etc/rc.conf , 228which is enabled by default; see 229.Xr rc.conf 5 . 230.\""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" 231.Sh LIMITATIONS 232Some people worry about recovery from state compromise \(em that is, 233ensuring that even if an attacker sees the entire state of the 234operating system, then the attacker will be unable to predict any new 235future outputs as long as the operating system gathers fresh entropy 236quickly enough. 237.Pp 238But if an attacker has seen the entire state of your machine, 239refreshing entropy is probably the least of your worries, so we do not 240address that threat model here. 241.Pp 242The 243.Nm 244subsystem does 245.Em not 246automatically defend against hardware colluding with an attacker to 247influence entropy sources based on the state of the operating system. 248.Pp 249For example, a PCI device or CPU instruction for random number 250generation which has no side channel to an attacker other than the 251.Pa /dev/urandom 252device could be bugged to observe all other entropy sources, and to 253carefully craft 254.Sq observations 255that cause a certain number of bits of 256.Pa /dev/urandom 257output to be ciphertext that either is predictable to an attacker or 258conveys a message to an attacker. 259.Pp 260No amount of scrutiny by the system's operator could detect this. 261The only way to prevent this attack would be for the operator to 262disable all entropy sources that may be colluding with an attacker. 263If you're not sure which ones are not, you can always disable all of 264them and fall back to the coin in your pocket. 265.\""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" 266.Sh IOCTLS 267The 268.Pa /dev/random 269and 270.Pa /dev/urandom 271devices support a number of ioctls, defined in the 272.In sys/rndio.h 273header file, for querying and controlling the entropy pool. 274.Pp 275Since timing between hardware events contributes to the entropy pool, 276statistics about the entropy pool over time may serve as a side channel 277for the state of the pool, so access to such statistics is restricted 278to the super-user and should be used with caution. 279.Pp 280Several ioctls are concerned with particular entropy sources, described 281by the following structure: 282.Bd -literal 283typedef struct { 284 char name[16]; /* symbolic name */ 285 uint32_t total; /* estimate of entropy provided */ 286 uint32_t type; /* RND_TYPE_* value */ 287 uint32_t flags; /* RND_FLAG_* mask */ 288} rndsource_t; 289 290#define RND_TYPE_UNKNOWN 291#define RND_TYPE_DISK /* disk device */ 292#define RND_TYPE_ENV /* environment sensor (temp, fan, &c.) */ 293#define RND_TYPE_NET /* network device */ 294#define RND_TYPE_POWER /* power events */ 295#define RND_TYPE_RNG /* hardware RNG */ 296#define RND_TYPE_SKEW /* clock skew */ 297#define RND_TYPE_TAPE /* tape drive */ 298#define RND_TYPE_TTY /* tty device */ 299#define RND_TYPE_VM /* virtual memory faults */ 300 301#define RND_TYPE_MAX /* value of highest-numbered type */ 302 303#define RND_FLAG_COLLECT_TIME /* use timings of samples */ 304#define RND_FLAG_COLLECT_VALUE /* use values of samples */ 305#define RND_FLAG_ESTIMATE_TIME /* estimate entropy of timings */ 306#define RND_FLAG_ESTIMATE_VALUE /* estimate entropy of values */ 307#define RND_FLAG_NO_COLLECT /* ignore samples from this */ 308#define RND_FLAG_NO_ESTIMATE /* do not estimate entropy */ 309.Ed 310.Pp 311The following ioctls are supported: 312.Bl -tag -width abcd 313.It Dv RNDGETENTCNT Pq Vt uint32_t 314Return the number of bits of entropy the system is estimated to have. 315.It Dv RNDGETSRCNUM Pq Vt rndstat_t 316.Bd -literal 317typedef struct { 318 uint32_t start; 319 uint32_t count; 320 rndsource_t source[RND_MAXSTATCOUNT]; 321} rndstat_t; 322.Ed 323.Pp 324Fill the 325.Fa sources 326array with information about up to 327.Fa count 328entropy sources, starting at 329.Fa start . 330The actual number of sources described is returned in 331.Fa count . 332At most 333.Dv RND_MAXSTATCOUNT 334sources may be requested at once. 335.It Dv RNDGETSRCNAME Pq Vt rndstat_name_t 336.Bd -literal 337typedef struct { 338 char name[16]; 339 rndsource_t source; 340} rndstat_name_t; 341.Ed 342.Pp 343Fill 344.Fa source 345with information about the entropy source named 346.Fa name , 347or fail with 348.Dv ENOENT 349if there is none. 350.It Dv RNDCTL Pq Vt rndctl_t 351.Bd -literal 352typedef struct { 353 char name[16]; 354 uint32_t type; 355 uint32_t flags; 356 uint32_t mask; 357} rndctl_t; 358.Ed 359.Pp 360For each entropy source of the type 361.Fa type , 362or if 363.Fa type 364is 365.Li 0xff 366then for the entropy source named 367.Fa name , 368replace the flags in 369.Fa mask 370by 371.Fa flags . 372.It Dv RNDADDDATA Pq Vt rnddata_t 373.Bd -literal 374typedef struct { 375 uint32_t len; 376 uint32_t entropy; 377 unsigned char data[RND_SAVEWORDS * sizeof(uint32_t)]; 378} rnddata_t; 379.Ed 380.Pp 381Feed 382.Fa len 383bytes of data to the entropy pool. 384The sample is expected to have been drawn with at least 385.Fa entropy 386bits of entropy. 387.Pp 388This ioctl can be used only once per boot. 389It is intended for a system that saves entropy to disk on shutdown and 390restores it on boot, so that the system can immediately be 391unpredictable without having to wait to gather entropy. 392.Pp 393This ioctl is the only way for userland to directly change the system's 394entropy estimate. 395.It Dv RNDGETPOOLSTAT Pq Vt rndpoolstat_t 396.Bd -literal 397typedef struct { 398 uint32_t poolsize; /* size of each LFSR in pool */ 399 uint32_t threshold; /* no. bytes of pool hash returned */ 400 uint32_t maxentropy; /* total size of pool in bits */ 401 uint32_t added; /* no. bits of entropy ever added */ 402 uint32_t curentropy; /* current entropy `balance' */ 403 uint32_t discarded; /* no. bits dropped when pool full */ 404 uint32_t generated; /* no. bits yielded by pool while 405 curentropy is zero */ 406} rndpoolstat_t; 407.Ed 408.Pp 409Return various statistics about entropy. 410.El 411.\""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" 412.Sh SYSCTLS 413The following 414.Xr sysctl 8 415variables provided by 416.Nm 417can be set by privileged users: 418.Bl -tag -width abcd 419.It Dv kern.entropy.collection Pq Vt bool 420(Default on.) 421Enables entering data into the entropy pool. 422If disabled, no new data can be entered into the entropy pool, whether 423by device drivers, by writes to 424.Pa /dev/random 425or 426.Pa /dev/urandom , 427or by the 428.Dv RNDADDDATA 429ioctl. 430.It Dv kern.entropy.depletion Pq Vt bool 431(Default off.) 432Enables 433.Sq entropy depletion , 434meaning that even after attaining full entropy, the kernel subtracts 435the number of bits read out of the entropy pool from its estimate of 436the system entropy. 437This is not justified by modern cryptography \(em an adversary will 438never guess the 256-bit secret in a Keccak sponge no matter how much 439output from the sponge they see \(em but may be useful for testing. 440.It Dv kern.entropy.consolidate Pq Vt int 441Trigger for entropy consolidation: executing 442.Dl # sysctl -w kern.entropy.consolidate=1 443causes the system to consolidate pending entropy from per-CPU pools 444into the global pool, and waits until done. 445.El 446.Pp 447The following read-only 448.Xr sysctl 8 449variables provide information to privileged users about the state of 450the entropy pool: 451.Bl -tag -width abcd 452.It Dv kern.entropy.needed Pq Vt unsigned int 453Number of bits of entropy the system is waiting for in the global pool 454before reads from 455.Pa /dev/random 456will return without blocking. 457When zero, the system is considered to have full entropy. 458.It Dv kern.entropy.pending Pq Vt unsigned int 459Number of bits of entropy pending in per-CPU pools. 460This is the amount of entropy that will be contributed to the global 461pool at the next consolidation, such as from triggering 462.Dv kern.entropy.consolidate . 463.It Dv kern.entropy.epoch Pq Vt unsigned int 464Number of times system has reached full entropy, or entropy has been 465consolidated with 466.Dv kern.entropy.consolidate , 467as an unsigned 32-bit integer. 468Consulted inside the kernel by subsystems such as 469.Xr cprng 9 470to decide whether to reseed. 471Initially set to 2^32 \- 1 472.Pq i.e., Li "(unsigned)\-1" 473meaning the system has never reached full entropy and the entropy has 474never been consolidated; never again set to 2^32 \- 1. 475Never zero, so applications can initialize a cache of the epoch to zero 476to ensure they reseed the next time they check whether it is different 477from the stored epoch. 478.El 479.\""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" 480.Sh IMPLEMENTATION NOTES 481(This section describes the current implementation of the 482.Nm 483subsystem at the time of writing. 484It may be out-of-date by the time you read it, and nothing in here 485should be construed as a guarantee about the behaviour of the 486.Pa /dev/random 487and 488.Pa /dev/urandom 489devices.) 490.Pp 491Device drivers gather samples from entropy sources and absorb them into 492a collection of per-CPU Keccak sponges called 493.Sq entropy pools 494using the 495.Xr rnd 9 496kernel API. 497The device driver furnishes an estimate for the entropy of the sampling 498process, under the assumption that each sample is independent. 499When the estimate of entropy pending among the per-CPU entropy pools 500reaches a threshold of 256 bits, the entropy is drawn from the per-CPU 501pools and consolidated into a global pool. 502Keys for 503.Pa /dev/random , 504.Pa /dev/urandom , 505.Li kern.arandom , 506and the in-kernel 507.Xr cprng 9 508subsystem are extracted from the global pool. 509.Pp 510Early after boot, before CPUs have been detected, device drivers 511instead enter directly into the global pool. 512If anything in the system extracts data from the pool before the 513threshold has been reached at least once, the system will print a 514warning to the console and reset the entropy estimate to zero. 515The reason for resetting the entropy estimate to zero in this case is 516that an adversary who can witness output from the pool with partial 517entropy \(em say, 32 bits \(em can undergo a feasible brute force 518search to ascertain the complete state of the pool; as such, the 519entropy of the adversary's state of knowledge about the pool is zero. 520.Pp 521If the operator is confident that the drivers' estimates of the entropy 522of the sampling processes are too conservative, the operator can issue 523.Dl # sysctl -w kern.entropy.consolidate=1 524to force consolidation into the ready pool. 525The operator can also fool the system into thinking it has more entropy 526than it does by feeding data from 527.Pa /dev/urandom 528into 529.Pa /dev/random , 530but this voids the security model and should be limited to testing 531purposes. 532.Pp 533.Em Short 534reads from 535.Pa /dev/urandom 536are served by a persistent per-CPU Hash_DRBG instance that is 537reseeded from the entropy pool after any entropy consolidation. 538Reads from 539.Pa /dev/random 540and 541.Em long 542reads from 543.Pa /dev/urandom 544are served by a temporary Hash_DRBG seeded from the entropy pool on 545each read. 546.Pp 547When 548.Sq entropy depletion 549is enabled by 550setting the sysctl variable 551.Dv kern.entropy.depletion Ns Li \&=1 , 552every read from 553.Pa /dev/random 554is limited to 256 bits, since reading more than that would nearly 555always block again. 556.\""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" 557.Sh FILES 558.Bl -tag -width /dev/urandom -compact 559.It Pa /dev/random 560Uniform random byte source. 561May block. 562.It Pa /dev/urandom 563Uniform random byte source. 564Never blocks. 565.El 566.\""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" 567.Sh DIAGNOSTICS 568The 569.Nm 570subsystem may print the following warnings to the console likely 571indicating security issues: 572.Bl -diag -offset indent 573.It entropy: WARNING: extracting entropy too early 574Something requested extraction of entropy from the pool before it 575has ever reached full entropy in the system's estimation. 576.Pp 577The entropy may be low enough that an adversary who sees the output 578could guess the state of the pool by brute force, so in this event the 579system resets its estimate of entropy to none. 580.Pp 581This message is rate-limited to happen no more often than once per 582minute, so if you want to make sure it is gone you should consult 583.Dv kern.entropy.needed 584to confirm it is zero. 585.It entropy: WARNING: consolidating less than full entropy 586The operator triggered consolidation of entropy pending in per-CPU 587pools into the global pool when the system's estimate of the amount of 588entropy was still below the 256-bit threshold. 589.Pp 590This message can be safely ignored if the operator knows something the 591system doesn't, e.g. if the operator has flipped a coin 256 times and 592written the outcomes to 593.Pa /dev/random . 594.Pp 595This message is rate-limited to happen no more often than once per 596minute. 597.El 598.Pp 599The 600.Nm 601subsystem may print any of various messages about obtaining an entropy 602seed from the bootloader to diagnose saving and loading seeds on disk: 603.Bl -diag -offset indent 604.It entropy: entering seed from bootloader with N bits of entropy 605The bootloader provided an entropy seed to the kernel, which recorded 606an estimate of N bits of entropy in the process that generated it. 607.It entropy: no seed from bootloader 608The bootloader did not provide an entropy seed to the kernel before 609starting the kernel. 610This does not necessarily indicate a problem; not all bootloaders 611support the option, and the 612.Xr rc.conf 5 613setting 614.Li random_seed=YES 615can serve instead. 616.It entropy: invalid seed length N, expected sizeof(rndsave_t) = M 617The bootloader provided an entropy seed of the wrong size to the 618kernel. 619This may indicate a bug in 620.Xr rndctl 8 . 621The seed will be ignored. 622.It entropy: invalid seed checksum 623The entropy seed provided by the bootloader was malformed. 624The seed will be entered into the entropy pool, but it will be 625considered to contribute no entropy. 626.It entropy: double-seeded by bootloader 627A buggy bootloader tried to provide an entropy seed more than once to 628the kernel. 629Subsequent seeds will be entered into the entropy pool, but they will 630be considered to contribute no entropy. 631.It entropy: ready 632The system has full entropy for the first time. 633.El 634.\""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" 635.Sh SEE ALSO 636.Xr arc4random 3 , 637.Xr rndctl 8 , 638.Xr cprng 9 , 639.Xr rnd 9 640.Rs 641.%A Elaine Barker 642.%A John Kelsey 643.%T Recommendation for Random Number Generation Using Deterministic Random Bit Generators 644.%D June 2015 645.%Q United States Department of Commerce 646.%I National Institute of Standards and Technology 647.%O NIST Special Publication 800-90A, Revision 1 648.%U https://csrc.nist.gov/publications/detail/sp/800-90a/rev-1/final 649.Re 650.Rs 651.%A Meltem S\(:onmez Turan 652.%A Elaine Barker 653.%A John Kelsey 654.%A Kerry A. McKay 655.%A Mary L. Baish 656.%A Mike Boyle 657.%T Recommendations for the Entropy Sources Used for Random Bit Generation 658.%D January 2018 659.%Q United States Department of Commerce 660.%I National Institute of Standards and Technology 661.%O NIST Special Publication 800-90B 662.%U https://csrc.nist.gov/publications/detail/sp/800-90b/final 663.Re 664.Rs 665.%A Daniel J. Bernstein 666.%T Entropy Attacks! 667.%D 2014-02-05 668.%U http://blog.cr.yp.to/20140205-entropy.html 669.Re 670.Rs 671.%A Nadia Heninger 672.%A Zakir Durumeric 673.%A Eric Wustrow 674.%A J. Alex Halderman 675.%T Mining Your Ps and Qs: Detection of Widespread Weak Keys in Network Devices 676.%B Proceedings of the 21st USENIX Security Symposium 677.%I USENIX 678.%D August 2012 679.%P 205-220 680.%U https://www.usenix.org/conference/usenixsecurity12/technical-sessions/presentation/heninger 681.%U https://factorable.net/ 682.Re 683.Rs 684.%A Edwin T. Jaynes 685.%B Probability Theory: The Logic of Science 686.%I Cambridge University Press 687.%D 2003 688.%U https://bayes.wustl.edu/ 689.Re 690.\""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" 691.Sh HISTORY 692The 693.Pa /dev/random 694and 695.Pa /dev/urandom 696devices first appeared in 697.Nx 1.3 . 698.\""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" 699.Sh AUTHORS 700The 701.Nm 702subsystem was first implemented by 703.An Michael Graff Aq Mt explorer@flame.org , 704was then largely rewritten by 705.An Thor Lancelot Simon Aq Mt tls@NetBSD.org , 706and was most recently largely rewritten by 707.An Taylor R. Campbell Aq Mt riastradh@NetBSD.org . 708.\""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" 709.Sh BUGS 710Many people are confused about what 711.Pa /dev/random 712and 713.Pa /dev/urandom 714mean. 715Unfortunately, no amount of software engineering can fix that. 716