Secure random number generation.
There are several parts of this module:
- The signature of generator modules, together with a facility to convert such modules into actual generators, and functions that operate on this representation.
- A global generator instance, which needs to be initialized by calling
TL;DR Don't forget to seed; don't maintain your own
The RNGs here are merely the deterministic part of a full random number generation suite. For proper operation, they need to be seeded with a high-quality entropy source.
Suitable entropy feeding of generators are provided by other libraries mirage-crypto-rng-lwt (for Lwt), mirage-crypto-rng-async (for Async), mirage-crypto-rng-mirage (for MirageOS), mirage-crypto-rng.unix, and mirage-crypto-rng-eio (for Eio).
The intention is that "initialize" in the respective sub-library is called once, which sets the default generator and registers entropy harvesting asynchronous tasks. The semantics is that the entropy is always fed to the default generator, which is not necessarily the one set by "initialize". The reasoning behind this is that the default generator should be used in most setting, and that should be fed a constant stream of entropy.
mirage-crypto-rng-eio package differs slightly from other rng packages. Instead of the
initilize function a
run function is provided with similar behaviour, i.e. RNG setup, entropy collection and periodic reseeding.
Although this module exposes a more fine-grained interface, e.g. allowing manual seeding of generators, this is intended either for implementing entropy-harvesting modules, or very specialized purposes. Users of this library should almost certainly use one of the above entropy libraries, and avoid manually managing the generator seeding.
Similarly, although it is possible to swap the default generator and gain control over the random stream, this is also intended for specialized applications such as testing or similar scenarios where the RNG needs to be fully deterministic (RFC 6979, deterministic usage of DSA), or as a component of deterministic algorithms which internally rely on pseudorandom streams.
In the general case, users should not maintain their local instances of g. All of the generators in a process have to compete for entropy, and it is likely that the overall result will have lower effective unpredictability.
The recommended way to use these functions is either to accept an optional generator and pass it down, or to ignore the generator altogether, as illustrated in the examples.
Thrown when using an uninitialized generator.
set_default_generator has not been called.
module Entropy : sig ... end
Entropy sources and collection
module type Generator = sig ... end
A single PRNG algorithm.
Ready-to-use RNG algorithms.
HMAC_DRBG: A NIST-specified RNG based on HMAC construction over the provided hash.
g is the state to use, otherwise a fresh one is created.
seed can be provided to immediately reseed the generator with.
strict puts the generator into a more standards-conformant, but slighty slower mode. Useful if the outputs need to match published test-vectors.
time is used to limit the amount of reseedings. Fortuna uses at most once every second.
val default_generator : unit -> g
default_generator () is the default generator. Functions in this module use this generator when not explicitly supplied one.
val set_default_generator : g -> unit
set_default_generator g sets the default generator to
g. This function must be called once.
Generating a random 13-byte
let cs = Rng.generate 13
let rec f1 ?g ~n i = if i < 1 then  else Rng.generate ?g n :: f1 ?g ~n (i - 1)
Z.t smaller than
let f2 ?g () = Mirage_crypto_pk.Z_extra.gen ?g Z.(~$10)
Creating a local Fortuna instance and using it as a key-derivation function:
let f3 secret = let g = Rng.(create ~seed:secret (module Generators.Fortuna)) in Rng.generate ~g 32