Lines
90.2 %
Functions
29.63 %
Branches
100 %
//! Random number generation.
//!
//! For most purposes in Arti, we use one of two random number generators:
//! - `rand::rng()` (formerly called `rand::thread_rng()`, up till rand 0.9)
//! - The [`CautiousRng`] implemented here.
//! [`CautiousRng`] should be used whenever we are generating
//! a medium- or long-term cryptographic key:
//! one that will be stored to disk, or used for more than a single communication.
//! It is slower than [`rand::rng()`],
//! but is more robust against several kinds of failure.
//
// Note: Although we want to use CautiousRng
// whenever we generate a medium- or long-term key,
// we do not consider it a major
// security hole if we use rand::rng() instead:
// CautiousRng is a defense-in-depth mechanism.
use std::convert::Infallible;
use digest::{ExtendableOutput, Update};
use rand::rngs::SysRng;
use rand_core::TryRng;
use sha3::Shake256;
use zeroize::Zeroizing;
/// Trait representing an Rng where every output is derived from
/// supposedly strong entropy.
///
/// Implemented by [`CautiousRng`].
/// # Warning
/// Do not implement this trait for new Rngs unless you know what you are doing;
/// any Rng to which you apply this trait should be _at least_ as
/// unpredictable and secure as `SysRng`.
/// We recommend using [`CautiousRng`] when you need an instance of this trait.
pub trait EntropicRng: rand_core::CryptoRng {}
impl EntropicRng for CautiousRng {}
/// Functionality for testing Rng code that requires an EntropicRng.
#[cfg(feature = "testing")]
mod testing {
/// Testing only: Pretend that an inner RNG truly implements `EntropicRng`.
#[allow(clippy::exhaustive_structs)]
pub struct FakeEntropicRng<R>(pub R);
impl<R: rand_core::TryRng<Error = Infallible>> rand_core::TryRng for FakeEntropicRng<R> {
type Error = Infallible;
fn try_next_u32(&mut self) -> Result<u32, Infallible> {
self.0.try_next_u32()
}
fn try_next_u64(&mut self) -> Result<u64, Infallible> {
self.0.try_next_u64()
fn try_fill_bytes(&mut self, dst: &mut [u8]) -> Result<(), Infallible> {
self.0.try_fill_bytes(dst)
impl<R: rand_core::TryCryptoRng<Error = Infallible>> rand_core::TryCryptoRng
for FakeEntropicRng<R>
{
impl<R: rand_core::CryptoRng> super::EntropicRng for FakeEntropicRng<R> {}
pub use testing::FakeEntropicRng;
/// An exceptionally cautious wrapper for [`SysRng`]
/// Ordinarily, one trusts `SysRng`.
/// But we want Arti to run on a wide variety of platforms,
/// and the chances of a bogus SysRng increases the more places we run.
/// This Rng combines SysRng with several other entropy sources,
/// in an attempt to reduce the likelihood of creating compromised keys.[^scary]
/// This Rng is slower than `SysRng`.
/// # Panics
/// This rng will panic if `SysRng` fails;
/// but that's the only sensible behavior for a cryptographic-heavy application like ours.
/// [^scary]: Who else remembers [CVE-2008-0166](https://www.cve.org/CVERecord?id=CVE-2008-0166)?
#[derive(Default)]
pub struct CautiousRng;
impl TryRng for CautiousRng {
let mut buf = Zeroizing::new([0_u8; 4]);
self.try_fill_bytes(buf.as_mut())?;
Ok(u32::from_le_bytes(*buf))
let mut buf = Zeroizing::new([0_u8; 8]);
Ok(u64::from_le_bytes(*buf))
fn try_fill_bytes(&mut self, dest: &mut [u8]) -> Result<(), Infallible> {
let mut xof = Shake256::default();
let mut buf = Zeroizing::new([0_u8; 32]);
// According to some oldschool crypto wisdom,
// provided by cryptographers wearing tinfoil hats,
// when you're making a construction like this you should poll your RNGs
// from least trusted to most-trusted,
// in case one of the least trusted ones is secretly Pascal's Demon,
// providing the input deliberately tuned to make your Shake256 output predictable.
// The idea is somewhat ludicrous, but we have to poll in _some_ order,
// and just writing this code has put us into a world of tinfoil hats.
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
if let Ok(mut rdrand) = rdrand::RdRand::new() {
// We'll tolerate a failure from rdrand here,
// since it can indicate a few different error conditions,
// including a lack of hardware support, or exhausted CPU entropy
// (whatever that is supposed to mean).
// We only want to panic on a failure from SysRng.
let _ignore_failure = rdrand.try_fill_bytes(buf.as_mut());
// We add the output from rdrand unconditionally, since a partial return is possible,
// and since there's no real harm in doing so.
// (Performance is likely swamped by syscall overhead, and call to our BackupRng.)
// In the worst case, we just add some NULs in this case, which is fine.
xof.update(buf.as_ref());
// TODO: Consider using rndr on aarch64.
#[cfg(not(target_arch = "wasm32"))]
if let Some(mut rng) = backup::backup_rng() {
let _ignore_failure = rng.try_fill_bytes(buf.as_mut());
rand::rng().try_fill_bytes(buf.as_mut())?;
SysRng
.try_fill_bytes(buf.as_mut())
.expect("No strong entropy source was available: cannot proceed");
xof.finalize_xof_into(dest);
Ok(())
impl rand_core::TryCryptoRng for CautiousRng {}
/// A backup RNG, independent of other known sources.
/// Not necessarily strong, but hopefully random enough to cause an attacker some trouble
/// in the event of catastrophic failure.
/// A failure from this RNG _does not_ cause a panic.
mod backup {
use rand::TryRng;
use rand_chacha::ChaCha20Rng;
use reseeding_rng::ReseedingRng;
use std::sync::LazyLock;
use std::sync::{Mutex, MutexGuard};
/// The type we've chosen to use for our backup Rng.
/// (We need to box this because the default JitterRng is unnameable.)
/// We use JitterRng to reseed a ChaCha20 core
/// because it is potentially _very_ slow.
type BackupRng = ReseedingRng<ChaCha20Rng, Box<dyn TryRng<Error = Infallible> + Send>>;
/// Static instance of our BackupRng; None if we failed to construct one.
static JITTER_BACKUP: LazyLock<Option<Mutex<BackupRng>>> = LazyLock::new(new_backup_rng);
/// Construct a new instance of our backup Rng;
/// return None on failure.
fn new_backup_rng() -> Option<Mutex<BackupRng>> {
let jitter = rand_jitter::JitterRng::new().ok()?;
let jitter: Box<dyn TryRng<Error = Infallible> + Send> = Box::new(jitter);
// The "1024" here is chosen more or less arbitrarily;
// we might want to tune it if we find that it matters.
let reseeding = ReseedingRng::try_new(1024, jitter).ok()?;
Some(Mutex::new(reseeding))
/// Return a MutexGuard for our backup rng, or None if we couldn't construct one.
pub(super) fn backup_rng() -> Option<MutexGuard<'static, BackupRng>> {
JITTER_BACKUP
.as_ref()
.map(|mutex| mutex.lock().expect("lock poisoned"))