serai/coins/monero/src/ringct/clsag/mod.rs
Luke Parker af86b7a499
Support caching preprocesses in FROST (#190)
* Remove the explicit included participants from FROST

Now, whoever submits preprocesses becomes the signing set. Better separates
preprocess from sign, at the cost of slightly more annoying integrations
(Monero needs to now independently lagrange/offset its key images).

* Support caching preprocesses

Closes https://github.com/serai-dex/serai/issues/40.

I *could* have added a serialization trait to Algorithm and written a ton of
data to disk, while requiring Algorithm implementors also accept such work.
Instead, I moved preprocess to a seeded RNG (Chacha20) which should be as
secure as the regular RNG. Rebuilding from cache simply loads the previously
used Chacha seed, making the Algorithm oblivious to the fact it's being
rebuilt from a cache. This removes any requirements for it to be modified
while guaranteeing equivalency.

This builds on the last commit which delayed determining the signing set till
post-preprocess acquisition. Unfortunately, that commit did force preprocess
from ThresholdView to ThresholdKeys which had visible effects on Monero.

Serai will actually need delayed set determination for #163, and overall,
it remains better, hence it's inclusion.

* Document FROST preprocess caching

* Update ethereum to new FROST

* Fix bug in Monero offset calculation and update processor
2022-12-08 19:04:35 -05:00

325 lines
9.3 KiB
Rust

#![allow(non_snake_case)]
use core::ops::Deref;
use lazy_static::lazy_static;
use thiserror::Error;
use rand_core::{RngCore, CryptoRng};
use zeroize::{Zeroize, ZeroizeOnDrop, Zeroizing};
use subtle::{ConstantTimeEq, Choice, CtOption};
use curve25519_dalek::{
constants::ED25519_BASEPOINT_TABLE,
scalar::Scalar,
traits::{IsIdentity, VartimePrecomputedMultiscalarMul},
edwards::{EdwardsPoint, VartimeEdwardsPrecomputation},
};
use crate::{
Commitment, random_scalar, hash_to_scalar, wallet::decoys::Decoys, ringct::hash_to_point,
serialize::*,
};
#[cfg(feature = "multisig")]
mod multisig;
#[cfg(feature = "multisig")]
pub use multisig::{ClsagDetails, ClsagAddendum, ClsagMultisig};
#[cfg(feature = "multisig")]
pub(crate) use multisig::add_key_image_share;
lazy_static! {
static ref INV_EIGHT: Scalar = Scalar::from(8u8).invert();
}
/// Errors returned when CLSAG signing fails.
#[derive(Clone, Error, Debug)]
pub enum ClsagError {
#[error("internal error ({0})")]
InternalError(String),
#[error("invalid ring")]
InvalidRing,
#[error("invalid ring member (member {0}, ring size {1})")]
InvalidRingMember(u8, u8),
#[error("invalid commitment")]
InvalidCommitment,
#[error("invalid key image")]
InvalidImage,
#[error("invalid D")]
InvalidD,
#[error("invalid s")]
InvalidS,
#[error("invalid c1")]
InvalidC1,
}
/// Input being signed for.
#[derive(Clone, PartialEq, Eq, Debug, Zeroize, ZeroizeOnDrop)]
pub struct ClsagInput {
// The actual commitment for the true spend
pub(crate) commitment: Commitment,
// True spend index, offsets, and ring
pub(crate) decoys: Decoys,
}
impl ClsagInput {
pub fn new(commitment: Commitment, decoys: Decoys) -> Result<ClsagInput, ClsagError> {
let n = decoys.len();
if n > u8::MAX.into() {
Err(ClsagError::InternalError("max ring size in this library is u8 max".to_string()))?;
}
let n = u8::try_from(n).unwrap();
if decoys.i >= n {
Err(ClsagError::InvalidRingMember(decoys.i, n))?;
}
// Validate the commitment matches
if decoys.ring[usize::from(decoys.i)][1] != commitment.calculate() {
Err(ClsagError::InvalidCommitment)?;
}
Ok(ClsagInput { commitment, decoys })
}
}
#[allow(clippy::large_enum_variant)]
enum Mode {
Sign(usize, EdwardsPoint, EdwardsPoint),
Verify(Scalar),
}
// Core of the CLSAG algorithm, applicable to both sign and verify with minimal differences
// Said differences are covered via the above Mode
fn core(
ring: &[[EdwardsPoint; 2]],
I: &EdwardsPoint,
pseudo_out: &EdwardsPoint,
msg: &[u8; 32],
D: &EdwardsPoint,
s: &[Scalar],
A_c1: Mode,
) -> ((EdwardsPoint, Scalar, Scalar), Scalar) {
let n = ring.len();
let images_precomp = VartimeEdwardsPrecomputation::new([I, D]);
let D = D * *INV_EIGHT;
// Generate the transcript
// Instead of generating multiple, a single transcript is created and then edited as needed
const PREFIX: &[u8] = b"CLSAG_";
#[rustfmt::skip]
const AGG_0: &[u8] = b"agg_0";
#[rustfmt::skip]
const ROUND: &[u8] = b"round";
const PREFIX_AGG_0_LEN: usize = PREFIX.len() + AGG_0.len();
let mut to_hash = Vec::with_capacity(((2 * n) + 5) * 32);
to_hash.extend(PREFIX);
to_hash.extend(AGG_0);
to_hash.extend([0; 32 - PREFIX_AGG_0_LEN]);
let mut P = Vec::with_capacity(n);
for member in ring {
P.push(member[0]);
to_hash.extend(member[0].compress().to_bytes());
}
let mut C = Vec::with_capacity(n);
for member in ring {
C.push(member[1] - pseudo_out);
to_hash.extend(member[1].compress().to_bytes());
}
to_hash.extend(I.compress().to_bytes());
to_hash.extend(D.compress().to_bytes());
to_hash.extend(pseudo_out.compress().to_bytes());
// mu_P with agg_0
let mu_P = hash_to_scalar(&to_hash);
// mu_C with agg_1
to_hash[PREFIX_AGG_0_LEN - 1] = b'1';
let mu_C = hash_to_scalar(&to_hash);
// Truncate it for the round transcript, altering the DST as needed
to_hash.truncate(((2 * n) + 1) * 32);
for i in 0 .. ROUND.len() {
to_hash[PREFIX.len() + i] = ROUND[i];
}
// Unfortunately, it's I D pseudo_out instead of pseudo_out I D, meaning this needs to be
// truncated just to add it back
to_hash.extend(pseudo_out.compress().to_bytes());
to_hash.extend(msg);
// Configure the loop based on if we're signing or verifying
let start;
let end;
let mut c;
match A_c1 {
Mode::Sign(r, A, AH) => {
start = r + 1;
end = r + n;
to_hash.extend(A.compress().to_bytes());
to_hash.extend(AH.compress().to_bytes());
c = hash_to_scalar(&to_hash);
}
Mode::Verify(c1) => {
start = 0;
end = n;
c = c1;
}
}
// Perform the core loop
let mut c1 = CtOption::new(Scalar::zero(), Choice::from(0));
for i in (start .. end).map(|i| i % n) {
// This will only execute once and shouldn't need to be constant time. Making it constant time
// removes the risk of branch prediction creating timing differences depending on ring index
// however
c1 = c1.or_else(|| CtOption::new(c, i.ct_eq(&0)));
let c_p = mu_P * c;
let c_c = mu_C * c;
let L = (&s[i] * &ED25519_BASEPOINT_TABLE) + (c_p * P[i]) + (c_c * C[i]);
let PH = hash_to_point(P[i]);
// Shouldn't be an issue as all of the variables in this vartime statement are public
let R = (s[i] * PH) + images_precomp.vartime_multiscalar_mul(&[c_p, c_c]);
to_hash.truncate(((2 * n) + 3) * 32);
to_hash.extend(L.compress().to_bytes());
to_hash.extend(R.compress().to_bytes());
c = hash_to_scalar(&to_hash);
}
// This first tuple is needed to continue signing, the latter is the c to be tested/worked with
((D, c * mu_P, c * mu_C), c1.unwrap_or(c))
}
/// CLSAG signature, as used in Monero.
#[derive(Clone, PartialEq, Eq, Debug)]
pub struct Clsag {
pub D: EdwardsPoint,
pub s: Vec<Scalar>,
pub c1: Scalar,
}
impl Clsag {
// Sign core is the extension of core as needed for signing, yet is shared between single signer
// and multisig, hence why it's still core
pub(crate) fn sign_core<R: RngCore + CryptoRng>(
rng: &mut R,
I: &EdwardsPoint,
input: &ClsagInput,
mask: Scalar,
msg: &[u8; 32],
A: EdwardsPoint,
AH: EdwardsPoint,
) -> (Clsag, EdwardsPoint, Scalar, Scalar) {
let r: usize = input.decoys.i.into();
let pseudo_out = Commitment::new(mask, input.commitment.amount).calculate();
let z = input.commitment.mask - mask;
let H = hash_to_point(input.decoys.ring[r][0]);
let D = H * z;
let mut s = Vec::with_capacity(input.decoys.ring.len());
for _ in 0 .. input.decoys.ring.len() {
s.push(random_scalar(rng));
}
let ((D, p, c), c1) =
core(&input.decoys.ring, I, &pseudo_out, msg, &D, &s, Mode::Sign(r, A, AH));
(Clsag { D, s, c1 }, pseudo_out, p, c * z)
}
/// Generate CLSAG signatures for the given inputs.
/// inputs is of the form (private key, key image, input).
/// sum_outputs is for the sum of the outputs' commitment masks.
pub fn sign<R: RngCore + CryptoRng>(
rng: &mut R,
mut inputs: Vec<(Zeroizing<Scalar>, EdwardsPoint, ClsagInput)>,
sum_outputs: Scalar,
msg: [u8; 32],
) -> Vec<(Clsag, EdwardsPoint)> {
let mut res = Vec::with_capacity(inputs.len());
let mut sum_pseudo_outs = Scalar::zero();
for i in 0 .. inputs.len() {
let mut mask = random_scalar(rng);
if i == (inputs.len() - 1) {
mask = sum_outputs - sum_pseudo_outs;
} else {
sum_pseudo_outs += mask;
}
let mut nonce = Zeroizing::new(random_scalar(rng));
let (mut clsag, pseudo_out, p, c) = Clsag::sign_core(
rng,
&inputs[i].1,
&inputs[i].2,
mask,
&msg,
nonce.deref() * &ED25519_BASEPOINT_TABLE,
nonce.deref() *
hash_to_point(inputs[i].2.decoys.ring[usize::from(inputs[i].2.decoys.i)][0]),
);
clsag.s[usize::from(inputs[i].2.decoys.i)] =
(-((p * inputs[i].0.deref()) + c)) + nonce.deref();
inputs[i].0.zeroize();
nonce.zeroize();
debug_assert!(clsag
.verify(&inputs[i].2.decoys.ring, &inputs[i].1, &pseudo_out, &msg)
.is_ok());
res.push((clsag, pseudo_out));
}
res
}
/// Verify the CLSAG signature against the given Transaction data.
pub fn verify(
&self,
ring: &[[EdwardsPoint; 2]],
I: &EdwardsPoint,
pseudo_out: &EdwardsPoint,
msg: &[u8; 32],
) -> Result<(), ClsagError> {
// Preliminary checks. s, c1, and points must also be encoded canonically, which isn't checked
// here
if ring.is_empty() {
Err(ClsagError::InvalidRing)?;
}
if ring.len() != self.s.len() {
Err(ClsagError::InvalidS)?;
}
if I.is_identity() {
Err(ClsagError::InvalidImage)?;
}
let D = self.D.mul_by_cofactor();
if D.is_identity() {
Err(ClsagError::InvalidD)?;
}
let (_, c1) = core(ring, I, pseudo_out, msg, &D, &self.s, Mode::Verify(self.c1));
if c1 != self.c1 {
Err(ClsagError::InvalidC1)?;
}
Ok(())
}
pub(crate) fn fee_weight(ring_len: usize) -> usize {
(ring_len * 32) + 32 + 32
}
pub fn serialize<W: std::io::Write>(&self, w: &mut W) -> std::io::Result<()> {
write_raw_vec(write_scalar, &self.s, w)?;
w.write_all(&self.c1.to_bytes())?;
write_point(&self.D, w)
}
pub fn deserialize<R: std::io::Read>(decoys: usize, r: &mut R) -> std::io::Result<Clsag> {
Ok(Clsag { s: read_raw_vec(read_scalar, decoys, r)?, c1: read_scalar(r)?, D: read_point(r)? })
}
}