trailofbits / reverie

An efficient and generalized implementation of the IKOS-style KKW proof system (https://eprint.iacr.org/2018/475) for arbitrary rings.

Reverie

An efficient implementation of the NIZKPoK outlined in KKW 2018

Reverie is an implementation (prover and verifier) of the MPC-in-the-head NIZKPoK outlined in Improved Non-Interactive Zero Knowledge with Applications to Post-Quantum Signatures. Reverie seeks to offer concrete prover efficiency (linear proving time with small constants) for complex predicates. The implementation seeks to offer 128-bits of (classical) security and support arbitrary rings, most efficiently Z₂ and Z₆₄.

Reverie provides both a library (with a simplified and a streaming interface), in addition to a CLI program for proving/verifying statements specified in Bristol format to enable easy experimentation.

Running

Reverie requires a relatively recent nightly Rust.

Using SSE+AESNI

time RUSTFLAGS="-C target-cpu=native -C target-feature=+aes,+ssse3,+sse2" cargo run --release

Or even better with AVX2+AESNI

time RUSTFLAGS="-C target-cpu=native -C target-feature=+aes,+ssse3,+sse2,+avx2" cargo run --release

Improvements in 0.3+

• Pack 8 instances of 8 players over GF(2) into a single 64-bit integer (see gist for details).
• Switch to AES with AESNI
• Just-in-time preprocessing to condense proving into a single pass

reverie's Issues

Run `cargo fmt` in the CI

This is blocked on moving to a stable release, since nightly toolchains aren't guaranteed to have rustfmt available as a component.

Refactor use of rand/rand_core traits

We should pick one source of canonical "rand" traits, to avoid version hell.

Use a workspace structure for reverie and the companion

The reverie repository itself currently has a single top-level Cargo.toml for the reverie library. The companion is tucked away in the companion/ subdirectory, with its own Cargo.toml.

We should restructure the repository to use cargo workspaces. That'll allow us to unify our dependencies and publish both reverie and the companion to crates.io.

Remove our dependency on `aesni`

aesni is deprecated and replaced by aes, so we should switch to it.

Compile with a stable Rust

We should aim to support compilation and use with a stable Rust release, to avoid encountering nightly bugs and depending on unstable APIs.

Some items that need to be handled:

• Reverie currently uses the unstable std::test and #[bench]. We should probably use criterion instead.
• Reverie uses #[feature(stdsimd)] since the MMX intrinsics were never stabilized. Once they're removed in favor of SSE2 and AVX2, we can remove this #[feature]. (#4)