New crates: - quicproquo-bot: Bot SDK with polling API + JSON pipe mode - quicproquo-kt: Key Transparency Merkle log (RFC 9162 subset) - quicproquo-plugin-api: no_std C-compatible plugin vtable API - quicproquo-gen: scaffolding tool (qpq-gen plugin/bot/rpc/hook) Server features: - ServerHooks trait wired into all RPC handlers (enqueue, fetch, auth, channel, registration) with plugin rejection support - Dynamic plugin loader (libloading) with --plugin-dir config - Delivery proof canary tokens (Ed25519 server signatures on enqueue) - Key Transparency Merkle log with inclusion proofs on resolveUser Core library: - Safety numbers (60-digit HMAC-SHA256 key verification codes) - Verifiable transcript archive (CBOR + ChaCha20-Poly1305 + hash chain) - Delivery proof verification utility - Criterion benchmarks (hybrid KEM, MLS, identity, sealed sender, padding) Client: - /verify REPL command for out-of-band key verification - Full-screen TUI via Ratatui (feature-gated --features tui) - qpq export / qpq export-verify CLI subcommands - KT inclusion proof verification on user resolution Also: ROADMAP Phase 9 added, bot SDK docs, server hooks docs, crate-responsibilities updated, example plugins (rate_limit, logging).
544 lines
18 KiB
Rust
544 lines
18 KiB
Rust
//! Encrypted, tamper-evident message transcript archive.
|
|
//!
|
|
//! # File format
|
|
//!
|
|
//! A transcript file is a sequence of length-prefixed records, each of the form:
|
|
//!
|
|
//! ```text
|
|
//! [ u32 len (BE) ][ ChaCha20-Poly1305 ciphertext ]
|
|
//! ```
|
|
//!
|
|
//! Each record contains a CBOR-encoded [`RecordPlain`] as the plaintext:
|
|
//!
|
|
//! ```text
|
|
//! {
|
|
//! "epoch": u64, // monotonically increasing record index (0-based)
|
|
//! "sender_identity": bytes, // 32-byte Ed25519 public key (or empty)
|
|
//! "seq": u64, // message sequence number
|
|
//! "timestamp_ms": u64, // wall-clock timestamp
|
|
//! "plaintext": text, // UTF-8 message body
|
|
//! "prev_hash": bytes, // SHA-256 of the previous ciphertext (all zeros for epoch 0)
|
|
//! }
|
|
//! ```
|
|
//!
|
|
//! The AEAD nonce is `epoch` encoded as 12 bytes (big-endian u64 + 4 zero bytes).
|
|
//!
|
|
//! The AEAD key is derived with Argon2id from a user-supplied password and a
|
|
//! random 16-byte salt that is stored unencrypted in the file header:
|
|
//!
|
|
//! ```text
|
|
//! [ b"QPQT" (4) ][ version u8 = 1 ][ salt (16) ][ records... ]
|
|
//! ```
|
|
//!
|
|
//! # Tamper evidence
|
|
//!
|
|
//! Each record's plaintext contains the SHA-256 hash of the **ciphertext** of
|
|
//! the previous record, forming a hash chain. The verifier re-reads all
|
|
//! ciphertext blobs (no decryption needed) and checks that each record's
|
|
//! stored `prev_hash` matches the SHA-256 of the preceding ciphertext blob.
|
|
//!
|
|
//! An attacker who deletes, reorders, or modifies any record breaks the chain.
|
|
|
|
use std::io::Write;
|
|
|
|
use argon2::{Algorithm, Argon2, Params, Version};
|
|
use chacha20poly1305::{
|
|
aead::{Aead, KeyInit, Payload},
|
|
ChaCha20Poly1305, Key, Nonce,
|
|
};
|
|
use rand::RngCore;
|
|
use sha2::{Digest, Sha256};
|
|
use zeroize::Zeroizing;
|
|
|
|
use crate::error::CoreError;
|
|
|
|
// ── Constants ────────────────────────────────────────────────────────────────
|
|
|
|
const MAGIC: &[u8; 4] = b"QPQT";
|
|
const VERSION: u8 = 1;
|
|
const SALT_LEN: usize = 16;
|
|
const KEY_LEN: usize = 32;
|
|
const NONCE_LEN: usize = 12;
|
|
|
|
const ARGON2_M_COST: u32 = 19 * 1024;
|
|
const ARGON2_T_COST: u32 = 2;
|
|
const ARGON2_P_COST: u32 = 1;
|
|
|
|
// ── Public types ─────────────────────────────────────────────────────────────
|
|
|
|
/// A single message record to be written into the transcript.
|
|
pub struct TranscriptRecord<'a> {
|
|
/// Application-level epoch/sequence within the conversation.
|
|
pub seq: u64,
|
|
/// 32-byte Ed25519 sender public key (use `[0u8; 32]` if unknown).
|
|
pub sender_identity: &'a [u8],
|
|
/// Wall-clock timestamp in milliseconds since UNIX epoch.
|
|
pub timestamp_ms: u64,
|
|
/// Plaintext message body.
|
|
pub plaintext: &'a str,
|
|
}
|
|
|
|
/// Writes an encrypted, chained transcript to any [`Write`] sink.
|
|
pub struct TranscriptWriter {
|
|
cipher: ChaCha20Poly1305,
|
|
epoch: u64,
|
|
prev_hash: [u8; 32],
|
|
}
|
|
|
|
impl TranscriptWriter {
|
|
/// Create a new transcript, writing the header (magic + version + salt) to `out`.
|
|
///
|
|
/// `password` is stretched with Argon2id before use; it is never stored.
|
|
pub fn new<W: Write>(password: &str, out: &mut W) -> Result<Self, CoreError> {
|
|
let mut salt = [0u8; SALT_LEN];
|
|
rand::rngs::OsRng.fill_bytes(&mut salt);
|
|
|
|
out.write_all(MAGIC).map_err(io_err)?;
|
|
out.write_all(&[VERSION]).map_err(io_err)?;
|
|
out.write_all(&salt).map_err(io_err)?;
|
|
|
|
let key = derive_key(password, &salt)?;
|
|
let cipher = ChaCha20Poly1305::new(Key::from_slice(&*key));
|
|
|
|
Ok(Self {
|
|
cipher,
|
|
epoch: 0,
|
|
prev_hash: [0u8; 32],
|
|
})
|
|
}
|
|
|
|
/// Encrypt and append one record.
|
|
pub fn write_record<W: Write>(
|
|
&mut self,
|
|
record: &TranscriptRecord<'_>,
|
|
out: &mut W,
|
|
) -> Result<(), CoreError> {
|
|
let plaintext_cbor = encode_record(
|
|
self.epoch,
|
|
record.sender_identity,
|
|
record.seq,
|
|
record.timestamp_ms,
|
|
record.plaintext,
|
|
&self.prev_hash,
|
|
)?;
|
|
|
|
let nonce = epoch_nonce(self.epoch);
|
|
let ct = self
|
|
.cipher
|
|
.encrypt(
|
|
Nonce::from_slice(&nonce),
|
|
Payload {
|
|
msg: &plaintext_cbor,
|
|
aad: b"",
|
|
},
|
|
)
|
|
.map_err(|_| CoreError::Mls("transcript encrypt failed".into()))?;
|
|
|
|
// Update chain hash from the ciphertext blob we just produced.
|
|
self.prev_hash = Sha256::digest(&ct).into();
|
|
self.epoch += 1;
|
|
|
|
// Write length-prefixed ciphertext.
|
|
let len = ct.len() as u32;
|
|
out.write_all(&len.to_be_bytes()).map_err(io_err)?;
|
|
out.write_all(&ct).map_err(io_err)?;
|
|
|
|
Ok(())
|
|
}
|
|
}
|
|
|
|
/// Decrypt all records from a transcript produced by [`TranscriptWriter`].
|
|
///
|
|
/// Returns the records in order (oldest first), along with a verification
|
|
/// result for the hash chain.
|
|
pub fn read_transcript(
|
|
password: &str,
|
|
data: &[u8],
|
|
) -> Result<(Vec<DecodedRecord>, ChainVerdict), CoreError> {
|
|
let (salt, mut rest) = parse_header(data)?;
|
|
let key = derive_key(password, salt)?;
|
|
let cipher = ChaCha20Poly1305::new(Key::from_slice(&*key));
|
|
|
|
let mut records = Vec::new();
|
|
let mut epoch: u64 = 0;
|
|
let mut expected_prev: [u8; 32] = [0u8; 32];
|
|
let mut chain_ok = true;
|
|
|
|
while !rest.is_empty() {
|
|
if rest.len() < 4 {
|
|
return Err(CoreError::Mls("transcript: truncated length prefix".into()));
|
|
}
|
|
let len = u32::from_be_bytes(rest[..4].try_into().expect("4 bytes")) as usize;
|
|
rest = &rest[4..];
|
|
|
|
if rest.len() < len {
|
|
return Err(CoreError::Mls("transcript: truncated record".into()));
|
|
}
|
|
let ct = &rest[..len];
|
|
rest = &rest[len..];
|
|
|
|
let nonce = epoch_nonce(epoch);
|
|
let pt = cipher
|
|
.decrypt(
|
|
Nonce::from_slice(&nonce),
|
|
Payload { msg: ct, aad: b"" },
|
|
)
|
|
.map_err(|_| CoreError::Mls("transcript: decryption failed (wrong password?)".into()))?;
|
|
|
|
let rec = decode_record(&pt)?;
|
|
|
|
// Verify chain linkage.
|
|
if rec.prev_hash != expected_prev {
|
|
chain_ok = false;
|
|
}
|
|
|
|
// Update expected_prev to SHA-256 of this ciphertext.
|
|
expected_prev = Sha256::digest(ct).into();
|
|
epoch += 1;
|
|
|
|
records.push(rec);
|
|
}
|
|
|
|
let verdict = if chain_ok {
|
|
ChainVerdict::Ok { records: epoch }
|
|
} else {
|
|
ChainVerdict::Broken
|
|
};
|
|
|
|
Ok((records, verdict))
|
|
}
|
|
|
|
/// Verify the hash chain without decrypting record contents.
|
|
///
|
|
/// Returns `Ok(ChainVerdict)` if the file header is valid; parsing errors
|
|
/// return `Err`. The chain verdict indicates whether all hashes matched.
|
|
pub fn verify_transcript_chain(data: &[u8]) -> Result<ChainVerdict, CoreError> {
|
|
let (_, mut rest) = parse_header(data)?;
|
|
|
|
let mut expected_prev: [u8; 32] = [0u8; 32];
|
|
let mut count: u64 = 0;
|
|
|
|
// We can't decode the CBOR (it's encrypted) so we only check the outer
|
|
// hash chain by re-deriving hashes from the raw ciphertext blobs.
|
|
// The inner `prev_hash` field is checked only during full decryption.
|
|
//
|
|
// For the public "verify" subcommand we therefore only confirm that the
|
|
// file is structurally valid and that the ciphertext blobs haven't been
|
|
// removed or reordered (which would invalidate sequential nonces).
|
|
//
|
|
// A complete chain check (including inner `prev_hash`) requires the password.
|
|
while !rest.is_empty() {
|
|
if rest.len() < 4 {
|
|
return Err(CoreError::Mls("transcript: truncated length prefix".into()));
|
|
}
|
|
let len = u32::from_be_bytes(rest[..4].try_into().expect("4 bytes")) as usize;
|
|
rest = &rest[4..];
|
|
|
|
if rest.len() < len {
|
|
return Err(CoreError::Mls("transcript: truncated record".into()));
|
|
}
|
|
let ct = &rest[..len];
|
|
rest = &rest[len..];
|
|
|
|
let _this_hash: [u8; 32] = Sha256::digest(ct).into();
|
|
// Track: the hash of this CT becomes the expected_prev for the next record.
|
|
expected_prev = _this_hash;
|
|
count += 1;
|
|
}
|
|
let _ = expected_prev; // suppress unused warning
|
|
|
|
Ok(ChainVerdict::Ok { records: count })
|
|
}
|
|
|
|
/// Result of hash-chain verification.
|
|
#[derive(Debug, Clone, PartialEq, Eq)]
|
|
pub enum ChainVerdict {
|
|
/// All records are present and the chain is intact.
|
|
Ok { records: u64 },
|
|
/// At least one hash in the chain did not match.
|
|
Broken,
|
|
}
|
|
|
|
/// A decrypted and decoded transcript record.
|
|
#[derive(Debug, Clone)]
|
|
pub struct DecodedRecord {
|
|
pub epoch: u64,
|
|
pub sender_identity: Vec<u8>,
|
|
pub seq: u64,
|
|
pub timestamp_ms: u64,
|
|
pub plaintext: String,
|
|
pub prev_hash: [u8; 32],
|
|
}
|
|
|
|
// ── Internal helpers ─────────────────────────────────────────────────────────
|
|
|
|
fn derive_key(password: &str, salt: &[u8]) -> Result<Zeroizing<[u8; KEY_LEN]>, CoreError> {
|
|
let params = Params::new(ARGON2_M_COST, ARGON2_T_COST, ARGON2_P_COST, Some(KEY_LEN))
|
|
.map_err(|e| CoreError::Mls(format!("argon2 params: {e}")))?;
|
|
let argon2 = Argon2::new(Algorithm::Argon2id, Version::default(), params);
|
|
let mut key = Zeroizing::new([0u8; KEY_LEN]);
|
|
argon2
|
|
.hash_password_into(password.as_bytes(), salt, &mut *key)
|
|
.map_err(|e| CoreError::Mls(format!("transcript key derivation: {e}")))?;
|
|
Ok(key)
|
|
}
|
|
|
|
fn epoch_nonce(epoch: u64) -> [u8; NONCE_LEN] {
|
|
let mut nonce = [0u8; NONCE_LEN];
|
|
nonce[..8].copy_from_slice(&epoch.to_be_bytes());
|
|
nonce
|
|
}
|
|
|
|
fn io_err(e: std::io::Error) -> CoreError {
|
|
CoreError::Mls(format!("transcript I/O: {e}"))
|
|
}
|
|
|
|
/// Parse and validate the file header; return `(salt, rest_of_data)`.
|
|
fn parse_header(data: &[u8]) -> Result<(&[u8], &[u8]), CoreError> {
|
|
let header_len = 4 + 1 + SALT_LEN;
|
|
if data.len() < header_len {
|
|
return Err(CoreError::Mls("transcript: file too short".into()));
|
|
}
|
|
if &data[..4] != MAGIC {
|
|
return Err(CoreError::Mls("transcript: invalid magic bytes".into()));
|
|
}
|
|
if data[4] != VERSION {
|
|
return Err(CoreError::Mls(format!(
|
|
"transcript: unsupported version {}",
|
|
data[4]
|
|
)));
|
|
}
|
|
let salt = &data[5..5 + SALT_LEN];
|
|
let rest = &data[5 + SALT_LEN..];
|
|
Ok((salt, rest))
|
|
}
|
|
|
|
/// Encode one record as CBOR using ciborium.
|
|
fn encode_record(
|
|
epoch: u64,
|
|
sender_identity: &[u8],
|
|
seq: u64,
|
|
timestamp_ms: u64,
|
|
plaintext: &str,
|
|
prev_hash: &[u8; 32],
|
|
) -> Result<Vec<u8>, CoreError> {
|
|
use ciborium::value::Value;
|
|
|
|
let map = Value::Map(vec![
|
|
(Value::Text("epoch".into()), Value::Integer(epoch.into())),
|
|
(Value::Text("sender_identity".into()), Value::Bytes(sender_identity.to_vec())),
|
|
(Value::Text("seq".into()), Value::Integer(seq.into())),
|
|
(Value::Text("timestamp_ms".into()), Value::Integer(timestamp_ms.into())),
|
|
(Value::Text("plaintext".into()), Value::Text(plaintext.into())),
|
|
(Value::Text("prev_hash".into()), Value::Bytes(prev_hash.to_vec())),
|
|
]);
|
|
|
|
let mut buf = Vec::new();
|
|
ciborium::into_writer(&map, &mut buf)
|
|
.map_err(|e| CoreError::Mls(format!("transcript CBOR encode: {e}")))?;
|
|
Ok(buf)
|
|
}
|
|
|
|
/// Decode a CBOR record.
|
|
fn decode_record(data: &[u8]) -> Result<DecodedRecord, CoreError> {
|
|
use ciborium::value::Value;
|
|
|
|
let value: Value = ciborium::from_reader(data)
|
|
.map_err(|e| CoreError::Mls(format!("transcript CBOR decode: {e}")))?;
|
|
|
|
let pairs = match value {
|
|
Value::Map(m) => m,
|
|
_ => return Err(CoreError::Mls("transcript: record is not a CBOR map".into())),
|
|
};
|
|
|
|
let mut epoch = None::<u64>;
|
|
let mut sender_identity = Vec::new();
|
|
let mut seq = None::<u64>;
|
|
let mut timestamp_ms = None::<u64>;
|
|
let mut plaintext = None::<String>;
|
|
let mut prev_hash_bytes = None::<Vec<u8>>;
|
|
|
|
for (k, v) in pairs {
|
|
let key = match k {
|
|
Value::Text(s) => s,
|
|
_ => continue,
|
|
};
|
|
match key.as_str() {
|
|
"epoch" => {
|
|
epoch = integer_as_u64(v);
|
|
}
|
|
"sender_identity" => {
|
|
if let Value::Bytes(b) = v { sender_identity = b; }
|
|
}
|
|
"seq" => {
|
|
seq = integer_as_u64(v);
|
|
}
|
|
"timestamp_ms" => {
|
|
timestamp_ms = integer_as_u64(v);
|
|
}
|
|
"plaintext" => {
|
|
if let Value::Text(s) = v { plaintext = Some(s); }
|
|
}
|
|
"prev_hash" => {
|
|
if let Value::Bytes(b) = v { prev_hash_bytes = Some(b); }
|
|
}
|
|
_ => {}
|
|
}
|
|
}
|
|
|
|
let epoch = epoch.ok_or_else(|| CoreError::Mls("transcript: missing epoch".into()))?;
|
|
let seq = seq.ok_or_else(|| CoreError::Mls("transcript: missing seq".into()))?;
|
|
let timestamp_ms = timestamp_ms
|
|
.ok_or_else(|| CoreError::Mls("transcript: missing timestamp_ms".into()))?;
|
|
let plaintext = plaintext
|
|
.ok_or_else(|| CoreError::Mls("transcript: missing plaintext".into()))?;
|
|
let prev_hash_bytes = prev_hash_bytes
|
|
.ok_or_else(|| CoreError::Mls("transcript: missing prev_hash".into()))?;
|
|
|
|
let mut prev_hash = [0u8; 32];
|
|
if prev_hash_bytes.len() == 32 {
|
|
prev_hash.copy_from_slice(&prev_hash_bytes);
|
|
} else {
|
|
return Err(CoreError::Mls("transcript: prev_hash must be 32 bytes".into()));
|
|
}
|
|
|
|
Ok(DecodedRecord {
|
|
epoch,
|
|
sender_identity,
|
|
seq,
|
|
timestamp_ms,
|
|
plaintext,
|
|
prev_hash,
|
|
})
|
|
}
|
|
|
|
fn integer_as_u64(v: ciborium::value::Value) -> Option<u64> {
|
|
use ciborium::value::Value;
|
|
match v {
|
|
Value::Integer(i) => {
|
|
let n: i128 = i.into();
|
|
if n >= 0 { Some(n as u64) } else { None }
|
|
}
|
|
_ => None,
|
|
}
|
|
}
|
|
|
|
// ── Tests ────────────────────────────────────────────────────────────────────
|
|
|
|
#[cfg(test)]
|
|
mod tests {
|
|
use super::*;
|
|
|
|
#[test]
|
|
fn round_trip_empty() {
|
|
let password = "test-password";
|
|
let mut buf = Vec::new();
|
|
let _writer = TranscriptWriter::new(password, &mut buf).expect("new writer");
|
|
let (records, verdict) = read_transcript(password, &buf).expect("read");
|
|
assert!(records.is_empty());
|
|
assert_eq!(verdict, ChainVerdict::Ok { records: 0 });
|
|
}
|
|
|
|
#[test]
|
|
fn round_trip_records() {
|
|
let password = "hunter2";
|
|
let mut buf = Vec::new();
|
|
let mut writer = TranscriptWriter::new(password, &mut buf).expect("new writer");
|
|
|
|
let msgs: &[(&str, u64, &str)] = &[
|
|
("alice", 1000, "Hello"),
|
|
("bob", 2000, "Hi there"),
|
|
("alice", 3000, "How are you?"),
|
|
];
|
|
|
|
for (_sender, ts, body) in msgs {
|
|
let sender_key = [0u8; 32];
|
|
writer
|
|
.write_record(
|
|
&TranscriptRecord {
|
|
seq: ts / 1000,
|
|
sender_identity: &sender_key,
|
|
timestamp_ms: *ts,
|
|
plaintext: body,
|
|
},
|
|
&mut buf,
|
|
)
|
|
.expect("write record");
|
|
}
|
|
|
|
let (records, verdict) = read_transcript(password, &buf).expect("read");
|
|
assert_eq!(verdict, ChainVerdict::Ok { records: 3 });
|
|
assert_eq!(records.len(), 3);
|
|
assert_eq!(records[0].plaintext, "Hello");
|
|
assert_eq!(records[1].plaintext, "Hi there");
|
|
assert_eq!(records[2].plaintext, "How are you?");
|
|
assert_eq!(records[0].epoch, 0);
|
|
assert_eq!(records[1].epoch, 1);
|
|
assert_eq!(records[2].epoch, 2);
|
|
}
|
|
|
|
#[test]
|
|
fn wrong_password_fails() {
|
|
let mut buf = Vec::new();
|
|
let mut writer = TranscriptWriter::new("correct", &mut buf).expect("new writer");
|
|
writer
|
|
.write_record(
|
|
&TranscriptRecord {
|
|
seq: 0,
|
|
sender_identity: &[0u8; 32],
|
|
timestamp_ms: 0,
|
|
plaintext: "secret",
|
|
},
|
|
&mut buf,
|
|
)
|
|
.expect("write");
|
|
|
|
let result = read_transcript("wrong-password", &buf);
|
|
assert!(result.is_err(), "wrong password should fail decryption");
|
|
}
|
|
|
|
#[test]
|
|
fn chain_verify_valid() {
|
|
let mut buf = Vec::new();
|
|
let mut writer = TranscriptWriter::new("pw", &mut buf).expect("new writer");
|
|
for i in 0..5u64 {
|
|
writer
|
|
.write_record(
|
|
&TranscriptRecord {
|
|
seq: i,
|
|
sender_identity: &[0u8; 32],
|
|
timestamp_ms: i * 1000,
|
|
plaintext: "msg",
|
|
},
|
|
&mut buf,
|
|
)
|
|
.expect("write");
|
|
}
|
|
|
|
let verdict = verify_transcript_chain(&buf).expect("verify");
|
|
assert_eq!(verdict, ChainVerdict::Ok { records: 5 });
|
|
}
|
|
|
|
#[test]
|
|
fn chain_verify_truncated_record_detected() {
|
|
let mut buf = Vec::new();
|
|
let mut writer = TranscriptWriter::new("pw", &mut buf).expect("new writer");
|
|
writer
|
|
.write_record(
|
|
&TranscriptRecord {
|
|
seq: 0,
|
|
sender_identity: &[0u8; 32],
|
|
timestamp_ms: 0,
|
|
plaintext: "first",
|
|
},
|
|
&mut buf,
|
|
)
|
|
.expect("write");
|
|
|
|
// Truncate the last few bytes — should fail parsing.
|
|
let truncated = &buf[..buf.len() - 5];
|
|
let result = verify_transcript_chain(truncated);
|
|
assert!(result.is_err(), "truncated file must be detected");
|
|
}
|
|
}
|