feat(opaque): wire API

- POST /api/auth/opaque/login/ke1 (public) — takes {userIdentifier, startLoginRequest}, resolves the identifier via the same @-dispatch as legacy login (AuthApplicationService::lookup_user_for_login, factored out), fetches envelope, runs ServerLogin::start (real branch for known users, dummy branch for anti-enum on unknown/unregistered), stashes state under a random exchange_id in the moka cache, returns {exchangeId, loginResponse}.
- POST /api/auth/opaque/login/ke3 (public) — atomic take from the cache FIRST (anti-enum + anti-replay), then decodes the payload, runs ServerLogin::finish, stamps opaque_migrated_at (Phase 3 signal), and mints a session via the new AuthApplicationService::mint_session_for_authenticated_user helper — returns the same AuthResponseDto shape as legacy login so the SPA has one downstream handler.
- Session mint factored: mint_session_for_authenticated_user(User) extracted from login() so both the legacy password path and OPAQUE KE3 converge through one implementation.
- OpaqueRepositoryPort::mark_migrated with COALESCE-preserving idempotent stamp of opaque_migrated_at.
- opaque-setup CLI + Dockerfile wiring already shipped (Step 0 hygiene).
- Routing fix: sub-prefix split (/api/auth/opaque/register vs /api/auth/opaque/login) — axum composes middleware between sibling nests at the same prefix, which was cross-applying auth+CSRF to my public login routes. Distinct prefixes side-step that cleanly. Documented in both main.rs and the router builder doc.
- Rate-limit sharing: login KE1/KE3 layered with the SAME login_limiter instance as legacy POST /api/auth/login, so an attacker can't halve the per-IP budget by spraying both endpoints.

Anti-enum + anti-replay hardening in KE3: take runs BEFORE payload parse so:
- Unknown / expired / already-consumed exchange_id → 401 InvalidCredentials (same shape as wrong-passphrase, no payload-shape leak)
- Consumed handle can't be re-used to spam parse attempts
This commit is contained in:
Edouard Vanbelle
2026-07-27 22:23:15 +02:00
parent ae65c8475f
commit 7d7621e387
7 changed files with 696 additions and 49 deletions
@@ -125,6 +125,30 @@ impl OpaqueRepositoryPort for OpaquePgRepository {
}
}
async fn mark_migrated(&self, user_id: Uuid) -> Result<()> {
// COALESCE preserves the first-migration timestamp — same
// pattern as write_registration preserves opaque_registered_at.
// Ops dashboards read this to answer "what fraction of users
// have completed the OPAQUE cutover", so overwriting on every
// login would erase the signal.
let res = sqlx::query(
r#"
UPDATE auth.users
SET opaque_migrated_at = COALESCE(opaque_migrated_at, NOW())
WHERE id = $1
"#,
)
.bind(user_id)
.execute(self.pool())
.await
.map_err(|e| DomainError::internal_error("OpaquePg", format!("mark_migrated: {e}")))?;
if res.rows_affected() == 0 {
return Err(DomainError::not_found("User", user_id.to_string()));
}
Ok(())
}
async fn clear_registration(&self, user_id: Uuid) -> Result<()> {
// One UPDATE writes both the envelope invalidation AND the
// force-change flag — matches the atomicity we promise in the
@@ -438,6 +462,54 @@ mod integration_tests {
);
}
/// Mark_migrated is idempotent — a second call preserves the
/// first-migration timestamp, matching the ops-dashboard contract
/// ("when did this user first complete OPAQUE?").
#[tokio::test]
async fn mark_migrated_stamps_once_and_is_idempotent() {
let repo = test_repo().await;
let user =
seed_user(&repo, &format!("opaque-mig-{}@example.invalid", Uuid::new_v4())).await;
// Read the initial NULL state
let initial: (Option<chrono::DateTime<chrono::Utc>>,) = sqlx::query_as(
"SELECT opaque_migrated_at FROM auth.users WHERE id = $1",
)
.bind(user)
.fetch_one(repo.pool())
.await
.unwrap();
assert!(
initial.0.is_none(),
"new user starts with no opaque_migrated_at"
);
repo.mark_migrated(user).await.expect("first mark");
let first: (Option<chrono::DateTime<chrono::Utc>>,) = sqlx::query_as(
"SELECT opaque_migrated_at FROM auth.users WHERE id = $1",
)
.bind(user)
.fetch_one(repo.pool())
.await
.unwrap();
let first_ts = first.0.expect("timestamp set after first mark");
tokio::time::sleep(std::time::Duration::from_millis(50)).await;
repo.mark_migrated(user).await.expect("second mark");
let second: (Option<chrono::DateTime<chrono::Utc>>,) = sqlx::query_as(
"SELECT opaque_migrated_at FROM auth.users WHERE id = $1",
)
.bind(user)
.fetch_one(repo.pool())
.await
.unwrap();
assert_eq!(
second.0.unwrap(),
first_ts,
"second mark_migrated preserves the first timestamp (COALESCE)"
);
}
#[tokio::test]
async fn missing_user_surfaces_notfound_on_write_and_read_and_clear() {
let repo = test_repo().await;
@@ -449,6 +521,7 @@ mod integration_tests {
.unwrap_err(),
repo.read_registration(ghost).await.unwrap_err(),
repo.clear_registration(ghost).await.unwrap_err(),
repo.mark_migrated(ghost).await.unwrap_err(),
] {
assert_eq!(
err.kind,
@@ -71,10 +71,39 @@ pub const DEFAULT_MAX_INFLIGHT: u64 = 10_000;
/// correct passphrase).
pub type ExchangeId = Uuid;
/// Server-side state stashed between KE1 and KE3.
///
/// Holds the [`ServerLogin`] (opaque-ke handshake continuation) plus
/// the `user_id` KE1 resolved from the client's identifier — `Some`
/// for a real user, `None` for the anti-enum dummy branch (unknown
/// user or user without an envelope). KE3 uses this to know which
/// account to mint the session for on success; the dummy branch's
/// KE3 will fail the AKE check inside `ServerLogin::finish` and
/// never reach the session mint, so the `None` variant is a
/// belt-and-braces guard we never rely on for correctness.
///
/// `state` lives in a `Mutex<Option<_>>` because moka's `Cache::remove`
/// returns a *clone* of the stored `Arc` (not the original). That means
/// unwrapping the Arc to consume `ServerLogin` (which isn't `Clone`)
/// would race the briefly-lingering cache-side ref. `Mutex::lock().take()`
/// gives us ownership of the inner value regardless of how many Arc
/// clones exist.
struct StashedExchange {
state: std::sync::Mutex<Option<ServerLogin<OxiCloudSuite>>>,
user_id: Option<Uuid>,
}
/// Exchange state handed back to the KE3 handler — a plain owned
/// tuple so consumers work with the concrete types, not Arc<Mutex<_>>.
pub struct TakenExchange {
pub state: ServerLogin<OxiCloudSuite>,
pub user_id: Option<Uuid>,
}
/// In-memory cache holding server-side login state between KE1 and KE3.
#[derive(Clone)]
pub struct OpaqueLoginExchange {
inner: Arc<Cache<ExchangeId, ServerLogin<OxiCloudSuite>>>,
inner: Arc<Cache<ExchangeId, Arc<StashedExchange>>>,
}
impl OpaqueLoginExchange {
@@ -97,25 +126,43 @@ impl OpaqueLoginExchange {
}
}
/// Stash a fresh `ServerLogin` state and return the handle to
/// hand back to the client. The exchange_id is generated here so
/// callers can't accidentally reuse one — every KE1 gets its own.
pub fn store(&self, state: ServerLogin<OxiCloudSuite>) -> ExchangeId {
/// Stash a fresh exchange and return the handle to hand back to
/// the client. The exchange_id is generated here so callers
/// can't accidentally reuse one — every KE1 gets its own.
pub fn store(&self, state: ServerLogin<OxiCloudSuite>, user_id: Option<Uuid>) -> ExchangeId {
let id = Uuid::new_v4();
self.inner.insert(id, state);
self.inner.insert(
id,
Arc::new(StashedExchange {
state: std::sync::Mutex::new(Some(state)),
user_id,
}),
);
id
}
/// Atomically consume the state for `exchange_id`. Returns `None`
/// if the id is unknown, already consumed, or expired. Callers
/// must treat those three cases identically (anti-enum): a KE3
/// with a bad id, a replay, and a timeout should all surface as
/// the same `InvalidCredentials` shape to the client.
/// Atomically consume the exchange for `exchange_id`. Returns
/// `None` if the id is unknown, already consumed, or expired.
/// Callers must treat those three cases identically (anti-enum):
/// a KE3 with a bad id, a replay, and a timeout should all
/// surface as the same `InvalidCredentials` shape to the client.
///
/// Uses moka's atomic `remove` (verified single get-and-invalidate
/// in moka 0.12+, no race window between the two operations).
pub fn take(&self, exchange_id: ExchangeId) -> Option<ServerLogin<OxiCloudSuite>> {
self.inner.remove(&exchange_id)
/// Combines two atomicity guarantees:
///
/// * `moka::Cache::remove` is a single get-and-invalidate on
/// the cache side (verified in moka 0.12+).
/// * `Mutex::lock().take()` gives us ownership of the inner
/// `ServerLogin` even when moka returns a *clone* of the
/// stored `Arc` (which it does — `remove` yields a clone,
/// not the original), and prevents two concurrent takers
/// from both seeing `Some`.
pub fn take(&self, exchange_id: ExchangeId) -> Option<TakenExchange> {
let arc = self.inner.remove(&exchange_id)?;
let state = arc.state.lock().ok()?.take()?;
Some(TakenExchange {
state,
user_id: arc.user_id,
})
}
/// Force runtime maintenance (LRU eviction + TTL sweep). Moka runs
@@ -206,9 +253,10 @@ mod tests {
let cache = OpaqueLoginExchange::with_params(Duration::from_secs(60), 100);
let state = build_server_login_state();
let id = cache.store(state);
let id = cache.store(state, Some(Uuid::new_v4()));
// Second call after take must miss — single-use semantic.
assert!(cache.take(id).is_some(), "first take retrieves the state");
let taken = cache.take(id).expect("first take retrieves the state");
assert!(taken.user_id.is_some(), "user_id round-trips through the stash");
assert!(
cache.take(id).is_none(),
"second take must miss — exchange_id is single-use"
@@ -228,7 +276,7 @@ mod tests {
// (it runs pending tasks lazily), so `run_pending_tasks`
// forces a deterministic sweep.
let cache = OpaqueLoginExchange::with_params(Duration::from_millis(100), 100);
let id = cache.store(build_server_login_state());
let id = cache.store(build_server_login_state(), None);
std::thread::sleep(Duration::from_millis(150));
cache.run_pending_tasks();
@@ -245,8 +293,19 @@ mod tests {
// — reusing one would enable a KE3 to consume the wrong
// exchange's state.
let cache = OpaqueLoginExchange::new();
let a = cache.store(build_server_login_state());
let b = cache.store(build_server_login_state());
let a = cache.store(build_server_login_state(), None);
let b = cache.store(build_server_login_state(), None);
assert_ne!(a, b, "each store() must mint a fresh UUID");
}
#[test]
fn dummy_branch_stash_carries_no_user_id() {
// KE1 for an unknown user stashes `user_id: None` (anti-enum
// — dummy branch). KE3 for the dummy will fail at AKE, but
// the stash contract is: real user → Some(id), unknown → None.
let cache = OpaqueLoginExchange::new();
let id = cache.store(build_server_login_state(), None);
let taken = cache.take(id).expect("take dummy stash");
assert!(taken.user_id.is_none(), "dummy-branch stash carries no user_id");
}
}