Merge pull request 'Keep passkey sign-in available during anonymous challenge floods' (#536) from timmy/535-passkey-flood-protection into main

README.md

@@ -178,6 +178,11 @@ export STACKCHAIN_DASHBOARD_IDLE_TIMEOUT_SECONDS=900
 export STACKCHAIN_LOGIN_MAX_FAILURES=5
 export STACKCHAIN_LOGIN_WINDOW_SECONDS=300
 export STACKCHAIN_LOGIN_MAX_ENTRIES=10000
+# Public passkey ceremonies share the same window and durable source ledger.
+export STACKCHAIN_PASSKEY_OPTIONS_MAX_ATTEMPTS=10
+# Bound live one-time challenges even if anonymous clients rotate addresses.
+export STACKCHAIN_PASSKEY_MAX_CHALLENGES_PER_SOURCE=10
+export STACKCHAIN_PASSKEY_MAX_CHALLENGES=10000
 # Optional; defaults to STACKCHAIN_STATE_DIR/login-attempts.sqlite3.
 export STACKCHAIN_LOGIN_ATTEMPT_DB='/var/lib/stackchain-dashboard/login-attempts.sqlite3'
 # Trust forwarding headers only from these immediate reverse-proxy networks.
@@ -185,13 +190,16 @@ export STACKCHAIN_TRUSTED_PROXY_CIDRS='127.0.0.0/8'
 uvicorn src.main:app --host 127.0.0.1 --port 8000
 ```
 
-Sign-in failures are scoped to a hashed canonical client address and persisted across
+Token and passkey sign-in failures are scoped to a hashed canonical client address and
-workers and restarts. Once the budget is exhausted, the server returns `429` with
+persisted across workers and restarts. Public passkey option issuance has a separate
-`Retry-After`; the mobile login form disables retries for that interval. Expired
+fixed-window admission budget in the same ledger, and live challenges are bounded per
-source records are pruned and the ledger is size-bounded. Keep its SQLite file on
+source and globally in the session registry. Once either sign-in budget is exhausted,
-shared writable storage. `X-Forwarded-For` is ignored unless the immediate peer is
+the server returns `429` with `Retry-After`; a successful sign-in clears that source's
-inside `STACKCHAIN_TRUSTED_PROXY_CIDRS`; list only networks you operate. Without
+failure state. Expired source records and challenges are pruned, and both ledgers are
-that setting, a reverse proxy is safely treated as one shared source.
+size-bounded. Keep their SQLite files on shared writable storage. `X-Forwarded-For` is
+ignored unless the immediate peer is inside `STACKCHAIN_TRUSTED_PROXY_CIDRS`; list only
+networks you operate. Without that setting, a reverse proxy is safely treated as one
+shared source.
 
 Inbound API mutations are admitted through a body-size boundary before FastAPI
 parses JSON: sign-in is capped at 16 KiB and other `POST`, `PUT`, and `PATCH`

src/login_attempt_store.py

@@ -74,6 +74,17 @@ class LoginAttemptStore:
                 )
                 """
             )
+            connection.execute(
+                """
+                CREATE TABLE IF NOT EXISTS login_admissions (
+                    bucket TEXT NOT NULL,
+                    source_hash TEXT NOT NULL,
+                    attempts INTEGER NOT NULL,
+                    window_started_at REAL NOT NULL,
+                    PRIMARY KEY (bucket, source_hash)
+                )
+                """
+            )
             return connection
         except (OSError, sqlite3.Error) as exc:
             raise LoginAttemptStoreError(
@@ -97,6 +108,52 @@ class LoginAttemptStore:
         remaining = row[1] + self.window_seconds - now
         return max(0, math.ceil(remaining))
 
+    def admit(self, bucket: str, source: str, *, limit: int) -> int:
+        """Consume one fixed-window admission or return seconds until retry."""
+        now = self.clock()
+        source_hash = self._digest(source)
+        limit = max(1, limit)
+        try:
+            with self._connect() as connection:
+                connection.execute("BEGIN IMMEDIATE")
+                connection.execute(
+                    "DELETE FROM login_admissions WHERE window_started_at + ? <= ?",
+                    (self.window_seconds, now),
+                )
+                row = connection.execute(
+                    "SELECT attempts, window_started_at FROM login_admissions "
+                    "WHERE bucket = ? AND source_hash = ?",
+                    (bucket, source_hash),
+                ).fetchone()
+                if row is not None and row[0] >= limit:
+                    return max(1, math.ceil(row[1] + self.window_seconds - now))
+                if row is None:
+                    connection.execute(
+                        "INSERT INTO login_admissions VALUES (?, ?, 1, ?)",
+                        (bucket, source_hash, now),
+                    )
+                else:
+                    connection.execute(
+                        "UPDATE login_admissions SET attempts = attempts + 1 "
+                        "WHERE bucket = ? AND source_hash = ?",
+                        (bucket, source_hash),
+                    )
+                connection.execute(
+                    """
+                    DELETE FROM login_admissions
+                    WHERE (bucket, source_hash) NOT IN (
+                        SELECT bucket, source_hash FROM login_admissions
+                        ORDER BY window_started_at DESC, rowid DESC LIMIT ?
+                    )
+                    """,
+                    (self.max_entries,),
+                )
+                return 0
+        except (OSError, sqlite3.Error) as exc:
+            raise LoginAttemptStoreError(
+                "Sign-in throttling is temporarily unavailable"
+            ) from exc
+
     def record_failure(self, source: str) -> None:
         now = self.clock()
         source_hash = self._digest(source)

src/main.py

@@ -249,7 +249,14 @@ def _passkey_store() -> PasskeyStore:
     database = os.getenv(
         "STACKCHAIN_SESSION_DB", os.path.join(state_dir, "sessions.sqlite3")
     )
-    return PasskeyStore(database, clock=time.time)
+    return PasskeyStore(
+        database,
+        clock=time.time,
+        max_challenges=int(os.getenv("STACKCHAIN_PASSKEY_MAX_CHALLENGES", "10000")),
+        max_challenges_per_source=int(
+            os.getenv("STACKCHAIN_PASSKEY_MAX_CHALLENGES_PER_SOURCE", "10")
+        ),
+    )
 
 
 def _security_event_store() -> SecurityEventStore:
@@ -1339,6 +1346,34 @@ async def create_passkey_authentication_options(request: Request):
     credentials = await asyncio.to_thread(store.all)
     if not credentials:
         raise HTTPException(status_code=404, detail="No passkeys enrolled")
+    peer_host = request.client.host if request.client is not None else "unknown"
+    source = client_source(
+        peer_host,
+        request.headers.get("x-forwarded-for", ""),
+        os.getenv("STACKCHAIN_TRUSTED_PROXY_CIDRS", ""),
+    )
+    attempts = _login_attempt_store()
+    try:
+        retry_after = await asyncio.to_thread(attempts.retry_after, source)
+        if not retry_after:
+            retry_after = await asyncio.to_thread(
+                attempts.admit,
+                "passkey_options",
+                source,
+                limit=int(os.getenv("STACKCHAIN_PASSKEY_OPTIONS_MAX_ATTEMPTS", "10")),
+            )
+    except LoginAttemptStoreError:
+        return JSONResponse(
+            {"detail": "Sign-in throttling is temporarily unavailable"},
+            status_code=503,
+            headers={"Cache-Control": "no-store"},
+        )
+    if retry_after:
+        return JSONResponse(
+            {"detail": "Too many passkey sign-in attempts"},
+            status_code=429,
+            headers={"Cache-Control": "no-store", "Retry-After": str(retry_after)},
+        )
     rp_id, _origin = _passkey_relying_party(request)
     options, challenge = passkeys.authentication_options(
         rp_id=rp_id,
@@ -1351,6 +1386,7 @@ async def create_passkey_authentication_options(request: Request):
         purpose="authentication",
         action="sign_in",
         target="dashboard",
+        source=source,
     )
     return JSONResponse(options, headers={"Cache-Control": "no-store"})
 
@@ -1437,10 +1473,39 @@ async def verify_passkey_authentication(
 ):
     if payload.action != "sign_in" or payload.target != "dashboard":
         raise HTTPException(status_code=400, detail="Invalid passkey sign-in target")
+    peer_host = request.client.host if request.client is not None else "unknown"
+    source = client_source(
+        peer_host,
+        request.headers.get("x-forwarded-for", ""),
+        os.getenv("STACKCHAIN_TRUSTED_PROXY_CIDRS", ""),
+    )
+    attempts = _login_attempt_store()
+    try:
+        retry_after = await asyncio.to_thread(attempts.retry_after, source)
+    except LoginAttemptStoreError:
+        return JSONResponse(
+            {"detail": "Sign-in throttling is temporarily unavailable"},
+            status_code=503,
+            headers={"Cache-Control": "no-store"},
+        )
+    if retry_after:
+        return JSONResponse(
+            {"detail": "Too many passkey sign-in attempts"},
+            status_code=429,
+            headers={"Cache-Control": "no-store", "Retry-After": str(retry_after)},
+        )
     try:
         challenge = passkeys.decode(payload.challenge)
         credential_id = passkeys.decode(str(payload.credential.get("id", "")))
     except (ValueError, TypeError):
+        try:
+            await asyncio.to_thread(attempts.record_failure, source)
+        except LoginAttemptStoreError:
+            return JSONResponse(
+                {"detail": "Sign-in throttling is temporarily unavailable"},
+                status_code=503,
+                headers={"Cache-Control": "no-store"},
+            )
         raise HTTPException(status_code=400, detail="Invalid passkey ceremony")
     store = _passkey_store()
     valid = await asyncio.to_thread(
@@ -1453,6 +1518,14 @@ async def verify_passkey_authentication(
     )
     stored = await asyncio.to_thread(store.get, credential_id)
     if not valid or stored is None:
+        try:
+            await asyncio.to_thread(attempts.record_failure, source)
+        except LoginAttemptStoreError:
+            return JSONResponse(
+                {"detail": "Sign-in throttling is temporarily unavailable"},
+                status_code=503,
+                headers={"Cache-Control": "no-store"},
+            )
         raise HTTPException(status_code=401, detail="Passkey sign-in failed")
     rp_id, origin = _passkey_relying_party(request)
     try:
@@ -1478,7 +1551,24 @@ async def verify_passkey_authentication(
     except dashboard_auth.SessionStoreError:
         raise HTTPException(status_code=503, detail="Passkey registry is temporarily unavailable")
     except Exception as exc:
+        try:
+            await asyncio.to_thread(attempts.record_failure, source)
+        except LoginAttemptStoreError:
+            return JSONResponse(
+                {"detail": "Sign-in throttling is temporarily unavailable"},
+                status_code=503,
+                headers={"Cache-Control": "no-store"},
+            )
         raise HTTPException(status_code=401, detail="Passkey sign-in failed") from exc
+    try:
+        await asyncio.to_thread(attempts.clear, source)
+    except LoginAttemptStoreError:
+        await asyncio.to_thread(dashboard_auth.revoke_session, session)
+        return JSONResponse(
+            {"detail": "Sign-in throttling is temporarily unavailable"},
+            status_code=503,
+            headers={"Cache-Control": "no-store"},
+        )
     try:
         await asyncio.to_thread(
             _security_event_store().record,

src/passkey_store.py

@@ -20,9 +20,18 @@ class StoredPasskey:
 
 
 class PasskeyStore:
-    def __init__(self, path: str | Path, *, clock: Callable[[], float]) -> None:
+    def __init__(
+        self,
+        path: str | Path,
+        *,
+        clock: Callable[[], float],
+        max_challenges: int = 10_000,
+        max_challenges_per_source: int = 10,
+    ) -> None:
         self.path = Path(path)
         self.clock = clock
+        self.max_challenges = max(1, max_challenges)
+        self.max_challenges_per_source = max(1, max_challenges_per_source)
 
     @staticmethod
     def _digest(value: bytes | str) -> str:
@@ -50,6 +59,7 @@ class PasskeyStore:
                 CREATE TABLE IF NOT EXISTS passkey_challenges (
                     challenge_hash TEXT PRIMARY KEY,
                     session_hash TEXT,
+                    source_hash TEXT,
                     purpose TEXT NOT NULL,
                     action TEXT NOT NULL,
                     target TEXT NOT NULL,
@@ -57,6 +67,14 @@ class PasskeyStore:
                 )
                 """
             )
+            columns = {
+                row[1]
+                for row in connection.execute("PRAGMA table_info(passkey_challenges)")
+            }
+            if "source_hash" not in columns:
+                connection.execute(
+                    "ALTER TABLE passkey_challenges ADD COLUMN source_hash TEXT"
+                )
             return connection
         except (OSError, sqlite3.Error) as exc:
             raise SessionStoreError("Passkey registry is temporarily unavailable") from exc
@@ -69,25 +87,50 @@ class PasskeyStore:
         purpose: str,
         action: str,
         target: str,
+        source: str | None = None,
         ttl_seconds: int = 120,
     ) -> None:
         now = int(self.clock())
+        source_hash = self._digest(source or session_id or f"{purpose}:{action}:{target}")
         try:
             with self._connect() as connection:
+                connection.execute("BEGIN IMMEDIATE")
                 connection.execute("DELETE FROM passkey_challenges WHERE expires_at <= ?", (now,))
                 connection.execute(
                     "INSERT INTO passkey_challenges("
-                    "challenge_hash, session_hash, purpose, action, target, expires_at"
+                    "challenge_hash, session_hash, source_hash, purpose, action, target, expires_at"
-                    ") VALUES (?, ?, ?, ?, ?, ?)",
+                    ") VALUES (?, ?, ?, ?, ?, ?, ?)",
                     (
                         self._digest(challenge),
                         self._digest(session_id) if session_id else None,
+                        source_hash,
                         purpose,
                         action,
                         target,
                         now + max(1, ttl_seconds),
                     ),
                 )
+                connection.execute(
+                    """
+                    DELETE FROM passkey_challenges
+                    WHERE source_hash = ? AND challenge_hash NOT IN (
+                        SELECT challenge_hash FROM passkey_challenges
+                        WHERE source_hash = ?
+                        ORDER BY expires_at DESC, rowid DESC LIMIT ?
+                    )
+                    """,
+                    (source_hash, source_hash, self.max_challenges_per_source),
+                )
+                connection.execute(
+                    """
+                    DELETE FROM passkey_challenges
+                    WHERE challenge_hash NOT IN (
+                        SELECT challenge_hash FROM passkey_challenges
+                        ORDER BY expires_at DESC, rowid DESC LIMIT ?
+                    )
+                    """,
+                    (self.max_challenges,),
+                )
         except (OSError, sqlite3.Error) as exc:
             raise SessionStoreError("Passkey registry is temporarily unavailable") from exc
 

tests/test_dashboard_auth.py

@@ -167,6 +167,128 @@ async def test_enrolled_passkey_can_sign_in_without_the_operator_token(
     assert "correct horse battery staple" not in signed_in.text
 
 
+@pytest.mark.anyio
+async def test_passkey_options_are_source_limited_before_challenge_generation(
+    access_control, monkeypatch
+):
+    monkeypatch.setenv("STACKCHAIN_PASSKEY_OPTIONS_MAX_ATTEMPTS", "2")
+    await asyncio.to_thread(
+        main._passkey_store().register,
+        credential_id=b"phone-credential",
+        public_key=b"phone-public-key",
+        sign_count=0,
+        device_label="Phone",
+        management_id="phone-management-id",
+    )
+    first_source = httpx.ASGITransport(app=main.app, client=("203.0.113.7", 1234))
+    other_source = httpx.ASGITransport(app=main.app, client=("203.0.113.8", 1234))
+
+    async with httpx.AsyncClient(transport=first_source, base_url="https://test") as client:
+        first = await client.post("/api/v1/passkeys/authentication/options")
+        second = await client.post("/api/v1/passkeys/authentication/options")
+        blocked = await client.post("/api/v1/passkeys/authentication/options")
+    async with httpx.AsyncClient(transport=other_source, base_url="https://test") as client:
+        available = await client.post("/api/v1/passkeys/authentication/options")
+
+    assert first.status_code == second.status_code == available.status_code == 200
+    assert blocked.status_code == 429
+    assert blocked.json() == {"detail": "Too many passkey sign-in attempts"}
+    assert blocked.headers["retry-after"] == "60"
+    with sqlite3.connect(main._passkey_store().path) as connection:
+        assert connection.execute(
+            "SELECT COUNT(*) FROM passkey_challenges WHERE purpose = 'authentication'"
+        ).fetchone() == (3,)
+
+
+@pytest.mark.anyio
+async def test_failed_passkey_verification_consumes_the_shared_sign_in_budget(
+    access_control,
+):
+    await asyncio.to_thread(
+        main._passkey_store().register,
+        credential_id=b"phone-credential",
+        public_key=b"phone-public-key",
+        sign_count=0,
+        device_label="Phone",
+        management_id="phone-management-id",
+    )
+    transport = httpx.ASGITransport(app=main.app, client=("203.0.113.9", 1234))
+
+    async with httpx.AsyncClient(transport=transport, base_url="https://test") as client:
+        failures = []
+        for _attempt in range(3):
+            options = await client.post("/api/v1/passkeys/authentication/options")
+            failures.append(
+                await client.post(
+                    "/api/v1/passkeys/authentication/verify",
+                    json={
+                        "challenge": options.json()["challenge"],
+                        "credential": {"id": "d3JvbmctY3JlZGVudGlhbA"},
+                        "device_label": "Phone",
+                        "action": "sign_in",
+                        "target": "dashboard",
+                    },
+                )
+            )
+        blocked = await client.post("/api/v1/passkeys/authentication/options")
+
+    assert [response.status_code for response in failures] == [401, 401, 401]
+    assert blocked.status_code == 429
+    assert blocked.headers["retry-after"] == "60"
+
+
+@pytest.mark.anyio
+async def test_successful_passkey_sign_in_clears_prior_source_failures(
+    access_control, monkeypatch
+):
+    class VerifiedAuthentication:
+        new_sign_count = 1
+
+    monkeypatch.setattr(
+        main.passkeys,
+        "verify_authentication",
+        lambda **_kwargs: VerifiedAuthentication(),
+    )
+    _signed, _managed_session = await asyncio.to_thread(
+        main.dashboard_auth.issue_session,
+        device_label="Phone",
+        management_id="phone-management-id",
+    )
+    await asyncio.to_thread(
+        main._passkey_store().register,
+        credential_id=b"phone-credential",
+        public_key=b"phone-public-key",
+        sign_count=0,
+        device_label="Phone",
+        management_id="phone-management-id",
+    )
+    source = "203.0.113.10"
+    attempts = main._login_attempt_store()
+    await asyncio.to_thread(attempts.record_failure, source)
+    await asyncio.to_thread(attempts.record_failure, source)
+    transport = httpx.ASGITransport(app=main.app, client=(source, 1234))
+
+    async with httpx.AsyncClient(transport=transport, base_url="https://test") as client:
+        options = await client.post("/api/v1/passkeys/authentication/options")
+        signed_in = await client.post(
+            "/api/v1/passkeys/authentication/verify",
+            json={
+                "challenge": options.json()["challenge"],
+                "credential": {"id": "cGhvbmUtY3JlZGVudGlhbA"},
+                "device_label": "Phone",
+                "action": "sign_in",
+                "target": "dashboard",
+            },
+        )
+
+    assert signed_in.status_code == 200, signed_in.text
+    with sqlite3.connect(attempts.path) as connection:
+        assert connection.execute(
+            "SELECT COUNT(*) FROM login_attempts WHERE source_hash = ?",
+            (attempts._digest(source),),
+        ).fetchone() == (0,)
+
+
 @pytest.mark.anyio
 async def test_passkey_fresh_authorization_is_exact_target_bound_and_single_use(
     access_control, monkeypatch

tests/test_login_attempt_store.py

@@ -72,3 +72,28 @@ def test_failure_ledger_evicts_oldest_sources_at_its_size_limit(tmp_path):
 
     with sqlite3.connect(store.path) as connection:
         assert connection.execute("SELECT COUNT(*) FROM login_attempts").fetchone() == (2,)
+
+
+def test_named_admission_budget_is_atomic_across_instances_and_source_scoped(tmp_path):
+    now = [1_000.0]
+    database = tmp_path / "login-attempts.sqlite3"
+    first = LoginAttemptStore(
+        database,
+        clock=lambda: now[0],
+        max_failures=3,
+        window_seconds=60,
+    )
+    second = LoginAttemptStore(
+        database,
+        clock=lambda: now[0],
+        max_failures=3,
+        window_seconds=60,
+    )
+
+    assert first.admit("passkey_options", "203.0.113.7", limit=2) == 0
+    assert second.admit("passkey_options", "203.0.113.7", limit=2) == 0
+    assert first.admit("passkey_options", "203.0.113.7", limit=2) == 60
+    assert first.admit("passkey_options", "203.0.113.8", limit=2) == 0
+
+    now[0] = 1_060.0
+    assert second.admit("passkey_options", "203.0.113.7", limit=2) == 0

tests/test_passkey_store.py

@@ -0,0 +1,106 @@
+import sqlite3
+
+from src.passkey_store import PasskeyStore
+
+
+def test_active_challenges_are_bounded_per_source_and_globally_across_instances(tmp_path):
+    now = [1_000.0]
+    database = tmp_path / "passkeys.sqlite3"
+    first = PasskeyStore(
+        database,
+        clock=lambda: now[0],
+        max_challenges=3,
+        max_challenges_per_source=2,
+    )
+    second = PasskeyStore(
+        database,
+        clock=lambda: now[0],
+        max_challenges=3,
+        max_challenges_per_source=2,
+    )
+
+    for challenge in (b"first", b"second", b"third"):
+        first.issue_challenge(
+            challenge,
+            session_id=None,
+            purpose="authentication",
+            action="sign_in",
+            target="dashboard",
+            source="203.0.113.7",
+        )
+    second.issue_challenge(
+        b"other-one",
+        session_id=None,
+        purpose="authentication",
+        action="sign_in",
+        target="dashboard",
+        source="203.0.113.8",
+    )
+    second.issue_challenge(
+        b"other-two",
+        session_id=None,
+        purpose="authentication",
+        action="sign_in",
+        target="dashboard",
+        source="203.0.113.8",
+    )
+
+    with sqlite3.connect(database) as connection:
+        assert connection.execute("SELECT COUNT(*) FROM passkey_challenges").fetchone() == (3,)
+        counts = connection.execute(
+            "SELECT source_hash, COUNT(*) FROM passkey_challenges GROUP BY source_hash"
+        ).fetchall()
+    assert sorted(count for _source, count in counts) == [1, 2]
+
+    now[0] = 1_121.0
+    first.issue_challenge(
+        b"after-expiry",
+        session_id=None,
+        purpose="authentication",
+        action="sign_in",
+        target="dashboard",
+        source="203.0.113.9",
+    )
+    with sqlite3.connect(database) as connection:
+        assert connection.execute("SELECT COUNT(*) FROM passkey_challenges").fetchone() == (1,)
+
+
+def test_existing_challenge_registry_is_migrated_without_losing_live_challenges(tmp_path):
+    database = tmp_path / "passkeys.sqlite3"
+    with sqlite3.connect(database) as connection:
+        connection.execute(
+            """
+            CREATE TABLE passkey_challenges (
+                challenge_hash TEXT PRIMARY KEY,
+                session_hash TEXT,
+                purpose TEXT NOT NULL,
+                action TEXT NOT NULL,
+                target TEXT NOT NULL,
+                expires_at INTEGER NOT NULL
+            )
+            """
+        )
+        connection.execute(
+            "INSERT INTO passkey_challenges VALUES ('existing', NULL, "
+            "'authentication', 'sign_in', 'dashboard', 1120)"
+        )
+    store = PasskeyStore(database, clock=lambda: 1_000.0)
+
+    store.issue_challenge(
+        b"new",
+        session_id=None,
+        purpose="authentication",
+        action="sign_in",
+        target="dashboard",
+        source="203.0.113.7",
+    )
+
+    with sqlite3.connect(database) as connection:
+        columns = {
+            row[1] for row in connection.execute("PRAGMA table_info(passkey_challenges)")
+        }
+        rows = connection.execute(
+            "SELECT challenge_hash FROM passkey_challenges ORDER BY challenge_hash"
+        ).fetchall()
+    assert "source_hash" in columns
+    assert sorted(rows) == sorted([("existing",), (store._digest(b"new"),)])