fix(auth): harden OAuth/session for hosted multi-tenant deployment (#626)

Pre-launch hardening for the hosted Astrolabe Cloud offering. Addresses
all five findings raised in #626 (Tim Kaufmann, code review of v0.65.0).
Re-verified against master before fixing.

Finding 3 (LLM-controllable user_id) — drop user_id from the public
signatures of provision_nextcloud_access, revoke_nextcloud_access,
check_provisioning_status, check_logged_in. Tool wrappers now always
derive identity from the verified AccessToken; user_id is no longer
accepted as MCP input. Adds parameterized CI-guard test that locks the
schema.

Finding 2 (predictable session cookie) — replace mcp_session=<user_id>
cookie with a cryptographically random session_id mapped server-side
(new browser_sessions table, alembic 005). Cookie value is opaque,
expires, revocable. SessionAuthBackend looks up user_id via the new
mapping and additionally requires a refresh token to fail closed.

Finding 4 (logout doesn't revoke refresh token) — oauth_logout now
calls the IdP revocation_endpoint (RFC 7009) when advertised, deletes
the stored refresh token regardless, and clears the browser_sessions
row. Cleanup is best-effort: logout always 302s.

Finding 1 (unverified ID token decodes) — verify_id_token helper does
JWKS signature + issuer + audience + exp + nonce checks per OIDC core
3.1.3.7. Used by both OAuth callback handlers (browser + MCP). Removes
the four "verify_signature: False" decodes that previously trusted IdP
claims unconditionally. Drops dead-code _validate_token_audience in
token_broker. Refactors token_utils + provisioning_decorator to read
user_id from the verified AccessToken instead of re-decoding the JWT.

Finding 5 (hardcoded Fernet keys in docker-compose.yml) — replace the
three inline TOKEN_ENCRYPTION_KEY values with required env var
interpolation; document in env.sample.

Test coverage: 4 new unit test modules (signature pinning, browser
sessions, ID-token verification, logout + revoke + session backend).
693 unit tests pass; ruff/format/ty clean.

Migration note: existing browser admin-UI sessions become invalid on
rollout (cookies are looked up against the new browser_sessions table,
which starts empty). Users re-login. MCP API access is unaffected.

Tracked on Astrolabe Cloud POC board card #37.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Chris Coutinho
2026-05-02 17:03:57 +02:00
co-authored by Claude Opus 4.7
parent 83f2e88d2c
commit 15dbb26349
16 changed files with 1184 additions and 242 deletions
+75
View File
@@ -0,0 +1,75 @@
"""Unit tests for browser_sessions storage (issue #626 finding 2).
The browser admin UI no longer uses the raw user_id as the cookie value
— it uses a cryptographically random session_id mapped server-side to
user_id. These tests pin the storage contract.
"""
import secrets
import tempfile
import time
from pathlib import Path
import pytest
from cryptography.fernet import Fernet
from nextcloud_mcp_server.auth.storage import RefreshTokenStorage
pytestmark = pytest.mark.unit
@pytest.fixture
async def storage():
with tempfile.TemporaryDirectory() as tmpdir:
db_path = Path(tmpdir) / "test_browser_sessions.db"
s = RefreshTokenStorage(
db_path=str(db_path), encryption_key=Fernet.generate_key().decode()
)
await s.initialize()
yield s
async def test_create_and_get_browser_session(storage):
sid = secrets.token_urlsafe(32)
await storage.create_browser_session(session_id=sid, user_id="alice")
user_id = await storage.get_browser_session_user(sid)
assert user_id == "alice"
async def test_get_browser_session_unknown_returns_none(storage):
assert await storage.get_browser_session_user("does-not-exist") is None
async def test_delete_browser_session(storage):
sid = secrets.token_urlsafe(32)
await storage.create_browser_session(session_id=sid, user_id="alice")
deleted = await storage.delete_browser_session(sid)
assert deleted is True
assert await storage.get_browser_session_user(sid) is None
async def test_expired_browser_session_rejected_and_deleted(storage):
sid = secrets.token_urlsafe(32)
# ttl_seconds=0 so the row is immediately expired (now == expires_at)
await storage.create_browser_session(session_id=sid, user_id="alice", ttl_seconds=0)
# Make sure clock advances past expires_at
time.sleep(0.01)
assert await storage.get_browser_session_user(sid) is None
# Expired row should be deleted on encounter
assert await storage.delete_browser_session(sid) is False
async def test_replace_existing_session_id(storage):
"""INSERT OR REPLACE so re-using a session_id rebinds the user.
Not a recommended call pattern (session_ids are random), but the
storage layer must not raise UNIQUE constraint errors if it happens.
"""
sid = secrets.token_urlsafe(32)
await storage.create_browser_session(session_id=sid, user_id="alice")
await storage.create_browser_session(session_id=sid, user_id="bob")
assert await storage.get_browser_session_user(sid) == "bob"
+233
View File
@@ -0,0 +1,233 @@
"""Unit tests for OIDC ID token verification (issue #626 finding 1).
The OAuth callback handlers used to call
`jwt.decode(id_token, options={"verify_signature": False})` and trust the
result. They now go through `verify_id_token`, which checks signature
against JWKS and validates issuer / audience / exp / nonce per OIDC core
spec §3.1.3.7.
"""
import json
import time
from base64 import urlsafe_b64encode
from unittest.mock import patch
import httpx
import jwt
import pytest
from cryptography.hazmat.primitives import serialization
from cryptography.hazmat.primitives.asymmetric import rsa
from nextcloud_mcp_server.auth.token_utils import (
IdTokenVerificationError,
verify_id_token,
)
pytestmark = pytest.mark.unit
# Generated once per process — RSA keypair generation is slow.
_KEY = rsa.generate_private_key(public_exponent=65537, key_size=2048)
_PRIVATE_PEM = _KEY.private_bytes(
encoding=serialization.Encoding.PEM,
format=serialization.PrivateFormat.TraditionalOpenSSL,
encryption_algorithm=serialization.NoEncryption(),
)
_OTHER_KEY = rsa.generate_private_key(public_exponent=65537, key_size=2048)
_OTHER_PRIVATE_PEM = _OTHER_KEY.private_bytes(
encoding=serialization.Encoding.PEM,
format=serialization.PrivateFormat.TraditionalOpenSSL,
encryption_algorithm=serialization.NoEncryption(),
)
ISSUER = "https://idp.example.com"
DISCOVERY_URL = f"{ISSUER}/.well-known/openid-configuration"
JWKS_URI = f"{ISSUER}/jwks"
def _b64u_uint(n: int) -> str:
raw = n.to_bytes((n.bit_length() + 7) // 8, "big")
return urlsafe_b64encode(raw).rstrip(b"=").decode("ascii")
def _build_jwks() -> dict:
pub = _KEY.public_key().public_numbers()
return {
"keys": [
{
"kty": "RSA",
"use": "sig",
"kid": "test-key-1",
"alg": "RS256",
"n": _b64u_uint(pub.n),
"e": _b64u_uint(pub.e),
}
]
}
def _sign(
claims: dict, *, kid: str = "test-key-1", key_pem: bytes = _PRIVATE_PEM
) -> str:
return jwt.encode(claims, key_pem, algorithm="RS256", headers={"kid": kid})
def _idp_handler(request: httpx.Request) -> httpx.Response:
if str(request.url) == DISCOVERY_URL:
return httpx.Response(200, json={"issuer": ISSUER, "jwks_uri": JWKS_URI})
if str(request.url) == JWKS_URI:
return httpx.Response(
200,
content=json.dumps(_build_jwks()).encode(),
headers={"content-type": "application/json"},
)
return httpx.Response(404)
@pytest.fixture
def mock_idp():
"""Patch nextcloud_httpx_client used inside token_utils.verify_id_token."""
transport = httpx.MockTransport(_idp_handler)
def fake_client(**kwargs):
kwargs["transport"] = transport
return httpx.AsyncClient(**kwargs)
with patch(
"nextcloud_mcp_server.auth.token_utils.nextcloud_httpx_client",
side_effect=fake_client,
):
yield
async def test_verify_id_token_accepts_valid_token(mock_idp):
now = int(time.time())
token = _sign(
{
"iss": ISSUER,
"aud": "test-client",
"sub": "alice",
"iat": now,
"exp": now + 60,
}
)
payload = await verify_id_token(
token, discovery_url=DISCOVERY_URL, expected_audience="test-client"
)
assert payload["sub"] == "alice"
async def test_verify_id_token_rejects_wrong_audience(mock_idp):
now = int(time.time())
token = _sign(
{
"iss": ISSUER,
"aud": "other-client",
"sub": "alice",
"iat": now,
"exp": now + 60,
}
)
with pytest.raises(IdTokenVerificationError):
await verify_id_token(
token, discovery_url=DISCOVERY_URL, expected_audience="test-client"
)
async def test_verify_id_token_rejects_expired_token(mock_idp):
now = int(time.time())
token = _sign(
{
"iss": ISSUER,
"aud": "test-client",
"sub": "alice",
"iat": now - 120,
"exp": now - 60,
}
)
with pytest.raises(IdTokenVerificationError):
await verify_id_token(
token, discovery_url=DISCOVERY_URL, expected_audience="test-client"
)
async def test_verify_id_token_rejects_wrong_issuer(mock_idp):
now = int(time.time())
token = _sign(
{
"iss": "https://evil.example.com",
"aud": "test-client",
"sub": "alice",
"iat": now,
"exp": now + 60,
}
)
with pytest.raises(IdTokenVerificationError):
await verify_id_token(
token, discovery_url=DISCOVERY_URL, expected_audience="test-client"
)
async def test_verify_id_token_rejects_wrong_signature(mock_idp):
"""Token signed with a different key but matching kid header must fail."""
now = int(time.time())
forged = _sign(
{
"iss": ISSUER,
"aud": "test-client",
"sub": "alice",
"iat": now,
"exp": now + 60,
},
key_pem=_OTHER_PRIVATE_PEM,
)
with pytest.raises(IdTokenVerificationError):
await verify_id_token(
forged, discovery_url=DISCOVERY_URL, expected_audience="test-client"
)
async def test_verify_id_token_rejects_unknown_kid(mock_idp):
now = int(time.time())
token = _sign(
{
"iss": ISSUER,
"aud": "test-client",
"sub": "alice",
"iat": now,
"exp": now + 60,
},
kid="not-in-jwks",
)
with pytest.raises(IdTokenVerificationError, match="No JWKS key matches"):
await verify_id_token(
token, discovery_url=DISCOVERY_URL, expected_audience="test-client"
)
async def test_verify_id_token_nonce_mismatch_rejected(mock_idp):
now = int(time.time())
token = _sign(
{
"iss": ISSUER,
"aud": "test-client",
"sub": "alice",
"iat": now,
"exp": now + 60,
"nonce": "actual",
}
)
with pytest.raises(IdTokenVerificationError, match="nonce"):
await verify_id_token(
token,
discovery_url=DISCOVERY_URL,
expected_audience="test-client",
expected_nonce="expected",
)
async def test_verify_id_token_missing_token_rejected():
with pytest.raises(IdTokenVerificationError, match="missing"):
await verify_id_token(
"", discovery_url=DISCOVERY_URL, expected_audience="test-client"
)
+331
View File
@@ -0,0 +1,331 @@
"""Unit tests for OAuth logout (issue #626 finding 4) and the
SessionAuthBackend (finding 2).
These cover the new server-side session lifecycle:
- logout deletes refresh token + browser session
- logout calls IdP revocation_endpoint when available
- logout still succeeds when IdP/storage errors
- SessionAuthBackend resolves random session_id -> user_id, fails
closed when the session is unknown / expired / has no refresh token
"""
import json
import tempfile
from pathlib import Path
from types import SimpleNamespace
from unittest.mock import AsyncMock, MagicMock, patch
import httpx
import pytest
from cryptography.fernet import Fernet
from starlette.requests import HTTPConnection
from nextcloud_mcp_server.auth.browser_oauth_routes import (
_revoke_refresh_token_at_idp,
oauth_logout,
)
from nextcloud_mcp_server.auth.session_backend import SessionAuthBackend
from nextcloud_mcp_server.auth.storage import RefreshTokenStorage
pytestmark = pytest.mark.unit
# ---------------------------------------------------------------------------
# storage fixture (real SQLite backend; lighter than mocking every call)
# ---------------------------------------------------------------------------
@pytest.fixture
async def storage():
with tempfile.TemporaryDirectory() as tmpdir:
db_path = Path(tmpdir) / "test_logout.db"
s = RefreshTokenStorage(
db_path=str(db_path), encryption_key=Fernet.generate_key().decode()
)
await s.initialize()
yield s
def _build_request(*, cookie: str | None, oauth_context: dict | None):
"""Build a minimal Starlette-style request stub for oauth_logout."""
request = MagicMock()
request.query_params = {}
request.cookies = {"mcp_session": cookie} if cookie else {}
request.app.state.oauth_context = oauth_context
return request
# ---------------------------------------------------------------------------
# oauth_logout
# ---------------------------------------------------------------------------
async def test_logout_deletes_refresh_token_and_session(storage):
"""Happy path: logout removes the refresh token and the browser session."""
await storage.create_browser_session(session_id="sid-1", user_id="alice")
await storage.store_refresh_token(
user_id="alice", refresh_token="rt-abc", flow_type="browser"
)
request = _build_request(
cookie="sid-1",
oauth_context={"storage": storage, "discovery_url": None},
)
with patch(
"nextcloud_mcp_server.auth.browser_oauth_routes._revoke_refresh_token_at_idp",
new=AsyncMock(),
):
response = await oauth_logout(request)
assert response.status_code == 302
assert await storage.get_refresh_token("alice") is None
assert await storage.get_browser_session_user("sid-1") is None
async def test_logout_calls_revocation_when_refresh_token_present(storage):
"""The IdP revocation helper is called with the stored refresh token."""
await storage.create_browser_session(session_id="sid-2", user_id="bob")
await storage.store_refresh_token(
user_id="bob", refresh_token="rt-xyz", flow_type="browser"
)
revoke = AsyncMock()
request = _build_request(
cookie="sid-2",
oauth_context={"storage": storage, "discovery_url": "http://idp/.well-known"},
)
with patch(
"nextcloud_mcp_server.auth.browser_oauth_routes._revoke_refresh_token_at_idp",
new=revoke,
):
await oauth_logout(request)
revoke.assert_awaited_once()
args = revoke.await_args.args
# Second arg is the refresh token string
assert args[1] == "rt-xyz"
async def test_logout_no_session_cookie_returns_302(storage):
"""Without a cookie, logout still 302s and doesn't touch storage."""
request = _build_request(
cookie=None, oauth_context={"storage": storage, "discovery_url": None}
)
response = await oauth_logout(request)
assert response.status_code == 302
async def test_logout_swallows_storage_errors(storage):
"""Logout is best-effort — a storage failure must not 500 the response."""
await storage.create_browser_session(session_id="sid-3", user_id="carol")
broken_storage = MagicMock()
broken_storage.get_browser_session_user = AsyncMock(
side_effect=RuntimeError("db down")
)
broken_storage.delete_browser_session = AsyncMock()
request = _build_request(
cookie="sid-3",
oauth_context={"storage": broken_storage, "discovery_url": None},
)
response = await oauth_logout(request)
assert response.status_code == 302 # logout still succeeds
async def test_logout_handles_session_with_no_refresh_token(storage):
"""Cookie + session row exist but refresh token already gone — logout is idempotent."""
await storage.create_browser_session(session_id="sid-4", user_id="dave")
revoke = AsyncMock()
request = _build_request(
cookie="sid-4",
oauth_context={"storage": storage, "discovery_url": None},
)
with patch(
"nextcloud_mcp_server.auth.browser_oauth_routes._revoke_refresh_token_at_idp",
new=revoke,
):
await oauth_logout(request)
# Revoke not called — no token to revoke
revoke.assert_not_called()
# Browser session still cleared
assert await storage.get_browser_session_user("sid-4") is None
# ---------------------------------------------------------------------------
# _revoke_refresh_token_at_idp
# ---------------------------------------------------------------------------
def _httpx_handler(routes: dict[str, httpx.Response]):
def handler(request: httpx.Request) -> httpx.Response:
return routes.get(str(request.url), httpx.Response(404))
return handler
async def test_revoke_helper_posts_to_revocation_endpoint():
discovery_url = "http://idp.example/.well-known"
revocation_url = "http://idp.example/revoke"
received: list[httpx.Request] = []
def handler(request: httpx.Request) -> httpx.Response:
if str(request.url) == discovery_url:
return httpx.Response(
200,
content=json.dumps({"revocation_endpoint": revocation_url}).encode(),
headers={"content-type": "application/json"},
)
if str(request.url) == revocation_url:
received.append(request)
return httpx.Response(200)
return httpx.Response(404)
transport = httpx.MockTransport(handler)
def fake_client(**kwargs):
kwargs["transport"] = transport
return httpx.AsyncClient(**kwargs)
with patch(
"nextcloud_mcp_server.auth.browser_oauth_routes.nextcloud_httpx_client",
side_effect=fake_client,
):
await _revoke_refresh_token_at_idp(
{
"discovery_url": discovery_url,
"client_id": "test-client",
"client_secret": "test-secret",
},
"rt-secret",
)
assert len(received) == 1
body = received[0].content.decode()
assert "token=rt-secret" in body
assert "token_type_hint=refresh_token" in body
async def test_revoke_helper_skips_when_no_revocation_endpoint():
"""IdPs without a revocation_endpoint advertised: helper must no-op silently."""
discovery_url = "http://idp.example/.well-known"
def handler(request: httpx.Request) -> httpx.Response:
if str(request.url) == discovery_url:
return httpx.Response(200, json={}) # no revocation_endpoint
return httpx.Response(404)
transport = httpx.MockTransport(handler)
def fake_client(**kwargs):
kwargs["transport"] = transport
return httpx.AsyncClient(**kwargs)
with patch(
"nextcloud_mcp_server.auth.browser_oauth_routes.nextcloud_httpx_client",
side_effect=fake_client,
):
# Returns None and does not raise
result = await _revoke_refresh_token_at_idp(
{
"discovery_url": discovery_url,
"client_id": "x",
"client_secret": "y",
},
"rt",
)
assert result is None
async def test_revoke_helper_silent_on_idp_error():
"""If the IdP 500s, the helper must not raise — caller treats it as best-effort."""
def handler(request: httpx.Request) -> httpx.Response:
return httpx.Response(500, content=b"boom")
transport = httpx.MockTransport(handler)
def fake_client(**kwargs):
kwargs["transport"] = transport
return httpx.AsyncClient(**kwargs)
with patch(
"nextcloud_mcp_server.auth.browser_oauth_routes.nextcloud_httpx_client",
side_effect=fake_client,
):
result = await _revoke_refresh_token_at_idp(
{
"discovery_url": "http://x/.well-known",
"client_id": "x",
"client_secret": "y",
},
"rt",
)
assert result is None
# ---------------------------------------------------------------------------
# SessionAuthBackend
# ---------------------------------------------------------------------------
def _build_conn(*, cookie: str | None, oauth_context: dict | None):
conn = MagicMock(spec=HTTPConnection)
conn.cookies = {"mcp_session": cookie} if cookie else {}
conn.url = SimpleNamespace(path="/app")
conn.app = MagicMock()
conn.app.state.oauth_context = oauth_context
return conn
async def test_session_backend_authenticates_known_session_with_token(storage):
await storage.create_browser_session(session_id="sid-A", user_id="alice")
await storage.store_refresh_token(
user_id="alice", refresh_token="rt", flow_type="browser"
)
backend = SessionAuthBackend(oauth_enabled=True)
conn = _build_conn(cookie="sid-A", oauth_context={"storage": storage})
result = await backend.authenticate(conn)
assert result is not None
creds, user = result
assert "authenticated" in creds.scopes
assert user.username == "alice"
async def test_session_backend_rejects_unknown_session(storage):
backend = SessionAuthBackend(oauth_enabled=True)
conn = _build_conn(cookie="not-a-real-sid", oauth_context={"storage": storage})
assert await backend.authenticate(conn) is None
async def test_session_backend_rejects_session_without_refresh_token(storage):
"""Defense-in-depth: session row exists but user has no refresh token."""
await storage.create_browser_session(session_id="sid-B", user_id="bob")
# Note: NO refresh token stored for bob
backend = SessionAuthBackend(oauth_enabled=True)
conn = _build_conn(cookie="sid-B", oauth_context={"storage": storage})
assert await backend.authenticate(conn) is None
async def test_session_backend_rejects_when_no_cookie(storage):
backend = SessionAuthBackend(oauth_enabled=True)
conn = _build_conn(cookie=None, oauth_context={"storage": storage})
assert await backend.authenticate(conn) is None
async def test_session_backend_basicauth_mode_short_circuits(monkeypatch, storage):
"""In BasicAuth mode (oauth_enabled=False) the backend never touches storage."""
monkeypatch.setenv("NEXTCLOUD_USERNAME", "admin-user")
backend = SessionAuthBackend(oauth_enabled=False)
conn = _build_conn(cookie=None, oauth_context=None)
result = await backend.authenticate(conn)
assert result is not None
_, user = result
assert user.username == "admin-user"
+56
View File
@@ -0,0 +1,56 @@
"""Unit tests for OAuth tool input-schema hardening (issue #626 finding 3).
These tools must derive `user_id` from the verified MCP access token and
must never accept it as an MCP-level input. Otherwise an LLM (or any MCP
client) could supply an arbitrary user_id and reach cross-user revoke or
status-disclosure operations.
"""
import pytest
from mcp.server.fastmcp import FastMCP
from nextcloud_mcp_server.server.oauth_tools import register_oauth_tools
pytestmark = pytest.mark.unit
HARDENED_TOOLS = (
"provision_nextcloud_access",
"revoke_nextcloud_access",
"check_provisioning_status",
"check_logged_in",
)
@pytest.fixture
def registered_tools():
"""Register the OAuth tools against a fresh FastMCP and return them by name.
Uses FastMCP's `_tool_manager.list_tools()`; flagged as internal and may
break on SDK upgrades, but this is the supported way to inspect a tool's
JSON input schema in unit tests (see tests/unit/test_stdio.py).
"""
mcp = FastMCP("test-oauth-tools")
register_oauth_tools(mcp)
tools = mcp._tool_manager.list_tools()
return {t.name: t for t in tools}
def test_oauth_tools_registered(registered_tools):
for name in HARDENED_TOOLS:
assert name in registered_tools, f"{name} should be registered"
@pytest.mark.parametrize("tool_name", HARDENED_TOOLS)
def test_oauth_tool_schema_does_not_accept_user_id(tool_name, registered_tools):
"""user_id must not appear in the tool's JSON input schema."""
tool = registered_tools[tool_name]
properties = tool.parameters.get("properties", {})
required = tool.parameters.get("required", [])
assert "user_id" not in properties, (
f"{tool_name} accepts user_id as an MCP input — must be derived from "
f"the verified access token (issue #626 finding 3). "
f"properties={list(properties.keys())}"
)
assert "user_id" not in required
-18
View File
@@ -10,7 +10,6 @@ from datetime import datetime, timedelta, timezone
from unittest.mock import AsyncMock, MagicMock, patch
import httpx
import jwt
import pytest
from cryptography.fernet import Fernet
@@ -283,23 +282,6 @@ class TestTokenBrokerService:
# Verify cache was cleared
assert await token_broker.cache.get("user1") is None
async def test_validate_token_audience(self, token_broker):
"""Test token audience validation."""
# Create test token with audience
test_payload = {
"sub": "user1",
"aud": ["nextcloud", "other-service"],
"exp": datetime.now(timezone.utc) + timedelta(hours=1),
}
test_token = jwt.encode(test_payload, "secret", algorithm="HS256")
# Should not raise for correct audience
await token_broker._validate_token_audience(test_token, "nextcloud")
# Should raise for wrong audience
with pytest.raises(ValueError, match="doesn't include wrong-audience"):
await token_broker._validate_token_audience(test_token, "wrong-audience")
async def test_token_refresh_with_network_error(self, token_broker, mock_storage):
"""Test handling network errors during token refresh."""
# Storage returns already-decrypted refresh token