Files
stackpilot/backend/tests/test_route_authorization.py
T
menzeljandClaude Opus 5 9247ff9621
CI / check (push) Successful in 13m10s
CI / build-and-push (push) Successful in 3m50s
Deploy stacks from a Git repository (0.58.0)
StackPilot's stacks were already plain folders on disk, which makes GitOps less
of an architectural change than it would be elsewhere: a sync is "make these
files match that repo, then compose up". Almost all of the design effort went
into the word "these", because getting it wrong destroys data.

A stack folder is not just the compose file. Compose creates bind-mount
directories in it — ./config, ./data — and those hold the live state of whatever
is running. So the obvious implementation, clone into the stack folder and
git reset --hard, is a data-loss bug waiting for its first `git clean`. Instead
the clone lives in a cache under ${DATA_DIR}/git/<stack> where reset and clean
are safe, and the configured subtree is copied across. No .git ends up in the
stack folder, so backups and the file browser are unaffected too.

Deletion is the other half. Making a folder "match" a repo naively means
removing what the repo does not have, which is exactly the application data
above. So each sync records the paths it wrote, and the next sync may delete
only those — a file the repository never provided cannot be touched by any code
path here. Tested directly: a database file and a hand-written .env survive a
sync that replaces the compose file and removes a file the repo dropped.

What the repo does provide is overwritten, hand edits included. That is the
point of GitOps rather than a wart, but it is a surprise if you attach a repo to
a stack you have been editing, so the connect form says it before the first sync
and the first sync is never automatic.

The webhook is the only route in StackPilot with no bearer token, because a Git
forge has none to present. It authenticates with an HMAC over the body —
X-Hub-Signature-256 for GitHub/Gitea/Forgejo, X-Gitlab-Token for GitLab, both
compared in constant time — and answers 404, not 403, to anything unsigned. A
403 would confirm that a given stack exists and is connected to a repository,
which an unauthenticated caller has not earned. The authorization matrix test
caught this route being public and made me write that reasoning down in it,
which is exactly what that test is for.

Credentials never reach a command line: ps is readable by every process on the
host, and this runs in a container next to everything else. The HTTPS token goes
to git through GIT_ASKPASS and the environment, the SSH key through a 0600 file
kept outside the working tree, and everything git prints is scrubbed of both —
plus any credential-carrying URL — before it is stored in last_error or shown.

Auto-deploy takes the same per-stack lock as every other lifecycle action, so a
webhook firing mid-deploy reports "files synced, stack busy" instead of racing a
second compose run at the same project.

The image needed git and openssh-client, which is the only reason this release
touches the Dockerfile.

26 tests against real repositories created with the real git binary, none of
them touching the network — mocking git would mostly test the mock. Verified end
to end as well: connect, sync, a push that changes one file and deletes another,
a wrongly signed webhook, a correctly signed one, and the live data still there
afterwards.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-18 01:06:29 +02:00

300 lines
11 KiB
Python

"""The authorization matrix — StackPilot's most load-bearing test.
Authorization lives in the routers: each one picks ``require_admin`` or
``get_current_user`` per route, and nothing checks that the choice was right.
That is how 0.43.0 shipped a ``user`` role that could download the auth
database, every stack's ``.env`` and every ``.secrets/*`` file — a wrong default
on four routes, invisible in review.
So the policy is written down here instead of being implied by 171 individual
decisions:
every route requires admin, unless it is listed in USER_READABLE or PUBLIC.
Adding a route that the read-only role can reach means adding it to the list,
which is the review moment this test exists to force. The check is static — it
reads FastAPI's dependency graph rather than calling the routes — because
calling all 171 would mean constructing valid bodies for each and would happily
fire ``POST /stacks/{id}/down`` at whatever Docker is around.
The dynamic tests at the bottom are the direct regression net for the specific
leaks that were found: they use a real non-admin token and assert 403.
"""
from __future__ import annotations
import pytest
# --------------------------------------------------------------------------- #
# The policy
# --------------------------------------------------------------------------- #
#: Reachable without any token. Everything needed to log in, plus the health
#: probe the container's HEALTHCHECK hits.
PUBLIC = {
"GET /api/health",
"GET /api/auth/needs-setup",
"POST /api/auth/setup",
"POST /api/auth/login",
"POST /api/auth/refresh",
# Only drops the refresh cookie. Requiring a valid token would mean you
# cannot sign out once the session has already gone stale.
"POST /api/auth/logout",
# A Git forge has no StackPilot credentials to present, so this one cannot
# be behind a bearer token. It is authorized instead by an HMAC over the
# request body against a per-stack secret, and answers 404 — not 403 — to
# anything unsigned, so it cannot be used to discover which stacks exist or
# which are connected to a repository. See routers/git.py.
"POST /api/git/webhook/{stack_id}",
}
#: Reachable by the read-only ``user`` role. Everything here has been checked
#: for whether it can return a credential:
#:
#: * ``GET /api/settings`` returns the interval and counts — webhook URLs (which
#: carry tokens) come from the admin-only ``/api/settings/webhooks``.
#: * ``GET /api/stacks/{id}`` blanks out ``env`` for non-admins.
#: * ``GET /api/templates`` is metadata only; the detail route, which returns a
#: template's env, is admin-only.
#: * The ``/api/editor/*`` routes are pure YAML transformations — they take YAML
#: in and hand YAML back, touching nothing on disk.
USER_READABLE = {
"GET /api/auth/me",
# Mutating, but only ever on the caller's own account: it bumps their own
# token_version to sign every one of their devices out.
"POST /api/auth/logout-everywhere",
"GET /api/dashboard/fleet",
# Stacks: status, logs and the compose file. Not the .env, not the export.
"GET /api/stacks",
"GET /api/stacks/stats",
"GET /api/stacks/updates",
# The app-logo catalog and its images: public artwork from a public icon
# set, keyed by app name. Nothing here is derived from this install.
"GET /api/stacks/icons/search",
"GET /api/stacks/icons/logo/{slug}",
"GET /api/stacks/{stack_id}",
"GET /api/stacks/{stack_id}/auto-update",
# The stack's uploaded icon — an image the user already sees next to the
# stack's name in the list. Uploading and resetting it stay admin-only.
"GET /api/stacks/{stack_id}/icon",
"GET /api/stacks/{stack_id}/logs",
"GET /api/stacks/{stack_id}/services/{service}/logs",
"POST /api/stacks/convert",
# Editor helpers: stateless YAML rewriting.
"POST /api/editor/validate",
"POST /api/editor/services",
"POST /api/editor/add-volume",
"POST /api/editor/add-device",
"POST /api/editor/remove-device",
"POST /api/editor/set-gpu",
"POST /api/editor/set-privileged",
"POST /api/editor/set-resources",
# Read-only inventory.
"GET /api/containers/{container_id}",
"GET /api/images",
"GET /api/images/updates",
"GET /api/networks",
"GET /api/networks/{network_id}",
"GET /api/networks/{network_id}/containers",
"GET /api/volumes",
"GET /api/volumes/sizes",
"GET /api/volumes/orphaned",
"POST /api/volumes/generate-yaml",
"POST /api/ports/conflicts",
"GET /api/settings",
"GET /api/system/info",
"GET /api/system/gpus",
"GET /api/system/devices",
"GET /api/system/update",
"GET /api/templates",
}
# --------------------------------------------------------------------------- #
# Reading the declared authorization off the routes
# --------------------------------------------------------------------------- #
def _declared_auth(route) -> str:
"""What a route actually requires, per its dependency graph."""
from auth import get_current_user, require_admin
found = set()
def walk(dependant):
for sub in dependant.dependencies:
if sub.call is require_admin:
found.add("admin")
elif sub.call is get_current_user:
found.add("user")
walk(sub)
walk(route.dependant)
if "admin" in found:
return "admin"
if "user" in found:
return "user"
return "public"
def _expected_auth(key: str) -> str:
if key in PUBLIC:
return "public"
if key in USER_READABLE:
return "user"
return "admin"
def _all_routes(app):
from fastapi.routing import APIRoute
out = []
for route in app.routes:
if not isinstance(route, APIRoute):
continue
for method in sorted(route.methods - {"HEAD", "OPTIONS"}):
out.append((f"{method} {route.path}", route))
return sorted(out, key=lambda r: r[0])
@pytest.fixture(scope="session")
def routes(app):
return _all_routes(app)
# --------------------------------------------------------------------------- #
# The matrix
# --------------------------------------------------------------------------- #
def test_every_route_matches_the_declared_policy(routes):
"""Each route requires exactly what PUBLIC / USER_READABLE say it should.
A new route that nobody classified defaults to "admin" — which is the safe
direction. What this catches is the dangerous one: a route written with
``get_current_user`` that was never weighed against "can this return a
credential".
"""
wrong = []
for key, route in routes:
actual, expected = _declared_auth(route), _expected_auth(key)
if actual != expected:
wrong.append(f" {key}\n declared={actual} expected={expected}")
assert not wrong, (
"Route authorization does not match the policy in this file.\n\n"
+ "\n".join(wrong)
+ "\n\nIf the route is genuinely safe for the read-only role, add it to "
"USER_READABLE with a note on why it cannot return a credential. "
"Otherwise give it require_admin."
)
def test_no_unlisted_route_is_reachable_without_a_token(routes):
"""Only the login/health surface may skip authentication entirely."""
unauthenticated = {key for key, route in routes if _declared_auth(route) == "public"}
assert unauthenticated == PUBLIC
def test_policy_lists_have_no_stale_entries(routes):
"""Keep the lists honest when routes get renamed or removed."""
known = {key for key, _ in routes}
assert not (PUBLIC - known), f"PUBLIC lists routes that no longer exist: {PUBLIC - known}"
assert not (USER_READABLE - known), (
f"USER_READABLE lists routes that no longer exist: {USER_READABLE - known}"
)
def test_everything_touching_the_filesystem_requires_admin(routes):
"""The file browser, host-path picker and stack export, as one rule.
These reach whatever the backend container can see — which includes every
``.env`` and ``.secrets/*``. Reads are no less sensitive than writes here,
which is the mistake this test exists to prevent a repeat of.
"""
offenders = [
key
for key, route in routes
if ("/files" in key or "/host/paths" in key or key.endswith("/export"))
and _declared_auth(route) != "admin"
]
assert not offenders, f"Filesystem routes must be admin-only: {offenders}"
# --------------------------------------------------------------------------- #
# Regression tests for the leaks that were actually found (F1, F3)
# --------------------------------------------------------------------------- #
#: Routes a ``user`` token could reach before 0.44.0, each of which handed out
#: credentials. A 403 here is the whole point.
LEAKED_BEFORE_0_44 = [
("GET", "/api/files/list?path=/opt"),
("GET", "/api/files/read?path=/opt/stacks/x/.env"),
("GET", "/api/files/download?path=/opt/stacks/x/.env"),
("GET", "/api/host/paths?path=/opt"),
("GET", "/api/audit"),
("GET", "/api/stacks/anything/export"),
("GET", "/api/templates/jellyfin"),
]
@pytest.mark.parametrize("method,url", LEAKED_BEFORE_0_44)
def test_read_only_role_is_refused(as_user, method, url):
assert as_user.request(method, url).status_code == 403, (
f"{method} {url} is reachable by the read-only role again"
)
@pytest.mark.parametrize("method,url", LEAKED_BEFORE_0_44)
def test_admin_is_not_refused(as_admin, method, url):
"""The same routes must still work for admins.
Anything but 403 passes: without a Docker daemon or the referenced paths
these legitimately answer 400/404/502, and this test is about the
authorization layer, not the handler.
"""
assert as_admin.request(method, url).status_code != 403
def test_missing_token_is_401_not_403(client):
assert client.get("/api/files/read?path=/etc/hostname").status_code == 401
# --------------------------------------------------------------------------- #
# The .env is withheld from the read-only role (F1, via the stacks API)
# --------------------------------------------------------------------------- #
@pytest.fixture
def stack_with_env(app):
"""A real stack folder plus its DB row, so GET /api/stacks/{id} resolves."""
import os
from sqlmodel import Session
from database import engine
from models.stack import Stack
from services import compose_service
stack_id = "authz-fixture"
directory = compose_service.stack_dir(stack_id)
os.makedirs(directory, exist_ok=True)
compose_service.write_compose(stack_id, "services:\n app:\n image: alpine\n")
compose_service.write_env(stack_id, "DB_PASSWORD=super-secret-value\n")
with Session(engine) as session:
if not session.get(Stack, stack_id):
session.add(Stack(id=stack_id, name="authz fixture"))
session.commit()
return stack_id
def test_stack_detail_withholds_env_from_the_read_only_role(as_user, stack_with_env):
body = as_user.get(f"/api/stacks/{stack_with_env}").json()
assert body["env"] == ""
assert "super-secret-value" not in str(body)
# The compose file is still there — the read-only role keeps a useful view.
assert "image: alpine" in body["yaml"]
def test_stack_detail_gives_admins_the_env(as_admin, stack_with_env):
body = as_admin.get(f"/api/stacks/{stack_with_env}").json()
assert "super-secret-value" in body["env"]