feat: accept KeyCape approval clients with tested recovery
All checks were successful
CI Smoke / host-smoke (push) Successful in 0s
CI Smoke / container-smoke (push) Successful in 2s

Assistant: codex
Assistant-Model: gpt-5.6-luna
Assistant-Session: 01a07ff8-19d0-7820-b4d0-1353833cb7fc
This commit is contained in:
tegwick 2026-09-09 02:18:12 +02:00
parent ce826cfa8d
commit 303a584bd0
8 changed files with 810 additions and 20 deletions

View file

@ -4,6 +4,7 @@ from __future__ import annotations
import argparse
import base64
import copy
import hashlib
import importlib.util
import json
import os
@ -26,11 +27,18 @@ spec = importlib.util.spec_from_file_location('pin', Path(__file__).with_name('o
pin = importlib.util.module_from_spec(spec); spec.loader.exec_module(pin)
ISSUER = 'https://kc.coulomb.social'
IMAGE = 'forgejo.coulomb.social/coulomb/key-cape@sha256:7ff54c54e63ee172ae9e6e7fd2da96e427352f712343d74626ee6fe0f6f82611'
VERIFIER = Path('/home/worsch/.local/share/key-cape/verified-bin/dcebd46/keycape')
VERIFIER_SHA256 = '4bf93bbe9afe0bf2e21d51c03373a7e2b4be6864586bcc2d472eeb9abc913d92'
PRIOR_IMAGE = 'forgejo.coulomb.social/coulomb/key-cape:main-153258b'
IDS = ('secrets-engine-approval', 'approval-engine-operator')
SECRET_NAMES = ('keycape-secrets-engine-approval-client', 'keycape-approval-engine-operator-client')
def verify_artifact():
require(VERIFIER.is_file() and not VERIFIER.is_symlink()
and hashlib.sha256(VERIFIER.read_bytes()).hexdigest() == VERIFIER_SHA256, 'pinned_verifier_artifact_required')
def registrations():
source = yaml.safe_load((KEYCAPE / 'config/service-clients.example.yaml').read_text())['clients']
result = [next(c for c in source if c['clientId'] == name) for name in IDS]
@ -115,7 +123,9 @@ def ready(kube, image, timeout=150):
podlist = data(command(kube + ['-n', 'sso', 'get', 'pods', '-l', 'app.kubernetes.io/name=keycape', '-o', 'json']))['items']
if len(podlist) == 1 and not podlist[0]['metadata'].get('deletionTimestamp'):
containers = podlist[0].get('status', {}).get('containerStatuses', [])
if any(c['name'] == 'keycape' and c['ready'] and c['image'] == image for c in containers):
if any(c['name'] == 'keycape' and c['ready'] and
(c.get('imageID', '').removeprefix('docker-pullable://') == image
if '@sha256:' in image else c.get('image') == image) for c in containers):
return {'deployment_uid': dep['metadata']['uid'], 'generation': dep['metadata']['generation'],
'pod_uid': podlist[0]['metadata']['uid'], 'single_ready_replica': True}
time.sleep(3)
@ -141,7 +151,10 @@ def http(path, fields=None, credentials=None):
with opener.open(request, timeout=20) as response:
return response.status, json.load(response)
except urllib.error.HTTPError as error:
return error.code, json.load(error)
try:
return error.code, json.load(error)
except (ValueError, TypeError):
raise LaneError('http_' + str(error.code) + '_non_json_error') from None
def denied(result, status, feature):
@ -150,53 +163,79 @@ def denied(result, status, feature):
and body.get('feature') == feature, 'keycape_denial_inconclusive')
def verified_claims(token, public_key):
# Match authclient/client.go: tolerate at most 30s future iat, but no
# extension to expiry or not-before. PyJWT's global leeway would extend both.
claims = jwt.decode(token, public_key, algorithms=['RS256'], issuer=ISSUER, audience='approval-engine',
options={'verify_iat': False, 'require': ['exp', 'iat', 'sub', 'iss', 'aud']})
require(type(claims['iat']) is int and type(claims['exp']) is int
and claims['iat'] <= int(time.time()) + 30 and claims['iat'] < claims['exp'], 'issued_at_binding_failed')
return claims
def acceptance(kube, receipt):
receipt['acceptance_phase'] = 'discovery'
code, discovery = http('/.well-known/openid-configuration')
require(code == 200 and discovery['issuer'] == ISSUER
and discovery['token_endpoint'] == ISSUER + '/token'
and discovery['jwks_uri'].startswith(ISSUER + '/'), 'discovery_binding_mismatch')
receipt['acceptance_phase'] = 'jwks'
code, jwks = http(discovery['jwks_uri'].removeprefix(ISSUER))
require(code == 200, 'jwks_failed')
receipt['clients'] = []
for client, secret_name in zip(registrations(), SECRET_NAMES):
receipt['acceptance_client'] = client['clientId']
receipt['acceptance_phase'] = 'verifier_secret_read'
secret = get(kube, 'secret', secret_name)
credential = base64.b64decode(secret['data']['client-secret'], validate=True).decode()
scopes = ' '.join(client['allowedScopes'])
receipt['acceptance_phase'] = 'token_exchange'
result, response = http('/token', {'grant_type': 'client_credentials', 'scope': scopes}, (client['clientId'], credential))
require(result == 200 and response.get('token_type') == 'Bearer' and response.get('expires_in') == 900, 'service_issuance_failed')
receipt['acceptance_phase'] = 'signature_and_claims'
token = response['access_token']; header = jwt.get_unverified_header(token)
keys = [key for key in jwks['keys'] if key['kid'] == header.get('kid')]
require(header.get('alg') == 'RS256' and len(keys) == 1, 'signing_key_selection_failed')
claims = jwt.decode(token, jwt.PyJWK.from_dict(keys[0]).key, algorithms=['RS256'],
issuer=ISSUER, audience='approval-engine',
options={'require': ['exp', 'iat', 'sub', 'iss', 'aud']})
claims = verified_claims(token, jwt.PyJWK.from_dict(keys[0]).key)
require(claims['sub'] == client['serviceSubject'] and claims['tenant'] == 'tenant:platform'
and claims['aud'] == 'approval-engine'
and claims['roles'] == client['roles'] and claims['exp'] - claims['iat'] == 900
and claims['principal_type'] == 'service'
and set(claims['scope'].split()) == set(client['allowedScopes']), 'exact_claims_mismatch')
excessive = 'approval:approve' if client['clientId'] == IDS[0] else 'approval:consume'
receipt['acceptance_phase'] = 'excess_scope_denial'
denied(http('/token', {'grant_type': 'client_credentials', 'scope': excessive},
(client['clientId'], credential)), 400, 'scope')
receipt['acceptance_phase'] = 'wrong_secret_denial'
denied(http('/token', {'grant_type': 'client_credentials', 'scope': scopes},
(client['clientId'], secrets.token_urlsafe(48))), 401, 'Authorization')
args = kube + ['-n', 'sso', 'exec', 'deployment/keycape', '--', '/keycape', 'verify-client',
args = [str(VERIFIER), 'verify-client',
'-issuer', ISSUER, '-client-id', client['clientId'], '-audience', 'approval-engine', '-scope', scopes,
'-secret-env', client['secretRef'].removeprefix('env:'), '-expect-subject', client['serviceSubject'],
'-expect-tenant', 'tenant:platform', '-expect-roles', ','.join(client['roles']), '-deny-scope', excessive]
command(args)
receipt['acceptance_phase'] = 'pinned_artifact_verifier'
verify_artifact()
verifier_env = os.environ.copy()
verifier_env[client['secretRef'].removeprefix('env:')] = credential
command(args, env=verifier_env)
receipt['clients'].append({'client_id': client['clientId'], 'live_jwks_signature_verified': True,
'exact_claims_verified': True, 'lifetime_seconds': 900, 'excess_scope_denied': True,
'wrong_secret_denied': True, 'pod_verify_client_passed': True,
'exact_claims_verified': True, 'lifetime_seconds': 900, 'maximum_future_iat_seconds': 30, 'expiry_leeway_seconds': 0, 'excess_scope_denied': True,
'wrong_secret_denied': True, 'pinned_artifact_verifier_passed': True,
'verifier_location': 'attended owner process', 'verifier_sha256': VERIFIER_SHA256,
'real_predecessor_rotation_tested': False, 'observed_wall_clock_expiry': False})
receipt['acceptance_phase'] = 'human_consume_denial'
query = urllib.parse.urlencode({'client_id': 'openbao-admin', 'response_type': 'code',
'redirect_uri': 'http://localhost:8250/oidc/callback', 'scope': 'openid approval:consume',
'code_challenge_method': 'S256', 'code_challenge': 'A' * 43, 'state': secrets.token_urlsafe(32)})
denied(http('/authorize?' + query), 400, 'scope')
receipt['human_client_consume_denied'] = True
receipt['acceptance_phase'] = 'passed'
def rollout(kube, receipt, recovery_path):
assert_cluster(kube)
verify_artifact()
before = get(kube, 'secret', 'keycape-config')
deployment = get(kube, 'deployment', 'keycape')
require(before['metadata']['uid'] == '2e94519d-1550-41c7-9701-2efe47fe1fd3'
@ -223,7 +262,8 @@ def rollout(kube, receipt, recovery_path):
receipt.update(status='service_acceptance_passed_pending_fresh_human_login', image=IMAGE,
config_resource_version=after['metadata']['resourceVersion'], signing_key_unchanged=True,
unrelated_config_bytes_preserved=True, existing_human_login_after=False)
except Exception:
except Exception as failure:
receipt['failure_class'] = type(failure).__name__
# Read after uncertain API outcomes too; never assume a timeout means no write.
now = get(kube, 'secret', 'keycape-config')
require(now['metadata']['uid'] == before['metadata']['uid'], 'rollback_config_identity_drift')

View file

@ -1,8 +1,8 @@
# Deployment + Service — KeyCape (namespace: sso)
#
# KeyCape is the OIDC orchestration layer. It is stateless: all persistent
# state lives in Authelia (session), LLDAP (users), and privacyIDEA (MFA tokens).
# No PVC is required.
# KeyCape orchestrates OIDC. Pending logins and authorization codes are process-local;
# use one replica with Recreate during replacement. Persistent identity state remains
# in Authelia, LLDAP and privacyIDEA. No PVC is required.
#
# Configuration is stored entirely in the keycape-config Secret, which holds
# a complete config.yaml and the RSA private key used to sign OIDC tokens
@ -33,7 +33,7 @@ spec:
matchLabels:
app.kubernetes.io/name: keycape
strategy:
type: RollingUpdate # stateless — safe to roll
type: Recreate # one issuer instance; process-local login/code state
template:
metadata:
labels:
@ -50,7 +50,7 @@ spec:
- name: keycape
# Image published to the self-hosted Forgejo OCI registry (KEY-WP-0002).
# KEY-WP-0012: canonical OIDC subject resolution for /userinfo.
image: forgejo.coulomb.social/coulomb/key-cape:main-153258b
image: forgejo.coulomb.social/coulomb/key-cape@sha256:7ff54c54e63ee172ae9e6e7fd2da96e427352f712343d74626ee6fe0f6f82611
imagePullPolicy: IfNotPresent
ports:
@ -67,6 +67,17 @@ spec:
name: keycape-rapp-qonto-client
key: client-secret
- name: KEYCAPE_SECRETS_ENGINE_APPROVAL_CLIENT_SECRET
valueFrom:
secretKeyRef:
name: keycape-secrets-engine-approval-client
key: client-secret
- name: KEYCAPE_APPROVAL_ENGINE_OPERATOR_CLIENT_SECRET
valueFrom:
secretKeyRef:
name: keycape-approval-engine-operator-client
key: client-secret
volumeMounts:
# keycape-config Secret provides config.yaml and key.pem
- name: config-secret
@ -89,7 +100,7 @@ spec:
failureThreshold: 3
readinessProbe:
httpGet:
path: /healthz
path: /readyz
port: 8080
initialDelaySeconds: 0
periodSeconds: 10

View file

@ -0,0 +1,92 @@
import base64, importlib.util, json, os, secrets, socket, subprocess, tempfile, time
from pathlib import Path
from unittest.mock import patch
from cryptography.hazmat.primitives import serialization
from cryptography.hazmat.primitives.asymmetric import rsa
import yaml
import http.server, ssl, threading, urllib.request, urllib.error, datetime, ipaddress
from cryptography import x509
from cryptography.x509.oid import NameOID
import argparse
parser=argparse.ArgumentParser(description='Exercise the pinned image and verifier over loopback HTTPS using synthetic credentials only.')
parser.add_argument('--receipt', required=True, type=Path)
args=parser.parse_args()
spec=importlib.util.spec_from_file_location('rollout',Path(__file__).with_name('approval-clients-rollout.py'))
m=importlib.util.module_from_spec(spec); spec.loader.exec_module(m)
m.verify_artifact()
def call(args):
p=subprocess.run(args,capture_output=True,timeout=45)
if p.returncode: raise RuntimeError('contained_docker_command_failed')
return p
sock=socket.socket(); sock.bind(('127.0.0.1',0)); port=sock.getsockname()[1]; sock.close()
class Proxy(http.server.BaseHTTPRequestHandler):
def log_message(self, *args): pass
def handle_proxy(self):
body = self.rfile.read(int(self.headers.get('Content-Length', 0))) if self.command == 'POST' else None
req = urllib.request.Request(f'http://127.0.0.1:{port}'+self.path, data=body,
headers={k:self.headers[k] for k in ['Content-Type','Authorization'] if k in self.headers})
try:
response=urllib.request.urlopen(req,timeout=20)
except urllib.error.HTTPError as error: response=error
with response:
self.send_response(response.code)
self.send_header('Content-Type', response.headers.get('Content-Type','application/json'))
self.end_headers(); self.wfile.write(response.read())
do_GET=handle_proxy
do_POST=handle_proxy
server=http.server.ThreadingHTTPServer(('127.0.0.1',0),Proxy)
https_port=server.server_address[1]
name='keycape-approval-exercise-'+secrets.token_hex(5)
with tempfile.TemporaryDirectory(prefix='keycape-private-exercise-') as temp:
root=Path(temp)
key=rsa.generate_private_key(public_exponent=65537,key_size=2048)
(root/'key.pem').write_bytes(key.private_bytes(serialization.Encoding.PEM,serialization.PrivateFormat.PKCS8,serialization.NoEncryption()))
os.chmod(root/'key.pem',0o600)
name_attr=x509.Name([x509.NameAttribute(NameOID.COMMON_NAME,'KeyCape local exercise')])
cert=(x509.CertificateBuilder().subject_name(name_attr).issuer_name(name_attr).public_key(key.public_key())
.serial_number(x509.random_serial_number()).not_valid_before(datetime.datetime.now(datetime.timezone.utc)-datetime.timedelta(minutes=1))
.not_valid_after(datetime.datetime.now(datetime.timezone.utc)+datetime.timedelta(hours=1))
.add_extension(x509.SubjectAlternativeName([x509.IPAddress(ipaddress.ip_address('127.0.0.1'))]),False)
.add_extension(x509.BasicConstraints(ca=True,path_length=None),True))
from cryptography.hazmat.primitives import hashes
(root/'cert.pem').write_bytes(cert.sign(key,hashes.SHA256()).public_bytes(serialization.Encoding.PEM))
context=ssl.SSLContext(ssl.PROTOCOL_TLS_SERVER); context.load_cert_chain(root/'cert.pem',root/'key.pem')
server.socket=context.wrap_socket(server.socket,server_side=True)
threading.Thread(target=server.serve_forever,daemon=True).start()
config=yaml.safe_load((m.KEYCAPE/'config/dev-config.yaml').read_text())
config.update(issuer=f'https://127.0.0.1:{https_port}')
config['authelia']['issuer']='https://auth.coulomb.social'
config['clients'].extend(m.registrations())
(root/'config.yaml').write_text(yaml.safe_dump(config,sort_keys=False))
os.chmod(root/'config.yaml',0o600)
values={c['clientId']:secrets.token_urlsafe(48) for c in m.registrations()}
(root/'env').write_text('KEYCAPE_CONFIG=/etc/keycape/config.yaml\n'+''.join(c['secretRef'].removeprefix('env:')+'='+values[c['clientId']]+'\n' for c in m.registrations()))
os.chmod(root/'env',0o600)
created=False
try:
call(['docker','run','-d','--rm','--name',name,'--user',str(os.getuid()),'--publish',f'127.0.0.1:{port}:8080',
'--env-file',str(root/'env'),'--mount','type=bind,source='+temp+',target=/etc/keycape,readonly',m.IMAGE])
created=True
with patch.object(m,'ISSUER',config['issuer']), patch.dict(os.environ,{'SSL_CERT_FILE':str(root/'cert.pem')}):
for _ in range(30):
try:
status,_=m.http('/.well-known/openid-configuration')
if status==200: break
except Exception: pass
time.sleep(.5)
else: raise RuntimeError('scratch_keycape_not_ready')
def fake_get(kube,kind,name):
index=m.SECRET_NAMES.index(name)
return {'data': {'client-secret':base64.b64encode(values[m.IDS[index]].encode()).decode()}}
receipt={}
with patch.object(m,'get',side_effect=fake_get):
m.acceptance([],receipt)
receipt.update(target='disposable pinned KeyCape image; synthetic keys and clients only',
pinned_native_cli_verification='passed against loopback HTTPS with synthetic credentials',cleanup_complete=False)
finally:
if created: call(['docker','stop','--time=5',name])
server.shutdown(); server.server_close()
receipt['cleanup_complete']=True
fd=os.open(args.receipt,os.O_WRONLY|os.O_CREAT|os.O_EXCL,0o600)
with os.fdopen(fd,'w') as out: out.write(json.dumps(receipt,indent=2)+'\n')
print('Pinned-image scratch acceptance passed; synthetic credentials removed.')

View file

@ -83,7 +83,7 @@ class RolloutTests(unittest.TestCase):
changed['metadata']['resourceVersion'] = str(int(changed['metadata']['resourceVersion']) + 1)
state[kind] = changed
return copy.deepcopy(changed)
with tempfile.TemporaryDirectory() as directory, patch.object(m, 'assert_cluster'), \
with tempfile.TemporaryDirectory() as directory, patch.object(m, 'assert_cluster'), patch.object(m, 'verify_artifact'), \
patch.object(m, 'get', side_effect=lambda kube, kind, name: copy.deepcopy(state[kind])), \
patch.object(m, 'patch_object', side_effect=fake_patch), patch.object(m, 'ready', return_value={}), \
patch.object(m, 'acceptance', side_effect=m.LaneError('synthetic_acceptance_failure')):
@ -95,6 +95,29 @@ class RolloutTests(unittest.TestCase):
self.assertEqual(state['deployment']['spec'], original['deployment']['spec'])
self.assertEqual((Path(directory) / 'recovery.json').stat().st_mode & 0o777, 0o600)
def test_ready_matches_manifest_imageid_not_runtime_config_id(self):
dep = deployment(); dep['metadata']['generation'] = 30
dep['status'] = {'observedGeneration': 30, 'updatedReplicas': 1, 'readyReplicas': 1, 'availableReplicas': 1, 'replicas': 1}
pod = {'metadata': {'uid': 'fixture-pod'}, 'status': {'containerStatuses': [
{'name': 'keycape', 'ready': True, 'image': 'sha256:runtime-config-id', 'imageID': m.IMAGE}]}}
from types import SimpleNamespace
with patch.object(m, 'get', return_value=dep), patch.object(m, 'command', return_value=SimpleNamespace(stdout=json.dumps({'items': [pod]}).encode())):
self.assertTrue(m.ready([], m.IMAGE, timeout=1)['single_ready_replica'])
def test_iat_skew_matches_native_contract_without_extending_expiry(self):
from cryptography.hazmat.primitives.asymmetric import rsa
key = rsa.generate_private_key(public_exponent=65537, key_size=2048)
now = int(m.time.time())
claims = {'iss': m.ISSUER, 'aud': 'approval-engine', 'sub': 'synthetic-service', 'iat': now + 2, 'exp': now + 902}
encoded = m.jwt.encode(claims, key, algorithm='RS256')
self.assertEqual(m.verified_claims(encoded, key.public_key())['iat'], now + 2)
future = m.jwt.encode(dict(claims, iat=now+60), key, algorithm='RS256')
with self.assertRaisesRegex(m.LaneError, 'issued_at_binding_failed'):
m.verified_claims(future, key.public_key())
expired = m.jwt.encode(dict(claims, iat=now-900, exp=now-1), key, algorithm='RS256')
with self.assertRaises(m.jwt.ExpiredSignatureError):
m.verified_claims(expired, key.public_key())
def test_unrelated_refusal_never_passes(self):
for status, body in [(500, {}), (400, {'error': 'invalid_profile_usage', 'feature': 'client_id'}),
(401, {'error': 'invalid_profile_usage', 'feature': 'scope'})]: