CDT with bearer app token

Author: rayluo

msal/application.py

@@ -1,4 +1,6 @@
 from __future__ import annotations
+import base64
+import datetime
 import functools
 import json
 import time
@@ -166,6 +168,17 @@ def _preferred_browser():
     return None
 
 
+def _build_req_cnf(jwk:dict, remove_padding:bool = False) -> str:
+    """req_cnf usually requires base64url encoding.
+
+    https://datatracker.ietf.org/doc/html/draft-ietf-oauth-pop-key-distribution-07#section-4.2.1
+    https://learn.microsoft.com/en-us/openspecs/windows_protocols/ms-rdpbcgr/e967ebeb-9e9f-443e-857a-5208802943c2
+    """
+    raw = json.dumps(jwk)
+    encoded = base64.urlsafe_b64encode(raw.encode('utf-8')).decode('utf-8')
+    return encoded.rstrip('=') if remove_padding else encoded
+
+
 class _ClientWithCcsRoutingInfo(Client):
 
     def initiate_auth_code_flow(self, **kwargs):
@@ -232,6 +245,7 @@ class ClientApplication(object):
     _TOKEN_SOURCE_IDP = "identity_provider"
     _TOKEN_SOURCE_CACHE = "cache"
     _TOKEN_SOURCE_BROKER = "broker"
+    _XMS_DS_NONCE = "xms_ds_nonce"
 
     _enable_broker = False
     _AUTH_SCHEME_UNSUPPORTED = (
@@ -241,8 +255,17 @@ class ClientApplication(object):
 
     _TOKEN_CACHE_DATA: dict[str, str] = {  # field_in_data: field_in_cache
         "key_id": "key_id",  # Some token types (SSH-certs, POP) are bound to a key
+        "req_ds_cnf": "req_ds_cnf",  # Used in CDT scenario
     }
 
+    @functools.lru_cache(maxsize=2)
+    def __get_rsa_key(self, _bucket):  # _bucket is used with lru_cache pattern
+        from .crypto import _generate_rsa_key
+        return _generate_rsa_key()
+
+    def _get_rsa_key(self, _bucket=None):  # Return the same RSA key, cached for a day
+        return self.__get_rsa_key(_bucket or datetime.date.today())
+
     def __init__(
             self, client_id,
             client_credential=None, authority=None, validate_authority=True,
@@ -656,7 +679,12 @@ def __init__(
 
         self._decide_broker(allow_broker, enable_pii_log)
         self.token_cache = token_cache or TokenCache()
-        self.token_cache._set(data_to_at=self._TOKEN_CACHE_DATA)
+        self.token_cache._set(
+            data_to_at=self._TOKEN_CACHE_DATA,
+            response_to_at={  # field_in_resp: field_in_cache
+                "xms_ds_nonce": "xms_ds_nonce",
+            },
+        )
         self._region_configured = azure_region
         self._region_detected = None
         self.client, self._regional_client = self._build_client(
@@ -1559,6 +1587,9 @@ def _acquire_token_silent_from_cache_and_possibly_refresh_it(
                     "expires_in": int(expires_in),  # OAuth2 specs defines it as int
                     self._TOKEN_SOURCE: self._TOKEN_SOURCE_CACHE,
                     }
+                if self._XMS_DS_NONCE in entry:  # CDT needs this
+                    access_token_from_cache[self._XMS_DS_NONCE] = entry[
+                        self._XMS_DS_NONCE]
                 if "refresh_on" in entry:
                     access_token_from_cache["refresh_on"] = int(entry["refresh_on"])
                     if int(entry["refresh_on"]) < now:  # aging
@@ -2347,7 +2378,16 @@ class ConfidentialClientApplication(ClientApplication):  # server-side web app
     except that ``allow_broker`` parameter shall remain ``None``.
     """
 
-    def acquire_token_for_client(self, scopes, claims_challenge=None, **kwargs):
+    def acquire_token_for_client(
+        self,
+        scopes,
+        claims_challenge=None,
+        *,
+        delegation_constraints: Optional[list] = None,
+        delegation_confirmation_key=None,  # A Cyprtography's RSAPrivateKey-like object
+            # TODO: Support ECC key? https://github.com/pyca/cryptography/issues/4093
+        **kwargs
+    ):
         """Acquires token for the current confidential client, not for an end user.
 
         Since MSAL Python 1.23, it will automatically look for token from cache,
@@ -2370,8 +2410,36 @@ def acquire_token_for_client(self, scopes, claims_challenge=None, **kwargs):
             raise ValueError(  # We choose to disallow force_refresh
                 "Historically, this method does not support force_refresh behavior. "
             )
-        return _clean_up(self._acquire_token_silent_with_error(
-            scopes, None, claims_challenge=claims_challenge, **kwargs))
+        if delegation_constraints:
+            private_key = delegation_confirmation_key or self._get_rsa_key()
+            from .crypto import _convert_rsa_keys
+            _, jwk = _convert_rsa_keys(private_key)
+        result = _clean_up(self._acquire_token_silent_with_error(
+            scopes, None, claims_challenge=claims_challenge, data=dict(
+                kwargs.pop("data", {}),
+                req_ds_cnf=_build_req_cnf(jwk)  # It is part of token cache key
+                    if delegation_constraints else None,
+                ),
+            **kwargs))
+        if delegation_constraints and not result.get("error"):
+            if not result.get(self._XMS_DS_NONCE):  # Available in cached token, too
+                raise ValueError(
+                    "The resource did not opt in to xms_ds_cnf claim. "
+                    "After its opt-in, call this function again with "
+                    "a new app object or a new delegation_confirmation_key"
+                    # in order to invalidate the token in cache
+                    )
+            import jwt  # Lazy loading
+            cdt_envelope = jwt.encode({
+                "constraints": delegation_constraints,
+                self._XMS_DS_NONCE: result[self._XMS_DS_NONCE],
+                }, private_key, algorithm="PS256")
+            result["access_token"] = jwt.encode({
+                "t": result["access_token"],
+                "c": cdt_envelope,
+                }, None, algorithm=None, headers={"typ": "cdt+jwt"})
+            del result[self._XMS_DS_NONCE]  # Caller shouldn't need to know that
+        return result
 
     def _acquire_token_for_client(
         self,

msal/crypto.py

@@ -0,0 +1,37 @@
+from cryptography.hazmat.primitives.asymmetric import rsa
+
+
+def _urlsafe_b64encode(n:int, bit_size:int) -> str:
+    from base64 import urlsafe_b64encode
+    return urlsafe_b64encode(n.to_bytes(
+        length=int(bit_size/8),
+        byteorder="big",
+        )).decode("utf-8").rstrip("=")
+
+
+def _to_jwk(public_key: rsa.RSAPublicKey) -> dict:
+    """Equivalent to:
+
+    numbers = public_key.public_numbers()
+    result = {
+        "kty": "RSA",
+        "n": _urlsafe_b64encode(numbers.n, public_key.key_size),
+        "e": _urlsafe_b64encode(numbers.e, 24),
+        }
+    return result
+    """
+    import jwt
+    return jwt.get_algorithm_by_name(  # PyJWT 2.5.0 https://github.com/jpadilla/pyjwt/releases/tag/2.5.0
+        "RS256"
+        ).to_jwk(
+            public_key,
+            as_dict=True,  # PyJWT 2.7.0 https://github.com/jpadilla/pyjwt/releases/tag/2.7.0
+        )
+
+def _convert_rsa_keys(private_key: rsa.RSAPrivateKey):
+    return "pairs.private_bytes()", _to_jwk(private_key.public_key())
+
+def _generate_rsa_key() -> rsa.RSAPrivateKey:
+    # https://cryptography.io/en/latest/hazmat/primitives/asymmetric/rsa/#cryptography.hazmat.primitives.asymmetric.rsa.generate_private_key
+    return rsa.generate_private_key(public_exponent=65537, key_size=2048)
+

msal/token_cache.py

@@ -1,4 +1,5 @@
 from __future__ import annotations
+import hashlib
 import json
 import threading
 import time
@@ -82,6 +83,7 @@ def __init__(self):
                         realm=None, target=None,
                         # Note: New field(s) can be added here
                         #key_id=None,
+                        req_ds_cnf=None,
                         **ignored_payload_from_a_real_token:
                     "-".join([  # Note: Could use a hash here to shorten key length
                         home_account_id or "",
@@ -91,6 +93,13 @@ def __init__(self):
                         realm or "",
                         target or "",
                         #key_id or "",  # So ATs of different key_id can coexist
+                        hashlib.sha256(req_ds_cnf.encode()).hexdigest()
+                            # TODO: Could hash the entire key eventually.
+                            #       But before that project, we better first
+                            #       change the scope to use input scope
+                            #       instead of response scope,
+                            #       so that a search() can probably have O(1) hit.
+                            if req_ds_cnf else "",  # CDT
                         ]).lower(),
             self.CredentialType.ID_TOKEN:
                 lambda home_account_id=None, environment=None, client_id=None,

setup.cfg

@@ -44,7 +44,7 @@ install_requires =
 
     # MSAL does not use jwt.decode(),
     # therefore is insusceptible to CVE-2022-29217 so no need to bump to PyJWT 2.4+
-    PyJWT[crypto]>=1.0.0,<3
+    PyJWT[crypto]>=2.7.0,<3
 
     # load_key_and_certificates() is available since 2.5
     # https://cryptography.io/en/latest/hazmat/primitives/asymmetric/serialization/#cryptography.hazmat.primitives.serialization.pkcs12.load_key_and_certificates

tests/test_application.py

@@ -1,5 +1,6 @@
 # Note: Since Aug 2019 we move all e2e tests into test_e2e.py,
 # so this test_application file contains only unit tests without dependency.
+from __future__ import annotations
 import json
 import logging
 import sys
@@ -11,6 +12,7 @@
     ClientApplication, PublicClientApplication, ConfidentialClientApplication,
     _str2bytes, _merge_claims_challenge_and_capabilities,
 )
+from msal.oauth2cli.oidc import decode_part
 from tests import unittest
 from tests.test_token_cache import build_id_token, build_response
 from tests.http_client import MinimalHttpClient, MinimalResponse
@@ -856,3 +858,76 @@ def test_app_did_not_register_redirect_uri_should_error_out(self):
                 )
             self.assertEqual(result.get("error"), "broker_error")
 
+
+class CdtTestCase(unittest.TestCase):
+
+    def createConstraint(self, typ: str, action: str, targets: list[str]) -> dict:
+        return {"ver": "1.0", "typ": typ, "a": action, "targets": targets}
+
+    def test_constraint_format(self):
+        self.assertEqual([
+             self.createConstraint("ns:usr", "create", ["guid1", "guid2"]),
+             self.createConstraint("ns:app", "update", ["guid3", "guid4"]),
+             self.createConstraint("ns:subscription", "read", ["guid5", "guid6"]),
+         ], [  # Format defined in https://microsoft-my.sharepoint-df.com/:w:/p/rohitshende/EZgP9niwOvhKn-CUbj1NgG4BTZ6FSD9_16vXvsaXTiUzkg?e=j5DcQu&nav=eyJoIjoiODU5NDAyNjI4In0
+            {"ver": "1.0", "typ": "ns:usr", "a": "create", "targets": ["guid1", "guid2"]},
+            {"ver": "1.0", "typ": "ns:app", "a": "update", "targets": ["guid3", "guid4"]},
+            {"ver": "1.0", "typ": "ns:subscription", "a": "read", "targets": [
+                "guid5", "guid6",
+            ]},
+        ], "Constraint format is correct")  # MSAL actually accepts arbitrary JSON blob
+
+    def assertCdt(self, result: dict, constraints: list[dict]) -> None:
+        self.assertIsNotNone(
+            result.get("access_token"), "Encountered {}: {}".format(
+                result.get("error"), result.get("error_description")))
+        _expectancy = "The return value should look like a Bearer response"
+        self.assertEqual(result["token_type"], "Bearer", _expectancy)
+        self.assertNotIn("xms_ds_nonce", result, _expectancy)
+        headers = json.loads(decode_part(result["access_token"].split(".")[0]))
+        self.assertEqual(headers.get("typ"), "cdt+jwt", "typ should be cdt+jwt")
+        payload = json.loads(decode_part(result["access_token"].split(".")[1]))
+        self.assertIsNotNone(payload.get("t") and payload.get("c"))
+        cdt_envelope = json.loads(decode_part(payload["c"].split(".")[1]))
+        self.assertIn("xms_ds_nonce", cdt_envelope)
+        self.assertEqual(cdt_envelope["constraints"], constraints)
+
+    def assertAppObtainsCdt(self, client_app, scopes) -> None:
+        constraints1 = [self.createConstraint("ns:usr", "create", ["guid1"])]
+        result = client_app.acquire_token_for_client(
+            scopes, delegation_constraints=constraints1,
+        )
+        self.assertCdt(result, constraints1)
+
+        constraints2 = [self.createConstraint("ns:app", "update", ["guid2"])]
+        result = client_app.acquire_token_for_client(
+            scopes, delegation_constraints=constraints2,
+        )
+        self.assertEqual(result["token_source"], "cache", "App token Should hit cache")
+        self.assertCdt(result, constraints2)
+
+        result = client_app.acquire_token_for_client(
+            scopes, delegation_constraints=constraints2,
+            delegation_confirmation_key=client_app._get_rsa_key("new"),
+        )
+        self.assertEqual(
+            result["token_source"], "identity_provider",
+            "Different key should result in a new app token")
+        self.assertCdt(result, constraints2)
+
+    @patch("msal.authority.tenant_discovery", new=Mock(return_value={
+        "authorization_endpoint": "https://contoso.com/placeholder",
+        "token_endpoint": "https://contoso.com/placeholder",
+        }))
+    def test_acquire_token_for_client_should_return_a_cdt(self):
+        app = msal.ConfidentialClientApplication("id", client_credential="secret")
+        with patch.object(app.http_client, "post", return_value=MinimalResponse(
+            status_code=200, text=json.dumps({
+                "token_type": "Bearer",
+                "access_token": "app token",
+                "expires_in": 3600,
+                "xms_ds_nonce": "nonce",
+        }))) as mocked_post:
+            self.assertAppObtainsCdt(app, ["scope1", "scope2"])
+            self.assertEqual(mocked_post.call_count, 2)
+

tests/test_crypto.py

@@ -0,0 +1,12 @@
+from unittest import TestCase
+
+from msal.crypto import _generate_rsa_key, _convert_rsa_keys
+
+
+class CryptoTestCase(TestCase):
+    def test_key_generation(self):
+        key = _generate_rsa_key()
+        _, jwk = _convert_rsa_keys(key)
+        self.assertEqual(jwk.get("kty"), "RSA")
+        self.assertIsNotNone(jwk.get("n") and jwk.get("e"))
+

tests/test_e2e.py

@@ -5,6 +5,7 @@
 LAB_OBO_PUBLIC_CLIENT_ID=...
 LAB_OBO_CONFIDENTIAL_CLIENT_ID=...
 """
+from __future__ import annotations
 try:
     from dotenv import load_dotenv  # Use this only in local dev machine
     load_dotenv()  # take environment variables from .env.
@@ -27,8 +28,10 @@
 
 import msal
 from tests.http_client import MinimalHttpClient, MinimalResponse
+from tests.test_application import CdtTestCase
 from msal.oauth2cli import AuthCodeReceiver
 from msal.oauth2cli.oidc import decode_part
+from msal.application import _build_req_cnf
 
 try:
     import pymsalruntime
@@ -533,7 +536,7 @@ def tearDownClass(cls):
         cls.session.close()
 
     @classmethod
-    def get_lab_app_object(cls, client_id=None, **query):  # https://msidlab.com/swagger/index.html
+    def get_lab_app_object(cls, client_id=None, **query) -> dict:  # https://msidlab.com/swagger/index.html
         url = "https://msidlab.com/api/app/{}".format(client_id or "")
         resp = cls.session.get(url, params=query)
         result = resp.json()[0]
@@ -791,12 +794,12 @@ def test_user_account(self):
         self._test_user_account()
 
 
-def _data_for_pop(key):
-    raw_req_cnf = json.dumps({"kid": key, "xms_ksl": "sw"})
+def _data_for_pop(key_id):
     return {  # Sampled from Azure CLI's plugin connectedk8s
         'token_type': 'pop',
-        'key_id': key,
-        "req_cnf": base64.urlsafe_b64encode(raw_req_cnf.encode('utf-8')).decode('utf-8').rstrip('='),
+        'key_id': key_id,
+        "req_cnf": _build_req_cnf(
+            {"kid": key_id, "xms_ksl": "sw"}, remove_padding=True),
             # Note: Sending raw_req_cnf without base64 encoding would result in an http 500 error
     }  # See also https://github.com/Azure/azure-cli-extensions/blob/main/src/connectedk8s/azext_connectedk8s/_clientproxyutils.py#L86-L92
 
@@ -817,6 +820,23 @@ def test_user_account(self):
         self._test_user_account()
 
 
+class CdtTestCase(LabBasedTestCase, CdtTestCase):
+    def test_acquire_token_for_client_should_return_a_cdt(self):
+        resource = self.get_lab_app_object(  # This resource has opted in to CDT
+            publicClient="no", signinAudience="AzureAdMyOrg")
+        client_app = msal.ConfidentialClientApplication(
+            # Any CCA can use a CDT, as long as the resource opted in for a CDT
+            # Here we use the OBO app which is in same tenant as the resource.
+            os.getenv("LAB_OBO_CONFIDENTIAL_CLIENT_ID"),
+            client_credential=os.getenv("LAB_OBO_CLIENT_SECRET"),
+            authority="{}{}.onmicrosoft.com".format(
+                resource["authority"],
+                resource["labName"].lower().rstrip(".com"),
+                ),
+            )
+        self.assertAppObtainsCdt(client_app, [f"{resource['appId']}/.default"])
+
+
 class WorldWideTestCase(LabBasedTestCase):
 
     def test_aad_managed_user(self):  # Pure cloud