class Publish:
def __init__(self,
profiles,
login_method,
publisher_name,
discovery_url=None):
"""
@profiles list of cis_profiles.User
@login_method str a valid login_method for the user (such as "ad")
@publisher_name str of your publisher name (such as 'ldap' or 'mozilliansorg')
@discovery_url a discovery URL for CIS (CIS_DISCOVERY_URL env var will be used otherwise)
"""
self.profiles = profiles
self.login_method = login_method
self.publisher_name = publisher_name
if discovery_url is None:
discovery_url = os.environ.get(
"CIS_DISCOVERY_URL",
"https://auth.mozilla.com/.well-known/mozilla-iam")
self.__discovery_url = discovery_url
# Defaults
self.api_url = None
self.api_url_person = None
self.api_url_change = None
self.access_token = None
# Number of retries when calling CIS APIs, for robustness
self.max_retries = 3
self.max_threads = 50
# retry_delay is passed to time.sleep
self.retry_delay = 1
# known_cis_users is the output of the Person API query
# known_cis_users_by_user_id is a dict that maps user_id: email
# known_cis_users_by_email is a dict that maps email: user_id (instead of user_id: email)
self.cis_user_list = None
self.known_cis_users = None
self.known_cis_users_by_email = {}
self.known_cis_users_by_user_id = {}
self.all_known_profiles = {}
self.__inited = False
def __deferred_init(self):
"""
Init all data that requires external resources
"""
if self.__inited:
return
logger.info("Getting API URLs from well-known {}".format(
self.__discovery_url))
self.secret_manager = secret.Manager()
self.config = common.get_config()
self.__well_known = WellKnown(self.__discovery_url)
wk = self.__well_known.get_well_known()
self.api_url = wk["api"]["endpoints"]
# XXX These are not currently used
# self.api_audience = wk["api"]["audience"]
# self.api_url_person = self.api_url["person"]
# self.api_url_change = self.api_url["change"]
self.api_audience = self.config("api_identifier",
namespace="cis",
default="api.dev.sso.allizom.org")
self.api_url_person = "https://" + self.config(
"person_api_url",
namespace="cis",
default="person.api.dev.sso.allizom.org")
self.api_url_change = "https://" + self.config(
"change_api_url",
namespace="cis",
default="change.api.dev.sso.allizom.org")
self.publisher_rules = self.__well_known.get_publisher_rules()
self.__inited = True
def post_all(self, user_ids=None):
"""
Post all profiles
@user_ids list of str which are user ids like 'ad|test'
Returns list of failed users (empty if no failure)
"""
self.__deferred_init()
qs = "/v2/user"
threads = []
failed_users = queue.Queue()
logger.info("Received {} user profiles to post".format(
len(self.profiles)))
if user_ids is not None:
logger.info(
"Requesting a specific list of user_id's to post {} (total user_ids: {}, total profiles: {})"
.format(user_ids, len(user_ids), len(self.profiles)))
if not isinstance(user_ids, list):
raise PublisherError("user_ids must be a list", user_ids)
# list what to delete, then delete instead of slower copy list operations or filters
# This is because some data sets are huge / GBs of data
xlist = []
for idx, profile in enumerate(self.profiles):
if profile.user_id.value is None:
if self.known_cis_users_by_email.get(
profile.primary_email.value) not in user_ids:
xlist.append(idx)
elif profile.user_id.value not in user_ids:
xlist.append(idx)
for i in reversed(xlist):
if self.profiles[i].active.value:
del self.profiles[i]
logger.info(
"After filtering, we have {} user profiles to post".format(
len(self.profiles)))
# XXX - we already validate in the API, is this needed?
# self.validate()
for profile in self.profiles:
# If we have no user_id provided we need to find it here
# These are always considered updated users, not new users
if profile.user_id.value is None:
user_id = self.known_cis_users_by_email[
profile.primary_email.value]
else:
user_id = profile.user_id.value
# Filter out non-updatable attributes as needed
self.filter_known_cis_users(profiles=[profile])
threads.append(
threading.Thread(target=self._really_post,
args=(user_id, qs, profile, failed_users)))
threads[-1].start()
num_threads = len(threading.enumerate())
while num_threads >= self.max_threads:
time.sleep(1)
num_threads = len(threading.enumerate())
logger.info("Too many concurrent threads, waiting a bit...")
logger.debug("Waiting for threads to terminate...")
for t in threads:
t.join()
logger.debug("Retrieving results from the queue...")
ret = []
while not failed_users.empty():
ret.append(failed_users.get())
failed_users.task_done()
return ret
def _really_post(self, user_id, qs, profile, failed_users):
response_ok = False
retries = 0
access_token = self._get_authzero_token()
# Existing users (i.e. users to update) have to be passed as argument
if user_id in self.known_cis_users_by_user_id:
qs = "/v2/user?user_id={}".format(user_id)
# New users do not
else:
qs = "/v2/user"
# We don't always get a user_id set
identifier = user_id
if identifier is None:
identifier = profile.primary_email.value
if identifier is None:
logger.critical("Could not find profile identifier!")
logger.debug("Posting user profile: {}".format(profile.as_dict()))
while not response_ok:
logger.info(
"Attempting to post profile (user_id: {}, primary_email: {} to API {}{}"
.format(profile.user_id.value, profile.primary_email.value,
self.api_url_change, qs))
response = self._request_post(
url="{}{}".format(self.api_url_change, qs),
payload=profile.as_dict(),
headers={"authorization": "Bearer {}".format(access_token)},
)
response_ok = response.ok
if not response_ok:
logger.warning(
"Posting profile {} to API failed, retry is {} retry_delay is {} status_code is {} reason is {}"
"contents were {}".format(identifier, retries,
self.retry_delay,
response.status_code,
response.reason, response.text))
retries = retries + 1
time.sleep(self.retry_delay)
if retries >= self.max_retries:
logger.error(
"Maximum retries reached ({}), profile is not to be sent {}"
.format(retries, identifier))
failed_users.put(identifier)
break
else:
logger.info(
"Profile successfully posted to API {}, status_code: {}".
format(identifier, response.status_code))
def _request_post(self, url, payload, headers):
return requests.post(url, json=payload, headers=headers)
def _request_get(self, url, qs, headers):
return requests.get("{}{}".format(url, qs), headers=headers)
def _get_authzero_client(self):
authzero = secret.AuthZero(
client_id=self.secret_manager.secret("client_id"),
client_secret=self.secret_manager.secret("client_secret"),
api_identifier=self.api_audience,
authzero_tenant=self.config("authzero_tenant",
namespace="cis",
default="auth.mozilla.auth0.com"),
)
return authzero
def _get_authzero_token(self):
# This could check for expiration
if self.access_token is not None:
return self.access_token
else:
authzero = self._get_authzero_client()
self.access_token = authzero.exchange_for_access_token()
return self.access_token
def get_known_cis_users_paginated(self):
"""
Call CIS Person API and return a list of all known profiles
return: list of dict JSON profiles
"""
self.__deferred_init()
if len(self.all_known_profiles) > 0:
return self.all_known_profiles
logger.info(
"Requesting CIS Person API for a list of all user profiles")
qs = "/v2/users"
access_token = self._get_authzero_token()
nextPage = ""
while nextPage is not None:
if nextPage != "":
real_qs = "{}?nextPage={}".format(qs, nextPage)
else:
real_qs = qs
response = self._request_get(
self.api_url_person,
real_qs,
headers={"authorization": "Bearer {}".format(access_token)})
if not response.ok:
logger.error(
"Failed to query CIS Person API: {}{} response: {}".format(
self.api_url_person, real_qs, response.text))
raise PublisherError("Failed to query CIS Person API",
response.text)
response_json = response.json()
for p in response_json["Items"]:
self.all_known_profiles[p["user_id"]["value"]] = p
nextPage = response_json.get("nextPage")
logger.info("Got {} users known to CIS".format(
len(self.all_known_profiles)))
return self.all_known_profiles
def get_known_cis_users(self):
"""
Call CIS Person API and return a list of existing user ids and/or remails
return: list of str: cis user ids
"""
self.__deferred_init()
if self.known_cis_users is not None:
return self.known_cis_users
logger.info(
"Requesting CIS Person API for a list of existing users for method {}"
.format(self.login_method))
qs = "/v2/users/id/all?connectionMethod={}&active=True".format(
self.login_method)
access_token = self._get_authzero_token()
response = self._request_get(
self.api_url_person,
qs,
headers={"authorization": "Bearer {}".format(access_token)})
if not response.ok:
logger.error(
"Failed to query CIS Person API: {}{} response: {}".format(
self.api_url_person, qs, response.text))
raise PublisherError("Failed to query CIS Person API",
response.text)
self.known_cis_users = response.json()
logger.info("Got {} users known to CIS".format(
len(self.known_cis_users)))
# Also save copies that are easier to query directly
for u in self.known_cis_users:
self.known_cis_users_by_user_id[u["user_id"]] = u["primary_email"]
self.known_cis_users_by_email[u["primary_email"]] = u["user_id"]
return self.known_cis_users
def get_cis_user(self, user_id):
"""
Call CIS Person API and return the matching user profile
@user_id str a user_id
"""
self.__deferred_init()
logger.info(
"Requesting CIS Person API for a user profile {}".format(user_id))
access_token = self._get_authzero_token()
qs = "/v2/user/user_id/{}".format(
urlencode(user_id, quote_via=quote_plus))
response = self._request_get(
self.api_url_person,
qs,
headers={"authorization": "Bearer {}".format(access_token)})
if not response.ok:
logger.error(
"Failed to query CIS Person API: {}{} response: {}".format(
self.api_url_person, qs, response.text))
raise PublisherError("Failed to query CIS Person API",
response.text)
return User(response.json())
def filter_known_cis_users(self, profiles=None, save=True):
"""
Filters out fields that are not allowed to be updated by this publisher from the profile before posting
This is for "new" users
"""
self.__deferred_init()
self.get_known_cis_users()
if profiles is None:
profiles = self.profiles
# Never NULL/None these fields during filtering as they're used for knowing where to post
whitelist = ["user_id", "active"]
allowed_updates = self.publisher_rules["update"]
for n in range(0, len(profiles)):
p = profiles[n]
if p.user_id.value is None:
user_id = self.known_cis_users_by_email[p.primary_email.value]
else:
user_id = p.user_id.value
if user_id in self.known_cis_users_by_user_id:
logger.debug(
"Filtering out non-updatable values from user {} because it already exist in CIS"
.format(user_id))
for pfield in p.__dict__:
# Skip? (see below for sub item)
if pfield in whitelist:
continue
if pfield not in allowed_updates:
continue
# sub-item?
elif pfield in [
"identities", "staff_information",
"access_information"
]:
for subpfield in p.__dict__[pfield]:
# Skip?
if subpfield in whitelist:
continue
# XXX access_information.{hris,ldap, ...} - this needs refactor
exit_loop = False
if isinstance(allowed_updates[pfield], dict):
for sub_au in allowed_updates[pfield]:
if (p.__dict__[pfield][subpfield]
["signature"]["publisher"]["name"] ==
self.publisher_name):
exit_loop = True
break
if exit_loop:
continue
if allowed_updates[pfield] != self.publisher_name:
p.__dict__[pfield][subpfield]["signature"][
"publisher"]["value"] = ""
if "value" in p.__dict__[pfield][
subpfield].keys():
p.__dict__[pfield][subpfield][
"value"] = None
elif "values" in p.__dict__[pfield][
subpfield].keys():
p.__dict__[pfield][subpfield][
"values"] = None
else:
if allowed_updates[pfield] != self.publisher_name:
p.__dict__[pfield]["signature"]["publisher"][
"value"] = ""
if "value" in p.__dict__[pfield].keys():
p.__dict__[pfield]["value"] = None
elif "values" in p.__dict__[pfield].keys():
p.__dict__[pfield]["values"] = None
logger.debug("Filtered fields for user {}".format(user_id))
profiles[n] = p
if save:
self.profiles = profiles
return profiles
def validate(self):
"""
Validates all profiles are from the correct provider
"""
logger.info("Validating {} profiles".format(len(self.profiles)))
# XXX ensure ldap2s3 use the right login_method
# then remove this
lm_map = {"ad": ["Mozilla-LDAP", "Mozilla-LDAP-Dev"]}
if self.login_method in lm_map:
local_login_method = lm_map[self.login_method]
else:
local_login_method = [self.login_method]
for profile in self.profiles:
if profile.login_method.value not in local_login_method:
logger.error(
"Incorrect login method for this user {} - looking for {} but got {}"
.format(profile.user_id.value, local_login_method,
profile.login_method.value))
logger.info("Validation completed for all profiles")
def test_rules_retrieve(self):
wk = WellKnown()
data = wk.get_publisher_rules()
assert isinstance(data, dict)
assert isinstance(data.get("create"), dict)
class User(object):
"""
A Mozilla IAM Profile "v2" user structure.
It is loaded a configuration file (JSON) and dynamically generated.
If you wish to change the structure, modify the JSON file!
By default this will load the JSON file with its defaults.
You can use this like a normal class:
```
from cis_profile import User
skel_user = User(user_id="bobsmith")
skel_user.user_id.value = "notbobsmith"
if skel_user.validate():
profile = skel_user.as_json()
```
"""
def __init__(self,
user_structure_json=None,
user_structure_json_file=None,
discovery_url=None,
**kwargs):
"""
@user_structure_json an existing user structure to load in this class
@user_structure_json_file an existing user structure to load in this class, from a JSON file
@discovery_url the well-known Mozilla IAM URL
@kwargs any user profile attribute name to override on initializing, eg "user_id='test'"
"""
if discovery_url is None:
discovery_url = os.environ.get(
"CIS_DISCOVERY_URL",
"https://auth.mozilla.com/.well-known/mozilla-iam")
self.__well_known = WellKnown(discovery_url)
if user_structure_json is not None:
# Auto-detect if the passed struct is a JSON string or JSON dict
if isinstance(user_structure_json, str):
self.load(json.loads(user_structure_json))
else:
self.load(user_structure_json)
elif user_structure_json_file is not None:
self.load(self.get_profile_from_file(user_structure_json_file))
else:
# Load builtin defaults
self.load(
self.get_profile_from_file("data/user_profile_null.json"))
# Insert defaults from kwargs
for kw in kwargs:
if kw in self.__dict__.keys():
try:
self.__dict__[kw]["value"] = kwargs[kw]
except KeyError:
self.__dict__[kw]["values"] += [kwargs[kw]]
else:
logger.error("Unknown user profile attribute {}".format(kw))
raise Exception("Unknown user profile attribute {}".format(kw))
self.__signop = cis_crypto.operation.Sign()
self.__verifyop = cis_crypto.operation.Verify()
self.__verifyop.well_known = self.__well_known.get_well_known()
def load(self, profile_json):
"""
Load an existing JSON profile
@profile_json: dict (e.g. from json.load() or json.loads())
"""
logger.debug("Loading profile JSON data structure into class object")
self.__dict__.update(DotDict(profile_json))
def get_profile_from_file(self, user_structure_json_path):
"""
Load the json structure into a 'DotDict' so that attributes appear as addressable object values
Usually used with load().
"""
logger.debug("Loading default profile JSON structure from {}".format(
user_structure_json_path))
if not os.path.isfile(user_structure_json_path):
dirname = os.path.dirname(os.path.realpath(__file__))
path = dirname + "/" + user_structure_json_path
else:
path = user_structure_json_path
return DotDict(json.load(open(path)))
def merge(self, user_to_merge_in, level=None, _internal_level=None):
"""
Recursively merge user attributes:
Merge a User object with another User object. This will override all fields from the current user by non-null
(non-None) fields from `user_to_merge_in`.
Ex:
u_orig = User()
u_patch = User()
u_patch.timezone.value = "GMT -07:00 America/Los Angeles"
print(u_patch.uuid.value)
# >>> None
u_orig.merge(u_patch)
print(u_orig.timezone.value)
# >>> GMT -07:00 America/Los Angeles
print(u_orig.uuid.value)
# >>> None # <= unchanged because null/None
This is useful when updating existing users, rather than manually matching fields
!!! WARNING !!! This function will not do a validation of signatures or publishers. YOU must perform these tasks
by calling the appropriate functions.
@user_to_merge_in User object the user to merge into the current user object
@level dict of an attribute. This can be user_to_merge_in.__dict__ for the top level (recurses through all
attributes). The level MUST be from user_to_merge_in, not from the original/current user.
@_internal_level str attribute name of previous level attribute from the current user. Used internally.
This function returns a diff of what was changed and is always successful.
"""
tomerge = []
different_attrs = []
# Default to top level
if level is None:
level = user_to_merge_in.__dict__
if _internal_level is None:
_internal_level = self.__dict__
for attr in level.keys():
# If this is an internal class object, skip it
if attr.startswith("_") or not isinstance(level[attr], dict):
continue
# If we have no signature (or metadata in theory), this is not a "User attribute", keep doing deeper
if "signature" not in level[attr].keys():
different_attrs.extend(
self.merge(user_to_merge_in,
level=level[attr],
_internal_level=_internal_level[attr]))
# We will merge this attribute back (granted its not null/None)
else:
tomerge.append(attr)
for attr in tomerge:
# _internal_level is the original user attr
# level is the patch/merged in user attr
# This is where we skip null/None attributes even if the original/current
# user does not match (ie is not null)
# What is considered equivalent:
# Same `value` or `values`
# Same `metadata.display` or `metadata.verified`
# Different `signature.*` is ignored if the rest matches (re-signing existing values is ignored)
# Different `metadata.{last_modified,created,classification}` is ignored if the rest matches
if level[attr].get("value") is not None or level[attr].get(
"values") is not None:
if (_internal_level[attr].get("value") !=
level[attr].get("value")) or (
_internal_level[attr].get("values") !=
level[attr].get("values") or
(_internal_level[attr]["metadata"]["display"] !=
level[attr]["metadata"]["display"]) or
(_internal_level[attr]["metadata"]["verified"] !=
level[attr]["metadata"]["verified"])):
logger.debug("Merging in attribute {}".format(attr))
different_attrs.append(attr)
_internal_level[attr] = level[attr]
return different_attrs
def initialize_uuid_and_primary_username(self):
try:
salt = cis_crypto.secret.AWSParameterstoreProvider().uuid_salt()
except ClientError as e:
logger.critical(
"No salt set for uuid generation. This is very very dangerous: {}"
.format(e))
salt = ""
now = self._get_current_utc_time()
uuid = uuid5(NAMESPACE_URL,
"{}#{}".format(salt, self.__dict__["user_id"]["value"]))
self.__dict__["uuid"]["value"] = str(uuid)
self.__dict__["uuid"]["metadata"]["created"] = now
self.__dict__["uuid"]["metadata"]["last_modified"] = now
primary_username = "******".format(
urlsafe_b64encode(uuid.bytes).decode("utf-8"))
self.__dict__["primary_username"]["value"] = primary_username
self.__dict__["primary_username"]["metadata"]["created"] = now
self.__dict__["primary_username"]["metadata"]["last_modified"] = now
def initialize_timestamps(self):
now = self._get_current_utc_time()
logger.debug(
"Setting all profile metadata fields and profile modification timestamps to now: {}"
.format(now))
for item in self.__dict__:
if type(self.__dict__[item]) is not DotDict:
continue
if "metadata" in self.__dict__[item]:
self.__dict__[item]["metadata"]["created"] = now
self.__dict__[item]["metadata"]["last_modified"] = now
else:
# This is a 2nd level attribute such as `access_information`
# Note that we do not have a 3rd level so this is sufficient
for subitem in self.__dict__[item]:
if isinstance(
self.__dict__[item][subitem], dict
) and "metadata" in self.__dict__[item][subitem]:
self.__dict__[item][subitem]["metadata"][
"created"] = now
self.__dict__[item][subitem]["metadata"][
"last_modified"] = now
# XXX Hard-coded special profile values
self.__dict__["last_modified"].value = now
self.__dict__["created"].value = now
def update_timestamp(self, req_attr):
"""
Updates metadata timestamps for that attribute
@attr a valid user profile attribute
"""
req_attrs = req_attr.split(
"."
) # Support subitems/subattributes such as 'access_information.ldap'
if len(req_attrs) == 1:
attr = self.__dict__[req_attr]
else:
attr = self.__dict__[req_attrs[0]][req_attrs[1]]
if "metadata" not in attr:
raise KeyError("This attribute does not have metadata to update")
now = self._get_current_utc_time()
logger.debug(
"Updating to metadata.last_modified={} for attribute {}".format(
now, req_attr))
attr["metadata"]["last_modified"] = now
def _get_current_utc_time(self):
"""
returns str of current time that is valid for the CIS user profiles
"""
now = time.strftime("%Y-%m-%dT%H:%M:%S.000Z", time.gmtime())
return now
def _clean_dict(self):
"""
Removes non-user-attrs from internal dict
"""
user = self.__dict__.copy()
todel = []
classname = self.__class__.__name__
for k in user:
# Anything that is within the class namespace is whitelisted
if k.startswith("_{}".format(classname)) or k.startswith("_User"):
todel.append(k)
for d in todel:
del user[d]
return user
def as_json(self):
"""
Outputs a JSON version of this user
Filters out reserved values
"""
user = self._clean_dict()
return json.dumps(user)
def as_dict(self):
"""
Outputs a real dict version of this user (not a DotDict)
Filters out reserved values
"""
user = self._clean_dict()
return dict(user)
def as_dynamo_flat_dict(self):
"""
Flattens out User.as_dict() output into a simple structure without any signature or metadata.
Effectively, this outputs something like this:
```{'uuid': '11c8a5c8-0305-4524-8b41-95970baba84c', 'user_id': 'email|c3cbf9f5830f1358e28d6b68a3e4bf15', ...```
`flatten()` is recursive.
Note that this form cannot be verified or validated back since it's missing all attributes!
Return: dynamodb serialized low level dict of user in a "flattened" form for dynamodb consumption in particular
"""
user = self._clean_dict()
def sanitize(attrs):
# Types whose values need no sanitization to serialize.
supported_base_types = [type(None), bool, int, float]
# Empty strings cannot be sanitized.
def is_nonempty_str(s):
return isinstance(s, str) and len(s) > 0
def not_empty_str(v):
return not isinstance(v, str) or is_nonempty_str(v)
if type(attrs) in supported_base_types or is_nonempty_str(attrs):
return attrs
# We want to remove empty strings from lists and sanitize everything else.
if isinstance(attrs, list):
cleaned = filter(not_empty_str, attrs)
return list(map(sanitize, cleaned))
# We are dealing with a dictionary.
cleaned = {
key: sanitize(value)
for key, value in attrs.items()
if not_empty_str(key) and not_empty_str(value)
}
# If we have a dictionary, we want to ensure it only has one of either
# the "value" key or "values" key.
has_value = "value" in cleaned
has_values = "values" in cleaned
if (has_value and not has_values) or (has_values
and not has_value):
return cleaned.get("value", cleaned.get("values"))
return cleaned
serializer = TypeSerializer()
return {k: serializer.serialize(v) for k, v in sanitize(user).items()}
def filter_scopes(self,
scopes=MozillaDataClassification.PUBLIC,
level=None):
"""
Filter in place/the current user profile object (self) to only contain attributes with scopes listed in @scopes
@scopes list of str
"""
self._filter_all(level=self.__dict__,
valid=scopes,
check="classification")
def filter_display(self,
display_levels=[DisplayLevel.PUBLIC, DisplayLevel.NULL],
level=None):
"""
Filter in place/the current user profile object (self) to only contain attributes with display levels listed
in @display_levels
@display_levels list of str
"""
self._filter_all(level=self.__dict__,
valid=display_levels,
check="display")
def validate(self):
"""
Validates against a JSON schema
"""
return jsonschema.validate(self.as_dict(),
self.__well_known.get_schema())
def verify_all_publishers(self, previous_user):
"""
Verifies all child nodes have an allowed publisher set according to the rules
@previous_user profile.User object is the previous user we're updating fields from. This allows for checking if
fields are being updated (value is already set) or created (values are changed from `null`). It defaults to an
empty profile (with a bunch of `null` values).
Ex: user.verify_all_publishers(cis_profile.profile.User()) #this checks against a brand new user
Returns True on success, False if validation fails.
"""
for item in self.__dict__:
if type(self.__dict__[item]) is not DotDict:
continue
try:
attr = self.__dict__[item]
ret = self.verify_can_publish(
attr,
attr_name=item,
previous_attribute=previous_user.as_dict()[item])
except (AttributeError, KeyError):
# This is the 2nd level attribute match, see also initialize_timestamps()
for subitem in self.__dict__[item]:
attr = self.__dict__[item][subitem]
ret = self.verify_can_publish(
attr,
attr_name=subitem,
parent_name=item,
previous_attribute=previous_user.as_dict()[item]
[subitem],
)
if ret is not True:
logger.warning(
"Verification of publisher failed for attribute {}".format(
attr))
return False
return True
def verify_can_publish(self,
attr,
attr_name,
parent_name=None,
previous_attribute=None):
"""
Verifies that the selected publisher is allowed to change this attribute.
This works for both first-time publishers ('created' permission) and subsequent updates ('update' permission).
Note that this does NOT VERIFY SIGNATURE and that you MUST call verify_attribute_signature()
separately.
If you do not, any publisher can pass a fake publisher name and this function will answer that the publisher is
allowed to publish, if the correct one is passed.
@attr dict requested attribute to verify the publisher of.
@attr_name str the name of the requested attribute as we cannot look this up.
@parent_name str the name of the requested attribute's parent name as we cannot look this up.
@previous_attribute dict the previous attribute if the user is being updated. If None, the value is always
considered to be "updated" from the current value stored in self.__dict__. Otherwise, it will check against
the passed value if it's `null` or set, and consider it "created" if it's `null`, "updated" otherwise. Makes
sense? Good!
Return bool True on publisher allowed to publish, raise Exception otherwise.
"""
# SPECIAL publisher that can override `display` and `verified` in metadata IF AND ONLY IF value/values are
# matching the stored attribute. This "IF" is very important for security here.
# XXX Note that this is an exception and prone to race-conditions where a user might not get the expected
# result if unlucky with timing, e.g. they save a profile display value but instead of seeing the new value,
# they see the old value because the original publisher ran at the same time and "won" the timing battle.
publisher_that_can_override_metadata = ["mozilliansorg"]
# If there was no change made always allow publishing - this also helps if for example:
# - user contains a "created" attribute
# - user has an update merged in
# - any non-updated attribute will pass here (but may otherwise fail because their attribute will be understood
# as "updated" otherwise
if attr == previous_attribute:
logger.debug(
"Both attribute match and were not modified, allowing to publish (skipped verification logic)"
)
return True
publisher_name = attr.signature.publisher.name # The publisher that attempts the change is here
logger.debug(
"Verifying that {} is allowed to publish field {} (previous publisher {})"
.format(publisher_name, attr_name,
previous_attribute.signature.publisher.name))
operation = "create"
# Rules JSON structure:
# { "create": { "user_id": [ "publisherA", "publisherB"], ...}, "update": { "user_id": "publisherA",... }
# I know `updators` is not English :)
# DO NOTE: "create" is a list while "update" is a single item/str (ie check in the list of creators, but check
# equality against updators). This is because we explicitely do not support multiple update mechanisms, while we
# do support multiple create mechanisms.
rules = self.__well_known.get_publisher_rules()
if parent_name is None:
allowed_creators = rules["create"][attr_name]
allowed_updators = rules["update"][attr_name]
else:
try:
allowed_creators = rules["create"][parent_name][attr_name]
except TypeError: # This is not access_information, this is identities or staff_information
allowed_creators = rules["create"][parent_name]
try:
allowed_updators = rules["update"][parent_name][attr_name]
except TypeError: # This is not access_information, this is identities or staff_information
allowed_updators = rules["update"][parent_name]
# Do we have an attribute to check against?
if previous_attribute is not None:
# Creators are only allowed if there is no previous value set
if (self._attribute_value_set(previous_attribute) is
False) and (self._attribute_value_set(attr) is True):
operation = "create"
if publisher_name in allowed_creators:
logger.debug(
"[create] {} is allowed to publish field {}".format(
publisher_name, attr_name))
return True
else:
operation = "update"
# Find if the value changed at all, else don't bother checking and allow it. Otherwise, we'd fail when a
# publisher tries to validate all fields, but has not actually modified them all.
# Figure out where values are stored first:
if "value" in attr:
value = "value"
else:
value = "values"
if attr[value] == previous_attribute[value]:
if (attr["metadata"]["display"]
== previous_attribute["metadata"]["display"]
and attr["metadata"]["verified"]
== previous_attribute["metadata"]["verified"]):
logger.debug(
"[noop] {} skipped verification for {} (no changes)"
.format(publisher_name, attr_name))
return True
elif publisher_name in publisher_that_can_override_metadata:
logger.debug(
f"[{operation}] allowed {publisher_name} for {attr_name} because "
"it's part of `publisher_that_can_override_metadata`"
)
return True
if publisher_name == allowed_updators:
logger.debug(
"[update] {} is allowed to publish field {}".format(
publisher_name, attr_name))
return True
# No previous attribute set, just check we're allowed to change the field
else:
if not self._attribute_value_set(attr):
operation = "create"
logger.debug(
"[create] {} is allowed to publish field {}".format(
publisher_name, attr_name))
return True
elif publisher_name == allowed_updators:
logger.debug(
"[update] {} is allowed to publish field {}".format(
publisher_name, attr_name))
return True
else:
logger.warning(
"[noop] no previous_attribute passed, but trying to to compare against an existing value"
"({}, {})".format(publisher_name, attr_name))
raise ValueError(
"previous_attribute must be set when calling this function, if updating an attribute"
)
# None of the checks allowed this change, bail!
logger.warning(
"[{}] {} is NOT allowed to publish field {} (value: {}, previous_attribute value: {})"
.format(operation, publisher_name, attr_name, attr,
previous_attribute))
raise cis_profile.exceptions.PublisherVerificationFailure(
"[{}] {} is NOT allowed to publish field {}".format(
operation, publisher_name, attr_name))
def verify_all_signatures(self):
"""
Verifies all child nodes with a non-null value's signature against a publisher signature
"""
for item in self.__dict__:
if type(self.__dict__[item]) is not DotDict:
continue
try:
attr = self.__dict__[item]
if self._attribute_value_set(attr):
logger.debug("Verifying attribute {}".format(item))
attr = self._verify_attribute_signature(attr)
except KeyError:
# This is the 2nd level attribute match, see also initialize_timestamps()
for subitem in self.__dict__[item]:
attr = self.__dict__[item][subitem]
if self._attribute_value_set(attr):
logger.debug("Verifying attribute {}.{}".format(
item, subitem))
attr = self._verify_attribute_signature(attr)
if attr is None:
logger.warning(
"Verification failed for attribute {}".format(attr))
return False
return True
def verify_attribute_signature(self, req_attr):
"""
Verify the signature of an attribute
@req_attr str this is this user's attribute name, which will be looked up and verified in place
"""
req_attrs = req_attr.split(
"."
) # Support subitems/subattributes such as 'access_information.ldap'
if len(req_attrs) == 1:
attr = self.__dict__[req_attr]
else:
attr = self.__dict__[req_attrs[0]][req_attrs[1]]
return self._verify_attribute_signature(attr)
def _verify_attribute_signature(self, attr, publisher_name=None):
"""
Verify the signature of an attribute
@attr dict a structure of this user to be verified
@publisher_name str this is the name of the publisher that should be signing this attribute. If None, the
publisher_name from the current user structure is used instead and no check is performed.
"""
if not self._attribute_value_set(attr):
logger.error(
"Disallowing verification of NULL (None) value(s) attribute: {} for publisher {}"
.format(attr, publisher_name))
raise cis_profile.exceptions.SignatureVerificationFailure(
"Cannot verify attribute with NULL value(s)")
if publisher_name is not None and attr["signature"]["publisher"][
"name"] != publisher_name:
raise cis_profile.exceptions.SignatureVerificationFailure(
"Incorrect publisher")
else:
# If publisher name is not passed, we default to the attribute's built-in publisher
publisher_name = attr["signature"]["publisher"]["name"]
logger.debug(
"Attempting signature verification for publisher: {} and attribute: {}"
.format(publisher_name, attr))
self.__verifyop.load(attr["signature"]["publisher"]["value"])
try:
signed = json.loads(self.__verifyop.jws(publisher_name))
except jose.exceptions.JWSError as e:
logger.warning(
"Attribute signature verification failure: {} ({})".format(
attr, publisher_name))
raise cis_profile.exceptions.SignatureVerificationFailure(
"Attribute signature verification failure for {}"
"({}) ({})".format(attr, publisher_name, e))
# Finally check our object matches the stored data
attrnosig = attr.copy()
del attrnosig["signature"]
if signed is None:
raise cis_profile.exceptions.SignatureVerificationFailure(
"No data returned by jws() call for "
"attribute {}".format(attr))
elif signed != attrnosig:
raise cis_profile.exceptions.SignatureVerificationFailure(
"Signature data in jws does not match "
"attribute data => {} != {}".format(attrnosig, signed))
return attr
def sign_all(self, publisher_name, safety=True):
"""
Sign all child nodes with a non-null value(s) OR empty values (strict=False)
This requires cis_crypto to be properly setup (i.e. with keys)
To sign empty values, manually call sign_attribute()
@publisher_name str a publisher name (will be set in signature.publisher.name at signing time)
@safety bool if off will not raise a SignatureRefused error when your publisher_name does not match (will not
sign either!)
"""
logger.debug(
"Signing all profile fields that have a value set with publisher {}"
.format(publisher_name))
for item in self.__dict__:
if type(self.__dict__[item]) is not DotDict:
continue
try:
attr = self.__dict__[item]
if self._attribute_value_set(attr, strict=True):
if attr["signature"]["publisher"][
"name"] == publisher_name:
logger.debug("Signing attribute {}".format(item))
attr = self._sign_attribute(attr, publisher_name)
else:
logger.error(
"Attribute has value set but wrong publisher set, cannot sign: {} (publisher: {})"
.format(attr, publisher_name))
if safety:
raise cis_profile.exceptions.SignatureRefused(
"Attribute has value set but wrong publisher set, cannot"
" sign", attr)
except KeyError:
# This is the 2nd level attribute match, see also initialize_timestamps()
for subitem in self.__dict__[item]:
attr = self.__dict__[item][subitem]
if self._attribute_value_set(attr, strict=True):
if attr["signature"]["publisher"][
"name"] == publisher_name:
logger.debug("Signing attribute {}.{}".format(
item, subitem))
attr = self._sign_attribute(attr, publisher_name)
else:
logger.error(
"Attribute has value set but wrong publisher set, cannot sign: {} (publisher: "
"{})".format(attr, publisher_name))
if safety:
raise cis_profile.exceptions.SignatureRefused(
"Attribute has value set but wrong publisher set, cannot"
" sign", attr)
def sign_attribute(self, req_attr, publisher_name):
"""
Sign a single attribute, including null/empty/unset attributes
@req_attr str this user's attribute to sign in place
@publisher_name str a publisher name (will be set in signature.publisher.name) which corresponds to the
signing key
"""
req_attrs = req_attr.split(
"."
) # Support subitems/subattributes such as 'access_information.ldap'
if len(req_attrs) == 1:
attr = self.__dict__[req_attr]
else:
attr = self.__dict__[req_attrs[0]][req_attrs[1]]
return self._sign_attribute(attr, publisher_name)
def _attribute_value_set(self, attr, strict=True):
"""
Checks if an attribute is used/set, ie not null
@attr dict a complete CIS Profilev2 attribute (such as {'test': {'value': null}})
@strict bool if True then only null values will be ignored, if False then empty strings/Lists/Dicts will also be
ignored
returns: True if the attribute has a value, False if not
"""
# Note that None is the JSON `null` equivalent (and not "null" is not the string "null")
if "value" in attr:
if attr["value"] is None:
return False
elif isinstance(attr["value"], bool):
return True
elif not strict and len(attr["value"]) == 0:
return False
elif "values" in attr:
if attr["values"] is None:
return False
elif not strict and len(attr["values"]) == 0:
return False
else:
raise KeyError("Did not find value in attribute", attr)
return True
def _sign_attribute(self, attr, publisher_name):
"""
Perform the actual signature operation
See also https://github.com/mozilla-iam/cis/blob/profilev2/docs/Profiles.md
@attr: a CIS Profilev2 attribute
@publisher_name str a publisher name (will be set in signature.publisher.name) which corresponds to the
signing key
This method will not allow signing NULL (None) attributes
"""
if not self._attribute_value_set(attr):
logger.error(
"Disallowing signing of NULL (None) value(s) attribute: {} for publisher {}"
.format(attr, publisher_name))
raise cis_profile.exceptions.SignatureRefused(
"Signing NULL (None) attribute is forbidden")
logger.debug("Will sign {} for publisher {}".format(
attr, publisher_name))
# Extract the attribute without the signature structure itself
attrnosig = attr.copy()
del attrnosig["signature"]
self.__signop.load(attrnosig)
# Add the signed attribute back to the original complete attribute structure (with the signature struct)
# This ensure we also don't touch any existing non-publisher signatures
sigattr = attr["signature"]["publisher"]
sigattr["name"] = publisher_name
sigattr["alg"] = "RS256" # Currently hardcoded in cis_crypto
sigattr["typ"] = "JWS" # ""
sigattr["value"] = self.__signop.jws(publisher_name)
return attr
def _filter_all(self, level, valid, check):
"""
Recursively filters out (i.e. deletes) attribute values.
@level dict of an attribute. This can be self.__dict__ for the top level (recurses through all attributes)
@valid list of valid attributes values, i.e. attribute values that will be retained
@check str the attribute to check
"""
todel = []
for attr in level.keys():
if attr.startswith("_") or not isinstance(level[attr], dict):
continue
if "metadata" not in level[attr].keys():
self._filter_all(valid=valid, level=level[attr], check=check)
elif level[attr]["metadata"][check] not in valid:
todel.append(attr)
for _ in todel:
logger.debug("Removing attribute {} because it's not in {}".format(
_, valid))
del level[_]
class User(object):
"""
A Mozilla IAM Profile "v2" user structure.
It is loaded a configuration file (JSON) and dynamically generated.
If you wish to change the structure, modify the JSON file!
By default this will load the JSON file with its defaults.
You can use this like a normal class:
```
from cis_profile import User
skel_user = User(user_id="bobsmith")
skel_user.user_id.value = "notbobsmith"
if skel_user.validate():
profile = skel_user.as_json()
```
"""
def __init__(
self,
user_structure_json=None,
user_structure_json_file=None,
discovery_url="https://auth.mozilla.com/.well-known/mozilla-iam",
**kwargs):
"""
@user_structure_json an existing user structure to load in this class
@user_structure_json_file an existing user structure to load in this class, from a JSON file
@discovery_url the well-known Mozilla IAM URL
@kwargs any user profile attribute name to override on initializing, eg "user_id='test'"
"""
self.__well_known = WellKnown()
if user_structure_json is not None:
self.load(user_structure_json)
elif user_structure_json_file is not None:
self.load(self.get_profile_from_file(user_structure_json_file))
else:
# Load builtin defaults
self.load(
self.get_profile_from_file("data/user_profile_null.json"))
# Insert defaults from kwargs
for kw in kwargs:
if kw in self.__dict__.keys():
try:
self.__dict__[kw]["value"] = kwargs[kw]
except KeyError:
self.__dict__[kw]["values"] += [kwargs[kw]]
else:
logger.error("Unknown user profile attribute {}".format(kw))
raise Exception("Unknown user profile attribute {}".format(kw))
self.__signop = cis_crypto.operation.Sign()
self.__verifyop = cis_crypto.operation.Verify()
self.__verifyop.well_known = self.__well_known.get_well_known()
def load(self, profile_json):
"""
Load an existing JSON profile
@profile_json: dict (e.g. from json.load() or json.loads())
"""
logger.debug("Loading profile JSON data structure into class object")
self.__dict__.update(DotDict(profile_json))
def get_profile_from_file(self, user_structure_json_path):
"""
Load the json structure into a 'DotDict' so that attributes appear as addressable object values
Usually used with load().
"""
logger.debug("Loading default profile JSON structure from {}".format(
user_structure_json_path))
if not os.path.isfile(user_structure_json_path):
dirname = os.path.dirname(os.path.realpath(__file__))
path = dirname + "/" + user_structure_json_path
else:
path = user_structure_json_path
return DotDict(json.load(open(path)))
def merge(self,
user_to_merge_in,
publisher,
level=None,
_internal_level=None):
"""
Recursively merge user attributes:
Merge a User object with another User object. This will override all fields from the current user by non-null
(non-None) fields from `user_to_merge_in` which have a publisher set to `publisher`.
Ex:
u_orig = User()
u_patch = User()
u_patch.timezone.value = "GMT -07:00 America/Los Angeles"
u_orig.merge(u_patch, publisher="hris")
print(u_orig.timezone.value)
# >>> GMT -07:00 America/Los Angeles
This is useful when updating existing users, rather than manually matching fields. Passing the publisher ensures
that we only patch the fields that are allowed for this publisher, as a safeguard, which ensures validation will
work
@user_to_merge_in User object the user to merge into the current user object
@publisher str the publisher name that attempts to merge data in. any data from `user_to_merge_in` that is not
originating from this publisher will be ignored
@level dict of an attribute. This can be user_to_merge_in.__dict__ for the top level (recurses through all
attributes). The level MUST be from user_to_merge_in, not from the original/current user.
@_internal_level str attribute name of previous level attribute from the current user. Used internally.
This function always returns None and is always successful.
"""
tomerge = []
# Default to top level
if level is None:
level = user_to_merge_in.__dict__
if _internal_level is None:
_internal_level = self.__dict__
for attr in level.keys():
# If this is an internal class object, skip it
if attr.startswith("_") or not isinstance(level[attr], dict):
continue
# If we have no signature (or metadata in theory), this is not a "User attribute", keep doing deeper
if "signature" not in level[attr].keys():
self.merge(user_to_merge_in,
publisher=publisher,
level=level[attr],
_internal_level=_internal_level[attr])
# Check if this is an attribute we want to merge back
elif level[attr]["signature"]["publisher"]["name"] == publisher:
tomerge.append(attr)
for _ in tomerge:
logger.debug("Merging in attribute {} for publisher {}".format(
_, publisher))
# _internal_level is the original user attr
# level is the patch/merged in user attr
# We skip null/None attributes even if the original/current user does not match (ie is not null)
if level[_].get("value") is not None or level[_].get(
"values") is not None:
_internal_level[_] = level[_]
def initialize_timestamps(self):
now = self._get_current_utc_time()
logger.debug(
"Setting all profile metadata fields and profile modification timestamps to now: {}"
.format(now))
for item in self.__dict__:
if type(self.__dict__[item]) is not DotDict:
continue
if "metadata" in self.__dict__[item]:
self.__dict__[item]["metadata"]["created"] = now
self.__dict__[item]["metadata"]["last_modified"] = now
else:
# This is a 2nd level attribute such as `access_information`
# Note that we do not have a 3rd level so this is sufficient
for subitem in self.__dict__[item]:
if isinstance(
self.__dict__[item][subitem], dict
) and "metadata" in self.__dict__[item][subitem]:
self.__dict__[item][subitem]["metadata"][
"created"] = now
self.__dict__[item][subitem]["metadata"][
"last_modified"] = now
# XXX Hard-coded special profile values
self.__dict__["last_modified"].value = now
self.__dict__["created"].value = now
def update_timestamp(self, req_attr):
"""
Updates metadata timestamps for that attribute
@attr a valid user profile attribute
"""
req_attrs = req_attr.split(
"."
) # Support subitems/subattributes such as 'access_information.ldap'
if len(req_attrs) == 1:
attr = self.__dict__[req_attr]
else:
attr = self.__dict__[req_attrs[0]][req_attrs[1]]
if "metadata" not in attr:
raise KeyError("This attribute does not have metadata to update")
now = self._get_current_utc_time()
logger.debug(
"Updating to metadata.last_modified={} for attribute {}".format(
now, req_attr))
attr["metadata"]["last_modified"] = now
def _get_current_utc_time(self):
"""
returns str of current time that is valid for the CIS user profiles
"""
# instruct libc that we want UTC
os.environ["TZ"] = "UTC"
now = time.strftime("%Y-%m-%dT%H:%M:%S.000Z")
return now
def _clean_dict(self):
"""
Removes non-user-attrs from internal dict
"""
user = self.__dict__.copy()
todel = []
classname = self.__class__.__name__
for k in user:
# Anything that is within the class namespace is whitelisted
if k.startswith("_{}".format(classname)) or k.startswith("_User"):
todel.append(k)
for d in todel:
del user[d]
return user
def as_json(self):
"""
Outputs a JSON version of this user
Filters out reserved values
"""
user = self._clean_dict()
return json.dumps(user)
def as_dict(self):
"""
Outputs a real dict version of this user (not a DotDict)
Filters out reserved values
"""
user = self._clean_dict()
return dict(user)
def filter_scopes(self,
scopes=MozillaDataClassification.PUBLIC,
level=None):
"""
Filter in place/the current user profile object (self) to only contain attributes with scopes listed in @scopes
@scopes list of str
"""
self._filter_all(level=self.__dict__,
valid=scopes,
check="classification")
def filter_display(self, display_levels=[DisplayLevel.PUBLIC], level=None):
"""
Filter in place/the current user profile object (self) to only contain attributes with display levels listed
in @display_levels
@display_levels list of str
"""
self._filter_all(level=self.__dict__,
valid=display_levels,
check="display")
def validate(self):
"""
Validates against a JSON schema
"""
return jsonschema.validate(self.as_dict(),
self.__well_known.get_schema())
def verify_all_publishers(self, previous_user):
"""
Verifies all child nodes have an allowed publisher set according to the rules
@previous_user profile.User object is the previous user we're updating fields from. This allows for checking if
fields are being updated (value is already set) or created (values are changed from `null`). It defaults to an
empty profile (with a bunch of `null` values).
Ex: user.verify_all_publishers(cis_profile.profile.User()) #this checks against a brand new user
Returns True on success, False if validation fails.
"""
for item in self.__dict__:
if type(self.__dict__[item]) is not DotDict:
continue
try:
attr = self.__dict__[item]
ret = self.verify_can_publish(
attr,
attr_name=item,
previous_attribute=previous_user.as_dict()[item])
except (AttributeError, KeyError):
# This is the 2nd level attribute match, see also initialize_timestamps()
for subitem in self.__dict__[item]:
attr = self.__dict__[item][subitem]
ret = self.verify_can_publish(
attr,
attr_name=subitem,
parent_name=item,
previous_attribute=previous_user.as_dict()[item]
[subitem],
)
if ret is not True:
logger.warning(
"Verification of publisher failed for attribute {}".format(
attr))
return False
return True
def verify_can_publish(self,
attr,
attr_name,
parent_name=None,
previous_attribute=None):
"""
Verifies that the selected publisher is allowed to change this attribute.
This works for both first-time publishers ('created' permission) and subsequent updates ('update' permission).
Note that this does NOT VERIFY SIGNATURE and that you MUST call verify_attribute_signature()
separately.
If you do not, any publisher can pass a fake publisher name and this function will answer that the publisher is
allowed to publish, if the correct one is passed.
@attr dict requested attribute to verify the publisher of.
@attr_name str the name of the requested attribute as we cannot look this up.
@parent_name str the name of the requested attribute's parent name as we cannot look this up.
@previous_attribute dict the previous attribute if the user is being updated. If None, the value is always
considered to be "updated" from the current value stored in self.__dict__. Otherwise, it will check against
the passed value if it's `null` or set, and consider it "created" if it's `null`, "updated" otherwise. Makes
sense? Good!
Return bool True on publisher allowed to publish, raise Exception otherwise.
"""
publisher_name = attr.signature.publisher.name # The publisher that attempts the change is here
logger.debug("Verifying that {} is allowed to publish field {}".format(
publisher_name, attr_name))
operation = "create"
# Rules JSON structure:
# { "create": { "user_id": [ "publisherA", "publisherB"], ...}, "update": { "user_id": "publisherA",... }
# I know `updators` is not English :)
# DO NOTE: "create" is a list while "update" is a single item/str (ie check in the list of creators, but check
# equality against updators). This is because we explicitely do not support multiple update mechanisms, while we
# do support multiple create mechanisms.
rules = self.__well_known.get_publisher_rules()
if parent_name is None:
allowed_creators = rules["create"][attr_name]
allowed_updators = rules["update"][attr_name]
else:
try:
allowed_creators = rules["create"][parent_name][attr_name]
allowed_updators = rules["update"][parent_name][attr_name]
except TypeError: # This is not access_information, this is identities or staff_information
allowed_creators = rules["create"][parent_name]
allowed_updators = rules["create"][parent_name]
# Do we have an attribute to check against?
if previous_attribute is not None:
# Creators are only allowed if there is no previous value set
if (self._attribute_value_set(previous_attribute) is
False) and (self._attribute_value_set(attr) is True):
operation = "create"
if publisher_name in allowed_creators:
logger.debug(
"[create] {} is allowed to publish field {}".format(
publisher_name, attr_name))
return True
else:
operation = "update"
# Find if the value changed at all, else don't bother checking and allow it. Otherwise, we'd fail when a
# publisher tries to validate all fields, but has not actually modified them all.
# Figure out where values are stored first:
if "value" in attr:
value = "value"
else:
value = "values"
if attr[value] == previous_attribute[value]:
logger.debug(
"[noop] {} skipped verification for {} (no changes)".
format(publisher_name, attr_name))
return True
elif publisher_name == allowed_updators:
logger.debug(
"[update] {} is allowed to publish field {}".format(
publisher_name, attr_name))
return True
# No previous attribute set, just check we're allowed to change the field
else:
if not self._attribute_value_set(attr):
operation = "create"
logger.debug(
"[create] {} is allowed to publish field {}".format(
publisher_name, attr_name))
return True
elif publisher_name == allowed_updators:
logger.debug(
"[update] {} is allowed to publish field {}".format(
publisher_name, attr_name))
return True
# None of the checks allowed this change, bail!
logger.warning(
"[{}] {} is NOT allowed to publish field {} (value: {}, previous_attribute value: {})"
.format(operation, publisher_name, attr_name, attr,
previous_attribute))
raise cis_profile.exceptions.PublisherVerificationFailure(
"[{}] {} is NOT allowed to publish field {}".format(
operation, publisher_name, attr_name))
def verify_all_signatures(self):
"""
Verifies all child nodes with a non-null value's signature against a publisher signature
"""
for item in self.__dict__:
if type(self.__dict__[item]) is not DotDict:
continue
try:
attr = self.__dict__[item]
if self._attribute_value_set(attr):
attr = self._verify_attribute_signature(attr)
except KeyError:
# This is the 2nd level attribute match, see also initialize_timestamps()
for subitem in self.__dict__[item]:
attr = self.__dict__[item][subitem]
if self._attribute_value_set(attr):
attr = self._verify_attribute_signature(attr)
if attr is not True:
logger.warning(
"Verification failed for attribute {}".format(attr))
return False
return True
def verify_attribute_signature(self, req_attr):
"""
Verify the signature of an attribute
@req_attr str this is this user's attribute name, which will be looked up and verified in place
"""
req_attrs = req_attr.split(
"."
) # Support subitems/subattributes such as 'access_information.ldap'
if len(req_attrs) == 1:
attr = self.__dict__[req_attr]
else:
attr = self.__dict__[req_attrs[0]][req_attrs[1]]
return self._verify_attribute_signature(attr)
def _verify_attribute_signature(self, attr, publisher_name=None):
"""
Verify the signature of an attribute
@attr dict a structure of this user to be verified
@publisher_name str this is the name of the publisher that should be signing this attribute. If None, the
publisher_name from the current user structure is used instead and no check is performed.
"""
if publisher_name is not None and attr["signature"]["publisher"][
"value"] != publisher_name:
raise cis_profile.exceptions.SignatureVerificationFailure(
"Incorrect publisher")
self.__verifyop.load(attr["signature"]["publisher"]["value"])
try:
signed = json.loads(self.__verifyop.jws(publisher_name))
except jose.exceptions.JWSError as e:
logger.warning(
"Attribute signature verification failure: {} ({})".format(
attr, publisher_name))
raise cis_profile.exceptions.SignatureVerificationFailure(
"Attribute signature verification failure for {}"
"({}) ({})".format(attr, publisher_name, e))
# Finally check our object matches the stored data
attrnosig = attr.copy()
del attrnosig["signature"]
if signed is None:
raise cis_profile.exceptions.SignatureVerificationFailure(
"No data returned by jws() call for "
"attribute {}".format(attr))
elif signed != attrnosig:
raise cis_profile.exceptions.SignatureVerificationFailure(
"Signature data in jws does not match "
"attribute data => {} != {}".format(attrnosig, signed))
def sign_all(self, publisher_name):
"""
Sign all child nodes with a non-null value(s) OR empty values (strict=False)
To sign empty values, manually call sign_attribute()
This requires cis_crypto to be properly setup (i.e. with keys)
WARNING: Since it signs all attributes, this is to be used only when CREATING a new profile. Signing all fields
while UPDATING a profile is a sure way to cause a validation failure (since no publisher is allowed to modify
all fields)
@publisher_name str a publisher name (will be set in signature.publisher.name at signing time)
"""
logger.debug(
"Signing all profile fields that have a value set with publisher {}"
.format(publisher_name))
for item in self.__dict__:
if type(self.__dict__[item]) is not DotDict:
continue
try:
attr = self.__dict__[item]
if self._attribute_value_set(attr, strict=False):
attr = self._sign_attribute(attr, publisher_name)
except KeyError:
# This is the 2nd level attribute match, see also initialize_timestamps()
for subitem in self.__dict__[item]:
attr = self.__dict__[item][subitem]
if self._attribute_value_set(attr, strict=False):
attr = self._sign_attribute(attr, publisher_name)
def sign_attribute(self, req_attr, publisher_name):
"""
Sign a single attribute, including null/empty/unset attributes
@req_attr str this user's attribute to sign in place
@publisher_name str a publisher name (will be set in signature.publisher.name) which corresponds to the
signing key
"""
req_attrs = req_attr.split(
"."
) # Support subitems/subattributes such as 'access_information.ldap'
if len(req_attrs) == 1:
attr = self.__dict__[req_attr]
else:
attr = self.__dict__[req_attrs[0]][req_attrs[1]]
return self._sign_attribute(attr, publisher_name)
def _attribute_value_set(self, attr, strict=True):
"""
Checks if an attribute is used/set, ie not null
@attr dict a complete CIS Profilev2 attribute (such as {'test': {'value': null}})
@strict bool if True then only null values will be ignored, if False then empty strings/Lists/Dicts will also be
ignored
returns: True if the attribute has a value, False if not
"""
# Note that None is the JSON `null` equivalent (and not "null" is not the string "null")
if "value" in attr:
if attr["value"] is None:
return False
elif isinstance(attr["value"], bool):
return True
elif not strict and len(attr["value"]) == 0:
return False
elif "values" in attr:
if attr["values"] is None:
return False
elif not strict and len(attr["values"]) == 0:
return False
else:
raise KeyError(attr)
return True
def _sign_attribute(self, attr, publisher_name):
"""
Perform the actual signature operation
See also https://github.com/mozilla-iam/cis/blob/profilev2/docs/Profiles.md
@attr: a CIS Profilev2 attribute
@publisher_name str a publisher name (will be set in signature.publisher.name) which corresponds to the
signing key
"""
# Extract the attribute without the signature structure itself
attrnosig = attr.copy()
del attrnosig["signature"]
self.__signop.load(attrnosig)
# Add the signed attribute back to the original complete attribute structure (with the signature struct)
# This ensure we also don't touch any existing non-publisher signatures
sigattr = attr["signature"]["publisher"]
sigattr["name"] = publisher_name
sigattr["alg"] = "RS256" # Currently hardcoded in cis_crypto
sigattr["typ"] = "JWS" # ""
sigattr["value"] = self.__signop.jws()
return attr
def _filter_all(self, level, valid, check):
"""
Recursively filters out (i.e. deletes) attribute values.
@level dict of an attribute. This can be self.__dict__ for the top level (recurses through all attributes)
@valid list of valid attributes values, i.e. attribute values that will be retained
@check str the attribute to check
"""
todel = []
for attr in level.keys():
if attr.startswith("_") or not isinstance(level[attr], dict):
continue
if "metadata" not in level[attr].keys():
self._filter_all(valid=valid, level=level[attr], check=check)
elif level[attr]["metadata"][check] not in valid:
todel.append(attr)
for _ in todel:
logger.debug("Removing attribute {} because it's not in {}".format(
_, valid))
del level[_]