Simple LDAP Authenticator Plugin for JupyterHub
You can install it from pip with:
pip install jupyterhub-ldapauthenticator
...or using conda with:
conda install -c conda-forge jupyterhub-ldapauthenticator
If you make any changes to JupyterHub's authentication setup that changes
which group of users is allowed to login (such as changing allowed_groups
or even just turning on LDAPAuthenticator), you must change the
jupyterhub cookie secret, or users who were previously logged in and did
not log out would continue to be able to log in!
You can do this by deleting the jupyterhub_cookie_secret
file. Note
that this will log out all users who are currently logged in.
You can enable this authenticator by adding lines to your jupyterhub_config.py
.
Note: This file may not exist in your current installation! In TLJH, it is located in /opt/tljh/config/jupyterhub_config.d. Create it there if you don't already have one.
c.JupyterHub.authenticator_class = 'ldap'
At minimum, the following two configuration options must be set before the LDAP Authenticator can be used:
Address of the LDAP Server to contact. Just use a bare hostname or IP, without a port name or protocol prefix.
To authenticate a user we need the corresponding DN to bind against the LDAP server. The DN can be acquired by either:
- setting
bind_dn_template
, which is a list of string template used to generate the full DN for a user from the human readable username, or - setting
lookup_dn
toTrue
, which does a reverse lookup to obtain the user's DN. This is because some LDAP servers, such as Active Directory, don't always bind with the true DN.
If lookup_dn = False
, then bind_dn_template
is required to be a non-empty
list of templates the users belong to. For example, if some of the users in your
LDAP database have DN of the form uid=Yuvipanda,ou=people,dc=wikimedia,dc=org
and some other users have DN like uid=Mike,ou=developers,dc=wikimedia,dc=org
where Yuvipanda
and Mike
are the usernames, you would set this config item
to be:
c.LDAPAuthenticator.bind_dn_template = [
"uid={username},ou=people,dc=wikimedia,dc=org",
"uid={username},ou=developers,dc=wikimedia,dc=org",
]
Don't forget the preceeding c.
for setting configuration parameters! JupyterHub
uses traitlets for configuration, and the
c
represents the config object.
The {username}
is expanded into the username the user provides.
c.LDAPAuthenticator.lookup_dn = True
If bind_dn_template
isn't explicitly configured, i.e. the empty list, the
dynamically acquired value for DN from the username lookup will be used
instead. If bind_dn_template
is configured it will be used just like in the
lookup_dn = False
case.
The {username}
is expanded to the full path to the LDAP object returned by the
LDAP lookup. For example, on an Active Directory system {username}
might
expand to something like CN=First M. Last,OU=An Example Organizational Unit,DC=EXAMPLE,DC=COM
.
Also, when using lookup_dn = True
the options user_search_base
,
user_attribute
, lookup_dn_user_dn_attribute
and lookup_dn_search_filter
are required, although their defaults might be sufficient for your use case.
LDAP groups whose members are allowed to log in. This must be
set to either empty []
(the default, to disable) or to a list of
full DNs that have a member
attribute that includes the current
user attempting to log in.
As an example, to restrict access only to people in groups
researcher
or operations
,
c.LDAPAuthenticator.allowed_groups = [
"cn=researcher,ou=groups,dc=wikimedia,dc=org",
"cn=operations,ou=groups,dc=wikimedia,dc=org",
]
The LDAP group search filter.
The default value is an LDAP OR search that looks like the following:
(|(member={userdn})(uniqueMember={userdn})(memberUid={uid}))
So it basically compares the userdn
attribute against the member
attribute,
then against the uniqueMember
, and finally checks the memberUid
against
the uid
.
If you modify this value, you probably want to change group_attributes
too.
Here is an example that should work with OpenLDAP servers.
(member={userdn})
A list of attributes used when searching for LDAP groups.
By default, it uses member
, uniqueMember
, and memberUid
. Certain
servers may reject invalid values causing exceptions during
authentication.
All usernames will be checked against this before being sent to LDAP. This acts as both an easy way to filter out invalid usernames as well as protection against LDAP injection attacks.
By default it looks for the regex ^[a-z][.a-z0-9_-]*$
which
is what most shell username validators do.
use_ssl
is deprecated since 2.0. use_ssl=True
translates to configuring
tls_strategy="on_connect"
, but use_ssl=False
(previous default) doesn't
translate to anything.
When LDAPAuthenticator connects to the LDAP server, it can establish a SSL/TLS connection directly, or do it before binding, which is LDAP terminology for authenticating and sending sensitive credentials.
The LDAP v3 protocol deprecated establishing a SSL/TLS connection
directly (tls_strategy="on_connect"
) in favor of upgrading the
connection to SSL/TLS before binding (tls_strategy="before_bind"
).
Supported tls_strategy
values are:
- "before_bind" (default)
- "on_connect" (deprecated in LDAP v3, associated with use of port 636)
- "insecure"
When configuring tls_strategy="on_connect"
, the default value of
server_port
becomes 636.
A dictionary that will be used as keyword arguments for the constructor of the ldap3 package's Tls object, influencing encrypted connections to the LDAP server.
For details on what can be configured and its effects, refer to the ldap3 package's documentation and code:
- ldap3 documentation: https://ldap3.readthedocs.io/en/latest/ssltls.html#the-tls-object
- ldap3 code: https://github.com/cannatag/ldap3/blob/v2.9.1/ldap3/core/tls.py#L59-L82
You can for example configure this like:
c.LDAPAuthenticator.tls_kwargs = {
"ca_certs_file": "file/path.here",
}
Port on which to contact the LDAP server.
Defaults to 636
if tls_strategy="on_connect"
is set, 389
otherwise.
Only used with lookup_dn=True
or with a configured search_filter
.
Defines the search base for looking up users in the directory.
c.LDAPAuthenticator.user_search_base = 'ou=People,dc=example,dc=com'
LDAPAuthenticator will search all objects matching under this base where
the user_attribute
is set to the current username to form the userdn.
For example, if all users objects existed under the base
ou=people,dc=wikimedia,dc=org
, and the username users use is set with
the attribute uid
, you can use the following config:
c.LDAPAuthenticator.lookup_dn = True
c.LDAPAuthenticator.lookup_dn_search_filter = '({login_attr}={login})'
c.LDAPAuthenticator.lookup_dn_search_user = 'ldap_search_user_technical_account'
c.LDAPAuthenticator.lookup_dn_search_password = 'secret'
c.LDAPAuthenticator.user_search_base = 'ou=people,dc=wikimedia,dc=org'
c.LDAPAuthenticator.user_attribute = 'uid'
c.LDAPAuthenticator.lookup_dn_user_dn_attribute = 'cn'
Only used with lookup_dn=True
or with a configured search_filter
.
Together with user_search_base
, this attribute will be searched to
contain the username provided by the user in JupyterHub's login form.
# Active Directory
c.LDAPAuthenticator.user_attribute = 'sAMAccountName'
# OpenLDAP
c.LDAPAuthenticator.user_attribute = 'uid'
Only used with lookup_dn=True
.
How to query LDAP for user name lookup.
Default value '({login_attr}={login})'
should be good enough for most
use cases.
Only used with lookup_dn=True
.
Technical account for user lookup. If both lookup_dn_search_user
and
lookup_dn_search_password
are None, then anonymous LDAP query will be
done.
Only used with lookup_dn=True
.
Attribute containing user's name needed for building DN string.
See user_search_base
for info on how this attribute is used.
For most LDAP servers, this is username. For Active Directory, it is cn.
An optional list of attributes to be fetched for a user after login.
If found, these will be available as auth_state["user_attributes"]
.
Only used with lookup_dn=True
.
If configured True, the lookup_dn_user_dn_attribute
value used to
build the LDAP user's DN string is also used as the authenticated user's
JuptyerHub username.
If this is configured True, its important to ensure that the values of
lookup_dn_user_dn_attribute
are unique even after the are normalized
to be lowercase, otherwise two LDAP users could end up sharing the same
JupyterHub username.
With ldapauthenticator 2, the default value was changed to False.
LDAP3 Search Filter to limit allowed users.
That a unique LDAP user is identified with the search_filter is
necessary but not sufficient to grant access. Grant access by setting
one or more of allowed_users
, allow_all
, allowed_groups
, etc.
Users who do not match this filter cannot be allowed by any other configuration.
The search filter string will be expanded, so that:
{userattr}
is replaced with theuser_attribute
config's value.{username}
is replaced with an escaped username, either provided directly or previously looked up withlookup_dn
configured.
List of attributes to be passed in the LDAP search with search_filter
.
This has been tested against an OpenLDAP server, with the client running Python 3.4. Verifications of this code working well with other LDAP setups are welcome, as are bug reports and patches to make it work with other LDAP setups!
Please use following options for AD integration. This is useful especially in two cases:
- LDAP Search requires valid user account in order to query user database
- DN does not contain login but some other field, like CN (actual login is present in sAMAccountName, and we need to lookup CN)
c.LDAPAuthenticator.lookup_dn = True
c.LDAPAuthenticator.lookup_dn_search_filter = '({login_attr}={login})'
c.LDAPAuthenticator.lookup_dn_search_user = 'ldap_search_user_technical_account'
c.LDAPAuthenticator.lookup_dn_search_password = 'secret'
c.LDAPAuthenticator.user_search_base = 'ou=people,dc=wikimedia,dc=org'
c.LDAPAuthenticator.user_attribute = 'sAMAccountName'
c.LDAPAuthenticator.lookup_dn_user_dn_attribute = 'cn'
In setup above, first LDAP will be searched (with account ldap_search_user_technical_account) for users that have sAMAccountName=login Then DN will be constructed using found CN value.
Currently, local user creation by the LDAPAuthenticator is unsupported as this is insecure since there's no cleanup method for these created users. As a result, users who are disabled in LDAP will have access to this for far longer.
Alternatively, there's good support in Linux for integrating LDAP into the system user setup directly, and users can just use PAM (which is supported in not just JupyterHub, but ssh and a lot of other tools) to log in. You can see http://www.tldp.org/HOWTO/archived/LDAP-Implementation-HOWTO/pamnss.html and lots of other documentation on the web on how to set up LDAP to provide user accounts for your system. Those methods are very widely used, much more secure and more widely documented. We recommend you use them rather than have JupyterHub create local accounts using the LDAPAuthenticator.
Issue #19 provides additional discussion on local user creation.
If you have received a SSL/TLS handshake error, it could be that no cipher suite accepted by LDAPAuthenticator is also accepted by the LDAP server. This is likely because LDAPAuthenticator is stricter than the LDAP server and only accepts modern cipher suites than the LDAP server doesn't accept. Due to this, you should from a security perspective ideally modernize the LDAP server's accepted cipher suites rather than expand the LDAPAuthenticator accepted cipher suites to include older cipher suites.
The cipher suites that LDAPAuthenticator accepted by default come from ssl.create_default_context().get_ciphers(), which in turn can change with Python version. Upgrading Python from 3.7 - 3.9 to 3.10 - 3.13 is known to strictly reduce the set of accepted cipher suites from 30 to 17 for example. Due to this, upgrading Python could lead to observing a handshake error previously not observed.
If you want to configure LDAPAuthenticator to accept older cipher suites instead
of updating the LDAP server to accept modern cipher suites, you can do it using
LDAPAuthenticator.tls_kwargs
as demonstrated below.
# default cipher suites accepted by LDAPAuthenticator in Python 3.7 - 3.9
# it includes 30 cipher suites, where 13 of them were considered less secure
# and removed as default cipher suites in Python 3.10
old_ciphers_list_considered_less_secure = "AES128-SHA:AES256-SHA:DHE-RSA-AES128-SHA:DHE-RSA-AES256-SHA:ECDHE-ECDSA-AES128-SHA:ECDHE-ECDSA-AES256-SHA:ECDHE-RSA-AES128-SHA:ECDHE-RSA-AES256-SHA:AES128-GCM-SHA256:AES128-SHA256:AES256-GCM-SHA384:AES256-SHA256:DHE-RSA-AES128-GCM-SHA256:DHE-RSA-AES128-SHA256:DHE-RSA-AES256-GCM-SHA384:DHE-RSA-AES256-SHA256:DHE-RSA-CHACHA20-POLY1305:ECDHE-ECDSA-AES128-GCM-SHA256:ECDHE-ECDSA-AES128-SHA256:ECDHE-ECDSA-AES256-GCM-SHA384:ECDHE-ECDSA-AES256-SHA384:ECDHE-ECDSA-CHACHA20-POLY1305:ECDHE-RSA-AES128-GCM-SHA256:ECDHE-RSA-AES128-SHA256:ECDHE-RSA-AES256-GCM-SHA384:ECDHE-RSA-AES256-SHA384:ECDHE-RSA-CHACHA20-POLY1305:TLS_AES_128_GCM_SHA256:TLS_AES_256_GCM_SHA384:TLS_CHACHA20_POLY1305_SHA256"
# default cipher suites accepted by LDAPAuthenticator in Python 3.10 - 3.13
# this list includes 17 cipher suites out of the 30 in the old list, with no
# new additions
new_ciphers_list = "DHE-RSA-AES128-GCM-SHA256:DHE-RSA-AES128-SHA256:DHE-RSA-AES256-GCM-SHA384:DHE-RSA-AES256-SHA256:ECDHE-ECDSA-AES128-GCM-SHA256:ECDHE-ECDSA-AES128-SHA256:ECDHE-ECDSA-AES256-GCM-SHA384:ECDHE-ECDSA-AES256-SHA384:ECDHE-ECDSA-CHACHA20-POLY1305:ECDHE-RSA-AES128-GCM-SHA256:ECDHE-RSA-AES128-SHA256:ECDHE-RSA-AES256-GCM-SHA384:ECDHE-RSA-AES256-SHA384:ECDHE-RSA-CHACHA20-POLY1305:TLS_AES_128_GCM_SHA256:TLS_AES_256_GCM_SHA384:TLS_CHACHA20_POLY1305_SHA256"
c.LDAPAuthenticator.tls_kwargs = {
"ciphers": old_ciphers_list_considered_less_secure,
}
For reference, you can use a command like below to see what the default cipher suites LDAPAuthenticator will use in various Python versions.
docker run -it --rm python:3.13 python -c 'import ssl; c = ssl.create_default_context(); print(":".join(sorted([c["name"] for c in c.get_ciphers()])))'
This script can be written to a file such as test_ldap_auth.py
, and run with
python test_ldap_auth.py
, to test use of LDAPAuthenticator with a given config
without involving JupyterHub.
If the authenticator works, this script should print either None or a username depending if the user was considered allowed access.
import asyncio
import getpass
from traitlets.config import Config
from ldapauthenticator import LDAPAuthenticator
# Configure LDAPAuthenticator below to work against your ldap server
c = Config()
c.LDAPAuthenticator.server_address = "ldap.organisation.org"
c.LDAPAuthenticator.server_port = 636
c.LDAPAuthenticator.bind_dn_template = "uid={username},ou=people,dc=organisation,dc=org"
c.LDAPAuthenticator.user_attribute = "uid"
c.LDAPAuthenticator.user_search_base = "ou=people,dc=organisation,dc=org"
c.LDAPAuthenticator.attributes = ["uid", "cn", "mail", "ou", "o"]
# The following is an example of a search_filter which is build on LDAP AND and OR operations
# here in this example as a combination of the LDAP attributes 'ou', 'mail' and 'uid'
sf = "(&(o={o})(ou={ou}))".format(o="yourOrganisation", ou="yourOrganisationalUnit")
sf += "(&(o={o})(mail={mail}))".format(o="yourOrganisation", mail="yourMailAddress")
c.LDAPAuthenticator.search_filter = f"(&({{userattr}}={{username}})(|{sf}))"
# Run test
authenticator = LDAPAuthenticator(config=c)
username = input("Username: ")
password = getpass.getpass()
data = dict(username=username, password=password)
return_value = asyncio.run(authenticator.authenticate(None, data))
print(return_value)