Draft LDAP Single Sign On Token Processing

The need for new, simple single sign-on capable systems has arisen with the development of new technologies and systems. For these systems we should be able to provide a simple, localised and complete single sign-on service. This does not aim to replace Kerberos V5. It is designed for when Kerberos is too invasive for installation in an environment. Tokens generated by this system should be able to be transmitted over different protocols allowing middleware to relay tokens to clients. Clients can then contact the middleware natively and the middleware can negotiate the client authentication with the LDAP server. This implementation will provide an LDAP extended operation to create tokens which a client may cache, or relay to a further client. The token can then be sent in a SASL bind request to the LDAP server. The token remains valid over many binds. Finally, Tokens for a client are always able to be revoked at the LDAP Server using an LDAP extended operation, allowing global logout by the user or administrator.

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119.

This document has two components. A SASL Mechanism, and LDAP extended operations. There is no strict requirement for the two to coexist: The LDAP Operation is an implementation of the service providing tokens, and the SASL Mechanism to authenticate them. In theory, an alternate protocol and database could generate and authenticate these tokens.

Token formats are server implementation specific: As they are the only entity that will decrypt and consume them, they have the option to provide these in any format they wish. This means the client will only see an opaque data structure, and will only need to transmit this opaque structure as part of the authentication request. For the token system to operate correctly the server MUST generate tokens that contain at least these three values: Date Time Issued Date Time Until User Unique Id As the client does not ever see the contents the User Unique Id can be anything within the database that uniquely identifies the user that is the holder of the token. The User Unique Id MUST be an UTF8 String. The token format MUST be encrypted. The token format can be decrypted with either a asymmetric or symmetric keying system. The token format MUST have a form of data authentication. This can be through authenticated encryption, or validation of a hash. The Date Time Issued MUST be a complete timestamp in UTC, to prevent issues with changing timezones. Without these guarantees, the token system is not secure, and is vulnerable to credential forgery attacks. Here is an EXAMPLE ASN.1 format that would be encrypted and sent to the client:

This would be encrypted with AES-GCM and transmitted to the client. Another example would be to use a fernet token Fernet Specification.

Timestamp can be considered to be the DateTimeIssued as: "This field is a 64-bit unsigned big-endian integer. It records the number of seconds elapsed between January 1, 1970 UTC and the time the token was created." We can then create a Cipher text containing:

The Date Time Until is a 64-bit unsigned big-endian integer. It is, like Date Time Issued, the number of seconds since January 1, 1970 UTC, and the token creation time added to the number of seconds of the requested life time. This example format satisfies all of our data requirements for the sso token system.

The client will request a token from the authentication server. The acquisition method for the token is discussed in section XXX. For authentication, the client MUST send the token as it was received. IE changes to formatting are not permitted. The client MUST send the an appropriate authid in RFC 2078 form. This authid MUST internally match the User Unique Id in the token. The server is responsible for this validation. The client MAY transform the token if acting in a proxy fashion. However this transformation must be deterministic and able to be reversed to satisfy the previous requirement.

The client issues a SASL bind request with the mechanism name LDAPSSOTOKEN. The client sends an appropriate authid in RFC 2078 form. The client provides the encrypted token that was provided in the LDAPSSOTokenResponse Token Field. The token is decrypted and authenticated based on the token format selected by the server. The server MAY attempt multiple token keys and or formats to find the correct issuing format and key. If the token decryption fails, the attempt with this key and format MUST be considered to fail. If the values have been tampered with, IE hash authentication fails, the attempt with the key and format MUST be considered to fail. The token decryption MUST return a valid DateTimeUntil, DateTimeIssued and User Unique Id. If this is not returned, the decryption MUST be considered to fail. If all token formats and keys fail to decrypt, this MUST cause an invalidCredentials error. The DateTimeUntil field is checked against the servers current time. If the current time exceeds or is equal to DateTimeUntil, invalidCredentials MUST be returned. The User Unique Id is validated to exist on the server. If the User Unique Id does not exist, invalidCredentials MUST be returned. The authid provided by the SASL client is verified with the User Unique Id. For example if the authid is william@EXAMPLE.COM, the server maps this to an identity. Once this identity is validated, the identity is check to match the User Unique Id. If they do not match, the authentication MUST fail. The DateTimeIssued field is validated against the User Unique Id object's attribute or related attribute that contains "Valid Not Before". If the value of "Valid Not Before" exceeds or is equal to DateTimeIssued, invalidCredentials MUST be returned. Only if all of these steps have succeeded, then the authentication is considered successful.

The management and details of the "Valid Not Before" attribute are left to the implementation to decide how to implement and manage. The implementation should consider how an administrator or responsible party could revoke tokens for users other than their own. The Valid Not Before SHOULD be replicated between LDAP servers to allow correct revocation across many LDAP servers. For example, Valid Not Before MAY be an attribute on the User Unique Id object, or MAY be on another object with a unique relation to the User Unique Id.

An ldap extended operation is issued as per Section 4.12 of RFC 4511. The LDAP OID to be used for the LDAPSSOTokenRequest is 2.16.840.1.113730.3.5.14. The LDAP OID to be used for the LDAPSSOTokenResponse is 2.16.840.1.113730.3.5.15. A User Unique Id is selected. This may be the Bind DN, UUID or other utf8 identifier that uniquely determines an object. The extended operation must fail if the LDAP connection security stregth factors is 0. Tokens must not be generated for Anonymous binds. This means, tokens may only be generated for connections with a valid bind dn set. Token requests MUST contain a requested lifetime in seconds. The server MAY choose to ignore this lifetime and set it's own value. A token request of a negative or zero value SHOULD default to a server definied minimum lifetime. The token is created as per an example token format in 4.1. This value is then encrypted with an encryption algorithm of the servers choosing. The client does not need to be aware of the encryption algorithm. The DateTimeIssued, DateTimeUntil and User Unique Id are collected in the format required by the token format we are choosing to use in the server. The token is then generated by the chosen algorithm. The encrypted token is sent to the client in the LDAPSSOTokenResponse structure, along with the servers chosen valid life time as a guide for the client to approximate the expiry of the token. This valid life time value is in seconds. If the token cannot be generated due to a server error, LDAP_OPERATION_ERROR MUST be returned.

An ldap extended operation is issued as per Section 4.12 RFC 4511. The LDAP OID to be used for LDAPSSOTOKENRevokeRequest is 2.16.840.1.113730.3.5.16. The extended operation MUST fail if the connection is anonymous. The extended operation MUST fail if the LDAP connection security strength factors is 0. The extended operation MUST only act upon the "Valid Not Before" attribute related to the bind DN of the connection. Upon recieving the extended operation to revoke tokens, the directory server MUST set the current BindDN's related "Valid Not Before" attribute timestamp to the current datetime. This will have the effect, that all previously issued tokens are invalidated. This revocation option must work regardless of directory server access controls on the attribute containing "Valid Not Before".

The extended operation requestValue MUST not be set for LDAP SSO Token revocation. The extended operation does not provide a response OID. The result is set in the LDAPResult.

The SASL bind attempt MUST fail if the LDAP connection security strength factors is 0. The SASL Authentication is attempted as per Section 4.3. If this does not succeed, the bind attempt MUST fail. The LDAP Object is retrived from the User Unique Id, and a Bind DN Determined. If no Bind DN can be determined, the bind attempt MUST fail. The current Bind DN MUST be set to the Bind DN of the LDAP object that is determined, and the result ldap success is returned to the LDAP client.

Due to the design of this token, it is possible to use it in a replay attack. Notable threats are storage on the client and man in the middle attacks. To minimise the man in the middle attack thread, LDAP security strength factor of greater than 0 is a requirement. Client security is not covered by this document.

The SASL mechanism, LDAPSSOTOKEN, MUST be registered to IANA as per RFC 2222 Section 6.4