]> Cisco Systems GmbH

Richtistrasse 7 Wallisellen CH-8304 ZH Switzerland +41 44 878 9200 lear@cisco.com

Nokia Siemens Networks

Linnoitustie 6 Espoo 02600 Finland +358 (50) 4871445 Hannes.Tschofenig@gmx.net http://www.tschofenig.priv.at

Cisco Systems, Inc.

170 West Tasman Drive San Jose CA 95134 USA +1 (800) 553-6387 hmauldin@cisco.com

SJD AB

Hagagatan 24 Stockholm 113 47 SE simon@josefsson.org http://josefsson.org/

OpenID has found its usage on the Internet for Web Single Sign-On. Simple Authentication and Security Layer (SASL) and the Generic Security Service Application Program Interface (GSS-API) are application frameworks to generalize authentication. This memo specifies a SASL and GSS-API mechanism for OpenID that allows the integration of existing OpenID Identity Providers with applications using SASL and GSS-API.

OpenID is a three-party protocol that provides a means for a user to offer identity assertions and other attributes to a web server (Relying Party) via the help of an identity provider. The purpose of this system is to provide a way to verify that an end user controls an identifier. Simple Authentication and Security Layer (SASL) (SASL) is used by application protocols such IMAP, POP and XMPP, with the goal of modularizing authentication and security layers, so that newer mechanisms can be added as needed. This memo specifies just such a mechanism. The Generic Security Service Application Program Interface (GSS-API) provides a framework for applications to support multiple authentication mechanisms through a unified interface. This document defines a pure SASL mechanism for OpenID, but it conforms to the new bridge between SASL and the GSS-API called GS2. This means that this document defines both a SASL mechanism and a GSS-API mechanism. We want to point out that the GSS-API interface is optional for SASL implementers, and the GSS-API considerations can be avoided in environments that uses SASL directly without GSS-API. As currently envisioned, this mechanism is to allow the interworking between SASL and OpenID in order to assert identity and other attributes to relying parties. As such, while servers (as relying parties) will advertise SASL mechanisms, clients will select the OpenID mechanism. The OpenID mechanism described in this memo aims to re-use the available OpenID specification to a maximum extent and therefore does not establish a separate authentication, integrity and confidentiality mechanism. It is anticipated that existing security layers, such as Transport Layer Security (TLS), will continued to be used. describes the interworking between OpenID and SASL. This document requires enhancements to the Relying Party and to the Client (as the two SASL communication end points) but no changes to the OpenID Provider (OP) are necessary. To accomplish this goal indirect messaging required by the OpenID specification is tunneled within SASL.

| Relying | / | Party | // | | // +-----------+ // ^ OpenID // +--|--+ // | O| | / S | p| | // A | e| | // S | n| | // L | I| | // | D| | | Client | | | | | +------------+ +----------+ ]]>

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 . The reader is assumed to be familiar with the terms used in the OpenID 2.0 specification.

Because this mechanism transports information that should not be controlled by an attacker, the OpenID mechanism MUST only be used over channels protected by TLS, and the client MUST successfully validate the server certificate, or similar integrity protected and authenticated channels.

OpenID was originally envisioned for HTTP/HTML based communications, and with the associated semantic, the idea being that the user would be redirected by the Relying Party to an identity provider who authenticates the user, and then sends identity information and other attributes (either directly or indirectly) to the Relying Party. The identity provider in the OpenID specifications is referred to as an OpenID Provider (OP). The actual protocol flow, as copied from the OpenID 2.0 specification, is as follows: The end user initiates authentication by presenting a User-Supplied Identifier to the Relying Party via their User-Agent (e.g., http://user.example.com). After normalizing the User-Supplied Identifier, the Relying Party performs discovery on it and establishes the OP Endpoint URL that the end user uses for authentication. It should be noted that the User-Supplied Identifier may be an OP Identifier, which allows selection of a Claimed Identifier at the OP or for the protocol to proceed without a Claimed Identifier if something else useful is being done via an extension. The Relying Party and the OP optionally establish an association -- a shared secret established using Diffie-Hellman Key Exchange. The OP uses an association to sign subsequent messages and the Relying Party to verify those messages; this removes the need for subsequent direct requests to verify the signature after each authentication request/response. The Relying Party redirects the end user's User-Agent to the OP with an OpenID Authentication request. This occurs as stated in Section 10.3 of . The OP authenticates the end user and establishes whether the end user will authenticate to, and share specific attributes with, the Relying Party. For instance, the OP often asks the user what to do. The manner in which the end user authenticates to their OP and any policies surrounding such authentication is out of scope of OpenID. The OP redirects the end user's User-Agent back to the Relying Party with either an assertion that authentication is approved or a message that authentication failed. The Relying Party verifies the information received from the OP including checking the Return URL, verifying the discovered information, checking the nonce, and verifying the signature by using either the shared key established during the association or by sending a direct request to the OP. When considering this flow in the context of SASL, we note that while the RP and the client both must change their code to implement this SASL mechanism, the OP must remain untouched. Hence, an analog flow that interfaces the three parties needs to be created. In the analog, we note that unlike a web server, the SASL server already has some sort of session (probably a TCP connection) established with the client. However, it may be necessary to redirect a SASL client to another application. This will be discussed below. By doing so, we externalize much of the authentiction from SASL. The steps are shown from below: The Relying Party or SASL server advertises support for the SASL OpenID mechanism to the client. The client initiates a SASL authentiation and transmits the User-Supplied Identifier as well as an optional return_to parameter. After normalizing the User-Supplied Identifier, the Relying Party performs discovery on it and establishes the OP Endpoint URL that the end user uses for authentication. The Relying Party and the OP optionally establish an association -- a shared secret established using Diffie-Hellman Key Exchange. The OP uses an association to sign subsequent messages and the Relying Party to verify those messages; this removes the need for subsequent direct requests to verify the signature after each authentication request/response. The Relying Party transmits an authentication request to the OP to obtain an assertion in the form of an indirect request. These messages are passed through the client rather than directly between the RP and the OP. OpenID defines two methods for indirect communication, namely HTTP redirects and HTML form submission. Both mechanisms are not directly applicable for usage with SASL. To ensure that a standard OpenID 2.0 capable OP can be used a new method is defined in this document that requires the OpenID message content to be encoded using a Universal Resource Idenitifier (URI). The SASL client now sends an empty response, as authentication continues via the normal OpenID flow. At this point the client application MUST construct a URL containing the content received in the previous message from the RP. This URL is transmitted to the OP either by the SASL client application or an appropriate handler, such as a browser. Next the client optionally authenticates to the OP and then approves or disapproves authentication to the Relying Party. The manner in which the end user is authenticated to their respective OP and any policies surrounding such authentication is out of scope of OpenID and and hence also out of scope for this specification. This step happens out of band from SASL. The OP will convey information about the success or failure of the authentication phase back to the RP, again using an indirect response via the client browser or handler. The client transmits over HTTP the redirect of the OP result to the RP. This step happens out of band from SASL. The RP MAY send an OpenID check_authentication request directly to the OP, if no association has been established, and the OP should be expected to respond. Again this step happens out of band from SASL. The SASL server sends an appropriate SASL response to the client, with optional Open Simple Registry (SREG) attributes.

-----(1)----->| | Advertisement | | | |<-----(2)-----<| | Initiation | | | |> - - (3) - - - - - - - - - ->| Discovery | | |>- - -(4)- - - - - - - - - - >| Association |<- - -(4)- - - - - - - - - - <| | | | |>-----(5)----->| | Indirect Auth Request | | | |<-----(6)-----<| | Client Empty Response | | | | |>- - (7)- - ->| Client GET to the OP (ext) | | | | |<- - (8)- - ->| Client / OP Auth. (ext.) | | | |<- - -(9)- - - + - - - - - - <| HTTP(s) Indirect id_res | | | |<- - -(10)- - - - - - - - - ->| Optional check_authenticate | | | |>-----(11)---->| | SASL completion with status ----- = SASL - - - = HTTP or SSL ]]>

Note the directionality in SASL is such that the client MUST send an empty response. Specifically, it processes the redirect and then awaits a final SASL decision, while the rest of the OpenID authentication process continues.

To ensure that a specific request is bound, and in particular to ease interprocess communication, it may be necessary for the relying party to encode some sort of nonce in the URIs it transmits through the client for success or failure. This can be done in any number of ways. Examples would include making changes to the base URI or otherwise including an additional fragment.

As mentioned above OpenID is primarily designed to interact with web-based applications. Portions of the authentication stream are only defined in the crudest sense. That is, when one is prompted to approve or disapprove an authentication, anything that one might find on a browser is allowed, including JavaScript, fancy style-sheets, etc. Because of this lack of structure, implementations will need to invoke a fairly rich browser in order to insure that the authentication can be completed. Once there is an outcome, the SASL server needs to know about it. The astute will hopefully by now have noticed an empty client SASL challenge. This is not to say that nothing is happening, but rather that authentication flow has shifted from SASL to OpenID, and will return when the server has an outcome to hand to the client. The alternative to this flow is some signal from the HTML browser to the SASL client of the results that is in turn passed to the SASL server. The IPC issue this raises is substantial. Better, we conclude, to externalize the authentication to the browser, and have an empty client challenge.

Based on the previous figure, the following operations are performed with the OpenId SASL mechanism:

To advertise that a server supports OpenID, during application session initiation, it displays the name "OPENID2.0" in the list of supported SASL mechanisms.

A client initiates an OpenID authentication with SASL by sending the GS2 header followed by the XRI or URI, as specified in the OpenID specification. The GS2 header carries the optional authorization identity.

The "gs2-header" is specified in , and it is used as follows. The "gs2-nonstd-flag" MUST NOT be present. The "gs2-cb-flag" MUST be "n" because channel binding is not supported by this mechanism. The "gs2-authzid" carries the optional authorization identity. The XRI syntax is defined in . URI is specified in .

The SASL Server sends an OpenID message that contains an openid.mode of either "checkid_immediate" or "checkid_setup", as specified in Section 9.1 of the OpenID 2.0 specification. The client now sends that request via an HTTP GET to the OP, as if redirected to do so from an HTTP server. The client MUST handle both user authentication to the OP and confirmation or rejection of the authentiation of the RP. After all authentication has been completed by the OP, and after the response has been sent to the client, the client will relay the response to the Relying Party via HTTP or SSL.

The Relying Party now validates the response it received from the client via HTTP or SSL, as specified in the OpenID specification. The response by the Relying Party consists of an application specific response code indicating success or failure of authentication. In the additional data, the server MAY include OpenID Simple Registry (SREG) attributes that are listed in Section 4 of . They are encoded as follows: Strip "openid.sreg." from each attribute name. Treat the concatentation of results as URI parameters that are separated by an ambersand (&) and encode as one would a URI, absent the scheme, authority, and the question mark. For example: email=lear@example.com&fullname=Eliot%20Lear More formally:

If the application protocol allows, openid.error and openid.error_code and any other useful diagnostic information SHOULD be included in authentication failures.

This section and its sub-sections and all normative references of it not referenced elsewhere in this document are INFORMATIONAL for SASL implementors, but they are NORMATIVE for GSS-API implementors. The OpenID SASL mechanism is actually also a GSS-API mechanism. The messages are the same, but a) the GS2 header on the client's first message and channel binding data is excluded when OpenID is used as a GSS-API mechanism, and b) the RFC2743 section 3.1 initial context token header is prefixed to the client's first authentication message (context token). The GSS-API mechanism OID for OpenID is 1.3.6.1.4.1.11591.4.5. OpenID security contexts always have the mutual_state flag (GSS_C_MUTUAL_FLAG) set to TRUE. OpenID does not support credential delegation, therefore OpenID security contexts alway have the deleg_state flag (GSS_C_DELEG_FLAG) set to FALSE. The OpenID mechanism does not support per-message tokens or GSS_Pseudo_random.

OpenID supports standard generic name syntaxes for acceptors such as GSS_C_NT_HOSTBASED_SERVICE (see , Section 4.1). OpenID supports only a single name type for initiators: GSS_C_NT_USER_NAME. GSS_C_NT_USER_NAME is the default name type for OpenID. OpenID name normalization is covered by the OpenID specification, see section 7.2. The query, display, and exported name syntaxes for OpenID principal names are all the same. There are no OpenID-specific name syntaxes -- applications should use generic GSS-API name types such as GSS_C_NT_USER_NAME and GSS_C_NT_HOSTBASED_SERVICE (see , Section 4). The exported name token does, of course, conform to , Section 3.2, but the "NAME" part of the token should be treated as a potential input string to the OpenID name normalization rules. GSS-API name attributes may be defined in the future to hold the normalized OpenID Identifier.

Suppose one has an OpenID of http://openid.example, and wishes to authenticate his IMAP connection to mail.example (where .example is the top level domain specified in ). The user would input his Openid into his mail user agent, when he configures the account. In this case, no association is attempted between the OpenID Consumer and the OP. The client will make use of the return_to attribute to capture results of the authentication to be redirected to the server. The authentication on the wire would then look something like the following:

S: * OK C: C1 CAPABILITY S: * CAPABILITY IMAP4rev1 SASL-IR SORT [...] AUTH=OPENID20 S: C1 OK Capability Completed C: C2 AUTHENTICATE OPENID biwsaHR0cDovL29wZW5pZC5leGFtcGxlLw== [ This is the base64 encoding of "n,,http://openid.example/". Server performs discovery on http://openid.example/ ] S: + aHR0cDovL29wZW5pZC5leGFtcGxlL29wZW5pZC8/b3BlbmlkLm5z PWh0dHA6Ly9zcGVjcy5vcGVuaWQubmV0L2F1dGgvMi4wJm9wZW5p ZC5yZXR1cm5fdG89aHR0cHM6Ly9tYWlsLmV4YW1wbGUvY29uc3Vt ZXIvMWVmODg4YyZvcGVuaWQuY2xhaW1lZF9pZD1odHRwczovL29w ZW5pZC5leGFtcGxlLyZvcGVuaWQuaWRlbnRpdHk9aHR0cHM6Ly9v cGVuaWQuZXhhbXBsZS8mb3BlbmlkLnJlYWxtPWltYXA6Ly9tYWls LmV4YW1wbGUmb3BlbmlkLm1vZGU9Y2hlY2tpZF9zZXR1cA== [ This is the base64 encoding of "http://openid.example/openid/ ?openid.ns=http://specs.openid.net/auth/2.0 &openid.return_to=https://mail.example/consumer/1ef888c &openid.claimed_id=https://openid.example/ &openid.identity=https://openid.example/ &openid.realm=imap://mail.example &openid.mode=checkid_setup" with line breaks and spaces added here for readibility. ] C: [ The client now sends the URL it received to a browser for processing. The user logs into http://openid.example, and agrees to authenticate imap://mail.example. A redirect is passed back to the client browser who then connects to https://imap.example/consumer via SSL with the results. From an IMAP perspective, however, the client sends an empty response, and awaits mail.example. Server mail.example would now contact openid.example with an openid.check_authenticate message. After that... ] S: + ZW1haWw9bGVhckBtYWlsLmV4YW1wbGUsZnVsbG5hbWU9RWxp b3QlMjBMZWFy [ Here the IMAP server has returned an SREG attribute of email=lear@mail.example,fullname=Eliot%20Lear. Line break added in this example for clarity. ] C: [ In IMAP client must send a blank response to receive data that is included in a success response. ] S: C2 OK ]]>

This section will address only security considerations associated with the use of OpenID with SASL applications. For considerations relating to OpenID in general, the reader is referred to the OpenID specification and to other literature. Similarly, for general SASL Security Considerations, the reader is referred to that specification.

As specified in , the server is responsible for binding credentials to a specific authorization identity. It is therefore necessary that either some sort of registration process takes place to register specific OpenIDs, or that only specific trusted OpenID Providers be allowed. Some out of band knowledge may help this process along. For instance, users of a particular domain may utilize a particular OP that enforces a mapping.

In the initial SASL client response a user or host can transmit a malicious response to the RP for purposes of taking advantage of weaknesses in the RP's OpenID implementation. It is possible to add port numbers to the URL so that the outcome is the RP does a port scan of the site. The URL could send the connection to an internal host or even the local host, which the attacker would not normally have access to. The URL could contain a protocol other than http or https, such as file or ftp. To mitigate this attack, implementations should carefully analyze URLs received, eliminating any that would in some way be privileged. A log of those sites that fail SHOULD be kept, and limitations on queries from clients should be imposed, just as with any other authentication attempt.

There is no defined mechanism in the OpenID protocol to bind the OpenID session to the user's browser. An attacker may forge a cross-site request in the log-in form, which has the user logging into a proper RP as the attacker. The user would not recognize they are logged into the site as the attacker, and so may reveal information at the RP. Cross-site request forgery is a widely exploited vulnerability at web sites. This is only concern in the context SASL in as much as the client is not configured with the Relying Party (e.g., SASL server) in a safe manner.

The OP is aware of each RP that a user logs into. There is nothing in the protocol to hide this information from the OP. It is not a requirement to track the visits, but there is nothing that prohibits the collection of information. SASL servers should be aware that OpenID Providers will be track - to some extent - user access to their services and any additional information that OP provides.

It is possible for RPs to link data that they have collected on you. By using the same identifier to log into every RP, collusion between RPs is possible. In OpenID 2.0, directed identity was introduced. Directed identity allows the OP to transform the identifier the user typed in to another identifier. This way the RP would never see the actual user identifier, but a randomly generated identifier. This is an option the user has to understand and decide to use if the OP is supporting it.

The IANA is requested to register the following SASL profile: SASL mechanism profile: OPENID20 Security Considerations: See this document Published Specification: See this document For further information: Contact the authors of this document. Owner/Change controller: the IETF Note: None

The authors would like to thank Alexey Melenkov, Joe Hildebrand, Mark Crispin, Chris Newman, Leif Johansson, and Klaas Wierenga for their review and contributions.

OpenID Foundation

http://www.openid.net

OpenID Foundation

http://www.openid.net

Cordance

drummond.reed@cordance.net

Epok

dave.mcalpin@epokinc.com

&RFC2119; &RFC2743; &RFC4422; &RFC2606; &RFC2616; &RFC3986; &RFC5246; &GS2;

This section to be removed prior to publication. 01 Add nonce discussion, add authorized identity, explain a definition. 00 Initial Revision.