OpenID Authentication 2.0 - Draft 10

Abstract

OpenID Authentication provides a way to prove that an End User owns an Identifier. It does this without the Relying Party needing access to password, email address, or other sensitive information.

OpenID is decentralized. No central authority must approve or register Relying Parties or Identity Providers. An End User can freely choose which Identity Provider to use. They can preserve their Identifier if they switch Identity Providers.

While nothing in the protocol requires JavaScript or modern browsers, the authentication scheme plays nicely with "AJAX"-style setups, so an End User can prove their Identity to a Relying Party without having to leave the page they are on.

OpenID Authentication uses only standard HTTP requests and responses, so does not require any special capabilities of the User-Agent or other software.

The exchange of profile information, and other features not covered in this specification, is addressed through additional Service Types built on top of OpenID.

Table of Contents

1.  Requirements Notation
2.  Terminology
3.  Protocol Overview
4.  Data Formats
    4.1.  Protocol Messages
        4.1.1.  Key-Value Form Encoding
        4.1.2.  HTTP Encoding
        4.1.3.  Example
    4.2.  Integer Representations
5.  Communication Types
    5.1.  Direct Communication
        5.1.1.  Direct Request
        5.1.2.  Direct Response
    5.2.  Indirect Communication
        5.2.1.  HTTP Redirect
        5.2.2.  HTML FORM Redirection
        5.2.3.  Indirect Error Responses
6.  Generating Signatures
7.  Signatures
    7.1.  Signature Algorithms
    7.2.  Procedure
8.  Initiation and Discovery
    8.1.  Initiation
    8.2.  Normalization
    8.3.  Discovery
        8.3.1.  Discovered Information
        8.3.2.  XRDS-Based Discovery
        8.3.3.  HTML-Based Discovery
9.  Establishing Associations
    9.1.  Association Session Request
    9.2.  Association Session Response
        9.2.1.  Common Response Parameters
        9.2.2.  Unencrypted Response Parameters
        9.2.4.  Unsuccessful Response Parameters
    9.3.  Association Types
        9.3.1.  HMAC-SHA1
        9.3.2.  HMAC-SHA256
    9.4.  Association Session Types
10.  Requesting Authentication
    10.1.  Request Parameters
    10.2.  Realms
    10.3.  Immediate Requests
11.  Responding to Authentication Requests
    11.1.  Positive Assertions
    11.2.  Negative Assertions
        11.2.1.  In Response to Immediate Requests
        11.2.2.  In Response to Non-Immediate Requests
12.  Verifying Assertions
    12.1.  Checking the Nonce
    12.2.  Verifying Signatures
    12.3.  Verifying Discovered Information
    12.4.  Identifying the End User
        12.4.1.  HTTP and HTTPS URL Identifiers
13.  OpenID Authentication 1.1 Compatibility
    13.1.  Relying Parties
    13.2.  Identity Providers
14.  Extensions
15.  Discovering OpenID Relying Parties
16.  Security Considerations
    16.1.  Preventing Attacks
        16.1.1.  Eavesdropping Attacks
        16.1.2.  Man-in-the-Middle Attcks
    16.2.  User Agents
    16.3.  User Interface Considerations
Appendix A.  Examples
Appendix A.1.  Delegation
Appendix A.2.  XRDS
Appendix A.3.  HTML Identifier Markup
Appendix A.4.  Login Form
Appendix A.5.  XRI CanonicalID
Appendix B.  Diffie-Hellman Key Exchange Default Value
Appendix C.  Changes from the Previous OpenID Authentication Specification
Appendix C.1.  Updated Initiation and Discovery
Appendix C.2.  Security improvements
Appendix C.3.  Extensions
17.  Normative References
§  Authors' Addresses

1.  Requirements Notation

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 [RFC2119] (Bradner, B., “Key words for use in RFCs to Indicate Requirement Levels,” .).

2.  Terminology

Identifier:

An Identifier is a URL or XRI. An "Identifier" may be a Claimed Identifier, Delegate Identifier, IdP Identifier, or Verified Identifier, depending on context.

End User:

The person who wants to prove ownership of an Identifier.

User-Agent:

The End User's Web browser. See [RFC2616] (Fielding, R., Gettys, J., Mogul, J., Frystyk, H., Masinter, L., Leach, P., and T. Berners-Lee, “Hypertext Transfer Protocol -- HTTP/1.1,” .).

Claimed Identifier:

An Identifier that the End User claims to own that has not yet been verified by the Relying Party.

Verified Identifier:

An Identifier that the End User has proven to a Relying Party that they own.

Delegate Identifier:

An alternate Identifier that can be included in the discovery response.

Relying Party:

RP. A Web application that wants proof that the End User owns an Identifier.

Identity Provider:

IdP. This is the OpenID Authentication server that a Relying Party contacts for cryptographic proof that the End User owns an Identifier.

IdP Identifier:

An Identifier which represents an IdP.

Diffie-Hellman Key Exchange:

Diffie-Hellman Key Exchange [RFC2631] (Rescorla, E., “Diffie-Hellman Key Agreement Method,” .) is a protocol that allows two parties to share a secret, while preventing eavesdroppers from learning the secret.

3.  Protocol Overview

1. The End User initiates authentication (Initiation) by supplying an Identifier to the Relying Party.
2. The Relying Party performs discovery (Discovery) on the Claimed Identifier and establishes the location of the IdP's endpoint that the End User uses for authentication.
3. (optional) The Relying Party and the IdP establish an association (Establishing Associations) -- a shared secret established using Diffie-Hellman Key Exchange. The IdP uses an association to sign subsequent messages and the Relying Party to verify those messages.
4. The Relying Party redirects the End User's User Agent to the IdP with an OpenID Authentication request (Requesting Authentication).
5. The IdP establishes whether the End User is authorized to perform OpenID Authentication and wishes to do so. The manner in which the End User authenticates to their IdP is beyond the scope of this document.
6. The IdP redirects the End User's User Agent back to the Relying Party with an indication either that authentication is approved (Positive Assertions) or authentication failed (Negative Assertions).
7. The Relying Party verifies (Verifying Assertions) the information received from the IdP.

4.  Data Formats

4.1.  Protocol Messages

The OpenID Authentication protocol messages are mappings of plain-text keys to plain-text values. The keys and values are Unicode strings. When the keys and values need to be converted to bytes, they MUST be encoded using UTF-8.

4.1.1.  Key-Value Form Encoding

A Key-Value form message is a sequence of lines. Each line begins with a key, followed by a colon, and the value associated with the key. The line is terminated by a single newline (codepoint 10, "/n"). A key or value MUST NOT contain a newline and a key MUST NOT either contain a colon.

Additional characters, including whitespace, MUST NOT be added before or after the colon or newline. The message MUST be encoded in UTF-8 to produce a byte string.

Key-Value Form encoding is used for signature calculation and for direct responses (Direct Response) to relying parties.

4.1.2.  HTTP Encoding

When a message is sent to an HTTP server, it MUST be encoded using a form encoding specified in section 17.13.4 of the [HTML401] (W3C, “HTML 4.01 Specification,” .). Likewise, if the "Content-Type" header is included in the request headers, its value MUST also be such an encoding.

All of the keys in the request message are prefixed with "openid.". This prefix prevents interference with other parameters that are passed along with the OpenID message. When a message is sent as a POST, the application processing the HTTP request MUST only use the values in the POST body and MUST ignore any "openid." parameters that are present on the request URL.

This model applies to messages from the User Agent to both the Relying Party and the IdP, as well as messages from the Relying Party to the IdP.

4.1.3.  Example

Non-normative

The following examples encode the following information:

Key     | Value
--------+---------------------------
mode    | error
error   | This is an example message

Key-Value Form encoded

mode:error
error:This is an example message

x-www-urlencoded, as in a HTTP POST body or in a URL's query ([RFC3986] (Berners-Lee, T., “Uniform Resource Identifiers (URI): Generic Syntax,” .) section 3).

openid.mode=error&openid.error=This%20is%20an%20example%20message

4.2.  Integer Representations

Arbitrary precision integers MUST be encoded as big-endian signed two's complement binary strings. Henceforth, "btwoc" is a function that takes an arbitrary precision integer and returns its shortest big-endian two's complement representation. All integers that are used with Diffie-Hellman are positive. This means that the left-most bit of the two's complement representation MUST be zero. If it is not, implementations MUST add a zero byte at the front of the string.

Non-normative example:

Base 10 number | btwoc string representation
---------------+----------------------------
0              | "\x00"
127            | "\x7F"
128            | "\x00\x80"
255            | "\x00\xFF"
32768          | "\x00\x80\x00"

5.  Communication Types

5.1.  Direct Communication

Direct communication is initiated by a Relying Party to an IdP endpoint URL. It is used for establishing associations (Establishing Associations) and verifying authentication assertions (Verifying Directly with the Identity Provider).

5.1.1.  Direct Request

The message MUST be encoded as a POST body, as specified by Section 4.1.2 (HTTP Encoding).

5.1.2.  Direct Response

The body of a response to a Direct Request (Direct Request) consists of an HTTP Response body in Key-Value Form (Key-Value Form Encoding). The content-type of the response SHOULD be "text/plain".

5.1.2.1.  Successful Responses

A server receiving a properly formed request MUST send a response with an HTTP status code of 200.

5.1.2.2.  Error Responses

If a request is malformed or contains invalid arguments, the server MUST send a response with a status code of 400. The response body MUST be a Section 4.1.1 (Key-Value Form Encoding) message with the following fields:

• error

  Value: Unstructured text error message.

• contact

  Value: (optional) Contact address for the administrator of the IdP. The contact address may take any form, as it is intended to be displayed to a person.

• reference

  Value: (optional) A reference identifier, such as a support ticket number or a URL to a news blog, etc.

The IdP MAY add additional keys to this response.

The particular wire format of these messages will depend on whether they are responses to direct or indirect requests.

5.2.  Indirect Communication

In indirect communication, messages are passed through the User-Agent. This can be initiated by either the Relying Party or the IdP. Indirect communication is used for authentication requests (Requesting Authentication) and authentication responses (Responding to Authentication Requests).

There are two methods for indirect communication: HTTP redirects and HTML form submission. Both form submission and redirection require that the sender know a recipient URL and that the recipient URL expect indirect messages, as specified in Section 4.1.2 (HTTP Encoding). The initiator of the communication chooses which method of indirect communication is appropriate.

5.2.1.  HTTP Redirect

Data can be transferred by issuing a 302, 303, or 307 HTTP Redirect to the End User's User-Agent. The redirect URL is the URL of the receiver with the OpenID Authentication message appended to the query string, as specified in Section 4.1.2 (HTTP Encoding).

This method is deprecated as of OpenID Authentication version 2.0 though is still required for implementation to aide in backwards compatibility.

5.2.2.  HTML FORM Redirection

A mapping of keys to values can be transferred by returning an HTML page to the User-Agent that contains an HTML form element. Form submission MAY be automated using JavaScript.

The <form> element's "action" attribute value MUST be the URL of the receiving Web site. Each Key-Value pair MUST be included in the form as an <input> element. The key MUST be encoded as the "name" attribute and the value as the "value" attribute, such that the User Agent will generate a message as specified in Section 4.1.2 (HTTP Encoding) when the form is submitted. The form MUST include a submit button.

5.2.3.  Indirect Error Responses

If a request is malformed or contains invalid arguments and there is an "openid.return_to" argument whose value is a valid URL, the IdP MUST redirect the User Agent to that URL with a message of the following form:

• openid.mode

  Value: "error"

• openid.error

  Value: Unstructured text error message.

• openid.contact

  Value: (optional) Contact address for the administrator of the IdP.

• openid.reference

  Value: (optional) A reference identifier, such as a support ticket number or a URL to a news blog, etc.

The IdP MAY add additional keys to this response.

If the "openid.return_to" value is missing or not a valid URL, the IdP SHOULD return a response to the End User indicating the error and that it is unable to return the End User to the Relying Party.

6.  Generating Signatures

The most common usage of an association is as a Message Authentication Code (MAC) key used to sign OpenID Authentication messages.

When generating MAC keys, the recommendations in [RFC1750] (Eastlake, D., Crocker, S., and J. Schiller, “Randomness Recommendations for Security,” .) SHOULD be followed.

6.1.  Signed List Algorithm

The input to the Signed List Algorithm are the message to be signed, and the list of message keys that are to be signed with the "openid." prefix removed.

To compute the list of key/value pairs to be signed:

1. Iterate through the list of keys to be signed in the order they appear in the input to the algorithm. For each key, find the value in the message whose key is equal to the signed list key prefixed with "openid."
2. Append the signed list key and the associated value to the message to be signed.

The output of this algorithm is the message to be signed, and the list of keys that were signed. A message signed using this algorithm MUST append the list of signed fields to message.

As the algorithm strips the "openid." prefix from message keys while looking for a match, it MUST only sign elements that have keys beginning with "openid."

7.  Signatures

7.1.  Signature Algorithms

OpenID Authentication supports two signature algorithms: HMAC-SHA1([RFC2104] (Krawczyk, H., Bellare, M., and R. Canetti, “HMAC: Keyed-Hashing for Message Authentication,” .), [RFC3174] (Eastlake, D. and P. Jones, “US Secure Hash Algorithm 1 (SHA1),” .)), and HMAC-SHA256 ([RFC2104] (Krawczyk, H., Bellare, M., and R. Canetti, “HMAC: Keyed-Hashing for Message Authentication,” .), [FIPS180‑2] (U.S. Department of Commerce and National Institute of Standards and Technology, “Secure Hash Signature Standard,” .)).

Keys for associations that use HMAC-SHA1 signatures are 160 bits. Keys for associations that use HMAC-SHA256 are 256 bits. The use of HMAC-SHA256 is RECOMENDED though at the time of writing this document, library support seems lacking.

7.2.  Procedure

To generate a message signature:

1. Determine the appropriate signature list and signature algorithm from the association type (Establishing Associations).
2. Generate the list to be signed using the correct list algorithm.
3. Convert the list to an octet string by encoding with Key Value Form (Key-Value Form Encoding)
4. Apply the correct signature algorithm to the octet string.

8.  Initiation and Discovery

8.1.  Initiation

To initiate OpenID Authentication, the Relying Party SHOULD present the End User with a form that has a field for entering an Identifier.

It is RECOMMENDED that every Relying Party place the OpenID logo at the beginning of the form field where the End User enters their Identifier. This aides in End User recognition that they can use their OpenID Identifier at the specific Relying Party.

The form field's "name" attribute SHOULD have the value "openid_identifier" as to allow User Agents to automatically prefill the End User's Identifier when visiting a Relying Party.

8.2.  Normalization

The End User's input MUST be normalized into an Identifier. If the End User supplies input that does not include a scheme (http, https, or xri), then the application needs to determine if the input is an XRI or a URL missing the "http://". To do so, the application SHOULD examine the first character of the input. If it is an XRI Global Context Symbol (=, @, +, $, or ! see Section 2.2.1.2 of (Reed, D. and D. McAlpin, “Extensible Resource Identifier (XRI) Syntax V2.0,” .) [XRI Syntax 2.0]), then the input SHOULD be treated as an XRI. If it is not, then the input SHOULD be treated as an http URL, and prefixed with the string "http://".

URL identifiers MUST be further normalized by applying the rules in section 6 of RFC 3986, following redirects when retrieving their content, and finally applying the rules in section 6 of RFC 3986 (Normalization and Comparison) to the final destination URL. This final URL should be noted by the Relying Party as the Claimed Identifier and used during authentication requests.

8.3.  Discovery

Discovery is the process where the Relying Party uses the Identifier to look up ("discover") the necessary information for initiating authentication. OpenID Authentication has three paths through which to do discovery:

1. If the identifier is an XRI, XRI resolution will yield an XRDS document that contains the necessary information.
2. If it is a URL, the Yadis protocol (Miller, J., “Yadis Specification 1.0,” .) [Yadis] is first attempted. If it succeeds, the result is again an XRDS document.
3. If the Yadis protocol fails, the URL is retrieved and HTML-based discovery is attempted.

8.3.1.  Discovered Information

Upon successful completion of discovery, the Relying Party will have the following information:

IdP Endpoint URL:

The URL that accepts authentication requests. This MUST be an absolute URL.

Claimed Identifier:

(optional) The normalized Identifier upon which discovery was performed. The Claimed Identifier is present unless the End User entered an IdP Identifier during initiation (Initiation).

Delegate Identifier:

(optional) The identifier that the Relying Party SHOULD perform authentication using. Upon successful authentication, the Relying Party SHOULD recognize the End User using the Claimed Identifier. The Delegate Identifier can only be present when the End User enters a Claimed Identifier.

8.3.2.  XRDS-Based Discovery

If XRI or Yadis discovery was used, the result will be an XRDS Document. This is a XML document with entries for services that are related to the identifier.

8.3.2.1.  Service Elements

8.3.2.1.1.  IdP Identifier Element

An IdP Identifier Element is a <xrd:Service> element with the following information:

An <xrd:Type> tag whose text content is "http://openid.net/server/2.0"

An <xrd:URI> tag whose text content is the IdP Endpoint URL

8.3.2.1.2.  Claimed Identifier Element

A Claimed Identifier Element is an <xrd:Service> element with the following information:

An <xrd:Type> tag whose text content is "http://openid.net/signon/2.0".

An <xrd:URI> tag whose text content is the IdP Endpoint URL.

An <openid:Delegate> tag (optional) whose text content is The Delegate Identifier.

8.3.2.2.  Extracting Authentication Data

Once the Relying Party has obtained an XRDS document, it MUST first search the document (following the rules described in the XRDS specification) for an IdP Identifier, and if none is found, search for a Claimed Identifier Element.

8.3.2.3.  XRI and the CanonicalID Element

When the identifier is an XRI, the <xrd:XRD> element that contains the OpenID <xrd:Service> element will also contain a <CanonicalID> element. The content of this element MUST be preserved for use after a successful authentication request. See Section 12.4 (Identifying the End User).

The Relying Party MUST confirm that the provider of the XRD that contains the <CanonicalID> element is authoritative for that canonical ID. The provider is identified by the contents of the <ProviderID> element that is a child of the <XRD> element. If the provider is not authoritative for the canonical ID, the Relying Party MUST resolve the canonical ID to confirm the OpenID Service Endpoint information that was discovered. The information discovered when resolving the canonical ID MUST match the information discovered when resolving the user-supplied identifier.

When using XRI resolution, the canonical ID MUST be used as the Claimed Identifier. For an XRI to be a valid identifier, both the <ProviderID> and <CanonicalID> MUST be present in the discovery result.

When using URL-based identifiers, the CanonicalID element SHOULD be ignored.

8.3.2.4.  Additional Information

The "openid" namespace is "http://openid.net/signon/2.0". The "xrd" namespace is "xri://$xrd*($v*2.0)".

For compatibility with deployed code, it is RECOMMENDED that a Relying Party also accept "http://openid.net/signon/1.0" or "http://openid.net/signon/1.1" for the value of <xrd:Type>. When one of these values is used, the Relying Party MUST use OpenID 1.1 compatibility (OpenID Authentication 1.1 Compatibility).

If an OpenID IdP supports extensions (Section 14 (Extensions)), the extensions SHOULD be listed as additional <xrd:Type> child elements of the <xrd:Service> element.

8.3.3.  HTML-Based Discovery

OpenID Authentication 1.1 HTML-based discovery MUST be supported by Relying Parties. If a Relying Party locates an IdP using HTML-based discovery, it MUST use OpenID Authentication 1.1 Compatibility (OpenID Authentication 1.1 Compatibility) when communicating with that IdP.

To use HTML-based discovery, an HTML document MUST be available at the URL of the Claimed Identifier. In the HEAD section of the document:

A <LINK> tag MUST be included with attributes "rel" set to "openid.server", and "href" set to an IdP Endpoint URL

A <LINK> tag MAY be included with attributes "rel" set to "openid.delegate" and "href" set to the End User's Delegate Identifier

The host of the HTML document MAY be different from the End User's IdP's host.

The "openid.server" and "openid.delegate" URLs MUST NOT include entities other than &amp;, &lt;, &gt;, and &quot;. Other characters that would not be valid in the HTML document or that cannot be represented in the document's character encoding MUST be escaped using the %xx mechanism as described in [RFC3986] (Berners-Lee, T., “Uniform Resource Identifiers (URI): Generic Syntax,” .).

9.  Establishing Associations

An "association" is a shared secret between the IdP and Relying Party. Once established, it is used to verify subsequent protocol messages.

It is RECOMMENDED that a Relying Party form associations if it is possible for it to do so. If a Relying Party is incapable of creating or storing associations, Section 12.2 (Verifying Signatures) provides an alternate verification mechanism.

9.1.  Association Session Request

An association session is initiated by a direct request (Direct Communication) from a Relying Party to an IdP Endpoint URL with "openid.mode" of "associate".

9.1.1.  Common Request Parameters

These parameters are common to all association requests:

• openid.mode

  Value: "associate"

• openid.assoc_type

  The preferred association type. The association type defines the algorithm to be used to sign subsequent messages.

  Value: A valid association type from Section 9.3 (Association Types)

  Default: "HMAC-SHA1".

• openid.session_type

  The preferred association session type. This defines the method used to encrypt the association's MAC key in transit.

  Value: A valid association session type from Section 9.4 (Association Session Types)

  Default: "no-encryption"

9.1.2.  Diffie-Hellman Request Parameters

The following parameters are common to requests whose requested session type is DH-SHA1 or DH-SHA256:

• openid.dh_modulus

  Value: base64(btwoc(p))

  Default: See Appendix B (Diffie-Hellman Key Exchange Default Value)

• openid.dh_gen

  Value: base64(btwoc(g))

  Default: g = 2

• openid.dh_consumer_public

  Value: base64(btwoc(g ^ xa mod p))

See Section 9.4.2 (Diffie-Hellman Association Sessions) for more information on these parameters.

NOTE: the 'btwoc' function is defined in Section 4.2 (Integer Representations).

9.2.  Association Session Response

An association session response is a direct response from the IdP to the Relying Party in Key-Value Form (Key-Value Form Encoding).

9.2.1.  Common Response Parameters

• session_type

  The session type for this session. Valid values are listed in Section 9.3 (Association Types), but the server MUST respond with the same value as the request's "openid.session_type" parameter. If the server is unable or unwilling to support either of these session types, it MUST return an unsuccessful response.

• assoc_handle

  The association handle is used as a key to refer to this association in subsequent messages.

  Value: A string 255 characters or less, consisting only of ASCII characters in the range 33-126 inclusive (printable non-whitespace characters).

• assoc_type

  The value of the "openid.assoc_type" parameter from the request. If the server is unwilling or unable to support this association type, it MUST return an unsuccessful response.

• expires_in

  The lifetime, in seconds, of this association. The Relying Party MUST NOT use the association after this time has expired.

  Value: an integer, represented in base 10 ASCII.

9.2.2.  Unencrypted Response Parameters

• mac_key

  The MAC key (shared secret) for this association, Base 64 (Josefsson, S., “The Base16, Base32, and Base64 Data Encodings,” .) [RFC3548] encoded.

9.2.3.  Diffie-Hellman Response Parameters

• dh_server_public

  Value: base64(btwoc(g ^ xb mod p))

  Description: The Provider's Diffie-Hellman public key.

• enc_mac_key

  Value: base64(H(btwoc(g ^ (xa * xb) mod p)) XOR MAC key)

  Description: The MAC key (shared secret), encrypted with the secret Diffie-Hellman value. H is either SHA1 or SHA256 depending on the session type.

NOTE: the 'btwoc' function is defined in Section 4.2 (Integer Representations)

9.2.4.  Unsuccessful Response Parameters

If the IdP does not support an association session type or association type, it MUST respond with a message indicating that the association session failed. If there is another association session type or association type that is supported, the IdP MAY include that information in the response.

• error

  Value: (optional) A human-readable message indicating why the association session failed

• error_code

  Value: "unsupported-type"

• session_type

  Value: (optional) A method of exchanging association information that the IdP supports.

• assoc_type

  Value: (optional) An association type supported by the IdP

Upon receipt of an "unsupported-type" response, the Relying Party MAY make another request with the specified association session type and association type. If no association is established, the Relying Party MAY continue the authentication process in stateless mode.

9.3.  Association Types

9.3.1.  HMAC-SHA1

An association of type "HMAC-SHA1" uses the HMAC-SHA1 (Signature Algorithms) signature algorithm, in combination with the Signed List (Signed List Algorithm) algorithm.

9.3.2.  HMAC-SHA256

An association of type "HMAC-256" uses the HMAC-SHA256 (Signature Algorithms) signature algorithm, in combination with the Signed List (Signed List Algorithm) algorithm.

9.4.  Association Session Types

OpenID Authentication defines three valid association session types: "no-encryption", "DH-SHA1", and "DH-SHA256".

9.4.1.  No-Encryption Association Sessions

In a "no-encryption" association session, the IdP sends the association MAC key in plain-text to the Relying Party. This makes it possible for an eavesdropper to intercept the key, and forge messages to this Relying Party. Therefore, no-encryption association sessions SHOULD NOT be used unless the messages are using transport-level encryption.

The MAC key sent by the IdP MUST be the length specified for this association in Section 7.1 (Signature Algorithms).

9.4.2.  Diffie-Hellman Association Sessions

The "DH-SHA1" and DH-SHA256" association types use Diffie-Hellman key exchange to securely transmit the shared secret.

The MAC key MUST be the same length as the output of H, the hash function - 160 bits (20 bytes) for DH-SHA1 or 256 bits (32 bytes) for DH-SHA256, as well as the output of the signature algorithm of this association.

The Relying Party specifies a modulus, p, and a generator, g. The Relying Party chooses a random private key xa and Identity Provider chooses a random private key xb, both in the range [1 .. p-1]. The shared secret used to encrypt the MAC key is thus g ^ (xa * xb) mod p = (g ^ xa) ^ xb mod p = (g ^ xb) ^ xa mod p. For more information, see [RFC2631] (Rescorla, E., “Diffie-Hellman Key Agreement Method,” .). For information on the selection of random values, see [RFC1750] (Eastlake, D., Crocker, S., and J. Schiller, “Randomness Recommendations for Security,” .).

10.  Requesting Authentication

Once the Relying Party has successfully performed discovery and (optionally) created an association with the discovered IdP Endpoint URL, it can send an authentication request to the IdP to obtain an assertion. An authentication request is an indirect request (Indirect Communication).

10.1.  Request Parameters

• openid.ns

  Value: "http://openid.net/signon/2.0"

  This value MUST be present for the request to be a valid OpenID Authentication 2.0 request.

  Note: If an IdP receives an authentication request with this parameter missing or with a lower version number, it SHOULD still respond to the request. If it does respond it MUST use OpenID Authentication 1.1 Compatibility (OpenID Authentication 1.1 Compatibility) when communicating with that Relying Party.

• openid.mode

  Value: "checkid_immediate" or "checkid_setup"

  Note: If the Relying Party wishes the End User to be able to interact with the IdP, "checkid_setup" should be used. An example of a situation where interaction between the End User and the IdP is not desired is when the authentication request is happening asynchronously in JavaScript.

• openid.identity

  Value: (optional) Delegate Identifier when available, otherwise the Claimed Identifier

  Note: If this is set to the special value "http://openid.net/identifier_select/2.0", the IdP MAY choose an identifier that belongs to the End User.

• openid.assoc_handle

  Value: (optional) A handle for an association between the Relying Party and the IdP that should be used to sign the response.

  Note: If no association handle is sent, the transaction will take place in stateless mode.

• openid.return_to

  Value: URL to which the IdP SHOULD return the User Agent with additional responses indicating the status of the request.

• openid.realm

  Value: (optional) URL pattern the Provider SHALL ask the End User to trust. See Section 10.2 (Realms).

  Default: return_to URL

10.2.  Realms

A "realm" is a pattern that represents the part of URL-space for which an OpenID Authentication request is valid. A realm SHOULD give the End User an indication of the scope of the authentication request. IdPs SHOULD present the realm when requesting the End User's approval for an authentication request. IdPs MAY use the realm to allow the End User to automate approval of authentication requests.

A realm pattern is a URL, with the following changes:

• A realm MUST NOT contain a URI fragment
• A realm MAY contain a wild-card at the beginning of the URL authority section. A wild-card consists of the characters "*." prepended to the DNS name in the authority section of the URL.

A URL matches a realm if:

• The URL scheme (RFC 3986, section 3.1 (Berners-Lee, T., “Uniform Resource Identifiers (URI): Generic Syntax,” .) [RFC3986]) and port of the URL are identical to those in the realm.
• The URL's path is equal to or a sub-directory of the realm's path.
• Either:
  1. The realm's domain contains the wild-card characters "*.", and the trailing part of the URL's domain is identical to the part of the realm following the "*." wildcard, or
  2. The URL's domain is identical to the realm's domain

The "openid.return_to" URL MUST match the "openid.realm", or the IdP MUST return an error.

It is RECOMMENDED that IdP's protect their End Users from requests with overly-general realms, like http://*.com/ or http://*.co.uk/. Determining if a realm is overly-general is at the discretion of the IdP.

10.3.  Immediate Requests

When requesting authentication, the Relying Party MAY request that the IdP not interact with the End User. In this case the IdP MUST respond immediately with either an assertion that authentication is successful, or a response indicating that the request cannot be completed without further user interaction. This is accomplished by an authentication request with "openid.mode" set to "checkid_immediate".

11.  Responding to Authentication Requests

When an authentication request comes from the User-Agent, the IdP SHOULD identify the User-Agent, and determine whether the End User wishes to complete the authentication. If the End User can be identified and wishes to complete the authentication, the IdP should send a positive assertion (Positive Assertions) to the Relying Party.

Methods of identifying the end user and getting approval to finish authentication are beyond the scope of this specification.

If no Identifier was specified in the request and there are Identifiers in the control of the End User, the IdP SHOULD allow the End User to choose which Identifier to use. If an Identifier was specified, the IdP SHOULD only issue assertions about the specified Identifier.

If the Relying Party supplied an association handle with the authentication request, the IdP SHOULD attempt to look up an association based on that handle. If the association is missing or expired, the IdP SHOULD send the "openid.invalidate_handle" parameter of the response to the requests "openid.assoc_handle", and SHOULD proceed as if no association handle was specified.

If no association handle is specified, the IdP SHOULD create a private association for signing the response. The IdP MUST store this association and MUST respond to later requests to check the signature of the response.

11.1.  Positive Assertions

Positive assertions are indirect responses (Indirect Communication) with the following fields:

• openid.ns

  Value: "http://openid.net/signon/2.0"

  Note: This defines the interpretation of the openid arguments without a namespace. To be an OpenID Authentication 2.0 response, the given value must be present.

• openid.mode

  Value: "id_res"

• openid.identity

  Value: (optional) The Identifier about which the IdP is making a positive authentication assertion.

  Note: The Identifier MAY be omitted if an extension is in use that makes the response meaningful without it.

• openid.return_to

  Value: Verbatim copy of the return_to URL parameter sent in the request.

  Note: Because the "openid.return_to" URL is signed by the IdP, a Relying Party can make sure outside parties haven't sent responses with query parameters that were not included in the "openid.return_to" URL.

• openid.response_nonce

  Value: A string that MUST be unique to this particular successful authentication response. The nonce MUST start with the current time on the server, and MAY have additional characters appended to the end as necessary to make each response unique. The date and time MUST be formatted as specified in section 5.6 of [RFC3339] (Newman, C. and G. Klyne, “Date and Time on the Internet: Timestamps,” .), with the following restrictions:

  • All times must be in the UTC timezone, indicated with a "Z".
  • No fractional seconds are allowed

  For example: 2005-05-15T17:11:51ZUNIQUE

• openid.invalidate_handle

  Value: (optional) If the Relying Party sent an invalid association handle with the request, it SHOULD be included here. If it is present, this field MUST be signed.

• openid.assoc_handle

  Value: The handle for the association that was used to sign this assertion.

• openid.signed

  Value: (optional) Comma-separated list of signed fields.

  Note: If the association used to sign this response used the Signed List (Signed List Algorithm) algorithm, this MUST be present. If it is present, this entry consists of the fields without the "openid." prefix that the signature covers. This list MUST contain at least "return_to" and "response_nonce". Additionally, if the IdP is making an assertion about an Identifier, the "identity" field MUST be present. For example, "identity,return_to,response_nonce".

• openid.sig

  Value: Base 64 encoded signature calculated as specified in Section 6 (Generating Signatures).

  Successful authentication messages from the Identity Provider to the Relying Party MUST be signed.

11.2.  Negative Assertions

If the IdP is unable to identify the End User or the End User does not or cannot approve the authentication request, the IdP SHOULD send a negative assertion to the Relying Party as an indirect response (Indirect Communication).

11.2.1.  In Response to Immediate Requests

If the request was an immediate request, there is no chance for the End User to interact with pages on the IdP to provide identifying credentials or approval of a request. A negative assertion of an immediate request takes the following form:

• openid.mode

  Value: "id_res"

• openid.user_setup_url

  Value: A URL that the End User may visit to complete the request. The Relying Party may redirect the End User to this URL, or provide the End User with a link that points to this URL. The request is no longer immediate.

11.2.2.  In Response to Non-Immediate Requests

Since the IdP may display pages to the End User and request credentials from the End User, a negative response to a request that is not immediate is definitive. It takes the following form:

• openid.mode

  Value: "cancel"

In a lot of cases, the Relying Party won't get a cancel mode response; the End User will just quit or press back within their User-Agent. But if it is returned, the Relying Party SHOULD return to what it was doing.

12.  Verifying Assertions

When the Relying Party receives a positive assertion, it MUST verify the following before accepting the assertion:

• An assertion has not yet been accepted from this IdP with the same value for "openid.response_nonce"
• The signature on the assertion is valid
• Discovered information from the Identifier matches the information in the assertion.

12.1.  Checking the Nonce

To prevent replay attacks, the agent checking the signature SHOULD keep track of the nonce values included in positive assertions and never accept the same value more than once for the same IdP Endpoint URL. When using "check_authentication", the IdP is responsible for preventing replay attacks. When the Relying Party checks the signature on an assertion, it is responsible for preventing replay attacks.

The time-stamp may be used to reject responses that are too far away from the current time, limiting the amount of time that nonces must be stored to prevent replays. The acceptable range is implementation dependent. A larger range requires storing more nonces for a longer time. A shorter range increases the chance that clock-skew and transaction time will cause a spurious rejection.

12.2.  Verifying Signatures

If the Relying Party has stored an association with the association handle specified in the assertion, it MUST check the signature on the assertion itself. If it does not have an association stored, it MUST request that the IdP verify the signature (Verifying Directly with the Identity Provider).

12.2.1.  Verifying with an Association

The Relying Party follows the same procedure that the IdP followed in generating the signature (Generating Signatures), and then compares the signature in the response to the signature it generates. If the signatures do not match, the assertion is invalid.

If an authentication request included an association handle for an association between the IdP and the Relying party, and the IdP no longer wishes to use that handle (because it has expired or the secret has been compromised, for instance), the IdP will send a response that must be verified directly with the IdP, as specified in Section 12.2.2 (Verifying Directly with the Identity Provider). In that instance, the IdP will include the field "openid.invalidate_handle" set to the association handle that the Relying Party included with the original request.

12.2.2.  Verifying Directly with the Identity Provider

To verify a signature directly with the IdP, the Relying Party sends a direct request to the IdP.

12.2.2.1.  Request Parameters

• openid.mode

  Value: "check_authentication"

• Exact copies of all fields from the authentication response, except for "openid.mode".

12.2.2.2.  Response Parameters

• mode

  Value: "id_res"

• is_valid

  Value: "true" or "false"

  Description: Boolean; whether the signature is valid.

• invalidate_handle

  Value: (optional) An association handle

  Description: The association handle sent in the request, if the server confirms that it is invalid.

An IdP MUST NOT verify signatures for associations that have shared MAC keys. If an IdP did verify signatures for associations with shared MAC keys, it would be possible for parties other than the IdP to create valid assertions that seemed to come from the IdP.

The IdP SHOULD only return is_valid once for each authentication request. An authentication request may be identified by its "openid.response_nonce" value.

If the IdP verifies responds with "is_valid" set to "true", and "invalidate_handle" is present, the Relying Party SHOULD NOT send further authentication requests with that handle. "invalidate_handle" will only be present when the original authentication request to the IdP included an association that the IdP deemed invalid. This implies that it will only be present in this response if it was also present in the "id_res" response (Positive Assertions). Including "invalidate_handle" in the direct verification is necessary to prevent an attacker from invalidating an association at will by adding it to an authentication response.

12.3.  Verifying Discovered Information

The Relying Party MUST have performed discovery (Discovery) on the Identifier that will be used and the information in the assertion MUST match the discovered information.

Specifically, one of the following MUST be true:

1. The Identifier in the assertion points to the IdP making the assertion.
2. The Claimed Identifier specifies the identifier in the assertion as a delegate, and the Delegate Identifier points to the IdP making the assertion.

Note: The Relying Party must be able to look up the Claimed Identifier in order to perform the second test. Relying Parties unable to store information might append the Claimed Identifier to the "openid.return_to" parameter in the request.

If the Claimed Identifier was not present in the request ("openid.identity" was "http://openid.net/identifier_select/2.0"), the Relying Party MUST perform discovery on the identifier in the response to make sure that the IdP is authorized to make assertions about the Identifier.

12.4.  Identifying the End User

A successful authentication response provides the Relying Party with a Verified Identifier, which MAY be used as a user-visible Identifier. Identifiers in OpenID MUST be URLs or XRIs. If the Identifier is a URL, its scheme MUST be "http" or "https". Except in the case that the Verified Identifier is an XRI, the Relying Party SHOULD use the Verified Identifier as a key for local storage of information about the End User. If the Verified Identifier is an XRI, the discovered CanonicalID field from the XRD SHOULD be used as a key for local storage of information about the End User.

If a request is using delegation, the Verified Identifier is the Identifier on which discovery was performed, and not the Identifier that is contained in the assertion. If an assertion is made for an Identifier on which discovery has not been performed, the Relying Party MUST perform discovery on that Identifier and compare the discovered information to that in the assertion.

12.4.1.  HTTP and HTTPS URL Identifiers

Relying Parties MUST differentiate between URL Identifiers that have different schemes. When user input is processed into a URL, it is processed into a HTTP URL. If the same End User controls the same URL, differing only by scheme, and it is desired that the Identifier be the HTTPS URL, it is RECOMMENDED that a redirect be issued from the HTTP URL to the HTTPS URL. Because the HTTP and HTTPS URLs are not equivalent and the Identifier that is used is the URL after following redirects, there is no reduction in security when using this scheme. If an attacker could gain control of the HTTP URL, it would have no effect on the HTTPS URL, since the HTTP URL is not ever used as an Identifier.

13.  OpenID Authentication 1.1 Compatibility

OpenID Authentication 2.0 attempts to retain maximum compatibility with earlier versions of the OpenID Authentication specification, but this is not universally possible. This section lists the behavioral changes required of an OpenID Authentication 2.0 IdP or Relying Party when communicating with an earlier-protocol peer.

OpenID Authentication 2.0 implementations SHOULD however support OpenID Authentication 1.1 compatibility, unless security considerations make it undesirable.

All messages in OpenID Authentication 1.1 omit the "openid.ns" parameter. OpenID Authentication 1.1 only supports HMAC-SHA1 associations.

13.1.  Relying Parties

• Relying Parties MUST implement HTML-Based Discovery (HTML-Based Discovery).
• In authentication requests (Requesting Authentication), the "openid.identity" parameter MUST NOT be the special value "http://openid.net/identifier_select/2.0", because OpenID 1.1 does not support IdP-driven identifier selection.
• The "openid.realm" parameter in authentication requests was known as "trust_root". The syntax and meaning were identical.
• The Relying Party MUST accept an authentication response (Positive Assertions) that is missing the 'nonce' parameter. It SHOULD however implement an out-of-band method for preventing replay attacks.

13.2.  Identity Providers

• "openid.identity" MUST be sent in a positive authentication assertion (Positive Assertions).
• Unsuccessful association responses (Unsuccessful Response Parameters) MUST NOT be sent, since they are not part of the OpenID Authentication 1.1 protocol.
• IdPs MAY choose to return a successful "no-encryption" response to any association request.
• Omit the session_type parameter when making "no-encryption" responses to "associate" requests.
• The "openid.realm" parameter in authentication requests was known as "trust_root". The syntax and meaning were identical.

14.  Extensions

An Extension to OpenID Authentication is a protocol that rides on top of the authentication request and response. Extensions are useful for providing extra information about an authentication request or response or providing extra information about the subject of the authentication response.

OpenID extensions are identified by a URI. The URI MAY be used as the value of an <xrd:Type> element of an OpenID <xrd:Service> element in an XRDS document associated with a Claimed Identifier. It is also used to associate key-value pairs in messages with the extension.

To associate keys and values in a message with an extension, the key MUST be associated with the Type URI. To associate keys with a Type URI, establish an alias by adding a key prefixed with "openid.ns." and ending with the alias text whose value is the Type URI. Once an alias has been established, all pairs in the message whose keys start with "openid." followed by the alias text, followed by a period or the end of the key are associated with that extension.

A namespace alias MUST NOT contain a period, MUST NOT be the name of a field in a message defined in this specification, and MUST NOT be the same as another namespace alias in the same message. A namespace MUST NOT be assigned more than one alias in the same message. If a message is a response to another message, the response MAY use a different alias to refer to the same namespace.

Non-normative example: An extension's type URI is <http://example.com/ext/1.0>.

openid.ns.x=http://example.com/ext/1.0

openid.x=example

openid.x.foo=bar

openid.xx=notx

In this example, the keys openid.x and openid.x.foo are associated with the extension. The openid.xx key is not.

15.  Discovering OpenID Relying Parties

Relying Parties are RECOMMENDED to use the Yadis protocol to publish their return_to URL. This allows for automated discovery of OpenID Relying Parties.

The Relying Party's XRDS document's <xrd:Service> entry should have the return_to URL as the content of the <xrd:URI> tag and should have http://openid.net/return_to/2.0 as the content of the <xrd:Type> tag.

For example:

<Service xmlns="xri://$xrd*($v*2.0)">
  <Type>http://openid.net/return_to/2.0</Type>
  <URI>http://consumer.example.com/return</URI>
</Service>

16.  Security Considerations

16.1.  Preventing Attacks

16.1.1.  Eavesdropping Attacks

There are two places in this protocol that are vulnerable to eavesdropping attacks. An eavesdropper could intercept an unencrypted association session and recover the shared secret, allowing an attacker to masquerade as the IdP to that relying party. An eavesdropper could also intercept a successful authentication assertion and re-use it, if the nonce is not checked.

Both of these attacks can be prevented by using SSL for these connections. The association session can also use Diffie-Hellman Key Exchange instead of "no-encryption" to protect from eavesdropping. If the nonce is checked in message verification, the positive authentication assertion cannot be re-used.

16.1.2.  Man-in-the-Middle Attcks

Associations prevent tampering of signed fields by a man in the middle, except during discovery, association sessions and stateless mode. Altering signed fields without the shared secret requires breaking the MAC. Currently, no tractable attack is known on the MACs used in this protocol. The quality of the protection provided by the MAC depends on the randomness of the shared MAC key, so it is important that an unguessable value be used.

If DNS resolution or the transport layer is compromised, signatures on messages are not adequate, since the attacker can impersonate the IdP and issue its own associations, or its own decisions in stateless mode. If an attacker can tamper with the discovery process, he can specify any IdP, and so does not have to impersonate the IdP.

Using SSL with certificates signed by a trusted authority prevents these kinds of attacks by verifying the results of the DNS look-up against the certificate. Once the validity of the certificate has been established, tampering is not possible. Impersonating an SSL server requires forging or stealing a certificate, which is significantly harder than the network attacks.

In order to get protection from SSL, SSL must be used for all parts of the interaction, including interaction with the End User through the User Agent.

16.2.  User Agents

Since this protocol is intended to be used interactively, User Agents will primarily be common Web browsers. Web browsers or their hosts may be infected with spyware or other malware, which limits the strength of the authentication assertion, since untrusted software makes it impossible to know whether the authentication decision has been made with the End User's approval. With this said, many web applications and protocols today rely on the security of the Web browser and their hosts.

Cross-site-scripting attacks against IdPs may be used to the same effect. For the best security, IdPs should not depend on scripting so that User Agents without scripting enabled can make authentication decisions.

16.3.  User Interface Considerations

The Relying Party SHOULD redirect the End User to the IdP Endpoint URL in a top-level browser window with all controls visible. This allows better protection for the End User against IdP look-alike sites (phishing).

Appendix A.  Examples

Non-normative

Appendix A.1.  Delegation

For example, an End User wants to use http://www.example.com/ as their Identifier, but http://www.example.com/ doesn't have the means, or desire, to run an IdP. LiveJournal is an Identity Provider, so if the End User has a LiveJournal OpenID Identifier, they can delegate their authentication to LiveJournal.

Appendix A.2.  XRDS

To use www.example.com as their Identifier, but have Relying Parties actually verify http://exampleuser.livejournal.com/ with the Identity Provider located at http://www.livejournal.com/openid/server.bml, the following XML snippet should be present in the final XRD in the XRDS file:

<Service xmlns="xri://$xrd*($v*2.0)">
  <Type>http://openid.net/signon/2.0</Type>
  <URI>http://www.livejournal.com/openid/server.bml</URI>
  <Delegate xmlns="http://openid.net/signon/2.0">
    http://exampleuser.livejournal.com/
  </Delegate>
</Service>

Appendix A.3.  HTML Identifier Markup

To use www.example.com as their Identifier, but have Relying Parties actually verify http://exampleuser.livejournal.com/ with the Identity Provider located at http://www.livejournal.com/openid/server.bml, the following markup should be present in the <head> of the HTML document located by the identifier URL:

<link rel="openid.server"
      href="http://www.livejournal.com/openid/server.bml"/>
<link rel="openid.delegate"
      href="http://exampleuser.livejournal.com/"/>

Appendix A.4.  Login Form

Continuing this example, the End User visits a Relying Party site which supports OpenID Authentication. The Relying Party presents the End User with a form field for them to enter their Identifier or their IdP's identifier.

For example:

              ----------------------------------
              |[logo]example.com               | [Login Button]
              ----------------------------------

Appendix A.5.  XRI CanonicalID

For example, if =example and =exmpl both yield an XRD document with the CanonicalID xri://(example)!1234 then those identifiers should be treated as equivalent. For applications with user accounts, those identifiers should both be attached to the same account.

Appendix B.  Diffie-Hellman Key Exchange Default Value

This is a confirmed-prime number, used as the default modulus for Diffie-Hellman Key Exchange. In hexadecimal:

DCF93A0B883972EC0E19989AC5A2CE310E1D37717E8D9571BB7623731866E61E
F75A2E27898B057F9891C2E27A639C3F29B60814581CD3B2CA3986D268370557
7D45C2E7E52DC81C7A171876E5CEA74B1448BFDFAF18828EFD2519F14E45E382
6634AF1949E5B535CC829A483B8A76223E5D490A257F05BDFF16F2FB22C583AB

Appendix C.  Changes from the Previous OpenID Authentication Specification

This specification is based on the original specification for OpenID Authentication as written by Brad Fitzpatrick. That specification did not have a version number, but was called OpenID 1.0, and then OpenID 1.1 when it was revised. The protocol outlined in this specification is intended to be backwards-compatible with the revised OpenID protocol. The most significant changes to the specification are outlined in this section.

Appendix C.1.  Updated Initiation and Discovery

• Supports IdP-driven identifier selection. This new variation of the protocol flow is initiated by entering an Identifier for an IdP instead of an Identifier for an End User, and allows the IdP to assist the End User in selecting an Identifier.
• Supports the use of XRIs as Identifiers. XRIs may be used as Identifiers for both End Users and IdPs.
• When URLs are used as Identifiers, they are normalized according to RFC 3986, for better compatibility with existing Web infrastructure.
• Uses the Yadis protocol for discovery. This allows for using multiple IdPs for a single Identifier, for load-balancing and fallback in the case of IdP failure. Additionally, it allows for discovery of supported extensions.

Appendix C.2.  Security improvements

A nonce is now part of the protocol for built-in protection against replay attacks.

A new association type, HMAC-SHA256, and a new association session type, DH-SHA256, allow for stronger signatures on authentication assertions.

Appendix C.3.  Extensions

Extensions are a new mechanism to support data exchange and other Relying Party-IdP communication along with the authentication exchange. Extensions allow for the exchange of arbitrary attributes, as well as for protocol extensions, such as the inclusion of additional information about the Relying Party in the authentication request.

Because extensions can transfer arbitrary data, the Identifier is now optional in the response.

17. Normative References

Authors' Addresses