This document provides a technical specification for Digital Signatures in Extensible Messaging and Presence Protocol (&xmpp;) based upon End-to-End Object Encryption (&E2EEncrypt;) "work in progress".

The S/MIME approach defined in &rfc3923; has never been implemented in XMPP clients to the best of our knowledge, but has some attractive features, especially the ability to store-and-forward a signed message at a user's server if the user is not online when the message is received (in the XMPP community this is called "offline storage" and the message is referred to as an "offline message"). The authors surmise that RFC 3923 has not been implemented mainly because it adds several new dependencies to XMPP clients, especially MIME (along with the CPIM and MSGFMT media types). This document explores the possibility of an approach that is similar to but simpler than RFC 3923.

The process that a sending agent follows for securing stanzas is very similar regardless of the form of stanza (i.e., <iq/>, <message/>, or <presence/>).

1. Constructs a cleartext version of the stanza, S.
2. Notes the current UTC date and time N when this stanza is constructed, formatted as described in Section 5.
3. Converts the stanza to a UTF-8, as defined by &rfc3269;, encoded string, optionally removing line breaks and other insignificant whitespace between elements and attributes, i.e., UTF8-encode(S) = S'. We call S' a "stanza-string" because for purposes of signing and verification it is treated not as XML but as an opaque string (this avoids the need for complex canonicalization of the XML input).
4. Constructs a plaintext envelope (E) <plain/> qualified by the "urn:xmpp:signed:0" namespace as follows:
   • The attribute 'timestamp' set to the UTC date and time value N
   • The XML character data set to the base64-encoded form of S' (where the encoding adheres to the definition in Section 4 of &BASE64; and where the padding bits are set to zero). This encoding is necessary to preserve a canonicalized form of S'.
5. Converts the envelope (E) to a UTF-8 encoded string, optionally removing line breaks and other insignificant whitespace between elements and attributes, i.e., E' = UTF8-encode(E).
6. Produce a signature of UTF8-encoded envelope (E') using the intended signature algorithm. T = signature(E'). (This step is underspecified and will be expanded upon in later revision of this document.)
7. Base64-encodes T to produce the signature data T'.
8. Constructs an <signed/> element qualified by the "urn:xmpp:signed:0" namespace as follows:
   • The child element <signature> (implicitly qualified by the "urn:xmpp:signed:0" namespace) as follows:
     • The attribute 'algorithm' set to a string identifying the signature algorithm used.
     • The XML character data T'.
   • The child element <data> (implicitly qualified by the "urn:xmpp:signed:0" namespace) as follows:
     • The XML character data E'.
9. Sends the <signed> element as the payload of a stanza that SHOULD match the stanza from step 1 in kind (e.g., <message/>), type (e.g., "chat"), and addressing (e.g. to="romeo@montague.net" from="juliet@capulet.net/balcony"). If the original stanza (S) has a value for the "id" attribute, this stanza MUST NOT use the same value for its "id" attribute.

The following limitations and caveats apply:

• Undirected <presence/> stanzas SHOULD NOT be signed.
• Stanzas directed to multiplexing services (e.g. multi-user chat) SHOULD NOT be signed, unless the sender has established the service supports the handling of signed stanzas.

Several scenarios are possible when an entity receives an encrypted stanza:

Case #1:

The receiving application does not understand the protocol.

Case #2:

The receiving application understands the protocol and is able to verify the signature.

Case #3:

The receiving application understands the protocol and is able to verify the signature, but the timestamps fail the checks specified under Checking of Timestamps.

Case #4:

The receiving application understands the protocol and is unable to verify the signature.

In Case #1, the receiving application MUST do one and only one of the following: (1) ignore the <signed/> extension, (2) ignore the entire stanza, or (3), except where precluded by the protocol (&rfc3920bis;), return a <service-unavailable/> error to the sender.

In Case #2, the receiving application MUST NOT return a stanza error to the sender, since this is the success case.

In Case #3, the receiving application MAY, except where precluded by the protocol, return a <not-acceptable/> error to the sender, optionally supplemented by an application-specific error condition element of <bad-timestamp/> as shown below:

In Case #4, the receiving application SHOULD, except as precluded by the protocol, return a <bad-request/> error to the sender, optionally supplemented by an application-specific error condition element of <bad-signature/> as shown below:

Additionally in Case #4, the receiving application SHOULD NOT present the stanza to the intended recipient (human or application) and SHOULD provide some explicit alternate processing of the stanza (which may be to display a message informing the recipient that it has received a stanza that cannot be verified).

Timestamps are included to help prevent replay attacks. All timestamps MUST conform to &DATETIME; and be presented as UTC with no offset, always including the seconds and fractions of a second to three digits (resulting in a datetime 24 characters in length). Absent a local adjustment to the sending agent's perceived time or the underlying clock time, the sending agent MUST ensure that the timestamps it sends to the receiver increase monotonically (if necessary by incrementing the seconds fraction in the timestamp if the clock returns the same time for multiple requests). The following rules apply to the receiving application:

• It MUST verify that the timestamp received is within five minutes of the current time, except as described below for offline messages.
• If the foregoing check fails, the timestamp SHOULD be presented to the receiving entity (human or application) marked with descriptive text indicating "old timestamp" or "future timestamp" and the receiving entity MAY return a stanza error to the sender (except as precluded in the protocol).

The foregoing timestamp checks assume that the recipient is online when the message is received. However, if the recipient is offline then the server will probably store the message for delivery when the recipient is next online (offline storage does not apply to <iq/> or <presence/> stanzas, only <message/> stanzas). As described in &xep0160;, when sending an offline message to the recipient, the server SHOULD include delayed delivery data as specified in &xep0203; so that the recipient knows that this is an offline message and also knows the original time of receipt at the server. In this case, the recipient SHOULD verify that the timestamp received in the encrypted message is within five minutes of the time stamped by the recipient's server in the <delay/> element.

All implementations MUST support the following algorithms. Implementations MAY support other algorithms as well.

• TBD (RSA/SHA1? RSASSA-RKCS1-v1_5? RSASSA-PSS?)

To participate in end-to-end signing using the methods defined in this document, a client needs to possess an X.509 certificate. It is expected that many clients will generate their own (self-signed) certificates rather than obtain a certificate issued by a certification authority (CA). In any case the certificate MUST include an XMPP address that is represented using the ASN.1 Object Identifier "id-on-xmppAddr" as specified in Section 5.1.1 of RFC 3920bis.

TBD.

This document borrows ideas and text from End-to-End Object Encryption "work in progress" by Matthew Miller and Peter Saint-Andre.