Jingle Encrypted Transports (JET) strives to provide a modular and easily extensible way to wrap Jingle Transports in an additional end-to-end encryption layer. The focus of this specification lays on being modular. It should be possible to extend existing Jingle use scenarios with end-to-end encryption by simply adding a JET element to the negotiation.

JET uses multiple encryption layers, so it is necessary to declare a distinct denomination for the different keys involved.

Lets assume Romeo wants to initiate an encrypted Jingle session with Juliet. Prior to the Jingle session initiation, an already existing, established and (ideally) authenticated end-to-end encryption session between Romeo and Juliet MUST exist. This session is needed to transfer the Transport Secret from Romeo to Juliet.

When this precondition is met, Romeo initially generates a transport key (TK) and associated initialization vector (IV). These will later be used by the sender to encrypt, and respectively by the recipient to decrypt data that is exchanged. This protocol defines a set of usable ciphers from which Romeo might choose. TK and IV together form the transport secret (TS).

Next Romeo uses his established encryption session with Juliet to encrypt TS. The resulting envelope element (EE) will be part of the Jingle session initiation as child of the JET &security; element.

When Juliet receives Romeos session request, she decrypts EE to retrieve TS, from which she can obtain TK and IV. Now she and Romeo can go on with the session negotiation. Once the session is established, data can be encrypted and exchanged. Both parties MUST keep a copy of TS in cache until the Jingle session is ended.

&xep0234; has the disadvantage, that transmitted files are not encrypted (aside from regular TLS transport encryption), which means that intermediate nodes like XMPP/proxy server(s) have access to the transferred data. Considering that end-to-end encryption becomes more and more important to protect free speech and personal expression, this is a major flaw that needs to be addressed.

In order to initiate an encrypted file transfer, the initiator includes a JET &security; element in the Jingle file transfer request.

In order to encrypt the transported bytestream, the initiator must transmit a cipher key to the responder. There are multiple options available:

To advertise its support for the Jingle Encrypted Transports, when replying to service discovery information ("disco#info") requests an entity MUST return URNs for any version, or extension of this protocol that the entity supports -- e.g., "urn:xmpp:jingle:jet:0" for this version, or "urn:xmpp:jingle:jet-stub:0" for a stub encryption method &VNOTE;.

In order for an application to determine whether an entity supports this protocol, where possible it SHOULD use the dynamic, presence-based profile of service discovery defined in &xep0115;. However, if an application has not received entity capabilities information from an entity, it SHOULD use explicit service discovery instead.

The initiator SHOULD NOT use the generated key TK as IV, but instead generate a seperate random IV.

Instead of falling back to unencrypted transfer in case something goes wrong, implementations MUST instead abort the Jingle session, informing the user.

IMPORTANT: This approach does not deal with metadata. In case of &xep0234;, an attacker with access to the sent stanzas can for example still see the name of the file and other information included in the <file/> element.

The responder MUST check, whether the envelope element belongs to the initiator to prevent MitM attacks

Big thanks to Florian Schmaus for mentoring my Google Summer of Code project, which resulted in this protocol. Also thanks to Andrey Gursky, Daniel Gultsch, Dave Cridland, Goffi, Jonas Wielicki and Sam Whited for their input and feedback.