The ability to multiplex multiple virtual hosts over a single XMPP session (historically known as "piggybacking") was originally defined in &rfc3920; and later pulled out into &xep0220; for use with &rfc6120;. With the advent of cheap or free TLS certificates the use of dialback began falling off on the public XMPP network as more secure authentication mechanisms such as SASL EXTERNAL began to become more common. However, multiplexing is still a useful technique in constrained environments regardless of the authentication mechanism being used.

Multiplexing is also useful for reusing connections for additional services associated with a domain but hosted at a subdomain. For example, both the "montague.example" and the "capulet.example" may be hosted by the same XMPP server which may also host &xep0369; services at "chat.montague.example" and "rooms.capulet.example" respectively. Without multiplexing this would require eight TCP connections for a bidirectional exchange of stanzas between two sending domains and two target domains. However, with multiplexing this can be reduced to two connections, or, at the operator's discretion, more than two for operational reasons. If multiplexing is not used, the number of server-to-server connections needed to exchange stanzas between virtual hosting providers or multi-service XMPP servers can increase signficantly. This can lead to the number of connections exceeding the maximum number of connections allowed from a single address as explained in &xep0205;.

This specification defines new mechanisms for advertising and negotiating multiple hosts over a single session. Furthermore it advances the state of the art over the multiplexing solution defined in &xep0220; by working on both client-to-server (c2s) and server-to-server (s2s) sessions.

If a server supports receiving multiplexed streams it SHOULD inform the connecting entity when returning stream features during the negotiation process. Two mechanisms exist for authenticating domains that can be multiplexed over a connection: domains may be authenticated using the TLS certificate (and client certificate if applicable), and domains may be authorized using the connection authorization mechanism described later in this document.

To advertise support for multiplexing all domains present in a TLS certificate the server includes a <mux/> element qualified by the 'urn:xmpp:mux:0' namespace in the stream features list. This feature MUST be advertised only after TLS has been negotiated (either by opportunistic TLS using the STARTTLS feature or by implicit TLS when establishing the TCP socket) an before authentication using SASL EXTERNAL has been performed. This feature is not mandatory to negotiate.

The mux feature may also be advertised after authentication with SASL EXTERNAL. If advertised after authentication the feature MUST include a list of supported hosts wrapped in <host/> elements.

If the initiating entity wishes to indicate that it intends to use multiplexing with SASL EXTERNAL it MUST respond to the empty <mux/> element by sending another empty <mux/> element qualified by the 'urn:xmpp:mux:0' namespace in reply. No stream restart is necessary.

After indicating support for multiplexing by negotiating the mux stream feature, authentication can proceed. When using SASL EXTERNAL this is done by validating the certificate as detailed in &xep0178; except that every domain that is present in the certificate is now eligible for multiplexing without further negotiation. Further stream features (such as resource binding) still use the JID from the original connection (or from the authorization identity).

If a bidirectional s2s connection has been negotiated for this session using &xep0288;, negotiation of the mux stream feature also implies that the receiving entity SHOULD multiplex stanzas sent back to the initiating entity for all domains in the verified client certificate. If bidi is not negotiated then mux will need to be negotiated again when the original receiving entity establishes a connection with the original initiating entity.

Often it is not desirable to have one certificate containing every XMPP address or host managed by the server, or the use of SASL EXTERNAL may be impossible. In these cases the initiating entity may request authorization to send stanzas over an existing connection.

If the initiating entity has an authenticated connection to a server and wishes to send stanzas to another server that was listed in the original servers post-auth <mux/> stream feature it MAY establish an XMPP connection with the new server and verify that new server also lists the original server in its post-auth mux stream feature. If it does the initiating entity replies with a <mux/> element qualified by the 'urn:xmpp:mux:0' namespace with a shared secret as the payload and the host being selected included in the 'host' attribute. The old server then sends an IQ over its existing connection with the initiating entity containing the same mux element and secret, thereby confirming its relationship to the new server. If the client verifies that the secrets match it sends an empty IQ of type "result" in response to indicate success, otherwise the IQ response should be a "not-acceptable" stanza error (see &rfc6120; §8.3.3.9).

For example, if the server montague.example wishes to establish a multiplexed connection with capulet.example and rooms.capulet.example the flow would look like this:

The XML for this exchange would look like the following:

The format of the secret is not specified however, see the Security Considerations section of this document for some suggestions.

Some clients may send stanzas with no "from" attribute specified and rely on the server to add the attribute before routing the stanza to its final destination. If multiplexing is used the lack of a from attribute indicates that the client is acting on behalf of the origin JID for the connection, just like normal, so clients MUST set the from attribute on any stanzas sent on behalf of any multiplexed host.

The format of mux secrets is undefined in this document, however, they MUST be unpredictable. Only the initiating entity should attribute any meaning (if indeed there is any) to the format of mux secrets. In particular the receiving entity MUST NOT rely on a specific format for the secret.

One suggestion for generating mux secrets is to generate a key that signs information about the stream. The format defined in &xep0185; is appropriate for this. If the mux secret is a signature it must protect against reuse by at least include a random secret generated with a cryptographically secure random number source, the origin JID, the JID of the server initially receiving the mux secret, and the stream ID for the stream the key will be authenticating (this is not the same stream as the receiving entity's JID). It is also RECOMMENDED that an expiration time be included in the key after which it is no longer valid.

This document requires no interaction with &IANA;.

Thanks to Jeremie Miller, Peter Saint-Andre, and Philipp Hancke for writing &xep0220; from which some of the text and techniques used in this document were borrowed.