<abstract>This document specifies the XMPP Transport Layer Security (XTLS) protocol. XTLS, which provides communications privacy for the Extensible Messaging and Presence Protocol (XMPP), enables XMPP applications to communicate in a way that is designed to prevent eavesdropping, tampering, and forgery of XML stanzas. XTLS is based on Transport Layer Security (TLS) and integrated as security layer into Jingle. The protocol can be used for secure end-to-end messages as well as any other Jingle application like file file transfer.</abstract>
&LEGALNOTICE;
<number>xxxx</number>
<status>ProtoXEP</status>
<type>Standards Track</type>
<sig>Standards</sig>
<approver>Council</approver>
<dependencies>
<spec>XMPP Core</spec>
<spec>RFC 4346</spec>
<spec>XEP-0250</spec>
</dependencies>
<supersedes/>
<supersededby/>
<shortname>NOT_YET_ASSIGNED</shortname>
<author>
<firstname>Dirk</firstname>
<surname>Meyer</surname>
<email>dmeyer@tzi.de</email>
<jid>dmeyer@jabber.org</jid>
</author>
&stpeter;
<revision>
<version>0.0.4</version>
<date>2009-02-21</date>
<initials>dm</initials>
<remark><p>Improved the documentation about exchanging passwords and certificates.</p></remark>
</revision>
<revision>
<version>0.0.3</version>
<date>2009-02-20</date>
<initials>psa/dm</initials>
<remark><p>Simplified the syntax; added sections on requirements and approach; more clearly specified methods of bootstrapping trust on the first communication session.</p></remark>
</revision>
<revision>
<version>0.0.2</version>
<date>2009-02-20</date>
<initials>dm</initials>
<remark><p>Clarified order of events to be consistent with jingle-s5b; used TLS-SRP for password based authentication; added some documentation about password authentication in general.</p></remark>
</revision>
<revision>
<version>0.0.1</version>
<date>2009-02-12</date>
<initials>dm</initials>
<remark><p>First draft, based on previous attempts for end-to-end security.</p></remark>
</revision>
</header>
<section1topic='Introduction'anchor='intro'>
<p>End-to-end encryption of traffic sent over the Extensible Messaging and Presence Protocol (XMPP) is a desirable goal. The Jabber/XMPP developer community has experimented with several such technologies, including OpenPGP (&xep0027;), S/MIME (&rfc3923;), and encrypted sessions or "ESessions" (&xep0218;). For various reasons, these technologies have not been widely implemented and deployed. When the &XSF; asked various Internet security experts to complete a security review of encrypted sessions, it was recommended to explore the possibility of instead using the Transport Layer Security (TLS; see &rfc4346;) as the base technology for XMPP. That possibility is explored in this document.</p>
<p>TLS is the most widely implemented protocol for securing network traffic. In addition to applications in the email infrastructure, the World Wide Web (HTTP Over TLS; see &rfc2818;), and datagram transport for multimedia session negotiation (DTLS; see &rfc4347;), TLS is used in XMPP to secure TCP connections from client to server and from server to server, as specified in &xmppcore;. Therefore TLS is already familiar to XMPP developers.</p>
<p>This specification, called "XTLS", defines a method whereby any &xep0166; application can use TLS semantics for end-to-end encryption, whether the application data is sent over a streaming transport (like TCP) or a datagram transport (like UDP). The most common use case is to tunnel XMPP stanzas between two clients for end-to-end secure chat with &xep0247;. However, XTLS is not limited to client-to-client encryption, since it can be used between two XMPP clients, between an XMPP client and a remote XMPP service (i.e., a service with which a client does not have a direct XML stream, such as a &xep0045; chatroom), or between two remote XMPP services. Furthermore, XTLS can be used for encrypted file transfer (see &xep0234;), for encrypted voice or video sessions using &xep0167; and &dtlssrtp;, and other applications.</p>
<p><em>Note: This specification is a work in progress and does not yet define all uses of XTLS.</em></p>
</section1>
<section1topic='Requirements'anchor='req'>
<p>This specification is intended to meet the requirements defined in &xep0210; using building blocks that are already widely supported in XMPP clients, such as TLS, Jingle, and &xep0047;. It is possible that some of those requirements can be met only with particular TLS cipher suites, or cannot be met at all without defining extensions to TLS itself. A full gap analysis has not yet been completed.</p>
</section1>
<section1topic='Approach'anchor='approach'>
<p>In broad outline, XTLS takes the following approach to end-to-end encryption of XMPP traffic:</p>
<ol>
<li><p>We assume that all XMPP entities will have X.509 certificates, typically self-signed and automatically generated by an XMPP client (although CA-issued certificates are acceptable).</p></li>
<li><p>We use the XMPP Jingle extensions as the negotiation framework.</p></li>
<li><p>We define a <security/> element that can be included in any Jingle negotiation, and a new "security-info" Jingle action for sending security-related information.</p></li>
<li><p>When an entity wishes to encrypt its communications with a second entity, it sends a Jingle session-initiate request that specifies the desired application type (e.g., file transfer), a possible transport (e.g., SOCKS5 bytestreams), and security information such as the sender's X.509 fingerprint and supported TLS methods (e.g., secure remote password).</p></li>
<li><p>If both parties support XTLS, the first data sent over the negotiated transport is TLS handshake data, not application data. Once the TLS handshake has finished, the parties can then send application data over the now-encrypted transport.</p></li>
<li><p>The simplest scenario is end-to-end encryption of traditional XMPP text chat using end-to-end XML streams, in-band bytestreams, and previously-accepted X.509 certificates.</p></li>
<li><p>On first use of end-to-end encryption between two entities, it is encouraged to use secure remote passwords rather than leap-of-faith to bootstrap the subsequent use of the client-generated X.509 certificates.</p></li>
<li><p>More complex scenarios are theoretically supported (e.g., encrypted file transfer using SOCKS5 bytestreams and encrypted voice chat using DTLS-SRTP) but have not yet been fully defined.</p></li>
<li><p>XTLS theoretically can be used to establish a TLS-encrypted streaming transport or a DTLS-encrypted datagram transport, but integration with DTLS (&rfc4347;) has not yet been prototyped so use with streaming transports is the more stable scenario.</p></li>
</ol>
<p>We expand on this approach in the following section.</p>
</section1>
<section1topic='Basic Protocol'anchor='proto'>
<p>The basic flow for an XTLS session is as follows, where traffic represented by single dashes (---) is sent over the XMPP signalling channel and traffic represented by double lines (===) is sent over the negotiated transport.</p>
<code><![CDATA[
Initiator Responder
| |
| session-initiate |
| (with security info) |
|--------------------------->|
| ack |
|<---------------------------|
| session-accept |
|<---------------------------|
| ack |
|--------------------------->|
| open transport |
|<==========================>|
| TLS ClientHello |
|===========================>|
| TLS ServerHello, [...] |
|<===========================|
| TLS [...], Finished |
|===========================>|
| TLS [...], Finished |
|<===========================|
| application data |
|<==========================>|
| session-terminate |
|<---------------------------|
| ack |
|--------------------------->|
| |
]]></code>
<p>To simplify the description we assume here that the parties already trust each other's certificates. See the next section for information about bootstrapping of certificate trust on the first communication.</p>
<p>First the initiator sends a Jingle session-initiate request (here the simple case of an end-to-end text chat session using the &xep0261;). This request includes a <security/> element that contains the fingerprint of the certificate that the initiator will use during the TLS negotiation.</p>
<p>Depending on the application type, a user agent controlled by a human user might need to wait for the user to affirm a desire to proceed with the session before continuing. When the user agent has received such affirmation (or if the user agent can automatically proceed for any reason, e.g. because no human intervention is expected or because a human user has configured the user agent to automatically accept sessions with a given entity), it returns a Jingle session-accept message. This message will typically contain the offered application type, transport method, and a <security/> element that includes the fingerprint of the responder's X.509 certificate as well as the responder's supported TLS methods.</p>
<examplecaption='Responder accepts the session and agrees to use XTLS'><![CDATA[
<p>The following rules apply to the responder's handling of the session-initiate message:</p>
<ol>
<li><p>If the responder does not support Jingle-XTLS it will silently ignore the <security/> element in the offer and therefore will return a session-accept message without a <security/> element.</p></li>
<li><p>If the responder supports Jingle-XTLS it SHOULD return a session-accept message that contains a <security/> element.</p></li>
<li><p>If the responder thinks it will be able to verify the initiator's certificate, it MUST include the fingerprint for the responder's certificate in the <security/> element of the session-accept message. This is the "happy path" and will occur when the parties have already verified each other's certificates.</p></li>
<li><p>If the responder thinks it will not be able to verify the initiator's certificate, it MAY omit the fingerprint for the responder's certificate in the <security/> element of the session-accept message. This indicates that certificate-based authentication is not possible. In this case the responder SHOULD signal that it wishes to use some other authentication method, such as secure remote passwords (see the next section).</p></li>
<li><p>If the responding client cannot verify the initiator's certificate, it SHOULD ask the responding user if a password was exchanged between the parties that can be used for TLS-SRP. If this is not the case, setting up a mutually-authenticated link will fail and the responder MAY terminate the session. Alternatively it could send its own fingerprint knowing it cannot authenticate the initiator, in which case the responder has to trust that there is no man-in-the-middle (see next section).</p></li>
</ol>
<p>When the responder sends the session-accept message, the initiator acknowledges receipt.</p>
<p>The following rules apply to the initiator's handling of the session-accept message:</p>
<ol>
<li><p>If the initiator receives a session-accept without a <security/> element, setting up a secure transport layer has failed. The initiator MAY terminate the session at this point or instead proceed without securing the transport. The client SHOULD ask the initiating user how to processed. This depends on the Jingle application and the initiator's preferences: it makes no sense to use end-to-end XML streams without encryption, but the initiator might continue a file transfer without encryption.</p></li>
<li><p>If the initiating client cannot verify the responder's certificate it SHOULD ask the initiating user if a password was exchanged between the parties that can be used for TLS-SRP. If this is not the case, setting up a mutually-authenticated link will fail and the responder MAY terminate the session or proceed with leap-of-faith (see next section).</p></li>
</ol>
<p>The initiator can now determine if X.509 certificate based authentication will work or if TLS-SRP will be used. It sends an additional security-info message to the responder to signal its choice. This step is not really necessary because the responder will see the initiator's choice in the first message of the TLS handshake, but it can help an implementation to set up its TLS library properly. Because in this section we assume that the parties already have validated each other's certificates, the security method signalled here is "x509".</p>
<p>Parallel to the security-info exchange the clients set up a transport for the Jingle session. This example uses &xep0261; which does not require any additional messages (other transport like &xep0260; and &xep0176;) need more time to set up). Each transport MUST define a specific time when both cients know that the transport is ready. When XTLS is not used, the Jingle implementation would signal to the using application that the transport is open when the session-accept is sent or received, or when connectivity checks determine media can flow over one of the transport candidates. When XTLS is used, the Jingle implementation starts a TLS handshake on the transport and signals to the using application that the transport is open only after the four-way TLS handshake is finished successfully.</p>
<p>During the TLS handshake, the responder MUST take the role of the TLS server and the initiator MUST take the role of the TLS client. During the handshake, the responder (acting as the TLS server) MUST send a CertificateRequest.</p>
<p>If either client cannot verify the certificate of the peer or receives an invalid message on the TLS layer, it shall terminate the Jingle session immediately, including a Jingle reason of <security-error/>.</p>
<examplecaption="Initiator terminates the session"><![CDATA[
<iqfrom='romeo@montague.lit/orchard'
id='hz81vf48'
to='juliet@capulet.lit/balcony'
type='set'>
<jinglexmlns='urn:xmpp:jingle:0'
action='session-terminate'
initiator='romeo@montague.lit/orchard'
sid='a73sjjvkla37jfea'>
<reason><security-error/></reason>
</jingle>
</iq>
]]></example>
<p>The other party then acknowledges the session-terminate and the Jingle session is finished.</p>
<p>If the TLS negotiation succeeds, then the Jingle implementation shall signal to the using application that the transport is open. The parties can then exchange application data over the encrypted transport.</p>
</section1>
<section1topic='Bootstrapping Trust on First Communication'anchor='bootstrapping'>
<p>When two parties first attempt to use XTLS, their certificates might not be accepted (e.g., because they are self-signed or issued by unknown certification authorities). Therefore each party needs to accept the other's certificate for use in future communication sessions. There are several ways to do so:</p>
<ul>
<li><p>Leap of faith. The recipient can hope that there is no man-in-the-middle during the first communication session. If the certificate does not change in future sessions, the recipient at least knows that it is talking with the same entity it talked with during the first session. However, that entity might be a man-in-the-middle rather than the assumed communication partner. Therefore, leap of faith is discouraged.</p></li>
<li><p>Check fingerprints. The parties could validate the certificate fingerprints via some trusted means outside the XMPP band, such as in person, via encrypted email, or over the phone. This is not user-friendly because certificate fingerprints consist of long strings of letters and numbers. As a result, few humans check certificate fingerprints in protocols such as Secure Shell.</p></li>
<li><p>One-time password. The parties can exchange a user-friendly password known only to themselves and verify it out of band before the TLS handshake finishes. For this purpose, it is REQUIRED for implementations to support at least one TLS cipher that uses Secure Remote Password (SRP) as defined in &rfc5054;.</p></li>
<li><p>Channel binding. It is possible that a future version will describe how to use an appropriate Simple Authentication and Security Layer (SASL) mechanism, such as &scram;, to authenticate the XTLS channel after the TLS handshake finishes using the concept of channel bindings (see &rfc5056;).</p></li>
</ul>
<p>If the parties use a password or channel binding via SASL, the process needs to be completed only once. After the clients have authenticated with the shared secret, they can exchange their certificates for future communication.</p>
<p>To retrieve the certificate of the peer for future communications, a client SHOULD request the certificate according to <cite>XEP-0189</cite> over the secure connection. This only works if XTLS was used to set up an end-to-end secure XML stream defined in <cite>XEP-0247</cite>; exchanging certificates if XTLS was used for other purposes like file transfer is not possible. A client MUST NOT request the certificate over the insecure stream based on the connection to the XMPP server.</p>
<examplecaption='Public keys request'><![CDATA[
<iqfrom='romeo@montague.lit/orchard'
id='hf7634k'
to='juliet@capulet.lit/balcony'
type='get'>
<pubkeysxmlns='urn:xmpp:tmp:pubkey'/>
</iq>
]]></example>
<p>The peer MUST return its own client certificate. If the user has different clients with different client certificates and one user certificate, the user certificate SHOULD also be returned. The user certificate allows it to verify other client certificates using public key retrieval described in <cite>XEP-0189</cite>.</p>
<examplecaption='Successful public keys response'><