DTLS-SRTP Key Transport (KTR)
draft-wing-avt-dtls-srtp-key-transport-03

Status of this Memo

This Internet-Draft is submitted to IETF in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts.

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as “work in progress.”

The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt.

The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html.

This Internet-Draft will expire on September 10, 2009.

Copyright Notice

Copyright (c) 2009 IETF Trust and the persons identified as the document authors. All rights reserved.

This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents in effect on the date of publication of this document (http://trustee.ietf.org/license-info). Please review these documents carefully, as they describe your rights and restrictions with respect to this document.

Abstract

The existing DTLS-SRTP specification allows SRTP keys to be established between a pair of SRTP endpoints. However, when there are more than two participants in an SRTP session, DTLS-SRTP is unable to provide a single key for all of the participants. This existing limitation of DTLS-SRTP prevents deploying DTLS-SRTP in certain scenarios.

This document describes an extension to DTLS-SRTP called Key Transport (KTR). This extension transports SRTP keying material from one DTLS-SRTP peer to another, so the same SRTP keying material can be used by multiple DTLS-SRTP peers. This extension eliminates the need to key each SRTP session individually, allowing cost-effective deployment of several DTLS-SRTP scenarios.

Table of Contents

1.  Introduction
2.  Notational Conventions
3.  Scenarios
    3.1.  Point to Multipoint using the RFC 3550 mixer model
    3.2.  Point to Multipoint using Multicast
    3.3.  Point to Multipoint Using Video Switching MCUs
    3.4.  Scaling to Large Groups
        3.4.1.  Rekeying SRTP Quickly with Encrypted Key Transport (EKT)
        3.4.2.  Distributed Key Servers
    3.5.  Interworking with Other SRTP Key Management Systems
        3.5.1.  Security Descriptions
4.  Protocol Description
    4.1.  key_transport (KTR) extension to DTLS-SRTP
    4.2.  KTR Primitives
    4.3.  Procedures for Network Elements
        4.3.1.  Speaker
        4.3.2.  Mixer
        4.3.3.  Switcher
        4.3.4.  Listener
    4.4.  Key Transport SSRC and RTP SSRC Collisions
    4.5.  Fragmentation, Reassembly, and Retransmission
    4.6.  SDP extensions
5.  Examples
6.  Security Considerations
    6.1.  Group Policy when Joining/Leaving
    6.2.  Two-Time Pad
7.  Acknowledgements
8.  IANA Considerations
9.  References
    9.1.  Normative References
    9.2.  Informational References
Appendix A.  Changes
    A.1.  Changes from -00 to -01
    A.2.  Changes from -01 to -02
    A.3.  Changes from -02 to -03
Appendix B.  LKH for More Efficient Rekeying
§  Author's Address

1.  Introduction

When DTLS-SRTP (McGrew, D. and E. Rescorla, “Datagram Transport Layer Security (DTLS) Extension to Establish Keys for Secure Real-time Transport Protocol (SRTP),” February 2009.) [I‑D.ietf‑avt‑dtls‑srtp] establishes Secure RTP (Baugher, M., McGrew, D., Naslund, M., Carrara, E., and K. Norrman, “The Secure Real-time Transport Protocol (SRTP),” March 2004.) [RFC3711] master keys, each peer contributes part of the keying material to derive the SRTP master key. To reduce cryptographic operations in some scenarios it is desirable for one peer to change its SRTP key and to transmit SRTP packets using an SRTP key that was not derived from the DTLS key exchange.

The extension described in this document allows transporting an SRTP master key from one DTLS peer to the other. Thus, DTLS Key Transport differs from normal DTLS-SRTP in that the SRTP master key is not derived from the TLS handshake.

2.  Notational Conventions

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, S., “Key words for use in RFCs to Indicate Requirement Levels,” March 1997.).

A "listener" is an endpoint that only receives an SRTP stream. A "speaker" is an endpoint that only transmits an SRTP stream. And endpoint can be both a listener and a speaker.

3.  Scenarios

KTR allows mixers and video switchers to avoid having to encrypt each packet multiple times under multiple SRTP keys, by allowing a single SRTP key to be shared with the multiple recipients that are receiving the SRTP stream.

Several SRTP scenarios that benefit from KTR are described in the following sections, using terminology from RTP Topologies (Westerlund, M. and S. Wenger, “RTP Topologies,” January 2008.) [RFC5117].

3.1.  Point to Multipoint using the RFC 3550 mixer model

This RTP scenario is described in Section 3.4 of RTP Topologies (Westerlund, M. and S. Wenger, “RTP Topologies,” January 2008.) [RFC5117].

With DTLS-SRTP, this topology is computationally expensive for the video switcher because it has to encrypt the payload uniquely for each SRTP listener. Additionally, the architecture of a typical mixer requires each listener's SRTP to be encrypted serially, incurring additional delay for each successive listener. This is depicted below in Figure 1 (Point to Multipoint Mixer, without DTLS Key Transport).

     +-------key=F-------+
     |                   |
     V               +-------+         +------------+
+----+----+          |       +--key=C->+ listener 1 |
| speaker +--key=A-->+       |         +------------+
+---------+          |       |         +------------+
                     | mixer +--key=D->+ listener 2 |
+---------+          |       |         +------------+
| speaker +--key=B-->+       |         +------------+
+----+----+          |       +--key=E->+ listener 3 |
     ^               +---+---+         +------------+
     |                   |
     +-------key=G-------+

With KTR, the mixer can maintain one outbound SRTP cryptographic context, and encrypt the SRTP once for all listeners. This is depicted below in Figure 2 (Point to Multipoint Mixer, with DTLS Key Transport).

In the following figure, "=" indicates sessions where DTLS-SRTP Key Transport is used, and "-" indicates where only DTLS-SRTP is necessary. In this topology, only the listeners need support KTR so that the switcher and the listeners can benefit from KTR. In this scenario with DTLS-SRTP Key Transport, the mixer assumes an additional role -- group's key server -- and provides a common group SRTP key ("C") to all of the listeners. This group SRTP key is shared between all of the listeners. The two speakers, however, receive a unique stream (just as in the scenario above), but to prevent a two-time (padSection 6.2 (Two-Time Pad)), their content is encrypted using a different SRTP keys ("D" and "E").

     +=======key=D=======+
     |                   |
     V               +---+---+         +------------+
+---------+          |       +==key=C=>+ listener 1 |
| speaker +--key=A-->+       |         +------------+
+---------+          |       |         +------------+
                     | mixer +==key=C=>+ listener 2 |
+---------+          |       |         +------------+
| speaker +--key=B-->+       |         +------------+
+----+----+          |       +==key=C=>+ listener 3 |
     ^               +---+---+         +------------+
     |                   |
     +=======key=E=======+

The mixer is aware of listeners leaving or joining, and the mixer can rekey the remaining active listeners.

3.2.  Point to Multipoint using Multicast

This RTP topology is described in Section 3.2 of RTP Topologies (Westerlund, M. and S. Wenger, “RTP Topologies,” January 2008.) [RFC5117].

With DTLS-SRTP, this scenario is not attainable because each listener has a unique SRTP key. For this reason, [I‑D.ietf‑msec‑gdoi‑srtp] (Baugher, M., Rueegsegger, A., and S. Rowles, “GDOI Key Establishment for the SRTP Data Security Protocol,” December 2007.) was developed by the MSEC working group.

With KTR, this scenario is attainable because the same key can be provided to multiple listeners, as depicted below in Figure 3 (Point to Multipoint using Multicast with Key Transport). This compares favorably with [I‑D.ietf‑msec‑gdoi‑srtp] (Baugher, M., Rueegsegger, A., and S. Rowles, “GDOI Key Establishment for the SRTP Data Security Protocol,” December 2007.) when the group size is small enough that the speaker can perform key server functions (i.e., perform KTR) for all of the listeners.

                        +-------+            +------------+
                       /         \==key=A===>+ listener 1 |
                      /           \          +------------+
+---------+           | multicast |          +------------+
| speaker +==key=A===>+  network  +==key=A==>+ listener 2 |
+---------+           |           |          +------------+
                      \           /          +------------+
                       \         /===key=A==>+ listener 3 |
                        +-------+            +------------+

3.3.  Point to Multipoint Using Video Switching MCUs

This RTP topology is described in Section 3.5 of RTP Topologies (Westerlund, M. and S. Wenger, “RTP Topologies,” January 2008.) [RFC5117].

With DTLS-SRTP, this topology is computationally expensive for the video switcher because it has to encrypt the payload uniquely for each SRTP listener. Additionally, the architecture of a typical video switcher requires each listener's SRTP to be encrypted serially, incurring additional delay for each successive listener. This is depicted below in Figure 4 (Point to Multipoint Video Switching, without DTLS Key Transport).

In the following figure, KTR is used on all sessions and depicted by "=". In this scenario, both the speakers and listeners must support KTR so that the switcher and the listeners can benefit from KTR.

     +-------key=F-------+
     |                   |
     V               +---+------+         +------------+
+---------+          |          +==key=C=>+ listener 1 |
| speaker +==key=A==>+selected  |         +------------+
+---------+          |          |         +------------+
                     | switcher +==key=D=>+ listener 2 |
+---------+          |          |         +------------+
| speaker +==key=B==>+dropped   |         +------------+
+----+----+          |          +==key=E=>+ listener 3 |
     ^               +---+------+         +------------+
     |                   |
     +-------key=G-------+

With DTLS key transport, this becomes easier; in fact, the video switcher doesn't need to decrypt the SRTP at all, but just make its decision (select the stream or drop the stream) and transmit the SRTP packets to the listeners. This is depicted below in Figure 5 (Point to Multipoint Video Switching, with DTLS Key Transport).

     +-------key=B-------+
     |                   |
     V               +---+------+         +------------+
+----+----+          |          +==key=A=>+ listener 1 |
| speaker +==key=A==>+selected  |         +------------+
+---------+          |          |         +------------+
                     | switcher +==key=A=>+ listener 2 |
+---------+          |          |         +------------+
| speaker +==key=B==>+prev.spkr |         +------------+
+---------+          |          +==key=A=>+ listener 3 |
     ^               +----------+         +------------+
     |                   |
     +-------key=A-------+

The video switcher is aware of listeners leaving or joining. The protocol described in this document allows the switcher to dictate, to the speaker, that the speaker use a new encryption key. This allows the switcher to enforce security, based on the switcher's policy (Section 6.1 (Group Policy when Joining/Leaving)). This is done by the video switcher sending a DTLS "new_srtp_key_request" message. The speaker will respond with a DTLS "new_srtp_key" message. The "new_srtp_key" message is relayed, by the switcher, to each of the active listeners.

When there are multiple speakers, as shown in Figure 5 (Point to Multipoint Video Switching, with DTLS Key Transport) above, each speaker transmits with his own SRTP key. That SRTP key is derived from the DTLS handshake with the switcher. Each speaker uses KTR to signal the SSRC that it will use.

3.4.  Scaling to Large Groups

This section describes how DTLS-SRTP-Key-Transport supports large groups of listeners, both for unicast and multicast scenarios.

3.4.1.  Rekeying SRTP Quickly with Encrypted Key Transport (EKT)

When a new listener is added, or an existing listener is removed, a new SRTP master key is necessary to retain the security of the SRTP media. Normally this causes "n" cryptographic operations for "n" listeners. These cryptographic operations take time, and if the group is large enough or the processor slow enough, there can be a considerable delay before all listeners receive the new SRTP key and can decrypt the stream.

A solution is to communicate a common Encrypted Key Transport (EKT) (McGrew, D., Andreasen, F., Wing, D., and L. Dondeti, “Encrypted Key Transport for Secure RTP,” October 2009.) [I‑D.mcgrew‑srtp‑ekt] key to all listeners at session establishment using the ekt_key primitive. Then, when a listener is added and SRTP needs to be rekeyed, an EKT message is sent to the listeners. This EKT message contains a new SRTP key encrypted using that EKT key. Upon receiving an SRTP packet, the previous SRTP decryption key or this new key can be used to decrypt the packet; the EKT message provides information that assists with such key rollover (specifically, its 'initial sequence number' field). If the participant needs to send SRTP or SRTCP, it uses its own key to encrypt and authenticate that traffic - not the shared SRTP key conveyed in EKT.

When a listener is ejected, the following procedure occurs:

1. The DTLS-SRTP-Key-Transport device sends a new EKTKey to each remaining participant. This is done using the ekt_key primitive defined in this document.
2. Once all participants have the new EKTKey, an EKT message is sent. This EKT message contains a new SRTP key encrypted using the new EKTKey communicated in the previous step.
3. Begin encrypting SRTP traffic with the new SRTP master key.

More efficient mechanisms to rekey large groups of participants are possible (Appendix B (LKH for More Efficient Rekeying)).

3.4.2.  Distributed Key Servers

Another problem with all group scenarios is that because each listener establishes a DTLS-SRTP session with the speaker, only a finite number of listeners can be supported (the speaker cannot handle millions of DTLS-SRTP sessions). This is especially problematic for multicast, but is also a problem for large groups.

One workaround to the problem is distributing the DTLS-SRTP keying to other devices in the network. In this scheme, one key server is responsible for a sensible number of listeners and has sufficient CPU power to update those listeners with new SRTP master keys. This is done with a new SDP attribute, dtls-srtp-ktr-server, which indicates the IP address and port of DTLS-SRTP server associated with the media line.

There would need to be some communication between the KTR servers to communicate new SRTP keys to the listeners. This communication is for future study.

3.5.  Interworking with Other SRTP Key Management Systems

3.5.1.  Security Descriptions

Today, Security Descriptions (Andreasen, F., Baugher, M., and D. Wing, “Session Description Protocol (SDP) Security Descriptions for Media Streams,” July 2006.) [RFC4568] is used for distributing SRTP keys in several different IP PBX systems and is expected to be used by 3GPP's Long Term Evolution (LTE). The IP PBX systems are typically used within a single enterprise, and LTE is used within the confines of a mobile operator's network. A Session Border Controller is a reasonable solution to interwork between Security Descriptions in one network and DTLS-SRTP in another network. For example, a mobile operator (or an Enterprise) could operate Security Descriptions within their network and DTLS-SRTP towards the Internet.

However, due to the way Security Descriptions and DTLS-SRTP manage their SRTP keys, such an SBC has to authenticate, decrypt, re-encrypt, and re-authenticate the SRTP (and SRTCP) packets in one direction, as shown in Figure 6 (Interworking Security Descriptions and DTLS-SRTP), below. This is computationally expensive.

RFC4568 endpoint             SBC               DTLS-SRTP endpoint
       |                      |                       |
  1.   |---key=A------------->|                       |
  2.   |                      |<-DTLS-SRTP handshake->|
  3.   |<--key=B--------------|                       |
  4.   |                      |<--SRTP, encrypted w/B-|
  5.   |<-SRTP, encrypted w/B-|                       |
  6.   |-SRTP, encrypted w/A->|                       |
  7.   |            (decrypt, re-encrypt)             |
  8.   |                      |-SRTP, encrypted w/C-->|
       |                      |                       |

The message flow is as follows (similar steps occur with SRTCP):

1. The Security Descriptions (Andreasen, F., Baugher, M., and D. Wing, “Session Description Protocol (SDP) Security Descriptions for Media Streams,” July 2006.) [RFC4568] endpoint discloses its SRTP key to the SBC, using a=crypto in its SDP.
2. SBC completes DTLS-SRTP handshake. From this handshake, the SBC derives the SRTP key for traffic from the DTLS-SRTP endpoint (key B) and to the DTLS-SRTP endpoint (key C).
3. The SBC communicates the SRTP encryption key (key B) to the Security Descriptions endpoint (using a=crypto). (There is no way, with DTLS-SRTP, to communicate the Security Descriptions key to the DTLS-SRTP key endpoint.)
4. The DTLS-SRTP endpoint sends an SRTP key, encrypted with its key B. This is received by the SBC.
5. The received SRTP packet is simply forwarded; the SBC does not need to do anything with this packet as its key (key B) was already communicated in step 3.
6. The Security Descriptions endpoint sends an SRTP packet, encrypted with its key A.
7. The SBC has to authenticate and decrypt the SRTP packet (using key A), and re-encrypt it and generate an HMAC (using key C).
8. The SBC sends the new SRTP packet.

KTR helps to avoid the computationally expensive operation so the SBC does not need not perform any per-packet operations on the SRTP (or SRTCP) packets in either direction. With KTR the SBC can simply forward the SRTP (and SRTCP) packets in both directions without per-packet HMAC or cryptographic operations.

To accomplish this interworking, KTR must be supported on the DTLS-SRTP endpoint, which allows the SBC to transport the Security Description key to the KTR endpoint and send the DTLS-SRTP key to the Security Descriptions endpoint. This works equally well for both incoming and outgoing calls. An abbreviated message flow is shown in Figure 7 (Interworking Security Descriptions and KTR), below.

RFC4568 endpoint             SBC               DTLS-SRTP endpoint
       |                      |                       |
  1.   |---key=A------------->|                       |
  2.   |                      |<-DTLS-SRTP handshake->|
  3.   |<--key=B--------------|                       |
  4.   |                      |--new_srtp_key:A------>|
  5.   |                      |<--SRTP, encrypted w/B-|
  5.   |<-SRTP, encrypted w/B-|                       |
  6.   |-SRTP, encrypted w/A->|                       |
  7.   |                      |-SRTP, encrypted w/A-->|
       |                      |                       |

The message flow is as follows (similar steps occur with SRTCP):

1. Security Descriptions endpoint discloses its SRTP key to the SBC (a=crypto).
2. SBC completes DTLS-SRTP handshake. From this handshake, the SBC derives the SRTP key for traffic from the DTLS-SRTP endpoint (key B) and to the DTLS-SRTP endpoint (key C).
3. The SBC communicates the SRTP encryption key (key B) to the Security Descriptions endpoint.
4. The SBC uses the KTR to indicate the key (key A) the SBC will encrypt packets with key A to the DTLS-SRTP endpoint.
5. The DTLS-SRTP endpoint sends an SRTP key, encrypted with its key B. This is received by the SBC.
6. The received SRTP packet is simply forwarded; the SBC does not need to do anything with this packet as its key (key B) was communicated in step 3.
7. The Security Descriptions endpoint sends an SRTP packet, encrypted with its key A.
8. The received SRTP packet is simply forwarded; the SBC does not need to do anything with this packet as its key (key A) was communicated in step 4.

4.  Protocol Description

This section describes the extension to the DTLS protocol for KTR, which allows securely communicating the SRTP key to the DTLS peer.

4.1.  key_transport (KTR) extension to DTLS-SRTP

This document adds a new negotiated extension called "key_transport", which MUST only be requested in conjunction with the "use_srtp" extension (Section 3.2 of [I‑D.ietf‑avt‑dtls‑srtp] (McGrew, D. and E. Rescorla, “Datagram Transport Layer Security (DTLS) Extension to Establish Keys for Secure Real-time Transport Protocol (SRTP),” February 2009.)). The DTLS server indicates its support for key_transport by including key_transport in its ServerHello message. If a DTLS client includes key_transport in its ClientHello, but does not receive key_transport in the ServerHello, the DTLS client MUST NOT send DTLS packets with the srtp_key_transport content-type.

Support for the DTLS Key Transport extension is indicated in SDP with the "srtp-kt" attribute. Advertising support for the extension is necessary in SDP because in some cases it is required to establish an SRTP call. For example, a mixer may be able to only support SRTP listeners if those listeners implement DTLS Key Transport (because it lacks the CPU cycles necessary to encrypt SRTP uniquely for each listener).

A message flow showing a DTLS client and DTLS server using the key_transport extension

Client                                               Server

ClientHello + use_srtp + key_transport
                             -------->
                     ServerHello + use_srtp + key_transport
                                               Certificate*
                                         ServerKeyExchange*
                                        CertificateRequest*
                             <--------      ServerHelloDone
Certificate*
ClientKeyExchange
CertificateVerify*
[ChangeCipherSpec]
Finished                     -------->
                                         [ChangeCipherSpec]
                             <--------             Finished
SRTP packets                 <------->      SRTP packets

After successful negotiation of the key_transport extension, the DTLS client and server MAY exchange normal SRTP packets (i.e., SRTP packets encrypted with keys derived from the KDF described in [I‑D.ietf‑avt‑dtls‑srtp] (McGrew, D. and E. Rescorla, “Datagram Transport Layer Security (DTLS) Extension to Establish Keys for Secure Real-time Transport Protocol (SRTP),” February 2009.)). This is normal and expected, even if Key Transport was negotiated by both sides, as neither side may (yet) have a need to alter the SRTP key. However, it is also possible that one (or both) peers will immediately send a new_srtp_key message before sending any SRTP.

4.2.  KTR Primitives

A new protocol is defined, called the srtp_key_transport protocol which uses srtp_key_transport content-type which consists of the following message types (primitives):

new_srtp_key_request:

request that the DTLS peer choose a new key. Valid responses are new_srtp_key and new_srtp_key_error.

new_srtp_key:

contains the new SRTP keying material, the master key, master salt, SSRC, rollover counter, and sequence number. This message is sent by a DTLS endpoint that wants to change its SRTP key beginning at the indicated sequence number. This does not change any cryptographic parameters (a new DTLS handshake is necessary for that), but only the DTLS key for the associated SRTP session. This message includes the SSRC that will be used for this key, which allows listeners to establish one SRTP crypto-context per speaker (necessary for the video switching scenario). The key chosen MUST be cryptographically random [RFC4086] (Eastlake, D., Schiller, J., and S. Crocker, “Randomness Requirements for Security,” June 2005.). This master keying material is processed by the standard SRTP key deriviation function (Section 4.3.1 of SRTP (Baugher, M., McGrew, D., Naslund, M., Carrara, E., and K. Norrman, “The Secure Real-time Transport Protocol (SRTP),” March 2004.) [RFC3711]) to provide the session keys.

new_key_activate:

indicates receiver is prepared to receive SRTP packets encrypted with the new key.

ekt_key

The key used by EKT to encrypt its RTCP payload.

new_srtp_key_failure:

indicates a failure.

At any time, the DTLS client or DTLS server MAY send a key_transport message, as shown in Figure 9 (New Key Message Flow). The sender of the new_srtp_key message MAY immediately start transmitting SRTP packets with this new key. However, to account for loss of the new_srtp_key message it is RECOMMENDED that the sender wait before changing to the new SRTP key until it receives the new_key_activate message or it times out waiting for the new_key_activate_message. The duration of this timeout may vary depending on the sensitivity of the content (e.g., 1 second or 10 seconds). In any case, the new_srtp_key message is retransmitted until acknowledged by receipt of a new_key_activate message.

Client / Server                             Server / Client

[new_srtp_key_request]        -------->
                             <--------         new_srtp_key
new_srtp_key_activiate        -------->

The following figure shows the state machine for the protocol.

      receive new_srtp_key_request from peer
         or decide to choose new SRTP key
                     |
                     |
send                 V
new_srtp_key  +---------------+    timeout
    +---------| Communicate   |--------+
    |         |     Key       |        |
    +-------->|               |        |
              +---------------+        |
                |           ^          |
     receive    |           |   +----------------+
new_key_activate|           +---| send SRTP using|
                |               |  new SRTP key  |
        +----------------+      +----------------+
        | send SRTP using|
        |  new SRTP key  |
        +----------------+
                |
                V
               done

Using the syntax described in DTLS (Rescorla, E. and N. Modadugu, “Datagram Transport Layer Security version 1.2,” October 2009.) [I‑D.ietf‑tls‑rfc4347‑bis], the following structures are used:

enum {
   new_srtp_key_request(0),
   new_srtp_key(1),
   new_srtp_key_activate(2),
   ekt_key(3),
   new_srtp_key_failure(128),
   (255)
} SRTPKeyTransportType;

struct {
   SRTPKeyTransportType keytrans_type;
   uint24 length;
   uint16 message_seq;
   uint24 fragment_offset;
   uint24 fragment_length;
   select (SRTPKeyTransportType) {
      case new_srtp_key_request:         NewSRTPKeyRequest;
      case new_srtp_key:                 NewSRTPKey;
      case new_srtp_key_activate:        NewSRTPKeyActivate;
      case new_srtp_key_failure:         NewSRTPKeyFailure;
      case ekt_key:                      EKTkey;
    };
} KeyTransport;

struct {
    uint  random<64>;           // additional entropy for peer
} NewSRTPKeyRequest;

struct {
    boolean any_ssrc;           // true=this key applies to any SSRC
    uint32 ssrc;                // SSRC used for this key
    uint   key<16..32>;         // change_cipher_spec decides
    uint   auth_tag<4..10>;     // the key and auth_tag length.
    uint   salt<112>;
    uint   roc<32>;             // SRTP rollover counter
    uint   sequence<16>;        // beginning SRTP sequence number
    uint   random<64>;          // additional entropy for peer
} NewSRTPKey;

struct {
    uint  random<64>;           // additional entropy for peer
} NewSRTPKeyActivate;

enum {
  AES_128(0),
  AESKW_128(1),
  AESKW_192(2),
  AESKW_256(3),
} ektcipher;

struct {
   ektcipher EKT_Cipher;
   uint EKT_Key_Value<1..256>;
   uint16 EKT_SPI;
} EKTkey;

struct { } NewSRTPKeyFailure;

4.3.  Procedures for Network Elements

A 'speaker' is an endpoint that terminates the DTLS-SRTP exchange and also sends SRTP data towards its peer(s). This is usually indicated by 'sendrecv' or 'sendonly'.

A 'listener' is an endpoint that terminates the DTLS-SRTP exchange and also receives SRTP data from its peer(s). This is usually indicated by 'sendrecv' or 'recvonly'.

As the Key Transport extension was negotiated during the DTLS-SRTP handshake, an endpoint can send Key Transport primitives, and can become a speaker or become a listener, at any point.

4.3.1.  Speaker

When a new speaker joins, the speaker can immediately send SRTP using the key derived from the DTLS-SRTP handshake. There is no scaling advantage to all of the speakers using the same key (because their content is different), and if the speakers did use the same key it would also introduce the risk of a two-time pad.

Once a speaker begins sending SRTP packets using a key communicated via KTR (i.e., the NewSRTPKey primitive), the speaker MUST NOT revert to using the SRTP key derived from the DTLS-SRTP handshake.

If the speaker wants to use KTR, or is requested to change its SRTP key (via the NewSRTPKeyRequest primitive), the speaker chooses a new SRTP master key and salt, and chooses a sequence number a reasonable distance in the future (1 second is recommended). The speaker then sends this new key using the NewSRTPKey primitive. The NewSRTPKey primitive message is re-transmitted until acknowledged with a NewKeyActivate message. No matter if a NewKeyActiviate is received or not, the speaker changes keys at its previously-chosen sequence number.

4.3.2.  Mixer

When a new speaker joins a mixer, the speaker does not need to support KTR, and no KTR procedures need to occur with the speaker. This is because the listener needs to decrypt and examine the speaker's stream, and the mixer will mix, re-originate (with its own SSRC) and re-encrypt the speaker's stream to the listeners.

The mixer functions as a speaker (Section 4.3.1 (Speaker)) towards the listeners connected to the mixer.

When a speaker leaves, there is no need to propagate that information beyond the mixer.

When a listener joins or leaves, the mixer MUST rekey all of the listeners based on the conference policy (Section 6.1 (Group Policy when Joining/Leaving)).

4.3.3.  Switcher

When a new speaker joins, the switcher communicates the speaker's key to all listeners using the NewSRTPKey primitive. In this way, whenever one of the speakers becomes the active speaker, the active speaker's SRTP can be immediately sent to all listeners.

In the event there are a large number of (potentially active) speakers and it is not feasible to inform all listeners of all speaker's keys, the switcher MAY decide to defer informing listeneners of a speaker's key until the speaker becomes the active speaker. This can cause some clipping when a speaker becomes the active speaker.

4.3.4.  Listener

When a listener joins, the listener is provided the same SRTP master key as the other listeners. This is done with the NewSRTPKey primitive. SRTP master keys are associated with both an SSRC and the RTP sequence number. A single SRTP stream might have multiple keys active at any point in time, such as when other listeners are joining or leaving. For example, two NewSRTPKey primitives can indicate that for a single SSRC value, key "A" is for sequence numbers 100-200, and key B is for 200-300.

If a listener is also a speaker, it also follows the rules of a speaker.

A listener can receive an SRTP packet with an unknown SSRC which could caused by either:

• the speaker changed its SSRC (due to SSRC collision)
• the speaker changed its SRTP master key

the listener can attempt to authenticate the packet using the most-recently-used SRTP master key, which helps in the first case. If the second case has occurred, the listener can only wait until the sender (the speaker, the mixer, or the switcher) sends a NewSRTPKey primitive.

4.4.  Key Transport SSRC and RTP SSRC Collisions

Per [RFC3550] (Schulzrinne, H., Casner, S., Frederick, R., and V. Jacobson, “RTP: A Transport Protocol for Real-Time Applications,” July 2003.), if an RTP source notices an RTP SSRC collision, it is required to change its SSRC. If it has negotiated support for KTR, it then MUST also send a NewSRTPKey message indicating the new SSRC. The communication of the new SSRC is necessary if there are multiple speakers in the video switching scenario. However, because a speaker is not able to determine if their audio or their video is being switched, a speaker MUST always indicate a change in SSRC by following the procedure in this section for any SRTP stream (audio, video, or other).

When this is done, in order to prevent clipping in listeners, it is RECOMMENDED that the speaker retain the same SRTP master key and salt.

4.5.  Fragmentation, Reassembly, and Retransmission

Much like the DTLS handshake itself, the KTR extension also needs to handle fragmentation and reassembly (to send a large key) and retransmission (to account for packet loss). This is to allow communicating SRTP keys which are longer than the network MTU. The same technique as DTLS's handshake are used to provide this function: message_seq, fragment_offset, and fragment_length.

When transmitting the key transport message, the sender divides the message into a series of N contiguous data ranges. These ranges MUST NOT be larger than the maximum handshake fragment size and MUST jointly contain the entire key transport message. The ranges SHOULD NOT overlap. The sender then creates N key transport messages, all with the same message_seq value as the original key transport message. Each new message is labelled with the fragment_offset (the number of bytes contained in previous fragments) and the fragment_length (the length of this fragment). The length field in all messages is the same as the length field of the original message. An unfragmented message is a degenerate case with fragment_offset=0 and fragment_length=length.

When a DTLS implementation receives a key transport message fragment, it MUST buffer it until it has the entire key transport message. DTLS implementations MUST be able to handle overlapping fragment ranges. This allows senders to retransmit key transport messages with smaller fragment sizes during path MTU discovery.

4.6.  SDP extensions

Two new SDP attributes are defined, dtls-srtp-ktr and dtls-srtp-ktr-server. dtls-srtp-ktr merely indicates the endpoint is capable of DTLS-SRTP-KTR, and is helpful to diagnose interoperability issues. dtls-srtp-ktr-server causes the DTLS handshake to occur with a different host than that indicated by the c/m lines, which is useful to help offload computational effort from the speaker (Section 3.4.2 (Distributed Key Servers)). Either attribute can appear at the media level or session level.

The ABNF (Crocker, D. and P. Overell, “Augmented BNF for Syntax Specifications: ABNF,” January 2008.) [RFC5234] for new SDP (Handley, M., Jacobson, V., and C. Perkins, “SDP: Session Description Protocol,” July 2006.) [RFC4566] attributes is as follows:

  ktr-server  = "dtls-srtp-ktr-server:" port
                [space nettype space addrtype
                 space connection-address]
  ktr-capable = "dtls-srtp-ktr"

Only the port is required; if the nettype is not indicated, the network type, address type, and connection-address are all the same as on the associated c= line.

5.  Examples

The following example shows how Key Transport would be requested in an offer, using "a=dtls-srtp-kt".

      v=0
      o=- 25678 753849 IN IP4 192.0.2.1
      s=
      c=IN IP4 192.0.2.1
      t=0 0
      m=audio 53456 UDP/TLS/RTP/SAVP 0
      a=fingerprint:SHA-1 \
        4A:AD:B9:B1:3F:82:18:3B:54:02:12:DF:3E:5D:49:6B:19:E5:7C:AB
      a=dtls-srtp-ktr

Using the SDP syntax described in [I‑D.ietf‑mmusic‑sdp‑capability‑negotiation] (Andreasen, F., “SDP Capability Negotiation,” March 2010.), the following figure shows an offerer that requires DTLS Key Transport in order to set up this call as an SRTP call, otherwise it can set up this call as an RTP call. This is indicated by the ",2" on the "a=pcfg" line. If the answerer does not understand "a=dtls-srtp-kt" but does understand DTLS-SRTP and [I‑D.ietf‑mmusic‑sdp‑capability‑negotiation] (Andreasen, F., “SDP Capability Negotiation,” March 2010.), this can cannot be established using DTLS-SRTP; however, it can be established using RTP.

      v=0
      o=- 25678 753849 IN IP4 192.0.2.1
      s=
      c=IN IP4 192.0.2.1
      t=0 0
      m=audio 53456 RTP/AVP 0
      a=tcap:1 UDP/TLS/RTP/SAVP
      a=acap:1 fingerprint:SHA-1 \
        4A:AD:B9:B1:3F:82:18:3B:54:02:12:DF:3E:5D:49:6B:19:E5:7C:AB
      a=acap:2 dtls-srtp-ktr
      a=pcfg:1 t=1 a=1,2

Using the SDP syntax described in [I‑D.ietf‑mmusic‑sdp‑capability‑negotiation] (Andreasen, F., “SDP Capability Negotiation,” March 2010.), the following figure shows an offerer that indicates support for DTLS Key Transport but does not require DTLS Key Transport in order to set up this call as an SRTP call. This is indicated by the ",[2]" on the "a=pcfg" line. If the answerer does not understand "a=dtls-srtp-kt" but does understand DTLS-SRTP and [I‑D.ietf‑mmusic‑sdp‑capability‑negotiation] (Andreasen, F., “SDP Capability Negotiation,” March 2010.), this call can still be established using DTLS-SRTP.

      v=0
      o=- 25678 753849 IN IP4 192.0.2.1
      s=
      c=IN IP4 192.0.2.1
      t=0 0
      m=audio 53456 RTP/AVP 0
      a=tcap:1 UDP/TLS/RTP/SAVP
      a=acap:1 fingerprint:SHA-1 \
          4A:AD:B9:B1:3F:82:18:3B:54:02:12:DF:3E:5D:49:6B:19:E5:7C:AB
      a=acap:2 dtls-srtp-ktr
      a=pcfg:1 t=1 a=1,[2]

The following example shows a Key Transport offer where the DTLS-SRTP-KTR exchange occurs with another server.

      v=0
      o=- 25678 753849 IN IP4 192.0.2.1
      s=
      c=IN IP4 192.0.2.1
      t=0 0
      m=audio 53456 UDP/TLS/RTP/SAVP 0
      a=fingerprint:SHA-1 \
        4A:AD:B9:B1:3F:82:18:3B:54:02:12:DF:3E:5D:49:6B:19:E5:7C:AB
      a=dtls-srtp-ktr
      a=dtls-srtp-ktr-server:37382 IN IP4 192.0.2.2

6.  Security Considerations

In the point-to-multipoint scenario, Section 3.1 (Point to Multipoint using the RFC 3550 mixer model), all of the listeners know the key being used by the mixer. Any of those listeners could create SRTP packets that are encrypted with this same key, and send those SRTP packets to other listeners. In order to reduce the vulnerability to this threat, it is RECOMMENDED that the source transport address of received SRTP packets be discarded if they do not match the source transport address of the associated DTLS-SRTP session. Additionally, the network SHOULD prevent IP address spoofing [RFC2827] (Ferguson, P. and D. Senie, “Network Ingress Filtering: Defeating Denial of Service Attacks which employ IP Source Address Spoofing,” May 2000.).

6.1.  Group Policy when Joining/Leaving

When sharing SRTP keys with several listeners, it is imperative that the SRTP is changed when a new listener is added or a listener is removed. This is because a legitimate listener should only be able to decrypt the SRTP stream while he is listening; he should not be able to decrypt the SRTP stream prior to joining the conference or after leaving the conference. Failing to change the key when a listener joins (or leaves) allows a listener to decrypt SRTP traffic prior to (or after) they are authorized participants in the conference.

Policies for a specific user's access to a conference may be different from conference to conference. For example, a company-internal event announcing promotions might be accessible to all employees and have no need for re-keying when listeners join or leave the conference. As another example, a conference where a job candidate is interviewed should be rekeyed when the job candidate joins the conference and again when the job candidate leaves the conference.

The protocol described in this document allows whichever policy is needed for a particular situation. The protocol itself does not enforce a certain policy; that is, the protocol itself does not ensure the SRTP key is changed when a listener leaves (or joins) the conference.

The RTP sequence number in the NewSRTPKey primitive allows the old key to be used for a predictable period of time before switching to the new key. This can provide sufficient time for all listeners to learn the new SRTP key before the sender switches to the new key.

6.2.  Two-Time Pad

In some scenarios, different data is sent to different participants. For example, in the audio mixer scenario, the active speaker receives a different stream than the other listeners; the active speaker's stream does not contain the active speaker's own input. It is critical that the same SRTP key is not used for the different content, or else a (so-called) "two-time pad" occurs (Section 9.1 of [RFC3711] (Baugher, M., McGrew, D., Naslund, M., Carrara, E., and K. Norrman, “The Secure Real-time Transport Protocol (SRTP),” March 2004.)).

With KTR, the sender is wholly responsible for choosing its own SRTP key. Thus, implementations should ensure that different SRTP keys are used whenever different data is sent.

7.  Acknowledgements

Thanks to David McGrew for his improvements to this document and to the underlying protocol. Thanks to Brian Weis, Sheela Rowles, and Mark Baugher for suggesting how GDOI-SRTP's key management could be used by DTLS-SRTP.

Thanks to Flemming Andreasen for the reminder regarding two-time pads, to John Floroiu for reminder of salting key.

8.  IANA Considerations

[[This section will be completed in a future version of this document.]]

To do:

• Register new SDP attribute "dtls-srtp-ktr"
• Register new SDP attribute "dtls-srtp-ktr-server"
• new TLS content-type "key_transport" (26?)

9.  References

9.1. Normative References

9.2. Informational References

Appendix A.  Changes

[[Note to RFC Editor: Please remove this section prior to publication.]]

A.1.  Changes from -00 to -01

• more closely aligned with RTP Topologies (Westerlund, M. and S. Wenger, “RTP Topologies,” January 2008.) [RFC5117]
• added multicast scenario
• added voicemail storage/retrieval scenario
• added delete_srtp_key
• added your_new_srtp_key
• aligned SDP for DTLS-SRTP with draft-ietf-mmusic-sdp-dtls-00
• key change rules are now discussed in Security Considerations

A.2.  Changes from -01 to -02

• removed voicemail storage/retrieval scenario -- SPEECHSC found such a scenario does not work
• Described what occurs when a speaker joins/leaves, a listener joins/leaves, and what a mixer/switcher does (Section 4.3 (Procedures for Network Elements)).
• Removed primitives that can allow two-time pad.
• Added scenario for interworking with Security Descriptions
• Describe relationship with EKT (McGrew, D., Andreasen, F., Wing, D., and L. Dondeti, “Encrypted Key Transport for Secure RTP,” October 2009.) [I‑D.mcgrew‑srtp‑ekt]

A.3.  Changes from -02 to -03

• removed latent your_new_key text that was missed during previous editing.
• removed LKH to an appendix.
• described how DTLS-SRTP-KTR works with EKT. Added new ekt_key primitive to support EKT.

Appendix B.  LKH for More Efficient Rekeying

A more efficient mechanism to rekey SRTP is to use a subset difference based key management scheme [I‑D.irtf‑smug‑subsetdifference] (Lotspiech, J., Naor, M., and D. Naor, “Subset-Difference based Key Management for Secure Multicast,” .). In this scheme, the key server (the speaker) can send a single message so that every authorized listener -- but no unauthorized listeners -- can decrypt the message. The message contains the new SRTP key. The advantage of this scheme is that subset difference allows the message to be encrypted just once, no matter how many listeners. An implementation of subset-difference based key management is Logical Key Heirarchy (LKH) (Wallner, D., Harder, E., and R. Agee, “Key Management for Multicast: Issues and Architectures,” June 1999.) [RFC2627]), which is useful for unicast and multicast.

Extending DTLS-SRTP-Key-Transport to support LKH is for future study.

Author's Address