Completely Encrypting RTP Header Extensions and Contributing Sources

justin@uberti.name

Cisco

fluffy@iii.ca

Millicast

sergio.garcia.murillo@cosmosoftware.io

ART AVTCORE SRTP While the Secure Real-time Transport Protocol (SRTP) provides confidentiality for the contents of a media packet, a significant amount of metadata is left unprotected, including RTP header extensions and contributing sources (CSRCs). However, this data can be moderately sensitive in many applications. While there have been previous attempts to protect this data, they have had limited deployment, due to complexity as well as technical limitations. This document updates RFC 3711, the SRTP specification, and defines Cryptex as a new mechanism that completely encrypts header extensions and CSRCs and uses simpler Session Description Protocol (SDP) signaling with the goal of facilitating deployment.

Introduction

Problem Statement The Secure Real-time Transport Protocol (SRTP) mechanism provides message authentication for the entire RTP packet but only encrypts the RTP payload. This has not historically been a problem, as much of the information carried in the header has minimal sensitivity (e.g., RTP timestamp); in addition, certain fields need to remain as cleartext because they are used for key scheduling (e.g., RTP synchronization source (SSRC) and sequence number). However, as noted in , the security requirements can be different for information carried in RTP header extensions, including the per-packet sound levels defined in and , which are specifically noted as being sensitive in the Security Considerations sections of those RFCs. In addition to the contents of the header extensions, there are now enough header extensions in active use that the header extension identifiers themselves can provide meaningful information in terms of determining the identity of the endpoint and/or application. Accordingly, these identifiers can be considered a fingerprinting issue. Finally, the CSRCs included in RTP packets can also be sensitive, potentially allowing a network eavesdropper to determine who was speaking and when during an otherwise secure conference call.

Previous Solutions Encryption of Header Extensions in SRTP was proposed in 2013 as a solution to the problem of unprotected header extension values. However, it has not seen significant adoption and has a few technical shortcomings. First, the mechanism is complicated. Since it allows encryption to be negotiated on a per-extension basis, a fair amount of signaling logic is required. And in the SRTP layer, a somewhat complex transform is required to allow only the selected header extension values to be encrypted. One of the most popular SRTP implementations had a significant bug in this area that was not detected for five years. Second, the mechanism only protects the header extension values and not their identifiers or lengths. It also does not protect the CSRCs. As noted above, this leaves a fair amount of potentially sensitive information exposed. Third, the mechanism bloats the header extension space. Because each extension must be offered in both unencrypted and encrypted forms, twice as many header extensions must be offered, which will in many cases push implementations past the 14-extension limit for the use of one-byte extension headers defined in . Accordingly, in many cases, implementations will need to use two-byte headers, which are not supported well by some existing implementations. Finally, the header extension bloat combined with the need for backward compatibility results in additional wire overhead. Because two-byte extension headers may not be handled well by existing implementations, one-byte extension identifiers will need to be used for the unencrypted (backward-compatible) forms, and two-byte for the encrypted forms. Thus, deployment of encryption for header extensions will typically result in multiple extra bytes in each RTP packet, compared to the present situation.

Goals From the previous analysis, the desired properties of a solution are:

Built on the existing SRTP framework (simple to understand)
Built on the existing header extension framework (simple to implement)
Protection of header extension identifiers, lengths, and values
Protection of CSRCs when present
Simple signaling
Simple crypto transform and SRTP interactions
Backward compatibility with unencrypted endpoints, if desired
Backward compatibility with existing RTP tooling

The last point deserves further discussion. While other possible solutions that would have encrypted more of the RTP header (e.g., the number of CSRCs) were considered, the inability to parse the resultant packets with current tools and a generally higher level of complexity outweighed the slight improvement in confidentiality in these solutions. Hence, a more pragmatic approach was taken to solve the problem described in .

Terminology The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here.

Design This specification proposes a mechanism to negotiate encryption of all RTP header extensions (ids, lengths, and values) as well as CSRC values. It reuses the existing SRTP framework, is accordingly simple to implement, and is backward compatible with existing RTP packet parsing code, even when support for the mechanism has been negotiated. Except when explicitly stated otherwise, Cryptex reuses all the framework procedures, transforms, and considerations described in .

SDP Considerations Cryptex support is indicated via a new "a=cryptex" SDP attribute defined in this specification. The new "a=cryptex" attribute is a property attribute as defined in ; it therefore takes no value and can be used at the session level or media level. The presence of the "a=cryptex" attribute in the SDP (in either an offer or an answer) indicates that the endpoint is capable of receiving RTP packets encrypted with Cryptex, as defined below. Once each peer has verified that the other party supports receiving RTP packets encrypted with Cryptex, senders can unilaterally decide whether or not to use the Cryptex mechanism on a per-packet basis. If BUNDLE is in use as per and the "a=cryptex" attribute is present for a media line, it MUST be present for all RTP-based "m=" sections belonging to the same bundle group. This ensures that the encrypted Media Identifier (MID) header extensions can be processed, allowing RTP streams to be associated with the correct "m=" section in each BUNDLE group as specified in . When used with BUNDLE, this attribute is assigned to the TRANSPORT category . Both endpoints can change the Cryptex support status by modifying the session as specified in . Generating subsequent SDP offers and answers MUST use the same procedures for including the "a=cryptex" attribute as the ones on the initial offer and answer.

RTP Header Processing A General Mechanism for RTP Header Extensions defines two values for the "defined by profile" field for carrying one-byte and two-byte header extensions. In order to allow a receiver to determine if an incoming RTP packet is using the encryption scheme in this specification, two new values are defined:

0xC0DE for the encrypted version of the one-byte header extensions (instead of 0xBEDE).
0xC2DE for the encrypted versions of the two-byte header extensions (instead of 0x100).

In the case of using two-byte header extensions, the extension identifier with value 256 MUST NOT be negotiated, as the value of this identifier is meant to be contained in the "appbits" of the "defined by profile" field, which are not available when using the values above. Note that as per , it is not possible to mix one-byte and two-byte headers on the same RTP packet. Mixing one-byte and two-byte headers on the same RTP stream requires negotiation of the "extmap-allow-mixed" SDP attribute as defined in . Peers MAY negotiate both Cryptex and the Encryption of Header Extensions mechanism defined in via SDP offer/answer as described in , and if both mechanisms are supported, either one can be used for any given packet. However, if a packet is encrypted with Cryptex, it MUST NOT also use header extension encryption , and vice versa. If one of the peers has advertised the ability to receive both Cryptex and header extensions encrypted as per in the SDP exchange, it is RECOMMENDED that the other peer use Cryptex rather than the mechanism in when sending RTP packets so that all the header extensions and CSRCS are encrypted. However, if there is a compelling reason to use the mechanism in (e.g., a need for some header extensions to be sent in the clear so that so they are processable by RTP middleboxes), the other peer SHOULD use the mechanism in instead.

Sending When the mechanism defined by this specification has been negotiated, sending an RTP packet that has any CSRCs or contains any header extensions follows the steps below. This mechanism MUST NOT be used with header extensions other than the variety described in . If the RTP packet contains one-byte headers, the 16-bit RTP header extension tag MUST be set to 0xC0DE to indicate that the encryption has been applied and the one-byte framing is being used. If the RTP packet contains two-byte headers, the header extension tag MUST be set to 0xC2DE to indicate encryption has been applied and the two-byte framing is being used. If the packet contains CSRCs but no header extensions, an empty extension block consisting of the 0xC0DE tag and a 16-bit length field set to zero (explicitly permitted by ) MUST be appended, and the X bit MUST be set to 1 to indicate an extension block is present. This is necessary to provide the receiver an indication that the CSRCs in the packet are encrypted. The RTP packet MUST then be encrypted as described in ("Encryption Procedure").

Receiving When receiving an RTP packet that contains header extensions, the "defined by profile" field MUST be checked to ensure the payload is formatted according to this specification. If the field does not match one of the values defined above, the implementation MUST instead handle it according to the specification that defines that value. Alternatively, if the implementation considers the use of this specification mandatory and the "defined by profile" field does not match one of the values defined above, it MUST stop the processing of the RTP packet and report an error for the RTP stream. If the RTP packet passes this check, it is then decrypted as described in ("Decryption Procedure") and passed to the next layer to process the packet and its extensions. In the event that a zero-length extension block was added as indicated above, it can be left as is and will be processed normally.

Encryption and Decryption

Packet Structure When this mechanism is active, the SRTP packet is protected as follows:

A Protected SRTP Packet +=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+ | | | contributing source (CSRC) identifiers | | | | .... | | +>+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | X | 0xC0 or 0xC2 | 0xDE | length | | +>+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | | RFC 8285 header extensions | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | | payload ... | | | | +-------------------------------+ | | | | RTP padding | RTP pad count | | +>+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+<+ | ~ SRTP Master Key Identifier (MKI) (OPTIONAL) ~ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | : authentication tag (RECOMMENDED) : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | | +- Encrypted Portion Authenticated Portion ---+ ]]> Note that, as required by , the 4 bytes at the start of the extension block are not encrypted. Specifically, the Encrypted Portion MUST include any CSRC identifiers, any RTP header extension (except for the first 4 bytes), and the RTP payload.

Encryption Procedure The encryption procedure is identical to that of except for the Encrypted Portion of the SRTP packet. The plaintext input to the cipher is as follows: Here "header extension data" refers to the content of the RTP extension field, excluding the first four bytes (the extension header ). The first 4 * CSRC count (CC) bytes of the ciphertext are placed in the CSRC field of the RTP header. The remainder of the ciphertext is the RTP payload of the encrypted packet. To minimize changes to surrounding code, the encryption mechanism can choose to replace a "defined by profile" field from with its counterpart defined in ("RTP Header Processing") and encrypt at the same time. For Authenticated Encryption with Associated Data (AEAD) ciphers (e.g., AES-GCM), the 12-byte fixed header and the four-byte header extension header (the "defined by profile" field and the length) are considered additional authenticated data (AAD), even though they are non-contiguous in the packet if CSRCs are present. Here "fixed header" refers to the 12-byte fixed portion of the RTP header, and "extension header" refers to the four-byte extension header ("defined by profile" and extension length). Implementations can rearrange a packet so that the AAD and plaintext are contiguous by swapping the order of the extension header and the CSRC identifiers, resulting in an intermediate representation of the form shown in . After encryption, the CSRCs (now encrypted) and extension header would need to be swapped back to their original positions. A similar operation can be done when decrypting to create contiguous ciphertext and AAD inputs.

An RTP Packet Transformed to Make Cryptex Cipher Inputs Contiguous +=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+<+ | | contributing source (CSRC) identifiers | | | | .... | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | | RFC 8285 header extensions | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | | payload ... | | | | +-------------------------------+ | | | | RTP padding | RTP pad count | | +>+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | | +- Plaintext Input AAD Input ---+ ]]> Note that this intermediate representation is only displayed as reference for implementations and is not meant to be sent on the wire.

Decryption Procedure The decryption procedure is identical to that of except for the Encrypted Portion of the SRTP packet, which is as shown in the section above. To minimize changes to surrounding code, the decryption mechanism can choose to replace the "defined by profile" field with its no-encryption counterpart from and decrypt at the same time.

Backward Compatibility This specification attempts to encrypt as much as possible without interfering with backward compatibility for systems that expect a certain structure from an RTPv2 packet, including systems that perform demultiplexing based on packet headers. Accordingly, the first two bytes of the RTP packet are not encrypted. This specification also attempts to reuse the key scheduling from SRTP, which depends on the RTP packet sequence number and SSRC identifier. Accordingly, these values are also not encrypted.

Security Considerations All security considerations in are applicable to this specification; the exception is Section , because confidentiality of the RTP Header is the purpose of this specification. The risks of using weak or NULL authentication with SRTP, described in , apply to encrypted header extensions as well. This specification extends SRTP by expanding the Encrypted Portion of the RTP packet, as shown in ("Packet Structure"). It does not change how SRTP authentication works in any way. Given that more of the packet is being encrypted than before, this is necessarily an improvement. The RTP fields that are left unencrypted (see rationale above) are as follows:

RTP version
padding bit
extension bit
number of CSRCs
marker bit
payload type
sequence number
timestamp
SSRC identifier
number of header extensions

These values contain a fixed set (i.e., one that won't be changed by extensions) of information that, at present, is observed to have low sensitivity. In the event any of these values need to be encrypted, SRTP is likely the wrong protocol to use and a fully encapsulating protocol such as DTLS is preferred (with its attendant per-packet overhead).

IANA Considerations This document updates the "attribute-name (formerly "att-field")" subregistry of the "Session Description Protocol (SDP) Parameters" registry (see ). Specifically, it adds the SDP "a=cryptex" attribute for use at both the media level and the session level.

Contact name:: IETF AVT Working Group or IESG if the AVT Working Group is closed
Contact email address:: avt@ietf.org
Attribute name:: cryptex
Attribute syntax:: This attribute takes no values.
Attribute semantics:: N/A
Attribute value:: N/A
Usage level:: session, media
Charset dependent:: No
Purpose:: The presence of this attribute in the SDP indicates that the endpoint is capable of receiving RTP packets encrypted with Cryptex as described in this document.
O/A procedures:: SDP O/A procedures are described in Section of this document.
Mux Category:: TRANSPORT

References Normative References Informative References

Test Vectors All values are in hexadecimal and represented in network order (big endian).

AES-CTR The following subsections list the test vectors for using Cryptex with AES-CTR as per . Common values are organized as follows:

RTP Packet with One-Byte Header Extension RTP Packet: Encrypted RTP Packet:

RTP Packet with Two-Byte Header Extension RTP Packet: Encrypted RTP Packet:

RTP Packet with One-Byte Header Extension and CSRC Fields RTP Packet: Encrypted RTP Packet:

RTP Packet with Two-Byte Header Extension and CSRC Fields RTP Packet: Encrypted RTP Packet:

RTP Packet with Empty One-Byte Header Extension and CSRC Fields RTP Packet: Encrypted RTP Packet:

RTP Packet with Empty Two-Byte Header Extension and CSRC Fields RTP Packet: Encrypted RTP Packet: