are used to pad shorter keys to 192 or 256 bits.
9.3. Confidentiality of the RTP Payload
SRTP’s pre-defined ciphers are "seekable" stream ciphers, i.e.,
ciphers able to efficiently seek to arbitrary locations in their
keystream (so that the encryption or decryption of one packet does
not depend on preceding packets). By using seekable stream ciphers,
SRTP avoids the denial of service attacks that are possible on stream
ciphers that lack this property. It is important to be aware that,
as with any stream cipher, the exact length of the payload is
revealed by the encryption. This means that it may be possible to
deduce certain "formatting bits" of the payload, as the length of the
codec output might vary due to certain parameter settings etc. This,
in turn, implies that the corresponding bit of the keystream can be
deduced. However, if the stream cipher is secure (counter mode and
f8 are provably secure under certain assumptions [BDJR] [KSYH] [IK]),
knowledge of a few bits of the keystream will not aid an attacker in
predicting subsequent keystream bits. Thus, the payload length (and
information deducible from this) will leak, but nothing else.
As some RTP packet could contain highly predictable data, e.g., SID,
it is important to use a cipher designed to resist known plaintext
attacks (which is the current practice).
9.4. Confidentiality of the RTP Header
In SRTP, RTP headers are sent in the clear to allow for header
compression. This means that data such as payload type,
synchronization source identifier, and timestamp are available to an
eavesdropper. Moreover, since RTP allows for future extensions of
headers, we cannot foresee what kind of possibly sensitive
information might also be "leaked".
SRTP is a low-cost method, which allows header compression to reduce
bandwidth. It is up to the endpoints’ policies to decide about the
security protocol to employ. If one really needs to protect headers,
and is allowed to do so by the surrounding environment, then one
should also look at alternatives, e.g., IPsec [RFC2401].
9.5. Integrity of the RTP payload and header
SRTP messages are subject to attacks on their integrity and source
identification, and these risks are discussed in Section 9.5.1. To
protect against these attacks, each SRTP stream SHOULD be protected
by HMAC-SHA1 [RFC2104] with an 80-bit output tag and a 160-bit key,
or a message authentication code with equivalent strength. Secure
RTP SHOULD NOT be used without message authentication, except under
the circumstances described in this section. It is important to note
that encryption algorithms, including AES Counter Mode and f8, do not
provide message authentication. SRTCP MUST NOT be used with weak (or
NULL) authentication.
SRTP MAY be used with weak authentication (e.g., a 32-bit
authentication tag), or with no authentication (the NULL
authentication algorithm). These options allow SRTP to be used to
provide confidentiality in situations where
* weak or null authentication is an acceptable security risk, and
* it is impractical to provide strong message authentication.
These conditions are described below and in Section 7.5. Note that
both conditions MUST hold in order for weak or null authentication to
be used. The risks associated with exercising the weak or null
authentication options need to be considered by a security audit
prior to their use for a particular application or environment given
the risks, which are discussed in Section 9.5.1.
Weak authentication is acceptable when the RTP application is such
that the effect of a small fraction of successful forgeries is
negligible. If the application is stateless, then the effect of a
single forged RTP packet is limited to the decoding of that
particular packet. Under this condition, the size of the
authentication tag MUST ensure that only a negligible fraction of the
packets passed to the RTP application by the SRTP receiver can be
forgeries. This fraction is negligible when an adversary, if given
control of the forged packets, is not able to make a significant
impact on the output of the RTP application (see the example of
Section 7.5).
Weak or null authentication MAY be acceptable when it is unlikely
that an adversary can modify ciphertext so that it decrypts to an
intelligible value. One important case is when it is difficult for
an adversary to acquire the RTP plaintext data, since for many
codecs, an adversary that does not know the input signal cannot
manipulate the output signal in a controlled way. In many cases it
may be difficult for the adversary to determine the actual value of
the plaintext. For example, a hidden snooping device might be
required in order to know a live audio or video signal. The
adversary’s signal must have a quality equivalent to or greater than
that of the signal under attack, since otherwise the adversary would
not have enough information to encode that signal with the codec used
by the victim. Plaintext prediction may also be especially difficult
for an interactive application such as a telephone call.
Weak or null authentication MUST NOT be used when the RTP application
makes data forwarding or access control decisions based on the RTP
data. In such a case, an attacker may be able to subvert
confidentiality by causing the receiver to forward data to an
attacker. See Section 3 of [B96] for a real-life example of such
attacks.
Null authentication MUST NOT be used when a replay attack, in which
an adversary stores packets then replays them later in the session,
could have a non-negligible impact on the receiver. An example of a
successful replay attack is the storing of the output of a
surveillance camera for a period of time, later followed by the
injection of that output to the monitoring station to avoid
surveillance. Encryption does not protect against this attack, and
non-null authentication is REQUIRED in order to defeat it.
If existential message forgery is an issue, i.e., when the accuracy
of the received data is of non-negligible importance, null
authentication MUST NOT be used.
9.5.1. Risks of Weak or Null Message Authentication
During a security audit considering the use of weak or null
authentication, it is important to keep in mind the following attacks
which are possible when no message authentication algorithm is used.
An attacker who cannot predict the plaintext is still always able to
modify the message sent between the sender and the receiver so that
it decrypts to a random plaintext value, or to send a stream of bogus
packets to the receiver that will decrypt to random plaintext values.
This attack is essentially a denial of service attack, though in the
absence of message authentication, the RTP application will have
inputs that are bit-wise correlated with the true value. Some
multimedia codecs and common operating systems will crash when such
data are accepted as valid video data. This denial of service attack
may be a much larger threat than that due to an attacker dropping,
delaying, or re-ordering packets.
An attacker who cannot predict the plaintext can still replay a
previous message with certainty that the receiver will accept it.
Applications with stateless codecs might be robust against this type
of attack, but for other, more complex applications these attacks may
be far more grave.
An attacker who can predict the plaintext can modify the ciphertext
so that it will decrypt to any value of her choosing. With an
additive stream cipher, an attacker will always be able to change
individual bits.
An attacker may be able to subvert confidentiality due to the lack of
authentication when a data forwarding or access control decision is
made on decrypted but unauthenticated plaintext. This is because the
receiver may be fooled into forwarding data to an attacker, leading
to an indirect breach of confidentiality (see Section 3 of [B96]).
This is because data-forwarding decisions are made on the decrypted
plaintext; information in the plaintext will determine to what subnet
(or process) the plaintext is forwarded in ESP [RFC2401] tunnel mode
(respectively, transport mode). When Secure RTP is used without
message authentication, it should be verified that the application
does not make data forwarding or access control decisions based on
the decrypted plaintext.
Some cipher modes of operation that require padding, e.g., standard
cipher block chaining (CBC) are very sensitive to attacks on
confidentiality if certain padding types are used in the absence of
integrity. The attack [V02] shows that this is indeed the case for
the standard RTP padding as discussed in reference to Figure 1, when
used together with CBC mode. Later transform additions to SRTP MUST
therefore carefully consider the risk of using this padding without
proper integrity protection.
9.5.2. Implicit Header Authentication
The IV formation of the f8-mode gives implicit authentication (IHA)
of the RTP header, even when message authentication is not used.
When IHA is used, an attacker that modifies the value of the RTP
header will cause the decryption process at the receiver to produce
random plaintext values. While this protection is not equivalent to
message authentication, it may be useful for some applications.
10. Interaction with Forward Error Correction mechanisms
The default processing when using Forward Error Correction (e.g., RFC
2733) processing with SRTP SHALL be to perform FEC processing prior
to SRTP processing on the sender side and to perform SRTP processing
prior to FEC processing on the receiver side. Any change to this
ordering (reversing it, or, placing FEC between SRTP encryption and
SRTP authentication) SHALL be signaled out of band.
11. Scenarios
SRTP can be used as security protocol for the RTP/RTCP traffic in
many different scenarios. SRTP has a number of configuration
options, in particular regarding key usage, and can have impact on
the total performance of the application according to the way it is
used. Hence, the use of SRTP is dependent on the kind of scenario
and application it is used with. In the following, we briefly
illustrate some use cases for SRTP, and give some guidelines for
recommended setting of its options.
11.1. Unicast
A typical example would be a voice call or video-on-demand
application.
Consider one bi-directional RTP stream, as one RTP session. It is
possible for the two parties to share the same master key in the two
directions according to the principles of Section 9.1. The first
round of the key derivation splits the master key into any or all of
the following session keys (according to the provided security
functions):
SRTP_encr_key, SRTP_auth_key, SRTCP_encr_key, and SRTCP_auth key.
(For simplicity, we omit discussion of the salts, which are also
derived.) In this scenario, it will in most cases suffice to have a
single master key with the default lifetime. This guarantees
sufficiently long lifetime of the keys and a minimum set of keys in
place for most practical purposes. Also, in this case RTCP
protection can be applied smoothly. Under these assumptions, use of
the MKI can be omitted. As the key-derivation in combination with
large difference in the packet rate in the respective directions may
require simultaneous storage of several session keys, if storage is
an issue, we recommended to use low-rate key derivation.
The same considerations can be extended to the unicast scenario with
multiple RTP sessions, where each session would have a distinct
master key.
11.2. Multicast (one sender)
Just as with (unprotected) RTP, a scalability issue arises in big
groups due to the possibly very large amount of SRTCP Receiver
Reports that the sender might need to process. In SRTP, the sender
may have to keep state (the cryptographic context) for each receiver,
or more precisely, for the SRTCP used to protect Receiver Reports.
The overhead increases proportionally to the size of the group. In
particular, re-keying requires special concern, see below.
Consider first a small group of receivers. There are a few possible
setups with the distribution of master keys among the receivers.
Given a single RTP session, one possibility is that the receivers
share the same master key as per Section 9.1 to secure all their
respective RTCP traffic. This shared master key could then be the
same one used by the sender to protect its outbound SRTP traffic.
Alternatively, it could be a master key shared only among the
receivers and used solely for their SRTCP traffic. Both alternatives
require the receivers to trust each other.
Considering SRTCP and key storage, it is recommended to use low-rate
(or zero) key_derivation (except the mandatory initial one), so that
the sender does not need to store too many session keys (each SRTCP
stream might otherwise have a different session key at a given point
in time, as the SRTCP sources send at different times). Thus, in
case key derivation is wanted for SRTP, the cryptographic context for
SRTP can be kept separate from the SRTCP crypto context, so that it
is possible to have a key_derivation_rate of 0 for SRTCP and a non-
zero value for SRTP.
Use of the MKI for re-keying is RECOMMENDED for most applications
(see Section 8.1).
If there are more than one SRTP/SRTCP stream (within the same RTP
session) that share the master key, the upper limit of 2^48 SRTP
packets / 2^31 SRTCP packets means that, before one of the streams
reaches its maximum number of packets, re-keying MUST be triggered on
ALL streams sharing the master key. (From strict security point of
view, only the stream reaching the maximum would need to be re-keyed,
but then the streams would no longer be sharing master key, which is
the intention.) A local policy at the sender side should force
rekeying in a way that the maximum packet limit is not reached on any
of the streams. Use of the MKI for re-keying is RECOMMENDED.
In large multicast with one sender, the same considerations as for
the small group multicast hold. The biggest issue in this scenario
is the additional load placed at the sender side, due to the state
(cryptographic contexts) that has to be maintained for each receiver,
sending back RTCP Receiver Reports. At minimum, a replay window
might need to be maintained for each RTCP source.
11.3. Re-keying and access control
Re-keying may occur due to access control (e.g., when a member is
removed during a multicast RTP session), or for pure cryptographic
reasons (e.g., the key is at the end of its lifetime). When using
SRTP default transforms, the master key MUST be replaced before any
of the index spaces are exhausted for any of the streams protected by
one and the same master key.
How key management re-keys SRTP implementations is out of scope, but
it is clear that there are straightforward ways to manage keys for a
multicast group. In one-sender multicast, for example, it is
typically the responsibility of the sender to determine when a new
key is needed. The sender is the one entity that can keep track of
when the maximum number of packets has been sent, as receivers may
join and leave the session at any time, there may be packet loss and
delay etc. In scenarios other than one-sender multicast, other
methods can be used. Here, one must take into consideration that key
exchange can be a costly operation, taking several seconds for a
single exchange. Hence, some time before the master key is
exhausted/expires, out-of-band key management is initiated, resulting
in a new master key that is shared with the receiver(s). In any
event, to maintain synchronization when switching to the new key,
group policy might choose between using the MKI and the <From, To>,
as described in Section 8.1.
For access control purposes, the <From, To> periods are set at the
desired granularity, dependent on the packet rate. High rate re-
keying can be problematic for SRTCP in some large-group scenarios.
As mentioned, there are potential problems in using the SRTP index,
rather than the SRTCP index, for determining the master key. In
particular, for short periods during switching of master keys, it may
be the case that SRTCP packets are not under the current master key
of the correspondent SRTP. Therefore, using the MKI for re-keying in
such scenarios will produce better results.
11.4. Summary of basic scenarios
The description of these scenarios highlights some recommendations on
the use of SRTP, mainly related to re-keying and large scale
multicast:
- Do not use fast re-keying with the <From, To> feature. It may, in
particular, give problems in retrieving the correct SRTCP key, if
an SRTCP packet arrives close to the re-keying time. The MKI
SHOULD be used in this case.
- If multiple SRTP streams in the same RTP session share the same
master key, also moderate rate re-keying MAY have the same
problems, and the MKI SHOULD be used.
- Though offering increased security, a non-zero key_derivation_rate
is NOT RECOMMENDED when trying to minimize the number of keys in
use with multiple streams.
12. IANA Considerations
The RTP specification establishes a registry of profile names for use
by higher-level control protocols, such as the Session Description
Protocol (SDP), to refer to transport methods. This profile
registers the name "RTP/SAVP".
SRTP uses cryptographic transforms which a key management protocol
signals. It is the task of each particular key management protocol
to register the cryptographic transforms or suites of transforms with
IANA. The key management protocol conveys these protocol numbers,
not SRTP, and each key management protocol chooses the numbering
scheme and syntax that it requires.
Specification of a key management protocol for SRTP is out of scope
here. Section 8.2, however, provides guidance on the parameters that
need to be defined for the default and mandatory transforms.
13. Acknowledgements
David Oran (Cisco) and Rolf Blom (Ericsson) are co-authors of this
document but their valuable contributions are acknowledged here to
keep the length of the author list down.
The authors would in addition like to thank Magnus Westerlund, Brian
Weis, Ghyslain Pelletier, Morgan Lindqvist, Robert Fairlie-
Cuninghame, Adrian Perrig, the AVT WG and in particular the chairmen
Colin Perkins and Stephen Casner, the Transport and Security Area
Directors, and Eric Rescorla for their reviews and support.
14. References
14.1. Normative References
[AES] NIST, "Advanced Encryption Standard (AES)", FIPS PUB 197,
http://www.nist.gov/aes/
[RFC2104] Krawczyk, H., Bellare, M. and R. Canetti, "HMAC: Keyed-
Hashing for Message Authentication", RFC 2104, February
1997.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
[RFC2401] Kent, S. and R. Atkinson, "Security Architecture for
Internet Protocol", RFC 2401, November 1998.
[RFC2828] Shirey, R., "Internet Security Glossary", FYI 36, RFC 2828,
May 2000.
[RFC3550] Schulzrinne, H., Casner, S., Frederick, R. and V. Jacobson,
"RTP: A Transport Protocol for Real-time Applications", RFC
3550, July 2003.
[RFC3551] Schulzrinne, H. and S. Casner, "RTP Profile for Audio and
Video Conferences with Minimal Control", RFC 3551, July
2003.
14.2. Informative References
[AES-CTR] Lipmaa, H., Rogaway, P. and D. Wagner, "CTR-Mode
Encryption", NIST, http://csrc.nist.gov/encryption/modes/
workshop1/papers/lipmaa-ctr.pdf
[B96] Bellovin, S., "Problem Areas for the IP Security
Protocols," in Proceedings of the Sixth Usenix Unix
Security Symposium, pp. 1-16, San Jose, CA, July 1996
(http://www.research.att.com/~smb/papers/index.html).
[BDJR] Bellare, M., Desai, A., Jokipii, E. and P. Rogaway, "A
Concrete Treatment of Symmetric Encryption: Analysis of DES
Modes of Operation", Proceedings 38th IEEE FOCS, pp. 394-
403, 1997.
[BS00] Biryukov, A. and A. Shamir, "Cryptanalytic Time/Memory/Data
Tradeoffs for Stream Ciphers", Proceedings, ASIACRYPT 2000,
LNCS 1976, pp. 1-13, Springer Verlag.
[C99] Crowell, W. P., "Introduction to the VENONA Project",
http://www.nsa.gov:8080/docs/venona/index.html.
[CTR] Dworkin, M., NIST Special Publication 800-38A,
"Recommendation for Block Cipher Modes of Operation:
Methods and Techniques", 2001.
http://csrc.nist.gov/publications/nistpubs/800-38a/sp800-
38a.pdf.
[f8-a] 3GPP TS 35.201 V4.1.0 (2001-12) Technical Specification 3rd
Generation Partnership Project; Technical Specification
Group Services and System Aspects; 3G Security;
Specification of the 3GPP Confidentiality and Integrity
Algorithms; Document 1: f8 and f9 Specification (Release
4).
[f8-b] 3GPP TR 33.908 V4.0.0 (2001-09) Technical Report 3rd
Generation Partnership Project; Technical Specification
Group Services and System Aspects; 3G Security; General
Report on the Design, Specification and Evaluation of 3GPP
Standard Confidentiality and Integrity Algorithms (Release
4).
[GDOI] Baugher, M., Weis, B., Hardjono, T. and H. Harney, "The
Group Domain of Interpretation, RFC 3547, July 2003.
[HAC] Menezes, A., Van Oorschot, P. and S. Vanstone, "Handbook
of Applied Cryptography", CRC Press, 1997, ISBN 0-8493-
8523-7.
[H80] Hellman, M. E., "A cryptanalytic time-memory trade-off",
IEEE Transactions on Information Theory, July 1980, pp.
401-406.
[IK] T. Iwata and T. Kohno: "New Security Proofs for the 3GPP
Confidentiality and Integrity Algorithms", Proceedings of
FSE 2004.
[KINK] Thomas, M. and J. Vilhuber, "Kerberized Internet
Negotiation of Keys (KINK)", Work in Progress.
[KEYMGT] Arrko, J., et al., "Key Management Extensions for Session
Description Protocol (SDP) and Real Time Streaming Protocol
(RTSP)", Work in Progress.
[KSYH] Kang, J-S., Shin, S-U., Hong, D. and O. Yi, "Provable
Security of KASUMI and 3GPP Encryption Mode f8",
Proceedings Asiacrypt 2001, Springer Verlag LNCS 2248, pp.
255-271, 2001.
[MIKEY] Arrko, J., et. al., "MIKEY: Multimedia Internet KEYing",
Work in Progress.
[MF00] McGrew, D. and S. Fluhrer, "Attacks on Encryption of
Redundant Plaintext and Implications on Internet Security",
the Proceedings of the Seventh Annual Workshop on Selected
Areas in Cryptography (SAC 2000), Springer-Verlag.
[PCST1] Perrig, A., Canetti, R., Tygar, D. and D. Song, "Efficient
and Secure Source Authentication for Multicast", in Proc.
of Network and Distributed System Security Symposium NDSS
2001, pp. 35-46, 2001.
[PCST2] Perrig, A., Canetti, R., Tygar, D. and D. Song, "Efficient
Authentication and Signing of Multicast Streams over Lossy
Channels", in Proc. of IEEE Security and Privacy Symposium