RFC 3830
MIKEY: Multimedia Internet KEYing
4. Selected Key Management Functions
MIKEY manages symmetric keys in two main ways. First, following key transport or key exchange of TGK(s) (and other parameters) as defined by any of the above three methods, MIKEY maintains a mapping between Data SA identifiers and Data SAs, where the identifiers used depend on the security protocol in question, see Section 4.4. Thus, when the security protocol requests a Data SA, given such a Data SA identifier, an up-to-date Data SA will be obtained. In particular,
correct keying material, TEK(s), might need to be derived. The derivation of TEK(s) (and other keying material) is done from a TGK and is described in Section 4.1.3. Second, for use within MIKEY itself, two key management procedures are needed: * in the pre-shared case, deriving encryption and authentication key material from a single pre-shared key, and * in the public key case, deriving similar key material from the transported envelope key. These two key derivation methods are specified in section 4.1.4. All the key derivation functionality mentioned above is based on a pseudo-random function, defined next.4.1. Key Calculation
In the following, we define a general method (pseudo-random function) to derive one or more keys from a "master" key. This method is used to derive: * TEKs from a TGK and the RAND value, * encryption, authentication, or salting key from a pre-shared/ envelope key and the RAND value.4.1.1. Assumptions
We assume that the following parameters are in place: csb_id : Crypto Session Bundle ID (32-bits unsigned integer) cs_id : the Crypto Session ID (8-bits unsigned integer) RAND : (at least) 128-bit (pseudo-)random bit-string sent by the Initiator in the initial exchange. The key derivation method has the following input parameters: inkey : the input key to the derivation function inkey_len : the length in bits of the input key label : a specific label, dependent on the type of the key to be derived, the RAND, and the session IDs outkey_len: desired length in bits of the output key.
The key derivation method has the following output: outkey: the output key of desired length.4.1.2. Default PRF Description
Let HMAC be the SHA-1 based message authentication function, see [HMAC] [SHA-1]. Similarly to [TLS], we define: P (s, label, m) = HMAC (s, A_1 || label) || HMAC (s, A_2 || label) || ... HMAC (s, A_m || label) where A_0 = label, A_i = HMAC (s, A_(i-1)) s is a key (defined below) m is a positive integer (also defined below). Values of label depend on the case in which the PRF is invoked, and values are specified in the following for the default PRF. Thus, note that other PRFs later added to MIKEY MAY specify different input parameters. The following procedure describes a pseudo-random function, denoted PRF(inkey,label), based on the above P-function, applied to compute the output key, outkey: * let n = inkey_len / 256, rounded up to the nearest integer if not already an integer * split the inkey into n blocks, inkey = s_1 || ... || s_n, where * all s_i, except possibly s_n, are 256 bits each * let m = outkey_len / 160, rounded up to the nearest integer if not already an integer (The values "256" and "160" equals half the input block-size and full output hash size, respectively, of the SHA-1 hash as part of the P- function.) Then, the output key, outkey, is obtained as the outkey_len most significant bits of PRF(inkey, label) = P(s_1, label, m) XOR P(s_2, label, m) XOR ... XOR P(s_n, label, m).
4.1.3. Generating keys from TGK
In the following, we describe how keying material is derived from a TGK, thus assuming that a mapping of the Data SA identifier to the correct TGK has already been done according to Section 4.4. The key derivation method SHALL be executed using the above PRF with the following input parameters: inkey : TGK inkey_len : bit length of TGK label : constant || cs_id || csb_id || RAND outkey_len : bit length of the output key. The constant part of label depends on the type of key that is to be generated. The constant 0x2AD01C64 is used to generate a TEK from TGK. If the security protocol itself does not support key derivation for authentication and encryption from the TEK, separate authentication and encryption keys MAY be created directly for the security protocol by replacing 0x2AD01C64 with 0x1B5C7973 and 0x15798CEF respectively, and outkey_len by the desired key-length(s) in each case. A salt key can be derived from the TGK as well, by using the constant 0x39A2C14B. Note that the Key data sub-payload (Section 6.13) can carry a salt. The security protocol in need of the salt key SHALL use the salt key carried in the Key data sub-payload (in the pre- shared and public-key case), when present. If that is not sent, then it is possible to derive the salt key via the key derivation function, as described above. The table below summarizes the constant values, used to generate keys from a TGK. constant | derived key from the TGK -------------------------------------- 0x2AD01C64 | TEK 0x1B5C7973 | authentication key 0x15798CEF | encryption key 0x39A2C14B | salting key Table 4.1.3: Constant values for the derivation of keys from TGK. Note that these 32-bit constant values (listed in the table above) are taken from the decimal digits of e (i.e., 2.7182...), where each constant consists of nine decimal digits (e.g., the first nine decimal digits 718281828 = 0x2AD01C64). The strings of nine
decimal digits are not chosen at random, but as consecutive "chunks" from the decimal digits of e.4.1.4. Generating keys for MIKEY messages from an envelope/pre-shared key
This derivation is to form the symmetric encryption key (and salting key) for the encryption of the TGK in the pre-shared key and public key methods. This is also used to derive the symmetric key used for the message authentication code in these messages, and the corresponding verification messages. Hence, this derivation is needed in order to get different keys for the encryption and the MAC (and in the case of the pre-shared key, it will result in fresh key material for each new CSB). The parameters for the default PRF are here: inkey : the envelope key or the pre-shared key inkey_len : the bit length of inkey label : constant || 0xFF || csb_id || RAND outkey_len : desired bit length of the output key. The constant part of label depends on the type of key that is to be generated from an envelope/pre-shared key, as summarized below. constant | derived key -------------------------------------- 0x150533E1 | encryption key 0x2D22AC75 | authentication key 0x29B88916 | salt key Table 4.1.4: Constant values for the derivation of keys from an envelope/pre-shared key.4.2. Pre-defined Transforms and Timestamp Formats
This section identifies default transforms for MIKEY. It is mandatory to implement and support the following transforms in the respective case. New transforms can be added in the future (see Section 4.2.9 for further guidelines).4.2.1. Hash functions
In MIKEY, it is MANDATORY to implement SHA-1 as the default hash function.
4.2.2. Pseudo-random number generator and PRF
A cryptographically secure random or pseudo-random number generator MUST be used for the generation of the keying material and nonces, e.g., [BMGL]. However, which one to use is implementation specific (as the choice will not affect the interoperability). For the key derivations, it is MANDATORY to implement the PRF specified in Section 4.1. Other PRFs MAY be added by writing standard-track RFCs specifying the PRF constructions and their exact use within MIKEY.4.2.3. Key data transport encryption
The default and mandatory-to-implement key transport encryption is AES in counter mode, as defined in [SRTP], using a 128-bit key as derived in Section 4.1.4, SRTP_PREFIX_LENGTH set to zero, and using the initialization vector IV = (S XOR (0x0000 || CSB ID || T)) || 0x0000, where S is a 112-bit salting key, also derived as in Section 4.1.4, and where T is the 64-bit timestamp sent by the Initiator. Note: this restricts the maximum size that can be encrypted to 2^23 bits, which is still enough for all practical purposes [SRTP]. The NULL encryption algorithm (i.e., no encryption) can be used (but implementation is OPTIONAL). Note that this MUST NOT be used unless the underlying protocols can guarantee security. The main reason for including this is for specific SIP scenarios, where SDP is protected end-to-end. For this scenario, MIKEY MAY be used with the pre-shared key method, the NULL encryption, and NULL authentication algorithm (see Section 4.2.4) while relying on the security of SIP. Use this option with caution! The AES key wrap function [AESKW] is included as an OPTIONAL implementation method. If the key wrap function is used in the public key method, the NULL MAC is RECOMMENDED to be used, as the key wrap itself will provide integrity of the encrypted content (note though that the NULL MAC SHOULD NOT be used in the pre-shared key case, as the MAC in that case covers the entire message). The 128- bit key and a 64-bit salt, S, are derived in accordance to Section 4.1.4 and the key wrap IV is then set to S.
4.2.4. MAC and Verification Message function
MIKEY uses a 160-bit authentication tag, generated by HMAC with SHA-1 as the MANDATORY implementation method, see [HMAC]. Authentication keys are derived according to Section 4.1.4. Note that the authentication key size SHOULD be equal to the size of the hash function's output (e.g., for HMAC-SHA-1, a 160-bit authentication key is used) [HMAC]. The NULL authentication algorithm (i.e., no MAC) can be used together with the NULL encryption algorithm (but implementation is OPTIONAL). Note that this MUST NOT be used unless the underlying protocols can guarantee security. The main reason for including this is for specific SIP scenarios, where SDP is protected end-to-end. For this scenario, MIKEY MAY be used with the pre-shared key method and the NULL encryption and authentication algorithm, while relying on the security of SIP. Use this option with caution!4.2.5. Envelope Key encryption
The public key encryption algorithm applied is defined by, and dependent on the certificate used. It is MANDATORY to support RSA PKCS#1, v1.5, and it is RECOMMENDED to also support RSA OAEP [PSS].4.2.6. Digital Signatures
The signature algorithm applied is defined by, and dependent on the certificate used. It is MANDATORY to support RSA PKCS#1, v1.5, and it is RECOMMENDED to also support RSA PSS [PSS].4.2.7. Diffie-Hellman Groups
The Diffie-Hellman key exchange, when supported, uses OAKLEY 5 [OAKLEY] as a mandatory implementation. Both OAKLEY 1 and OAKLEY 2 MAY be used (but these are OPTIONAL implementations). See Section 4.2.9 for the guidelines on specifying a new DH Group to be used within MIKEY.4.2.8. Timestamps
The timestamp is as defined in NTP [NTP], i.e., a 64-bit number in seconds relative to 0h on 1 January 1900. An implementation MUST be aware of (and take into account) the fact that the counter will overflow approximately every 136th year. It is RECOMMENDED that the time always be specified in UTC.
4.2.9. Adding new parameters to MIKEY
There are two different parameter sets that can be added to MIKEY. The first is a set of MIKEY transforms (needed for the exchange itself), and the second is the Data SAs. New transforms and parameters (including new policies) SHALL be added by registering with IANA (according to [RFC2434], see also Section 10) a new number for the concerned payload, and also if necessary, documenting how the new transform/parameter is used. Sometimes it might be enough to point to an already specified document for the usage, e.g., when adding a new, already standardized, hash function. In the case of adding a new DH group, the group MUST be specified in a companion standards-track RFC (it is RECOMMENDED that the specified group use the same format as used in [OAKLEY]). A number can then be assigned by IANA for such a group to be used in MIKEY. When adding support for a new data security protocol, the following MUST be specified: * A map sub-payload (see Section 6.1). This is used to be able to map a crypto session to the right instance of the data security protocol and possibly also to provide individual parameters for each data security protocol. * A policy payload, i.e., specification of parameters and supported values. * General guidelines of usage.4.3. Certificates, Policies and Authorization
4.3.1. Certificate handling
Certificate handling may involve a number of additional tasks not shown here, and effect the inclusion of certain parts of the message (c.f. [X.509]). However, the following observations can be made: * The Initiator typically has to find the certificate of the Responder in order to send the first message. If the Initiator does not already have the Responder's certificate, this may involve one or more roundtrips to a central directory agent. * It will be possible for the Initiator to omit its own certificate and rely on the Responder getting this certificate using other means. However, we only recommend doing this when it is reasonable to expect that the Responder has cached the certificate
from a previous connection. Otherwise accessing the certificate
would mean additional roundtrips for the Responder as well.
* Verification of the certificates using Certificate Revocation
Lists (CRLs) [X.509] or protocols such as OCSP [OCSP] may be
necessary. All parties in a MIKEY exchange should have a local
policy which dictates whether such checks are made, how they are
made, and how often they are made. Note that performing the
checks may imply additional messaging.
4.3.2. Authorization
In general, there are two different models for making authorization
decisions for both the Initiator and the Responder, in the context of
the applications targeted by MIKEY:
* Specific peer-to-peer configuration. The user has configured the
application to trust a specific peer.
When pre-shared secrets are used, this is pretty much the only
available scheme. Typically, the configuration/entering of the
pre-shared secret is taken to mean that authorization is implied.
In some cases, one could also use this with public keys, e.g., if
two peers exchange keys offline and configure them to be used for
the purpose of running MIKEY.
* Trusted root. The user accepts all peers that prove to have a
certificate issued by a specific CA. The granularity of
authorization decisions is not very precise in this method.
In order to make this method possible, all participants in the
MIKEY protocol need to configure one or more trusted roots. The
participants also need to be capable of performing certificate
chain validation, and possibly transfer more than a single
certificate in the MIKEY messages (see also Section 6.7).
In practice, a combination of both mentioned methods might be
advantageous. Also, the possibility for a user to explicitly exclude
a specific peer (or sub-tree) in a trust chain might be needed.
These authorization policies address the MIKEY scenarios a-c of
Section 2.1, where the Initiator acts as the group owner and is also
the only one that can invite others. This implies that for each
Responder, the distributed keys MUST NOT be re-distributed to other
parties.
In a many-to-many situation, where the group control functions are distributed (and/or where it is possible to delegate the group control function to others), a means of distributing authorization information about who may be added to the group MUST exist. However, it is out of scope of this document to specify how this should be done. For any broader communication situation, an external authorization infrastructure may be used (following the assumptions of [GKMARCH]).4.3.3. Data Policies
Included in the message exchange, policies (i.e., security parameters) for the Data security protocol are transmitted. The policies are defined in a separate payload and are specific to the security protocol (see also Section 6.10). Together with the keys, the validity period of these can also be specified. For example, this can be done with an SPI (or SRTP MKI) or with an Interval (e.g., a sequence number interval for SRTP), depending on the security protocol. New parameters can be added to a policy by documenting how they should be interpreted by MIKEY and by also registering new values in the appropriate name space in IANA. If a completely new policy is needed, see Section 4.2.9 for guidelines.4.4. Retrieving the Data SA
The retrieval of a Data SA will depend on the security protocol, as different security protocols will have different characteristics. When adding support for a security protocol to MIKEY, some interface of how the security protocol retrieves the Data SA from MIKEY MUST be specified (together with policies that can be negotiated). For SRTP, the SSRC (see [SRTP]) is one of the parameters used to retrieve the Data SA (while the MKI may be used to indicate the TGK/TEK used for the Data SA). However, the SSRC is not sufficient. For the retrieval of the Data SA from MIKEY, it is RECOMMENDED that the MIKEY implementation support a lookup using destination network address and port together with SSRC. Note that MIKEY does not send network addresses or ports. One reason for this is that they may not be known in advance. Also, if a NAT exists in-between, problems may arise. When SIP or RTSP is used, the local view of the destination address and port can be obtained from either SIP or RTSP. MIKEY can then use these addresses as the index for the Data SA lookup.
4.5. TGK re-keying and CSB updating
MIKEY provides a means of updating the CSB (e.g., transporting a new TGK/TEK or adding a new Crypto Session to the CSB). The updating of the CSB is done by executing MIKEY again, for example, before a TEK expires, or when a new Crypto Session is added to the CSB. Note that MIKEY does not provide re-keying in the GKMARCH sense, only updating of the keys by normal unicast messages. When MIKEY is executed again to update the CSB, it is not necessary to include certificates and other information that was provided in the first exchange, for example, all payloads that are static or optionally included may be left out (see Figure 4.1). The new message exchange MUST use the same CSB ID as the initial exchange, but MUST use a new timestamp. A new RAND MUST NOT be included in the message exchange (the RAND will only have effect in the Initial exchange). If desired, new Crypto Sessions are added in the update message. Note that a MIKEY update message does not need to contain new keying material (e.g., new TGK). In this case, the crypto session continues to use the previously established keying material, while updating the new information. As explained in Section 3.2, the envelope key can be "cached" as a pre-shared key (this is indicated by the Initiator in the first message sent). If so, the update message is a pre-shared key message with the cached envelope key as the pre-shared key; it MUST NOT be a public key message. If the public key message is used, but the envelope key is not cached, the Initiator MUST provide a new encrypted envelope key that can be used in the verification message. However, the Initiator does not need to provide any other keys. Figure 4.1 visualizes the update messages that can be sent, including the optional parts. The main difference from the original message is that it is optional to include TGKs (or DH values in the DH method). Also see Section 3 for more details on the specific methods. By definition, a CSB can contain several CSs. A problem that then might occur is to synchronize the TGK re-keying if an SPI (or similar functionality, e.g., MKI in [SRTP]) is not used. It is therefore RECOMMENDED that an SPI or MKI be used, if more than one CS is present.
Initiator Responder Pre-shared key method: I_MESSAGE = HDR, T, [IDi], [IDr], {SP}, KEMAC ---> R_MESSAGE = [<---] HDR, T, [IDr], V Public key method: I_MESSAGE = HDR, T, [IDi|CERTi], [IDr], {SP}, [KEMAC], [CHASH], PKE, SIGNi ---> R_MESSAGE = [<---] HDR, T, [IDr], V DH method: I_MESSAGE = HDR, T, [IDi|CERTi], [IDr], {SP}, [DHi], SIGNi ---> R_MESSAGE = <--- HDR, T, [IDr|CERTr], IDi, [DHr, DHi], SIGNr Figure 4.1: Update messages. Note that for the DH method, if the Initiator includes the DHi payload, then the Responder MUST include DHr and DHi. If the Initiator does not include DHi, the Responder MUST NOT include DHr or DHi.5. Behavior and message handling
Each message that is sent by the Initiator or the Responder is built by a set of payloads. This section describes how messages are created and also when they can be used.5.1. General
5.1.1. Capability Discovery
The Initiator indicates the security policy to be used (i.e., in terms of security protocol algorithms). If the Responder does not support it (for some reason), the Responder can together with an error message (indicating that it does not support the parameters), send back its own capabilities (negotiation) to let the Initiator
choose a common set of parameters. This is done by including one or more security policy payloads in the error message sent in response (see Section 5.1.2.). Multiple attributes can be provided in sequence in the response. This is done to reduce the number of roundtrips as much as possible (i.e., in most cases, where the policy is accepted the first time, one roundtrip is enough). If the Responder does not accept the offer, the Initiator must go out with a new MIKEY message. If the Responder is not willing/capable of providing security or the parties simply cannot agree, it is up to the parties' policies how to behave, for example, accepting or rejecting an insecure communication. Note that it is not the intention of this protocol to have a broad variety of options, as it is assumed that a denied offer should rarely occur. In the one-to-many and many-to-many scenarios using multicast communication, one issue is of course that there MUST be a common security policy for all the receivers. This limits the possibility of negotiation.5.1.2. Error Handling
Due to the key management protocol, all errors SHOULD be reported to the peer(s) by an error message. The Initiator SHOULD therefore always be prepared to receive such a message from the Responder. If the Responder does not support the set of parameters suggested by the Initiator, the error message SHOULD include the supported parameters (see also Section 5.1.1). The error message is formed as: HDR, T, {ERR}, {SP}, [V|SIGNr] Note that if failure is due to the inability to authenticate the peer, the error message is OPTIONAL, and does not need to be authenticated. It is up to local policy to determine how to treat this kind of message. However, if in response to a failed authentication a signed error message is returned, this can be used for DoS purposes (against the Responder). Similarly, an unauthenticated error message could be sent to the Initiator in order to fool the Initiator into tearing down the CSB. It is highly RECOMMENDED that the local policy take this into consideration. Therefore, in case of authentication failure, one recommendation would be not to authenticate such an error message, and when
receiving an unauthenticated error message view it only as a recommendation of what may have gone wrong.5.2. Creating a message
To create a MIKEY message, a Common Header payload is first created. This payload is then followed, depending on the message type, by a set of information payloads (e.g., DH-value payload, Signature payload, Security Policy payload). The defined payloads and the exact encoding of each payload are described in Section 6. 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ! version ! data type ! next payload ! ! +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+... + ~ Common Header... ~ ! ! +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ! next payload ! Payload 1 ... ! +-+-+-+-+-+-+-+-+ + ~ ~ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ : : : : : : +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ! next payload ! Payload x ... ! +-+-+-+-+-+-+-+-+ + ~ ~ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ! MAC/Signature ~ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Figure 5.1. MIKEY payload message example. Note that the payloads are byte aligned and not 32-bit aligned. The process of generating a MIKEY message consists of the following steps: * Create an initial MIKEY message starting with the Common Header payload. * Concatenate necessary payloads of the MIKEY message (see the exchange definitions for payloads that may be included, and the recommended order). * As a last step (for messages that must be authenticated, this also includes the verification message), create and concatenate the MAC/signature payload without the MAC/signature field filled in
(if a Next payload field is included in this payload, it is set to
Last payload).
* Calculate the MAC/signature over the entire MIKEY message, except
the MAC/Signature field, and add the MAC/signature in the field.
In the case of the verification message, the Identity_i ||
Identity_r || Timestamp MUST directly follow the MIKEY message in
the Verification MAC calculation. Note that the added identities
and timestamp are identical to those transported in the ID and T
payloads.
In the public key case, the Key data transport payload is generated
by concatenating the IDi with the TGKs. This is then encrypted and
placed in the data field. The MAC is calculated over the entire Key
data transport payload except the MAC field. Before calculating the
MAC, the Next payload field is set to zero.
Note that all messages from the Initiator MUST use a unique
timestamp. The Responder does not create a new timestamp, but uses
the timestamp used by the Initiator.
5.3. Parsing a message
In general, parsing of a MIKEY message is done by extracting payload
by payload and checking that no errors occur. The exact procedure is
implementation specific; however, for the Responder, it is
RECOMMENDED that the following procedure be followed:
* Extract the Timestamp and check that it is within the allowable
clock skew (if not, discard the message). Also check the replay
cache (Section 5.4) so that the message is not replayed (see
Section 5.4). If the message is replayed, discard it.
* Extract the ID and authentication algorithm (if not included,
assume the default).
* Verify the MAC/signature.
* If the authentication is not successful, an Auth failure Error
message MAY be sent to the Initiator. The message is then
discarded from further processing. See also Section 5.1.2 for
treatment of errors.
* If the authentication is successful, the message is processed and
also added to the replay cache; processing is implementation
specific. Note also that only successfully authenticated messages
are stored in the replay cache.
* If any unsupported parameters or errors occur during the
processing, these MAY be reported to the Initiator by sending an
error message. The processing is then aborted. The error message
can also include payloads to describe the supported parameters.
* If the processing was successful and in case the Initiator
requested it, a verification/response message MAY be created and
sent to the Initiator.
5.4. Replay handling and timestamp usage
MIKEY does not use a challenge-response mechanism for replay
handling; instead, timestamps are used. This requires that the
clocks are synchronized. The required synchronization is dependent
on the number of messages that can be cached (note though, that the
replay cache only contains messages that have been successfully
authenticated). If we could assume an unlimited cache, the terminals
would not need to be synchronized at all (as the cache could then
contain all previous messages). However, if there are restrictions
on the size of the replay cache, the clocks will need to be
synchronized to some extent. In short, one can in general say that
it is a tradeoff between the size of the replay cache and the
required synchronization.
Timestamp usage prevents replay attacks under the following
assumptions:
* Each host has a clock which is at least "loosely synchronized"
with the clocks of the other hosts.
* If the clocks are to be synchronized over the network, a secure
network clock synchronization protocol SHOULD be used, e.g.,
[ISO3].
* Each Responder utilizes a replay cache in order to remember the
successfully authenticated messages presented within an allowable
clock skew (which is set by the local policy).
* Replayed and outdated messages, for example, messages that can be
found in the replay cache or which have an outdated timestamp are
discarded and not processed.
* If the host loses track of the incoming requests (e.g., due to
overload), it rejects all incoming requests until the clock skew
interval has passed.
In a client-server scenario, servers may encounter a high workload, especially if a replay cache is necessary. However, servers that assume the role of MIKEY Initiators will not need to manage any significant replay cache as they will refuse all incoming messages that are not a response to a message previously sent by the server. In general, a client may not expect a very high load of incoming messages and may therefore allow the degree of looseness to be on the order of several minutes to hours. If a (D)DoS attack is launched and the replay cache grows too large, MIKEY MAY dynamically decrease the looseness so that the replay cache becomes manageable. However, note that such (D)DoS attacks can only be performed by peers that can authenticate themselves. Hence, such an attack is very easy to trace and mitigate. The maximum number of messages that a client will need to cache may vary depending on the capacity of the client itself and the network. The number of expected messages should be taken into account. For example, assume that we can at most spend 6kB on a replay cache. Assume further that we need to store 30 bytes for each incoming authenticated message (the hash of the message is 20 bytes). This implies that it is possible to cache approximately 204 messages. If the expected number of messages per minute can be estimated, the clock skew can easily be calculated. For example, in a SIP scenario where the client is expected, in the most extreme case, to receive 10 calls per minute, the clock skew needed is then approximately 20 minutes. In a not so extreme setting, where one could expect an incoming call every 5th minute, this would result in a clock skew on the order of 16.5 hours (approx 1000 minutes). Consider a very extreme case, where the maximum number of incoming messages are assumed to be on the order of 120 messages per minute, and a requirement that the clock skew is on the order of 10 minutes, a 48kB replay cache would be required. Hence, one can note that the required clock skew will depend largely on the setting in which MIKEY is used. One recommendation is to fix a size for the replay cache, allowing the clock skew to be large (the initial clock skew can be set depending on the application in which it is used). As the replay cache grows, the clock skew is decreased depending on the percentage of the used replay cache. Note that this is locally handled, which will not require interaction with the peer (even though it may indirectly effect the peer). However, exactly how to implement such functionality is out of the scope of this document and considered implementation specific.
In case of a DoS attack, the client will most likely be able to handle the replay cache. A more likely (and serious) DoS attack is a CPU DoS attack where the attacker sends messages to the peer, which then needs to expend resources on verifying the MACs/signatures of the incoming messages.
(page 32 continued on part 3)
