5. Security Considerations
5. Security Considerations
While this protocol is designed to minimize disclosure of configuration information to unauthenticated peers, some such disclosure is unavoidable. One peer or the other must identify itself first and prove its identity first. To avoid probing, the initiator of an exchange is required to identify itself first, and usually is required to authenticate itself first. The initiator can, however, learn that the responder supports IKE and what cryptographic protocols it supports. The responder (or someone impersonating the responder) not only can probe the initiator for its identity but may, by using CERTREQ payloads, be able to determine what certificates the initiator is willing to use.
Use of EAP authentication changes the probing possibilities somewhat. When EAP authentication is used, the responder proves its identity before the initiator does, so an initiator that knew the name of a valid initiator could probe the responder for both its name and certificates.
Repeated rekeying using CREATE_CHILD_SA without additional Diffie-Hellman exchanges leaves all SAs vulnerable to cryptanalysis of a single key. Implementers should take note of this fact and set a limit on CREATE_CHILD_SA exchanges between exponentiations. This document does not prescribe such a limit.
The strength of a key derived from a Diffie-Hellman exchange using any of the groups defined here depends on the inherent strength of the group, the size of the exponent used, and the entropy provided by the random number generator used. Due to these inputs, it is difficult to determine the strength of a key for any of the defined groups. Diffie-Hellman group number two, when used with a strong random number generator and an exponent no less than 200 bits, is common for use with 3DES. Group five provides greater security than group two. Group one is for historic purposes only and does not provide sufficient strength except for use with DES, which is also for historic use only. Implementations should make note of these estimates when establishing policy and negotiating security parameters.
Note that these limitations are on the Diffie-Hellman groups themselves. There is nothing in IKE that prohibits using stronger groups nor is there anything that will dilute the strength obtained from stronger groups (limited by the strength of the other algorithms negotiated including the PRF). In fact, the extensible framework of IKE encourages the definition of more groups; use of elliptic curve groups may greatly increase strength using much smaller numbers.
It is assumed that all Diffie-Hellman exponents are erased from memory after use.
The IKE_SA_INIT and IKE_AUTH exchanges happen before the initiator has been authenticated. As a result, an implementation of this protocol needs to be completely robust when deployed on any insecure network. Implementation vulnerabilities, particularly DoS attacks, can be exploited by unauthenticated peers. This issue is particularly worrisome because of the unlimited number of messages in EAP-based authentication.
The strength of all keys is limited by the size of the output of the negotiated PRF. For this reason, a PRF whose output is less than 128 bits (e.g., 3DES-CBC) MUST NOT be used with this protocol.
The security of this protocol is critically dependent on the randomness of the randomly chosen parameters. These should be generated by a strong random or properly seeded pseudorandom source (see [RANDOMNESS]). Implementers should take care to ensure that use of random numbers for both keys and nonces is engineered in a fashion that does not undermine the security of the keys.
For information on the rationale of many of the cryptographic design choices in this protocol, see [SIGMA] and [SKEME]. Though the security of negotiated Child SAs does not depend on the strength of the encryption and integrity protection negotiated in the IKE SA, implementations MUST NOT negotiate NONE as the IKE integrity protection algorithm or ENCR_NULL as the IKE encryption algorithm.
When using pre-shared keys, a critical consideration is how to assure the randomness of these secrets. The strongest practice is to ensure that any pre-shared key contain as much randomness as the strongest key being negotiated. Deriving a shared secret from a password, name, or other low-entropy source is not secure. These sources are subject to dictionary and social-engineering attacks, among others.
The NAT_DETECTION_*_IP notifications contain a hash of the addresses and ports in an attempt to hide internal IP addresses behind a NAT. Since the IPv4 address space is only 32 bits, and it is usually very sparse, it would be possible for an attacker to find out the internal address used behind the NAT box by trying all possible IP addresses and trying to find the matching hash. The port numbers are normally fixed to 500, and the SPIs can be extracted from the packet. This reduces the number of hash calculations to 2^32. With an educated guess of the use of private address space, the number of hash calculations is much smaller. Designers should therefore not assume that use of IKE will not leak internal address information.
When using an EAP authentication method that does not generate a shared key for protecting a subsequent AUTH payload, certain man-in-the-middle and server-impersonation attacks are possible [EAPMITM]. These vulnerabilities occur when EAP is also used in protocols that are not protected with a secure tunnel. Since EAP is a general-purpose authentication protocol, which is often used to provide single-signon facilities, a deployed IPsec solution that relies on an EAP authentication method that does not generate a shared key (also known as a non-key-generating EAP method) can become compromised due to the deployment of an entirely unrelated application that also happens to use the same non-key-generating EAP method, but in an unprotected fashion. Note that this vulnerability is not limited to just EAP, but can occur in other scenarios where an authentication infrastructure is reused. For example, if the EAP mechanism used by IKEv2 utilizes a token authenticator, a man-in-the-middle attacker could impersonate the web server, intercept the token authentication exchange, and use it to initiate an IKEv2 connection. For this reason, use of non-key-generating EAP methods SHOULD be avoided where possible. Where they are used, it is extremely important that all usages of these EAP methods SHOULD utilize a protected tunnel, where the initiator validates the responder's certificate before initiating the EAP authentication. Implementers should describe the vulnerabilities of using non-key-generating EAP methods in the documentation of their implementations so that the administrators deploying IPsec solutions are aware of these dangers.
An implementation using EAP MUST also use a public-key-based authentication of the server to the client before the EAP authentication begins, even if the EAP method offers mutual authentication. This avoids having additional IKEv2 protocol variations and protects the EAP data from active attackers.
If the messages of IKEv2 are long enough that IP-level fragmentation is necessary, it is possible that attackers could prevent the exchange from completing by exhausting the reassembly buffers. The chances of this can be minimized by using the Hash and URL encodings instead of sending certificates (see Section 3.6). Additional mitigations are discussed in [DOSUDPPROT].
Admission control is critical to the security of the protocol. For example, trust anchors used for identifying IKE peers should probably be different than those used for other forms of trust, such as those used to identify public web servers. Moreover, although IKE provides a great deal of leeway in defining the security policy for a trusted peer's identity, credentials, and the correlation between them, having such security policy defined explicitly is essential to a secure implementation.