Support for Multiple Hash Algorithms in Cryptographically Generated Addresses (CGAs)

Recent attacks to currently used hash functions have motivated a considerable amount of concern in the Internet community. The recommended approach to deal with this issue is first to analyze the impact of these attacks on the different Internet protocols that use hash functions and second to make sure that the different Internet protocols that use hash functions are capable of migrating to an alternative (more secure) hash function without a major disruption in the Internet operation. This document performs such analysis for the Cryptographically Generated Addresses (CGAs) defined in . The first conclusion of the analysis is that the security of the protocols using CGAs is not affected by the recently available attacks against hash functions. The second conclusion of the analysis is that the hash function used is hard coded in the CGA specification. This document updates the CGA specification to enable the support of alternative hash functions. In order to do so, this document creates a new registry managed by IANA to register the different hash algorithms used in CGAs.

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 RFC 2119 .

Recent advances in cryptography have resulted in simplified attacks against the collision-free property of certain commonly used hash functions , including SHA-1 that is the hash function used by CGAs . The result is that it is possible to obtain two messages, M1 and M2, that have the same hash value with much less than 2^(L/2) attempts. We will next analyze the impact of such attacks in the currently proposed usages of CGAs. As we understand it, the attacks against the collision-free property of a hash function mostly challenge the application of such hash functions, for the provision of non-repudiation capabilities. This is because an attacker would be capable to create two different messages that result in the same hash value and it can then present any of the messages interchangeably (for example after one of them has been signed by the other party involved in the transaction). However, it must be noted that both messages must be generated by the same party. As far as we understand, current usages of CGAs does not include the provision of non-repudiation capabilities, so attacks against the collision-free property of the hash function do not enable any useful attack against CGA-based protocols. Current usages of the CGAs are basically oriented to prove the ownership of a CGA and then bind it to alternative addresses that can be used to reach the original CGA. This type of application of the CGA include: The application of CGAs to protect the shim6 protocol . In this case, CGAs are used as identifiers for the established communications. CGA features are used to prove that the owner of the identifier is the one that is providing the alternative addresses that can be used to reach the initial identifier. This is achieved by signing the list of alternative addresses available in the multihomed host with the private key of the CGA. The application of CGAs to secure the IPv6 mobility support protocol as proposed in . In this case, the CGAs are used as Home Addresses and they are used to prove that the owner of the Home Address is the one creating the binding with the new Care-off Address. Similarly to the previous case, this is achieved by signing the Binding Update message carrying the Care-off Address with the private key of the CGA. The application of CGA to Secure Neighbour Discovery . In this case, the CGA features are used to prove the address ownership, so that it is possible to verify that the owner of the IP address is the one that is providing the layer 2 address information. This is achieved by signing the layer 2 address information with the private key of the CGA. Essentially, all the current applications of CGAs rely on CGAs to protect a communication between two peers from third party attacks and not to provide protection from the peer itself. Attacks against the collision-free property of the hash functions suppose that one of the parties is generating two messages with the same hash value in order to launch an attack against its communicating peer. Since CGAs are not currently used to providing this type of protection, it is then natural that no additional attacks are enabled by a weaker collision resistance of the hash function.

CGAs, as currently defined in , are intrinsically bound to the SHA-1 hash algorithm and no other hash function is supported. Even though the attacks against the collision-free property of the hash functions do not result in new vulnerabilities in the current applications of CGAs, it seems wise to enable multiple hash function support in CGAs. This is mainly for two reasons: first, potential future applications of the CGA technology may be susceptible to attacks against the collision-free property of SHA-1. Supporting alternative hash functions would allow applications that have stricter requirements on the collision-free property to use CGAs. Second, one lesson learned from the recent attacks against hash functions is that it is possible that one day we need to start using alternative hash functions because of successful attacks against other properties of the commonly used hash functions. Therefore, it seems wise to modify protocols in general and the CGAs in particular to support this transition to alternative hash functions as easy as possible.

The next question we need to answer is where to encode the hash function that is being used. There are several options that can be considered: One option would be to include the hash function used as an input to the hash function. This basically means to create an extension to the CGA Parameter Data Structure, as defined in , that codifies the hash function used. The problem is that this approach is vulnerable to bidding down attacks or downgrading attacks as defined in . This means that even if a strong hash function is used, an attacker could find a CGA Parameter Data Structure that uses a weaker function but results in an equal hash value. This happens when the original hash function H1 and CGA Parameters Data Structure indicating H1 result in value X, and another hash function H2 and CGA Parameters Data Structure indicating H2 also result in the same value X. In other words, the downgrading attack would work as follows: suppose that Alice generates a CGA CGA_A using the strong hash function HashStrong and using a CGA Parameter Data Structure CGA_PDS_A. The selected hash function HashStrong is encoded as an extension field in the CGA_PDS_A. Suppose that by using a brute force attack, an attacker X finds an alternative CGA Parameter Data Structure CGA_PDS_X whose hash value, by using a weaker hash function, is CGA_A. At this point, the attacker can pretend to be the owner of CGA_A and the stronger hash function has not provided additional protection. The conclusion from the previous analysis is that the hash function used in the CGA generation must be encoded in the address itself. Since we want to support several hash functions, we will likely need at least 2 or 3 bits for this. One option would be to use more bits from the hash bits of the interface identifier. However, the problem with this approach is that the resulting CGA is weaker because less hash information is encoded in the address. In addition, since those bits are currently used as hash bits, it is impossible to make this approach backward compatible with existent implementations. Another option would be to use the "u" and the "g" bits to encode this information, but this is probably not such a good idea since those bits have been honoured so far in all interface identifier generation mechanisms, which allow them to be used for the original purpose (for instance we can still create a global registry for unique interface identifiers). Finally, another option is to encode the hash value used in the Sec bits. The Sec bits are used to artificially introduce additional difficulty in the CGA generation process in order to provide additional protection against brute force attacks. The Sec bits have been designed in a way that the lifetime of CGAs are extended, when it is feasible to attack 59-bits long hash values. However, this is not the case today, so in general CGA will have a Sec value of 000. The proposal is to encode in the Sec bits, not only information about brute force attack protection but also to encode the hash function used to generate the hash. So for instance, the Sec value 000 would mean that the hash function used is SHA-1 and the 0 bits of hash2 (as defined in RFC 3972) must be 0. Sec value of 001 could be that the hash function used is SHA-1 and the 16 bits of hash2 (as defined in RFC 3972) must be zero. However, the other values of Sec could mean that an alternative hash function needs to be used and that a certain amount of bits of hash2 must be zero. The proposal is not to define any concrete hash function to be used for other Sec values, since it is not yet clear that we need to do so nor is it clear which hash function should be selected. Note that since there are only 8 Sec values, it may be necessary to reuse Sec values when we run out of unused Sec values. The scenario where such an approach makes sense is where there are some Sec values that are no longer being used because the resulting security has become weak. In this case, where the usage of the Sec value has long been abandoned, it would be possible to reassign the Sec values. However, this must be a last resource option, since it may affect interoperability. This is because two implementations using different meanings of a given Sec value would not be able to interoperate properly (i.e., if an old implementation receives a CGA generated with the new meaning of the Sec value, it will fail and the same for a new implementation receiving a CGA generated with the old meaning of the Sec value). In case the approach of reassigning a Sec value is followed, a long time is required between the deprecation of the old value and the reassignment in order to prevent misinterpretation of the value by old implementations. An erroneous interpretation of a reused Sec value, both on the CGA owner's side and the CGA verifier's side, would have the following result, CGA verification would fail in the worst case and both nodes would have to revert to unprotected IPv6 addresses. This can happen only with obsolete CGA parameter sets, which would be considered insecure anyway. In any case, an implementation must not simultaneously support two different meanings of a Sec value.

The SEC registry defined in the IANA considerations section of this document contains entries for the different Sec values. Each of these entries points to an RFC that defines the CGA generation procedure that MUST be used when generating CGAs with the associated Sec value. It should be noted that the CGA generation procedure may be changed by the new procedure not only in terms of the hash function used but also in other aspects, e.g., longer Modifier values may be required if the number of 0s required in hash2 exceed the currently defined bound of 112 bits. The new procedure (which potentially involves a longer Modifier value) would be described in the RFC pointed to by the corresponding Sec registry entry. In addition, the RFC that defines the CGA generation procedure for a Sec value MUST explicitly define the minimum key length acceptable for CGAs with that Sec value. This is to provide a coherent protection both in the hash and the public key techniques.

This document defines a new registry entitled "CGA SEC" for the Sec field defined in RFC 3972 that has been created and is maintained by IANA. The values in this name space are 3-bit unsigned integers. Initial values for the CGA Extension Type field are given below; future assignments are to be made through Standards Action . Assignments consist of a name, the value, and the RFC number where the CGA generation procedure is defined. The following initial values are assigned in this document:

Name | Value | RFCs -------------------+-------+------------ SHA-1_0hash2bits | 000 | 3972, 4982 SHA-1_16hash2bits | 001 | 3972, 4982 SHA-1_32hash2bits | 010 | 3972, 4982

This document is about security issues and, in particular, about protection against potential attacks against hash functions.

Russ Housley, James Kempf, Christian Vogt, Pekka Nikander, and Henrik Levkowetz reviewed and provided comments about this document. Marcelo Bagnulo worked on this document while visiting Ericsson Research Laboratory Nomadiclab.