Signature Authentication in the Internet Key Exchange Version 2 (IKEv2) using PQC

Introduction The Internet Key Exchange, or IKEv2 , is a key agreement and security negotiation protocol; it is used for key establishment in IPsec. In the initial set of exchanges, both parties independently select and use their preferred authentication method, which may even differ between the initiator and the responder. In IKEv2, it occurs in the exchange called IKE_AUTH. One option for the authentication method is digital signatures using public key cryptography. Currently, traditional digital signatures are defined for use within IKE_AUTH: RSA signatures, Digital Signature Algorithm (DSA) Digital Signature Standard (DSS) and ECDSA. The presence of a Cryptographically Relevant Quantum Computer (CRQC) would render state-of-the-art traditional public-key algorithms obsolete and insecure. This is because the assumptions about the intractability of the mathematical problems these algorithms rely on, which offer confident levels of security today, no longer apply in the presence of a CRQC. Consequently, there is a requirement to update protocols and infrastructure to use post-quantum algorithms. Post-quantum algorithms are public-key algorithms designed to be secure against CRQCs as well as classical computers. The traditional cryptographic primitives that need to be replaced by PQC algorithms are discussed in . This document defines a general approach to incorporating PQC digital signature algorithms into IKEv2 while maintaining interoperability and backward compatibility. Additionally, it outlines how Module-Lattice-Based Digital Signatures (ML-DSA) and Stateless Hash-Based Digital Signatures (SLH-DSA) can be employed as authentication methods within IKEv2, as they have been standardized the US National Institute of Standards and Technology (NIST) PQC project.

Conventions and Definitions 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. This document uses terms defined in . For the purposes of this document, it is helpful to be able to divide cryptographic algorithms into two classes: "Asymmetric Traditional Cryptographic Algorithm": An asymmetric cryptographic algorithm based on integer factorisation, finite field discrete logarithms or elliptic curve discrete logarithms, elliptic curve discrete logarithms, or related mathematical problems. "Post-Quantum Algorithm": An asymmetric cryptographic algorithm that is believed to be secure against attacks using quantum computers as well as classical computers. Post-quantum algorithms can also be called quantum-resistant or quantum-safe algorithms. Examples of quantum-resistant digital signature schemes include ML-DSA and SLH-DSA.

General Framework for PQC Authentication in IKEv2 IKEv2 authentication relies on digital signatures to verify the identity of communicating peers. The mechanism described in this document enables the use of any PQC digital signature algorithm without modifying core IKEv2 operations.

Specifying PQC Signature Algorithms

IKEv2 can use arbitrary signature algorithms as described in , where the "Digital Signature" authentication method supersedes previously defined signature authentication methods. Any PQC digital signature algorithm can be incorporated using the "Signature Algorithm" field in authentication payloads, as defined in .
AlgorithmIdentifier ASN.1 objects will be used to uniquely identify PQC signature algorithm scheme and the parameter set associated with it.

Signature Generation and Verification PQC signatures may be generated in either deterministic or hedged modes. In the deterministic mode, the signature is derived entirely from the message and the signer’s private key, without introducing fresh randomness at signing time. While this eliminates reliance on an external random number generator (RNG), it increases susceptibility to side-channel attacks, particularly fault injection attacks. The hedged mode mitigates this risk by incorporating both fresh randomness generated at signing time and precomputed randomness included in the signer’s private key. This approach ensures protection against side-channel attacks. In the context of signature-based authentication in IKEv2, the data used for generating a digital signature is unique for each session, as it includes session-specific information such as nonces, cryptographic parameters, and identifiers. Therefore, PQC signature algorithms can leverage the hedged variant within IKEv2 to enhance security against side-channel attacks. The choice between deterministic and hedged signing modes does not impact interoperability because the verification process remains the same for both variants. If the PQC signature algorithm uses a 'context' input parameter, it will be set to an empty string. Certain digital signature algorithms support two modes: pure mode and pre-hash mode. In contrast, pre-hash mode involves signing a digest of the message. This document specifies the use of pure mode for signature-based authentication in IKEv2, where the message is signed directly along with domain separation information. The data used for authentication in IKEv2, as described in Section 2.15 of , consists of elements such as nonces, SPIs, and initial exchange messages, which are typically within device memory constraints. As discussed in , while pre-hash mode was considered for integration into IKEv2, various practical challenges led to the adoption of pure mode.

Handling PQC Signatures in IKEv2 For integrating PQC signature algorithms into IKEv2, the approach used in is followed. The implementation MUST send a SIGNATURE_HASH_ALGORITHMS notify with an 'Identity' (5) hash function. PQC signature algorithms that inherently operate on the raw message without preprocessing are only defined with the 'Identity' hash and MUST NOT be used with a receiver that has not indicated support for the 'Identity' hash. When generating a signature with a PQC signature algorithm, the IKEv2 implementation takes the InitiatorSignedOctets string or the ResponderSignedOctets string (as appropriate), logically sends it to the identity hash (which leaves it unchanged), and then passes it into the PQC signer as the message to be signed (with no context string, if applicable). The resulting signature is placed into the Signature Value field of the Authentication Payload. When verifying a signature with a PQC signature algorithm, the IKEv2 implementation takes the InitiatorSignedOctets string or the ResponderSignedOctets string (as appropriate), logically sends it to the identity hash (which leaves it unchanged), and then passes it into the PQC signature verifier as the message to be verified (with no context string, if applicable).

Mechanisms for Signaling Supported Key Pair Types The following mechanisms can be used by peers to signal the types of public/private key pairs they support:

Certificate Request Payload: One method to ascertain that the key pair type the initiator wants the responder to use is through a Certificate Request payload sent by the initiator. For example, the initiator can specify that it trusts certificates issued by a certificate authority (CA) that signs with a particular post-quantum cryptographic (PQC) signature algorithm. This implies that the initiator can process signatures generated using that algorithm, thereby allowing the responder to authenticate itself using a key pair associated with the specified PQC signature scheme.
Authentication Method Announcement: Another method is to utilize ,
which enables peers to declare their supported authentication methods. This improves interoperability when IKEv2 peers are configured with multiple credential types of different type to authenticate each other. The responder includes a SUPPORTED_AUTH_METHODS notification in the IKE_SA_INIT response message, listing the PQC signature scheme(s) it supports. The initiator includes the SUPPORTED_AUTH_METHODS notification in either the IKE_AUTH request message or in the IKE_INTERMEDIATE request. This notification lists the PQC digital signature scheme(s) supported by the initiator, ordered by preference.

Specifying ML-DSA within IKEv2 ML-DSA is a digital signature algorithm (part of the CRYSTALS suite) based on the hardness lattice problems over module lattices (i.e., the Module Learning with Errors problem (MLWE)). The design of the algorithm is based on the "Fiat-Shamir with Aborts" framework introduced by Lyubashevsky, that leverages rejection sampling to render lattice based FS schemes compact and secure. ML-DSA uses uniform distribution over small integers for computing coefficients in error vectors, which makes the scheme easier to implement. ML-DSA is instantiated with 3 parameter sets for the security categories 2, 3 and 5. Security properties of ML-DSA are discussed in Section 9 of . This document specifies the use of the ML-DSA algorithm in IKEv2 at three security levels: ML-DSA-44, ML-DSA-65, and ML-DSA-87.

Specifying SLH-DSA within IKEv2 SLH-DSA utilizes the concept of stateless hash-based signatures. In contrast to stateful signature algorithms, SLH-DSA eliminates the need for maintaining state information during the signing process. SLH-DSA is designed to sign up to 2^64 messages and it offers three security levels. The parameters for each of the security levels were chosen to provide 128 bits of security, 192 bits of security, and 256 bits of security. This document specifies the use of the SLH-DSA algorithm in IKEv2 at three security levels. It includes the small (S) or fast (F) versions of the algorithm. For security level 1, SHA-256 () is used. For security levels 3 and 5, SHA-512 () is used. SHAKE256 () is applicable for all security levels. The small version prioritizes smaller signature sizes, making them suitable for resource-constrained environments IoT devices. Conversely, the fast version prioritizes speed over signature size, minimizing the time required to generate signatures. However, signature verification with the small version is faster than with the fast version. On the other hand, ML-DSA outperforms SLH-DSA in both signature generation and validation time, as well as signature size. SLH-DSA, in contrast, offers smaller key sizes but larger signature sizes. The following combinations are defined in SLH-DSA :

SLH-DSA-128S-SHA2
SLH-DSA-128F-SHA2
SLH-DSA-192S-SHA2
SLH-DSA-192F-SHA2
SLH-DSA-256S-SHA2
SLH-DSA-256F-SHA2
SLH-DSA-128S-SHAKE
SLH-DSA-128F-SHAKE
SLH-DSA-192S-SHAKE
SLH-DSA-192F-SHAKE
SLH-DSA-256S-SHAKE
SLH-DSA-256F-SHAKE

SLH-DSA does not introduce a new hardness assumption beyond those inherent to the underlying hash functions. It builds upon established foundations in cryptography, making it a reliable and robust digital signature scheme for a post-quantum world. While attacks on lattice-based schemes like ML-DSA can compromise their security, SLH-DSA will remain unaffected by these attacks due to its distinct mathematical foundations. This ensures the continued security of systems and protocols that utilize SLH-DSA for digital signatures.

Implementation Alternatives for ML-DSA With ML-DSA, there are two different approaches to implementing the signature process. The first one is to simply hand the SignedOctets string to the crypto library to generate the full signature; this works for SLH-DSA as well. The second approach involves using the ExternalMu-ML-DSA API defined in . In this method, the implementation calls the ExternalMU-ML-DSA.Prehash API with the SignedOctets string and the ML-DSA public key, generating an hash. This hash is then passed to the cryptographic library to execute the ExternalMU-ML-DSA.Sign API, which takes the hash and the ML-DSA private key to produce the signature. These methods are equivalent, and so either may be used.

Discussion of ML-DSA and SLH-DSA and Prehashing This section discusses various approaches for integrating ML-DSA and SLH-DSA into IKEv2, not just the method proposed above. The signature architecture within IKE was designed around RSA (and later extended to ECDSA). In this architecture, the actual message (the SignedOctets) are first hashed (using a hash that the verifier has indicated support for), and then passed for the remaining part of the signature generation processing. That is, it is designed for signature algorithms that first apply one of a number of hash functions to the message and then perform processing on that hash. Neither ML-DSA nor SLH-DSA fits cleanly into this architecture. We see three ways to address this mismatch. The first consideration is that both ML-DSA and SLH-DSA provide prehashed parameter sets, which are designed to sign messages that have already been hashed by an external source. At first glance, this might seem like an ideal solution. However, several practical challenges arise:

The prehashed versions of ML-DSA and SLH-DSA appear to be rarely used, making it likely that support for them in cryptographic libraries is limited or unavailable.
The public keys for the prehashed variants use different OIDs, which means that certificates for IKEv2 would differ from those used in other protocols. Additionally, some certificate authorities (CAs) may not support issuing certificates for prehashed ML-DSA or SLH-DSA due to their limited adoption.
Some users have explicitly indicated a preference not to use the prehashed parameter sets.

The second is to note that, while IKEv2 normally acts this way, it doesn't always. EdDSA has a similar constraint on not working cleanly with the standard 'hash and then sign' paradigm, and so the existing provides an alternative method, which ML-DSA would cleanly fit into. We could certainly adopt this same strategy; our concern would be that it might be more difficult for IKEv2 implementors which do not already have support for EdDSA. The third way is what we can refer to as 'fake prehashing'; IKEv2 would generate the hash as current, but instead of running ML-DSA or SLH-DSA in prehash mode, we have it sign it in pure mode as if it was the message. This is a violation of the spirit, if not the letter of FIPS 204, 205. However, it is secure (assuming the hash function is strong), and fits in cleanly with both the existing IKEv2 architecture, and what crypto libraries provide. Additionally, for SLH-DSA, this means that we're now dependent on collision resistance (while the rest of the SLH-DSA architecture was carefully designed not to be).

Use of ML-DSA and SLH-DSA Both ML-DSA and SLH-DSA offer deterministic and hedged signing modes. By default, ML-DSA uses a hedged approach, where the random value rnd is a 256-bit string generated by an Random Bit Generator (RBG). The signature generation function utilizes this randomness along with the private key and the preprocessed message. In the deterministic variant, rnd is instead set to a constant 256-bit zero string. Similarly, SLH-DSA can operate in either deterministic or hedged mode. The mode is determined by the value of opt_rand, when opt_rand is set to a fixed value (e.g., the public seed from the public key), SLH-DSA generates deterministic signatures, ensuring that signing the same message twice produces the same signature. In hedged mode, opt_rand is a fresh random value, introducing additional entropy to enhance security and mitigate potential side-channel risks. Both ML-DSA and SLH-DSA can utilize their hedged versions when used within IKEv2. In both cases, the 'context' input parameter for the signature generation algorithm is set to an empty string. The three security levels of ML-DSA are identified via AlgorithmIdentifier ASN.1 objects, as specified in . The different combinations of SLH-DSA are identified via AlgorithmIdentifier ASN.1 objects, as specified in . Both ML-DSA and SLH-DSA define two signature modes: pure mode and pre-hash mode, as specified in and , respectively. Both and prefer pure mode over pre-hash mode, and neither nor discusses pre-hash mode.

Security Considerations PQC signature algorithms are modeled under strong unforgeability against an adaptive chosen message attack (SUF-CMA). Examples include ML-DSA and SLH-DSA, which adhere to this security model. Different PQC signature schemes are designed to provide security levels comparable to well-established cryptographic primitives. For example, some schemes align with the security of AES-128, AES-192, and AES-256, while others correspond to the security levels of SHA-256 or SHA3-256. The choice of a PQC signature algorithm should be guided by the desired security level and performance requirements. ML-DSA-44, ML-DSA-65, and ML-DSA-87 are designed to offer security comparable with the SHA-256/SHA3-256, AES-192, and AES-256 respectively. Similarly, SLH-DSA-128{S,F}-{SHA2,SHAKE}, SLH-DSA-192{S,F}-{SHA2,SHAKE}, and SLH-DSA-256{S,F}-{SHA2,SHAKE} are designed to offer security comparable with the AES-128, AES-192, and AES-256 respectively. The Security Considerations section of and applies to this specification as well.

Acknowledgements Thanks to Stefaan De Cnodder, Loganaden Velvindron, Paul Wouters, Andreas Steffen, and Daniel Van Geest for the discussion and comments.