RFC 4945: The Internet IP Security PKI Profile of IKEv1/ISAKMP, IKEv2, and PKIX

Network Working Group
Request for Comments: 4945
Category: Standards Track

B. Korver
Network Resonance, Inc.
August 2007

The Internet IP Security PKI Profile of IKEv1/ISAKMP, IKEv2, and PKIX

Status of This Memo

This document specifies an Internet standards track protocol for the Internet community, and requests discussion and suggestions for improvements. Please refer to the current edition of the "Internet Official Protocol Standards" (STD 1) for the standardization state and status of this protocol. Distribution of this memo is unlimited.

Copyright Notice

Abstract

The Internet Key Exchange (IKE) and Public Key Infrastructure for X.509 (PKIX) certificate profile both provide frameworks that must be profiled for use in a given application. This document provides a profile of IKE and PKIX that defines the requirements for using PKI technology in the context of IKE/IPsec. The document complements protocol specifications such as IKEv1 and IKEv2, which assume the existence of public key certificates and related keying materials, but which do not address PKI issues explicitly. This document addresses those issues. The intended audience is implementers of PKI for IPsec.

   1. Introduction ....................................................4
   2. Terms and Definitions ...........................................4
   3. Use of Certificates in RFC 2401 and IKEv1/ISAKMP ................5
      3.1. Identification Payload .....................................5
           3.1.1. ID_IPV4_ADDR and ID_IPV6_ADDR .......................7
           3.1.2. ID_FQDN .............................................9
           3.1.3. ID_USER_FQDN .......................................10
           3.1.4. ID_IPV4_ADDR_SUBNET, ID_IPV6_ADDR_SUBNET,
                  ID_IPV4_ADDR_RANGE, ID_IPV6_ADDR_RANGE .............11
           3.1.5. ID_DER_ASN1_DN .....................................11
           3.1.6. ID_DER_ASN1_GN .....................................12
           3.1.7. ID_KEY_ID ..........................................12
           3.1.8. Selecting an Identity from a Certificate ...........12
           3.1.9. Subject for DN Only ................................12
           3.1.10. Binding Identity to Policy ........................13
      3.2. Certificate Request Payload ...............................13
           3.2.1. Certificate Type ...................................14
           3.2.2. X.509 Certificate - Signature ......................14
           3.2.3. Revocation Lists (CRL and ARL) .....................14
           3.2.4. PKCS #7 wrapped X.509 certificate ..................15
           3.2.5. Location of Certificate Request Payloads ...........15
           3.2.6. Presence or Absence of Certificate Request
                  Payloads ...........................................15
           3.2.7. Certificate Requests ...............................15
           3.2.8. Robustness .........................................18
           3.2.9. Optimizations ......................................18
      3.3. Certificate Payload .......................................19
           3.3.1. Certificate Type ...................................20
           3.3.2. X.509 Certificate - Signature ......................20
           3.3.3. Revocation Lists (CRL and ARL) .....................20
           3.3.4. PKCS #7 Wrapped X.509 Certificate ..................20
           3.3.5. Location of Certificate Payloads ...................21
           3.3.6. Certificate Payloads Not Mandatory .................21
           3.3.7. Response to Multiple Certification
                  Authority Proposals ................................21
           3.3.8. Using Local Keying Materials .......................21
           3.3.9. Multiple End-Entity Certificates ...................22
           3.3.10. Robustness ........................................22
           3.3.11. Optimizations .....................................23
   4. Use of Certificates in RFC 4301 and IKEv2 ......................24
      4.1. Identification Payload ....................................24
      4.2. Certificate Request Payload ...............................24
           4.2.1. Revocation Lists (CRL and ARL) .....................24
      4.3. Certificate Payload .......................................25
           4.3.1. IKEv2's Hash and URL of X.509 Certificate ..........25
           4.3.2. Location of Certificate Payloads ...................25
           4.3.3. Ordering of Certificate Payloads ...................25
   5. Certificate Profile for IKEv1/ISAKMP and IKEv2 .................26
      5.1. X.509 Certificates ........................................26
           5.1.1. Versions ...........................................26
           5.1.2. Subject ............................................26
           5.1.3. X.509 Certificate Extensions .......................27
      5.2. X.509 Certificate Revocation Lists ........................33
           5.2.1. Multiple Sources of Certificate Revocation
                  Information ........................................34
           5.2.2. X.509 Certificate Revocation List Extensions .......34
      5.3. Strength of Signature Hashing Algorithms ..................35
   6. Configuration Data Exchange Conventions ........................36
      6.1. Certificates ..............................................36
      6.2. CRLs and ARLs .............................................37
      6.3. Public Keys ...............................................37
      6.4. PKCS#10 Certificate Signing Requests ......................37
   7. Security Considerations ........................................37
      7.1. Certificate Request Payload ...............................37
      7.2. IKEv1 Main Mode ...........................................37
      7.3. Disabling Certificate Checks ..............................38
   8. Acknowledgements ...............................................38
   9. References .....................................................38
      9.1. Normative References ......................................38
      9.2. Informative References ....................................39
   Appendix A. The Possible Dangers of Delta CRLs ....................40
   Appendix B. More on Empty CERTREQs ................................40

1. Introduction

IKE [1], ISAKMP [2], and IKEv2 [3] provide a secure key exchange mechanism for use with IPsec [4] [14]. In many cases, the peers authenticate using digital certificates as specified in PKIX [5]. Unfortunately, the combination of these standards leads to an underspecified set of requirements for the use of certificates in the context of IPsec.

ISAKMP references the PKIX certificate profile but, in many cases, merely specifies the contents of various messages without specifying their syntax or semantics. Meanwhile, the PKIX certificate profile provides a large set of certificate mechanisms that are generally applicable for Internet protocols, but little specific guidance for IPsec. Given the numerous underspecified choices, interoperability is hampered if all implementers do not make similar choices, or at least fail to account for implementations that have chosen differently.

This profile of the IKE and PKIX frameworks is intended to provide an agreed-upon standard for using PKI technology in the context of IPsec by profiling the PKIX framework for use with IKE and IPsec, and by documenting the contents of the relevant IKE payloads and further specifying their semantics.

In addition to providing a profile of IKE and PKIX, this document attempts to incorporate lessons learned from recent experience with both implementation and deployment, as well as the current state of related protocols and technologies.

Material from ISAKMP, IKEv1, IKEv2, or PKIX is not repeated here, and readers of this document are assumed to have read and understood those documents. The requirements and security aspects of those documents are fully relevant to this document as well.

This document is organized as follows. Section 2 defines special terminology used in the rest of this document, Section 3 provides the profile of IKEv1/ISAKMP, Section 4 provides a profile of IKEv2, and Section 5 provides the profile of PKIX. Section 6 covers conventions for the out-of-band exchange of keying materials for configuration purposes.

2. Terms and Definitions

Except for those terms that are defined immediately below, all terms used in this document are defined in either the PKIX [5], ISAKMP [2], IKEv1 [1], IKEv2 [3], or Domain of Interpretation (DOI) [6] documents.

Peer source address: The source address in packets from a peer. This address may be different from any addresses asserted as the "identity" of the peer.

FQDN: Fully qualified domain name.

ID_USER_FQDN: IKEv2 renamed ID_USER_FQDN to ID_RFC822_ADDR. Both are referred to as ID_USER_FQDN in this document.

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 [7].

3. Use of Certificates in RFC 2401 and IKEv1/ISAKMP

3.1. Identification Payload

The Identification (ID) Payload indicates the identity claimed by the sender. The recipient can then use the ID as a lookup key for policy and for certificate lookup in whatever certificate store or directory that it has available. Our primary concern in this section is to profile the ID payload so that it can be safely used to generate or lookup policy. IKE mandates the use of the ID payload in Phase 1.

The DOI [6] defines the 11 types of Identification Data that can be used and specifies the syntax for these types. These are discussed below in detail.

The ID payload requirements in this document cover only the portion of the explicit policy checks that deal with the Identification Payload specifically. For instance, in the case where ID does not contain an IP address, checks such as verifying that the peer source address is permitted by the relevant policy are not addressed here, as they are out of the scope of this document.

Implementations SHOULD populate ID with identity information that is contained within the end-entity certificate. Populating ID with identity information from the end-entity certificate enables recipients to use ID as a lookup key to find the peer end-entity certificate. The only case where implementations may populate ID with information that is not contained in the end-entity certificate is when ID contains the peer source address (a single address, not a subnet or range).

Because implementations may use ID as a lookup key to determine which policy to use, all implementations MUST be especially careful to verify the truthfulness of the contents by verifying that they correspond to some keying material demonstrably held by the peer.

Failure to do so may result in the use of an inappropriate or insecure policy. The following sections describe the methods for performing this binding.

The following table summarizes the binding of the Identification Payload to the contents of end-entity certificates and of identity information to policy. Each ID type is covered more thoroughly in the following sections.

   ID type  | Support  | Correspond  | Cert     | SPD lookup
            | for send | PKIX Attrib | matching | rules
   -------------------------------------------------------------------
            |          |             |          |
   IP*_ADDR | MUST [a] | SubjAltName | MUST [b] | [c], [d]
            |          | iPAddress   |          |
            |          |             |          |
   FQDN     | MUST [a] | SubjAltName | MUST [b] | [c], [d]
            |          | dNSName     |          |
            |          |             |          |
   USER_FQDN| MUST [a] | SubjAltName | MUST [b] | [c], [d]
            |          | rfc822Name  |          |
            |          |             |          |
   IP range | MUST NOT | n/a         | n/a      | n/a
            |          |             |          |
   DN       | MUST [a] | Entire      | MUST [b] | MUST support lookup
            |          | Subject,    |          | on any combination
            |          | bitwise     |          | of C, CN, O, or OU
            |          | compare     |          |
            |          |             |          |
   GN       | MUST NOT | n/a         | n/a      | n/a
            |          |             |          |
   KEY_ID   | MUST NOT | n/a         | n/a      | n/a
            |          |             |          |

[a] = Implementation MUST have the configuration option to send this

ID type in the ID payload. Whether or not the ID type is used is a matter of local configuration.

[b] = The ID in the ID payload MUST match the contents of the

corresponding field (listed) in the certificate exactly, with no other lookup. The matched ID MAY be used for Security Policy Database (SPD) lookup, but is not required to be used for this.

[c] = At a minimum, Implementation MUST be capable of being

configured to perform exact matching of the ID payload contents to an entry in the local SPD.

[d] = In addition, the implementation MAY also be configurable to

perform substring or wildcard matches of ID payload contents to entries in the local SPD. (More on this in Section 3.1.5.)

When sending an IPV4_ADDR, IPV6_ADDR, FQDN, or USER_FQDN, implementations MUST be able to be configured to send the same string as it appears in the corresponding SubjectAltName extension. This document RECOMMENDS that deployers use this configuration option. All these ID types are treated the same: as strings that can be compared easily and quickly to a corresponding string in an explicit value in the certificate. Of these types, FQDN and USER_FQDN are RECOMMENDED over IP addresses (see discussion in Section 3.1.1).

When sending a Distinguished Name (DN) as ID, implementations MUST send the entire DN in ID. Also, implementations MUST support at least the C, CN, O, and OU attributes for SPD matching. See Section 3.1.5 for more details about DN, including SPD matching.

Recipients MUST be able to perform SPD matching on the exact contents of the ID, and this SHOULD be the default setting. In addition, implementations MAY use substrings or wildcards in local policy configuration to do the SPD matching against the ID contents. In other words, implementations MUST be able to do exact matches of ID to SPD, but MAY also be configurable to do substring or wildcard matches of ID to SPD.

3.1.1. ID_IPV4_ADDR and ID_IPV6_ADDR

Implementations MUST support at least the ID_IPV4_ADDR or ID_IPV6_ADDR ID type, depending on whether the implementation supports IPv4, IPv6, or both. These addresses MUST be encoded in "network byte order", as specified in IP [8]: The least significant bit (LSB) of each octet is the LSB of the corresponding byte in the network address. For the ID_IPV4_ADDR type, the payload MUST contain exactly four octets [8]. For the ID_IPV6_ADDR type, the payload MUST contain exactly sixteen octets [10].

Implementations SHOULD NOT populate ID payload with IP addresses due to interoperability issues such as problems with Network Address Translator (NAT) traversal, and problems with IP verification behavior.

Deployments may only want to consider using the IP address as ID if all of the following are true:

the peer's IP address is static, not dynamically changing

the peer is NOT behind a NAT'ing device

the administrator intends the implementation to verify that the peer source address matches the IP address in the ID received, and that in the iPAddress field in the peer certificate's SubjectAltName extension.

Implementations MUST be capable of verifying that the IP address presented in ID matches via bitwise comparison the IP address present in the certificate's iPAddress field of the SubjectAltName extension. Implementations MUST perform this verification by default. When comparing the contents of ID with the iPAddress field in the SubjectAltName extension for equality, binary comparison MUST be performed. Note that certificates may contain multiple address identity types -- in which case, at least one must match the source IP. If the default is enabled, then a mismatch between the two addresses MUST be treated as an error, and security association setup MUST be aborted. This event SHOULD be auditable. Implementations MAY provide a configuration option to (i.e., local policy configuration can enable) skip that verification step, but that option MUST be off by default. We include the "option-to-skip- validation" in order to permit better interoperability as current implementations vary greatly in how they behave on this topic.

In addition, implementations MUST be capable of verifying that the address contained in the ID is the same as the address contained in the IP header. Implementations SHOULD be able to check the address in either the outermost or innermost IP header and MAY provide a configuration option for specifying which is to be checked. If there is no configuration option provided, an implementation SHOULD check the peer source address contained in the outermost header (as is the practice of most of today's implementations). If ID is one of the IP address types, then implementations MUST perform this verification by default. If this default is enabled, then a mismatch MUST be treated as an error, and security association setup MUST be aborted. This event SHOULD be auditable. Implementations MAY provide a configuration option to (i.e. local policy configuration can enable) skip that verification step, but that option MUST be off by default. We include the "option-to-skip-validation" in order to permit better interoperability, as current implementations vary greatly in how they behave on the topic of verification of source IP.

If the default for both the verifications above are enabled, then, by transitive property, the implementation will also be verifying that the peer source IP address matches via a bitwise comparison the contents of the iPAddress field in the SubjectAltName extension in the certificate. In addition, implementations MAY allow administrators to configure a local policy that explicitly requires that the peer source IP address match via a bitwise comparison the contents of the iPAddress field in the SubjectAltName extension in the certificate. Implementations SHOULD allow administrators to configure a local policy that skips this validation check.

Implementations MAY support substring, wildcard, or regular expression matching of the contents of ID to look up the policy in the SPD, and such would be a matter of local security policy configuration.

Implementations MAY use the IP address found in the header of packets received from the peer to look up the policy, but such implementations MUST still perform verification of the ID payload. Although packet IP addresses are inherently untrustworthy and must therefore be independently verified, it is often useful to use the apparent IP address of the peer to locate a general class of policies that will be used until the mandatory identity-based policy lookup can be performed.

For instance, if the IP address of the peer is unrecognized, a VPN gateway device might load a general "road warrior" policy that specifies a particular Certification Authority (CA) that is trusted to issue certificates that contain a valid rfc822Name, which can be used by that implementation to perform authorization based on access control lists (ACLs) after the peer's certificate has been validated. The rfc822Name can then be used to determine the policy that provides specific authorization to access resources (such as IP addresses, ports, and so forth).

As another example, if the IP address of the peer is recognized to be a known peer VPN endpoint, policy may be determined using that address, but until the identity (address) is validated by validating the peer certificate, the policy MUST NOT be used to authorize any IPsec traffic.

3.1.2. ID_FQDN

Implementations MUST support the ID_FQDN ID type, generally to support host-based access control lists for hosts without fixed IP addresses. However, implementations SHOULD NOT use the DNS to map the FQDN to IP addresses for input into any policy decisions, unless that mapping is known to be secure, for example, if DNSSEC [11] were employed for that FQDN.

If ID contains an ID_FQDN, implementations MUST be capable of verifying that the identity contained in the ID payload matches identity information contained in the peer end-entity certificate, in the dNSName field in the SubjectAltName extension. Implementations MUST perform this verification by default. When comparing the contents of ID with the dNSName field in the SubjectAltName extension for equality, case-insensitive string comparison MUST be performed. Note that case-insensitive string comparison works on internationalized domain names (IDNs) as well (See IDN [12]). Substring, wildcard, or regular expression matching MUST NOT be performed for this comparison. If this default is enabled, then a mismatch MUST be treated as an error, and security association setup MUST be aborted. This event SHOULD be auditable. Implementations MAY provide a configuration option to (i.e., local policy configuration can enable) skip that verification step, but that option MUST be off by default. We include the "option-to-skip- validation" in order to permit better interoperability, as current implementations vary greatly in how they behave on this topic.

3.1.3. ID_USER_FQDN

Implementations MUST support the ID_USER_FQDN ID type, generally to support user-based access control lists for users without fixed IP addresses. However, implementations SHOULD NOT use the DNS to map the FQDN portion to IP addresses for input into any policy decisions, unless that mapping is known to be secure, for example, if DNSSEC [11] were employed for that FQDN.

Implementations MUST be capable of verifying that the identity contained in the ID payload matches identity information contained in the peer end-entity certificate, in the rfc822Name field in the SubjectAltName extension. Implementations MUST perform this verification by default. When comparing the contents of ID with the rfc822Name field in the SubjectAltName extension for equality, case- insensitive string comparison MUST be performed. Note that case- insensitive string comparison works on internationalized domain names (IDNs) as well (See IDN [12]). Substring, wildcard, or regular expression matching MUST NOT be performed for this comparison. If this default is enabled, then a mismatch MUST be treated as an error, and security association setup MUST be aborted. This event SHOULD be auditable. Implementations MAY provide a configuration option to (i.e., local policy configuration can enable) skip that verification step, but that option MUST be off by default. We include the "option-to-skip-validation" in order to permit better interoperability, as current implementations vary greatly in how they behave on this topic.

Implementations MAY support substring, wildcard, or regular expression matching of the contents of ID to look up policy in the SPD, and such would be a matter of local security policy configuration.

3.1.4. ID_IPV4_ADDR_SUBNET, ID_IPV6_ADDR_SUBNET, ID_IPV4_ADDR_RANGE,

ID_IPV6_ADDR_RANGE

Note that RFC 3779 [13] defines blocks of addresses using the certificate extension identified by:

id-pe-ipAddrBlock OBJECT IDENTIFIER ::= { id-pe 7 }

although use of this extension in IKE is considered experimental at this time.

3.1.5. ID_DER_ASN1_DN

Implementations MUST support receiving the ID_DER_ASN1_DN ID type. Implementations MUST be capable of generating this type, and the decision to do so will be a matter of local security policy configuration. When generating this type, implementations MUST populate the contents of ID with the Subject field from the end- entity certificate, and MUST do so such that a binary comparison of the two will succeed. If there is not a match, this MUST be treated as an error, and security association setup MUST be aborted. This event SHOULD be auditable.

Implementations MUST NOT populate ID with the Subject from the end- entity certificate if it is empty, even though an empty certificate Subject is explicitly allowed in the "Subject" section of the PKIX certificate profile.

Regarding SPD matching, implementations MUST be able to perform matching based on a bitwise comparison of the entire DN in ID to its entry in the SPD. However, operational experience has shown that using the entire DN in local configuration is difficult, especially in large-scale deployments. Therefore, implementations also MUST be able to perform SPD matches of any combination of one or more of the C, CN, O, OU attributes within Subject DN in the ID to the same in the SPD. Implementations MAY support matching using additional DN attributes in any combination, although interoperability is far from certain and is dubious. Implementations MAY also support performing substring, wildcard, or regular expression matches for any of its supported DN attributes from ID, in any combination, to the SPD. Such flexibility allows deployers to create one SPD entry on the gateway for an entire department of a company (e.g., O=Foobar Inc., OU=Engineering) while still allowing them to draw out other details from the DN (e.g., CN=John Doe) for auditing purposes. All the above is a matter of local implementation and local policy definition and enforcement capability, not bits on the wire, but will have a great impact on interoperability.

3.1.6. ID_DER_ASN1_GN

Implementations MUST NOT generate this type, because the recipient will be unlikely to know how to use it.

3.1.7. ID_KEY_ID

The ID_KEY_ID type used to specify pre-shared keys and thus is out of scope.

3.1.8. Selecting an Identity from a Certificate

Implementations MUST support certificates that contain more than a single identity, such as when the Subject field and the SubjectAltName extension are both populated, or the SubjectAltName extension contains multiple identities irrespective of whether or not the Subject is empty. In many cases, a certificate will contain an identity, such as an IP address, in the SubjectAltName extension in addition to a non-empty Subject.

Implementations should populate ID with whichever identity is likely to be named in the peer's policy. In practice, this generally means FQDN, or USER_FQDN, but this information may also be available to the administrator through some out-of-band means. In the absence of such out-of-band configuration information, the identity with which an implementation chooses to populate the ID payload is a local matter.

3.1.9. Subject for DN Only

If an FQDN is intended to be processed as an identity for the purposes of ID matching, it MUST be placed in the dNSName field of the SubjectAltName extension. Implementations MUST NOT populate the Subject with an FQDN in place of populating the dNSName field of the SubjectAltName extension.

While nothing prevents an FQDN, USER_FQDN, or IP address information from appearing somewhere in the Subject contents, such entries MUST NOT be interpreted as identity information for the purposes of matching with ID or for policy lookup.

3.1.10. Binding Identity to Policy

In the presence of certificates that contain multiple identities, implementations should select the most appropriate identity from the certificate and populate the ID with that. The recipient MUST use the identity sent as a first key when selecting the policy. The recipient MUST also use the most specific policy from that database if there are overlapping policies caused by wildcards (or the implementation can de-correlate the policy database so there will not be overlapping entries, or it can also forbid creation of overlapping policies and leave the de-correlation process to the administrator, but, as this moves the problem to the administrator, it is NOT RECOMMENDED).

For example, imagine that an implementation is configured with a certificate that contains both a non-empty Subject and a dNSName. The sender's policy may specify which of those to use, and it indicates the policy to the other end by sending that ID. If the recipient has both a specific policy for the dNSName for this host and generic wildcard rule for some attributes present in the Subject field, it will match a different policy depending on which ID is sent. As the sender knows why it wanted to connect the peer, it also knows what identity it should use to match the policy it needs to the operation it tries to perform; it is the only party who can select the ID adequately.

In the event that the policy cannot be found in the recipient's SPD using the ID sent, then the recipient MAY use the other identities in the certificate when attempting to match a suitable policy. For example, say the certificate contains a non-empty Subject field, a dNSName and an iPAddress. If an iPAddress is sent in ID but no specific entry exists for the address in the policy database, the recipient MAY search in the policy database based on the Subject or the dNSName contained in the certificate.

3.2. Certificate Request Payload

The Certificate Request (CERTREQ) Payload allows an implementation to request that a peer provide some set of certificates or certificate revocation lists (CRLs). It is not clear from ISAKMP exactly how that set should be specified or how the peer should respond. We describe the semantics on both sides.

3.2.1. Certificate Type

The Certificate Type field identifies to the peer the type of certificate keying materials that are desired. ISAKMP defines 10 types of Certificate Data that can be requested and specifies the syntax for these types. For the purposes of this document, only the following types are relevant:

o X.509 Certificate - Signature
o Revocation Lists (CRL and ARL)
o PKCS #7 wrapped X.509 certificate

The use of the other types are out of the scope of this document:

o X.509 Certificate - Key Exchange
o PGP (Pretty Good Privacy) Certificate
o DNS Signed Key
o Kerberos Tokens
o SPKI (Simple Public Key Infrastructure) Certificate
o X.509 Certificate Attribute

3.2.2. X.509 Certificate - Signature

This type requests that the end-entity certificate be a certificate used for signing.

3.2.3. Revocation Lists (CRL and ARL)

ISAKMP does not support Certificate Payload sizes over approximately 64K, which is too small for many CRLs, and UDP fragmentation is likely to occur at sizes much smaller than that. Therefore, the acquisition of revocation material is to be dealt with out-of-band of IKE. For this and other reasons, implementations SHOULD NOT generate CERTREQs where the Certificate Type is "Certificate Revocation List (CRL)" or "Authority Revocation List (ARL)". Implementations that do generate such CERTREQs MUST NOT require the recipient to respond with a CRL or ARL, and MUST NOT fail when not receiving any. Upon receipt of such a CERTREQ, implementations MAY ignore the request.

In lieu of exchanging revocation lists in-band, a pointer to revocation checking SHOULD be listed in either the CRLDistributionPoints (CDP) or the AuthorityInfoAccess (AIA) certificate extensions (see Section 5 for details). Unless other methods for obtaining revocation information are available, implementations SHOULD be able to process these attributes, and from them be able to identify cached revocation material, or retrieve the relevant revocation material from a URL, for validation processing. In addition, implementations MUST have the ability to configure validation checking information for each certification authority. Regardless of the method (CDP, AIA, or static configuration), the acquisition of revocation material SHOULD occur out-of-band of IKE. Note, however, that an inability to access revocation status data through out-of-band means provides a potential security vulnerability that could potentially be exploited by an attacker.

3.2.4. PKCS #7 wrapped X.509 certificate

This ID type defines a particular encoding (not a particular certificate type); some current implementations may ignore CERTREQs they receive that contain this ID type, and the editors are unaware of any implementations that generate such CERTREQ messages. Therefore, the use of this type is deprecated. Implementations SHOULD NOT require CERTREQs that contain this Certificate Type. Implementations that receive CERTREQs that contain this ID type MAY treat such payloads as synonymous with "X.509 Certificate - Signature".

3.2.5. Location of Certificate Request Payloads

In IKEv1 Main Mode, the CERTREQ payload MUST be in messages 4 and 5.

3.2.6. Presence or Absence of Certificate Request Payloads

When in-band exchange of certificate keying materials is desired, implementations MUST inform the peer of this by sending at least one CERTREQ. In other words, an implementation that does not send any CERTREQs during an exchange SHOULD NOT expect to receive any CERT payloads.

3.2.7. Certificate Requests

3.2.7.1. Specifying Certification Authorities

When requesting in-band exchange of keying materials, implementations SHOULD generate CERTREQs for every peer trust anchor that local policy explicitly deems trusted during a given exchange. Implementations SHOULD populate the Certification Authority field with the Subject field of the trust anchor, populated such that binary comparison of the Subject and the Certification Authority will succeed.

Upon receipt of a CERTREQ, implementations MUST respond by sending at least the end-entity certificate corresponding to the Certification Authority listed in the CERTREQ unless local security policy configuration specifies that keying materials must be exchanged out- of-band. Implementations MAY send certificates other than the end- entity certificate (see Section 3.3 for discussion).

Note that, in the case where multiple end-entity certificates may be available that chain to different trust anchors, implementations SHOULD resort to local heuristics to determine which trust anchor is most appropriate to use for generating the CERTREQ. Such heuristics are out of the scope of this document.

3.2.7.2. Empty Certification Authority Field

Implementations SHOULD generate CERTREQs where the Certificate Type is "X.509 Certificate - Signature" and where the Certification Authority field is not empty. However, implementations MAY generate CERTREQs with an empty Certification Authority field under special conditions. Although PKIX prohibits certificates with an empty Issuer field, there does exist a use case where doing so is appropriate, and carries special meaning in the IKE context. This has become a convention within the IKE interoperability tests and usage space, and so its use is specified, explained here for the sake of interoperability.

USE CASE: Consider the rare case where you have a gateway with multiple policies for a large number of IKE peers: some of these peers are business partners, some are remote-access employees, some are teleworkers, some are branch offices, and/or the gateway may be simultaneously serving many customers (e.g., Virtual Routers). The total number of certificates, and corresponding trust anchors, is very high -- say, hundreds. Each of these policies is configured with one or more acceptable trust anchors, so that in total, the gateway has one hundred (100) trust anchors that could possibly used to authenticate an incoming connection. Assume that many of those connections originate from hosts/gateways with dynamically assigned IP addresses, so that the source IP of the IKE initiator is not known to the gateway, nor is the identity of the initiator (until it is revealed in Main Mode message 5). In IKE main mode message 4, the responder gateway will need to send a CERTREQ to the initiator. Given this example, the gateway will have no idea which of the hundred possible Certification Authorities to send in the CERTREQ. Sending all possible Certification Authorities will cause significant processing delays, bandwidth consumption, and UDP fragmentation, so this tactic is ruled out.

In such a deployment, the responder gateway implementation should be able to do all it can to indicate a Certification Authority in the CERTREQ. This means the responder SHOULD first check SPD to see if it can match the source IP, and find some indication of which CA is associated with that IP. If this fails (because the source IP is not familiar, as in the case above), then the responder SHOULD have a configuration option specifying which CAs are the default CAs to indicate in CERTREQ during such ambiguous connections (e.g., send CERTREQ with these N CAs if there is an unknown source IP). If such a fall-back is not configured or impractical in a certain deployment scenario, then the responder implementation SHOULD have both of the following configuration options:

send a CERTREQ payload with an empty Certification Authority field, or

terminate the negotiation with an appropriate error message and audit log entry.

Receiving a CERTREQ payload with an empty Certification Authority field indicates that the recipient should send all/any end-entity certificates it has, regardless of the trust anchor. The initiator should be aware of what policy and which identity it will use, as it initiated the connection on a matched policy to begin with, and can thus respond with the appropriate certificate.

If, after sending an empty CERTREQ in Main Mode message 4, a responder receives a certificate in message 5 that chains to a trust anchor that the responder either (a) does NOT support, or (b) was not configured for the policy (that policy was now able to be matched due to having the initiator's certificate present), this MUST be treated as an error, and security association setup MUST be aborted. This event SHOULD be auditable.

Instead of sending an empty CERTREQ, the responder implementation MAY be configured to terminate the negotiation on the grounds of a conflict with locally configured security policy.

The decision of which to configure is a matter of local security policy; this document RECOMMENDS that both options be presented to administrators.

More examples and explanation of this issue are included in "More on Empty CERTREQs" (Appendix B).

3.2.8. Robustness

3.2.8.1. Unrecognized or Unsupported Certificate Types

Implementations MUST be able to deal with receiving CERTREQs with unsupported Certificate Types. Absent any recognized and supported CERTREQ types, implementations MAY treat them as if they are of a supported type with the Certification Authority field left empty, depending on local policy. ISAKMP [2] Section 5.10, "Certificate Request Payload Processing", specifies additional processing.

3.2.8.2. Undecodable Certification Authority Fields

Implementations MUST be able to deal with receiving CERTREQs with undecodable Certification Authority fields. Implementations MAY ignore such payloads, depending on local policy. ISAKMP specifies other actions which may be taken.

3.2.8.3. Ordering of Certificate Request Payloads

Implementations MUST NOT assume that CERTREQs are ordered in any way.

3.2.9. Optimizations

3.2.9.1. Duplicate Certificate Request Payloads

Implementations SHOULD NOT send duplicate CERTREQs during an exchange.

3.2.9.2. Name Lowest 'Common' Certification Authorities

When a peer's certificate keying material has been cached, an implementation can send a hint to the peer to elide some of the certificates the peer would normally include in the response. In addition to the normal set of CERTREQs that are sent specifying the trust anchors, an implementation MAY send CERTREQs specifying the relevant cached end-entity certificates. When sending these hints, it is still necessary to send the normal set of trust anchor CERTREQs because the hints do not sufficiently convey all of the information required by the peer. Specifically, either the peer may not support this optimization or there may be additional chains that could be used in this context but will not be if only the end-entity certificate is specified.

No special processing is required on the part of the recipient of such a CERTREQ, and the end-entity certificates will still be sent. On the other hand, the recipient MAY elect to elide certificates based on receipt of such hints.

CERTREQs must contain information that identifies a Certification Authority certificate, which results in the peer always sending at least the end-entity certificate. Always sending the end-entity certificate allows implementations to determine unambiguously when a new certificate is being used by a peer (perhaps because the previous certificate has just expired), which may result in a failure because a new intermediate CA certificate might not be available to validate the new end-entity certificate). Implementations that implement this optimization MUST recognize when the end-entity certificate has changed and respond to it by not performing this optimization if the exchange must be retried so that any missing keying materials will be sent during retry.

3.2.9.3. Example

Imagine that an IKEv1 implementation has previously received and cached the peer certificate chain TA->CA1->CA2->EE. If, during a subsequent exchange, this implementation sends a CERTREQ containing the Subject field in certificate TA, this implementation is requesting that the peer send at least three certificates: CA1, CA2, and EE. On the other hand, if this implementation also sends a CERTREQ containing the Subject field of CA2, the implementation is providing a hint that only one certificate needs to be sent: EE. Note that in this example, the fact that TA is a trust anchor should not be construed to imply that TA is a self-signed certificate.

3.3. Certificate Payload

The Certificate (CERT) Payload allows the peer to transmit a single certificate or CRL. Multiple certificates should be transmitted in multiple payloads. For backwards-compatibility reasons, implementations MAY send intermediate CA certificates in addition to the appropriate end-entity certificate(s), but SHOULD NOT send any CRLs, ARLs, or trust anchors. Exchanging trust anchors and especially CRLs and ARLs in IKE would increase the likelihood of UDP fragmentation, make the IKE exchange more complex, and consume additional network bandwidth.

Note, however, that while the sender of the CERT payloads SHOULD NOT send any certificates it considers trust anchors, it's possible that the recipient may consider any given intermediate CA certificate to be a trust anchor. For instance, imagine the sender has the certificate chain TA1->CA1->EE1 while the recipient has the certificate chain TA2->EE2 where TA2=CA1. The sender is merely including an intermediate CA certificate, while the recipient receives a trust anchor.

However, not all certificate forms that are legal in the PKIX certificate profile make sense in the context of IPsec. The issue of how to represent IKE-meaningful name-forms in a certificate is especially problematic. This document provides a profile for a subset of the PKIX certificate profile that makes sense for IKEv1/ ISAKMP.

3.3.1. Certificate Type

The Certificate Type field identifies to the peer the type of certificate keying materials that are included. ISAKMP defines 10 types of Certificate Data that can be sent and specifies the syntax for these types. For the purposes of this document, only the following types are relevant:

o X.509 Certificate - Signature
o Revocation Lists (CRL and ARL)
o PKCS #7 wrapped X.509 certificate

The use of the other types are out of the scope of this document:

o X.509 Certificate - Key Exchange
o PGP Certificate
o DNS Signed Key
o Kerberos Tokens
o SPKI Certificate
o X.509 Certificate Attribute

3.3.2. X.509 Certificate - Signature

This type specifies that Certificate Data contains a certificate used for signing.

3.3.3. Revocation Lists (CRL and ARL)

These types specify that Certificate Data contains an X.509 CRL or ARL. These types SHOULD NOT be sent in IKE. See Section 3.2.3 for discussion.

3.3.4. PKCS #7 Wrapped X.509 Certificate

This type defines a particular encoding, not a particular certificate type. Implementations SHOULD NOT generate CERTs that contain this Certificate Type. Implementations SHOULD accept CERTs that contain this Certificate Type because several implementations are known to generate them. Note that those implementations sometimes include entire certificate hierarchies inside a single CERT PKCS #7 payload, which violates the requirement specified in ISAKMP that this payload contain a single certificate.

3.3.5. Location of Certificate Payloads

In IKEv1 Main Mode, the CERT payload MUST be in messages 5 and 6.

3.3.6. Certificate Payloads Not Mandatory

An implementation that does not receive any CERTREQs during an exchange SHOULD NOT send any CERT payloads, except when explicitly configured to proactively send CERT payloads in order to interoperate with non-compliant implementations that fail to send CERTREQs even when certificates are desired. In this case, an implementation MAY send the certificate chain (not including the trust anchor) associated with the end-entity certificate. This MUST NOT be the default behavior of implementations.

Implementations whose local security policy configuration expects that a peer must receive certificates through out-of-band means SHOULD ignore any CERTREQ messages that are received. Such a condition has been known to occur due to non-compliant or buggy implementations.

Implementations that receive CERTREQs from a peer that contain only unrecognized Certification Authorities MAY elect to terminate the exchange, in order to avoid unnecessary and potentially expensive cryptographic processing, denial-of-service (resource starvation) attacks.

3.3.7. Response to Multiple Certification Authority Proposals

In response to multiple CERTREQs that contain different Certification Authority identities, implementations MAY respond using an end-entity certificate which chains to a CA that matches any of the identities provided by the peer.

3.3.8. Using Local Keying Materials

Implementations MAY elect to skip parsing or otherwise decoding a given set of CERTs if those same keying materials are available via some preferable means, such as the case where certificates from a previous exchange have been cached.

3.3.9. Multiple End-Entity Certificates

Implementations SHOULD NOT send multiple end-entity certificates and recipients SHOULD NOT be expected to iterate over multiple end-entity certificates.

If multiple end-entity certificates are sent, they MUST have the same public key; otherwise, the responder does not know which key was used in the Main Mode message 5.

3.3.10. Robustness

3.3.10.1. Unrecognized or Unsupported Certificate Types

Implementations MUST be able to deal with receiving CERTs with unrecognized or unsupported Certificate Types. Implementations MAY discard such payloads, depending on local policy. ISAKMP [2] Section 5.10, "Certificate Request Payload Processing", specifies additional processing.

3.3.10.2. Undecodable Certificate Data Fields

Implementations MUST be able to deal with receiving CERTs with undecodable Certificate Data fields. Implementations MAY discard such payloads, depending on local policy. ISAKMP specifies other actions that may be taken.

3.3.10.3. Ordering of Certificate Payloads

Implementations MUST NOT assume that CERTs are ordered in any way.

3.3.10.4. Duplicate Certificate Payloads

Implementations MUST support receiving multiple identical CERTs during an exchange.

3.3.10.5. Irrelevant Certificates

Implementations MUST be prepared to receive certificates and CRLs that are not relevant to the current exchange. Implementations MAY discard such extraneous certificates and CRLs.

Implementations MAY send certificates that are irrelevant to an exchange. One reason for including certificates that are irrelevant to an exchange is to minimize the threat of leaking identifying information in exchanges where CERT is not encrypted in IKEv1. It should be noted, however, that this probably provides rather poor protection against leaking the identity.

Another reason for including certificates that seem irrelevant to an exchange is that there may be two chains from the Certification Authority to the end entity, each of which is only valid with certain validation parameters (such as acceptable policies). Since the end- entity doesn't know which parameters the relying party is using, it should send the certificates needed for both chains (even if there's only one CERTREQ).

Implementations SHOULD NOT send multiple end-entity certificates and recipients SHOULD NOT be expected to iterate over multiple end-entity certificates.

3.3.11. Optimizations

3.3.11.1. Duplicate Certificate Payloads

Implementations SHOULD NOT send duplicate CERTs during an exchange. Such payloads should be suppressed.

3.3.11.2. Send Lowest 'Common' Certificates

When multiple CERTREQs are received that specify certification authorities within the end-entity certificate chain, implementations MAY send the shortest chain possible. However, implementations SHOULD always send the end-entity certificate. See Section 3.2.9.2 for more discussion of this optimization.

3.3.11.3. Ignore Duplicate Certificate Payloads

Implementations MAY employ local means to recognize CERTs that have already been received and SHOULD discard these duplicate CERTs.

3.3.11.4. Hash Payload

IKEv1 specifies the optional use of the Hash Payload to carry a pointer to a certificate in either of the Phase 1 public key encryption modes. This pointer is used by an implementation to locate the end-entity certificate that contains the public key that a peer should use for encrypting payloads during the exchange.

Implementations SHOULD include this payload whenever the public portion of the keypair has been placed in a certificate.

4. Use of Certificates in RFC 4301 and IKEv2

4.1. Identification Payload

The Peer Authorization Database (PAD) as described in RFC 4301 [14] describes the use of the ID payload in IKEv2 and provides a formal model for the binding of identity to policy in addition to providing services that deal more specifically with the details of policy enforcement, which are generally out of scope of this document. The PAD is intended to provide a link between the SPD and the security association management in protocols such as IKE. See RFC 4301 [14], Section 4.4.3 for more details.

Note that IKEv2 adds an optional IDr payload in the second exchange that the initiator may send to the responder in order to specify which of the responder's multiple identities should be used. The responder MAY choose to send an IDr in the third exchange that differs in type or content from the initiator-generated IDr. The initiator MUST be able to receive a responder-generated IDr that is a different type from the one the initiator generated.

4.2. Certificate Request Payload

4.2.1. Revocation Lists (CRL and ARL)

IKEv2 does not support Certificate Payload sizes over approximately 64K. See Section 3.2.3 for the problems this can cause.

4.2.1.1. IKEv2's Hash and URL of X.509 certificate

This ID type defines a request for the peer to send a hash and URL of its X.509 certificate, instead of the actual certificate itself. This is a particularly useful mechanism when the peer is a device with little memory and lower bandwidth, e.g., a mobile handset or consumer electronics device.

If the IKEv2 implementation supports URL lookups, and prefers such a URL to receiving actual certificates, then the implementation will want to send a notify of type HTTP_CERT_LOOKUP_SUPPORTED. From IKEv2 [3], Section 3.10.1, "This notification MAY be included in any message that can include a CERTREQ payload and indicates that the sender is capable of looking up certificates based on an HTTP-based URL (and hence presumably would prefer to receive certificate specifications in that format)". If an HTTP_CERT_LOOKUP_SUPPORTED notification is sent, the sender MUST support the http scheme. See Section 4.3.1 for more discussion of HTTP_CERT_LOOKUP_SUPPORTED.

4.2.1.2. Location of Certificate Request Payloads

In IKEv2, the CERTREQ payload must be in messages 2 and 3. Note that in IKEv2, it is possible to have one side authenticating with certificates while the other side authenticates with pre-shared keys.

4.3. Certificate Payload

4.3.1. IKEv2's Hash and URL of X.509 Certificate

This type specifies that Certificate Data contains a hash and the URL to a repository where an X.509 certificate can be retrieved.

An implementation that sends an HTTP_CERT_LOOKUP_SUPPORTED notification MUST support the http scheme and MAY support the ftp scheme, and MUST NOT require any specific form of the url-path, and it SHOULD support having user-name, password, and port parts in the URL. The following are examples of mandatory forms:

   o  http://certs.example.com/certificate.cer
   o  http://certs.example.com/certs/cert.pl?u=foo;a=pw;valid-to=+86400
   o  http://certs.example.com/%0a/../foo/bar/zappa

while the following is an example of a form that SHOULD be supported:

   o  http://user:password@certs.example.com:8888/certificate.cer

FTP MAY be supported, and if it is, the following is an example of the ftp scheme that MUST be supported:

   o  ftp://ftp.example.com/pub/certificate.cer

4.3.2. Location of Certificate Payloads

In IKEv2, the CERT payload MUST be in messages 3 and 4. Note that in IKEv2, it is possible to have one side authenticating with certificates while the other side authenticates with pre-shared keys.

4.3.3. Ordering of Certificate Payloads

For IKEv2, implementations MUST NOT assume that any but the first CERT is ordered in any way. IKEv2 specifies that the first CERT contain an end-entity certificate that can be used to authenticate the peer.

5. Certificate Profile for IKEv1/ISAKMP and IKEv2

Except where specifically stated in this document, implementations MUST conform to the requirements of the PKIX [5] certificate profile.

5.1. X.509 Certificates

Users deploying IKE and IPsec with certificates have often had little control over the capabilities of CAs available to them. Implementations of this specification may include configuration knobs to disable checks required by this specification in order to permit use with inflexible and/or noncompliant CAs. However, all checks on certificates exist for a specific reason involving the security of the entire system. Therefore, all checks MUST be enabled by default. Administrators and users ought to understand the security purpose for the various checks, and be clear on what security will be lost by disabling the check.

5.1.1. Versions

Although PKIX states that "implementations SHOULD be prepared to accept any version certificate", in practice, this profile requires certain extensions that necessitate the use of Version 3 certificates for all but self-signed certificates used as trust anchors. Implementations that conform to this document MAY therefore reject Version 1 and Version 2 certificates in all other cases.

5.1.2. Subject

Certification Authority implementations MUST be able to create certificates with Subject fields with at least the following four attributes: CN, C, O, and OU. Implementations MAY support other Subject attributes as well. The contents of these attributes SHOULD be configurable on a certificate-by-certificate basis, as these fields will likely be used by IKE implementations to match SPD policy.

See Section 3.1.5 for details on how IKE implementations need to be able to process Subject field attributes for SPD policy lookup.

5.1.2.1. Empty Subject Name

IKE Implementations MUST accept certificates that contain an empty Subject field, as specified in the PKIX certificate profile. Identity information in such certificates will be contained entirely in the SubjectAltName extension.

5.1.2.2. Specifying Hosts and not FQDN in the Subject Name

Implementations that desire to place host names that are not intended to be processed by recipients as FQDNs (for instance "Gateway Router") in the Subject MUST use the commonName attribute.

5.1.2.3. EmailAddress

As specified in the PKIX certificate profile, implementations MUST NOT populate X.500 distinguished names with the emailAddress attribute.

5.1.3. X.509 Certificate Extensions

Conforming IKE implementations MUST recognize extensions that must or may be marked critical according to this specification. These extensions are: KeyUsage, SubjectAltName, and BasicConstraints.

Certification Authority implementations SHOULD generate certificates such that the extension criticality bits are set in accordance with the PKIX certificate profile and this document. With respect to compliance with the PKIX certificate profile, IKE implementations processing certificates MAY ignore the value of the criticality bit for extensions that are supported by that implementation, but MUST support the criticality bit for extensions that are not supported by that implementation. That is, a relying party SHOULD processes all the extensions it is aware of whether the bit is true or false -- the bit says what happens when a relying party cannot process an extension.

          implements    bit in cert     PKIX mandate    behavior
          ------------------------------------------------------
          yes           true            true            ok
          yes           true            false           ok or reject
          yes           false           true            ok or reject
          yes           false           false           ok
          no            true            true            reject
          no            true            false           reject
          no            false           true            reject
          no            false           false           ok

5.1.3.1. AuthorityKeyIdentifier and SubjectKeyIdentifier

Implementations SHOULD NOT assume support for the AuthorityKeyIdentifier or SubjectKeyIdentifier extensions. Thus, Certification Authority implementations should not generate certificate hierarchies that are overly complex to process in the absence of these extensions, such as those that require possibly verifying a signature against a large number of similarly named CA certificates in order to find the CA certificate that contains the key that was used to generate the signature.

5.1.3.2. KeyUsage

IKE uses an end-entity certificate in the authentication process. The end-entity certificate may be used for multiple applications. As such, the CA can impose some constraints on the manner that a public key ought to be used. The KeyUsage (KU) and ExtendedKeyUsage (EKU) extensions apply in this situation.

Since we are talking about using the public key to validate a signature, if the KeyUsage extension is present, then at least one of the digitalSignature or the nonRepudiation bits in the KeyUsage extension MUST be set (both can be set as well). It is also fine if other KeyUsage bits are set.

A summary of the logic flow for peer cert validation follows:

If no KU extension, continue.

If KU present and doesn't mention digitalSignature or nonRepudiation (both, in addition to other KUs, is also fine), reject cert.

If none of the above, continue.

5.1.3.3. PrivateKeyUsagePeriod

The PKIX certificate profile recommends against the use of this extension. The PrivateKeyUsageExtension is intended to be used when signatures will need to be verified long past the time when signatures using the private keypair may be generated. Since IKE security associations (SAs) are short-lived relative to the intended use of this extension in addition to the fact that each signature is validated only a single time, the usefulness of this extension in the context of IKE is unclear. Therefore, Certification Authority implementations MUST NOT generate certificates that contain the PrivateKeyUsagePeriod extension. If an IKE implementation receives a certificate with this set, it SHOULD ignore it.

5.1.3.4. CertificatePolicies

Many IKE implementations do not currently provide support for the CertificatePolicies extension. Therefore, Certification Authority implementations that generate certificates that contain this extension SHOULD NOT mark the extension as critical. As is the case with all certificate extensions, a relying party receiving this extension but who can process the extension SHOULD NOT reject the certificate because it contains the extension.

5.1.3.5. PolicyMappings

Many IKE implementations do not support the PolicyMappings extension. Therefore, implementations that generate certificates that contain this extension SHOULD NOT mark the extension as critical.

5.1.3.6. SubjectAltName

Deployments that intend to use an ID of FQDN, USER_FQDN, IPV4_ADDR, or IPV6_ADDR MUST issue certificates with the corresponding SubjectAltName fields populated with the same data. Implementations SHOULD generate only the following GeneralName choices in the SubjectAltName extension, as these choices map to legal IKEv1/ISAKMP/ IKEv2 Identification Payload types: rfc822Name, dNSName, or iPAddress. Although it is possible to specify any GeneralName choice in the Identification Payload by using the ID_DER_ASN1_GN ID type, implementations SHOULD NOT assume support for such functionality, and SHOULD NOT generate certificates that do so.

5.1.3.6.1. dNSName

If the IKE ID type is FQDN, then this field MUST contain a fully qualified domain name. If the IKE ID type is FQDN, then the dNSName field MUST match its contents. Implementations MUST NOT generate names that contain wildcards. Implementations MAY treat certificates that contain wildcards in this field as syntactically invalid.

Although this field is in the form of an FQDN, IKE implementations SHOULD NOT assume that this field contains an FQDN that will resolve via the DNS, unless this is known by way of some out-of-band mechanism. Such a mechanism is out of the scope of this document. Implementations SHOULD NOT treat the failure to resolve as an error.

5.1.3.6.2. iPAddress

If the IKE ID type is IPV4_ADDR or IPV6_ADDR, then the iPAddress field MUST match its contents. Note that although PKIX permits CIDR [15] notation in the "Name Constraints" extension, the PKIX certificate profile explicitly prohibits using CIDR notation for conveying identity information. In other words, the CIDR notation MUST NOT be used in the SubjectAltName extension.

5.1.3.6.3. rfc822Name

If the IKE ID type is USER_FQDN, then the rfc822Name field MUST match its contents. Although this field is in the form of an Internet mail address, IKE implementations SHOULD NOT assume that this field contains a valid email address, unless this is known by way of some out-of-band mechanism. Such a mechanism is out of the scope of this document.

5.1.3.7. IssuerAltName

Certification Authority implementations SHOULD NOT assume that other implementations support the IssuerAltName extension, and especially should not assume that information contained in this extension will be displayed to end users.

5.1.3.8. SubjectDirectoryAttributes

The SubjectDirectoryAttributes extension is intended to convey identification attributes of the subject. IKE implementations MAY ignore this extension when it is marked non-critical, as the PKIX certificate profile mandates.

5.1.3.9. BasicConstraints

The PKIX certificate profile mandates that CA certificates contain this extension and that it be marked critical. IKE implementations SHOULD reject CA certificates that do not contain this extension. For backwards compatibility, implementations may accept such certificates if explicitly configured to do so, but the default for this setting MUST be to reject such certificates.

5.1.3.10. NameConstraints

Many IKE implementations do not support the NameConstraints extension. Since the PKIX certificate profile mandates that this extension be marked critical when present, Certification Authority implementations that are interested in maximal interoperability for IKE SHOULD NOT generate certificates that contain this extension.

5.1.3.11. PolicyConstraints

Many IKE implementations do not support the PolicyConstraints extension. Since the PKIX certificate profile mandates that this extension be marked critical when present, Certification Authority implementations that are interested in maximal interoperability for IKE SHOULD NOT generate certificates that contain this extension.

5.1.3.12. ExtendedKeyUsage

The CA SHOULD NOT include the ExtendedKeyUsage (EKU) extension in certificates for use with IKE. Note that there were three IPsec- related object identifiers in EKU that were assigned in 1999. The semantics of these values were never clearly defined. The use of these three EKU values in IKE/IPsec is obsolete and explicitly deprecated by this specification. CAs SHOULD NOT issue certificates for use in IKE with them. (For historical reference only, those values were id-kp-ipsecEndSystem, id-kp-ipsecTunnel, and id-kp- ipsecUser.)

The CA SHOULD NOT mark the EKU extension in certificates for use with IKE and one or more other applications. Nevertheless, this document defines an ExtendedKeyUsage keyPurposeID that MAY be used to limit a certificate's use:

   id-kp-ipsecIKE OBJECT IDENTIFIER ::= { id-kp 17 }

where id-kp is defined in RFC 3280 [5]. If a certificate is intended to be used with both IKE and other applications, and one of the other applications requires use of an EKU value, then such certificates MUST contain either the keyPurposeID id-kp-ipsecIKE or anyExtendedKeyUsage [5], as well as the keyPurposeID values associated with the other applications. Similarly, if a CA issues multiple otherwise-similar certificates for multiple applications including IKE, and it is intended that the IKE certificate NOT be used with another application, the IKE certificate MAY contain an EKU extension listing a keyPurposeID of id-kp-ipsecIKE to discourage its use with the other application. Recall, however, that EKU extensions in certificates meant for use in IKE are NOT RECOMMENDED.

Conforming IKE implementations are not required to support EKU. If a critical EKU extension appears in a certificate and EKU is not supported by the implementation, then RFC 3280 requires that the certificate be rejected. Implementations that do support EKU MUST support the following logic for certificate validation:

If no EKU extension, continue.

If EKU present AND contains either id-kp-ipsecIKE or anyExtendedKeyUsage, continue.

Otherwise, reject cert.

5.1.3.13. CRLDistributionPoints

Because this document deprecates the sending of CRLs in-band, the use of CRLDistributionPoints (CDP) becomes very important if CRLs are used for revocation checking (as opposed to, say, Online Certificate Status Protocol - OCSP [16]). The IPsec peer either needs to have a URL for a CRL written into its local configuration, or it needs to learn it from CDP. Therefore, Certification Authority implementations SHOULD issue certificates with a populated CDP.

Failure to validate the CRLDistributionPoints/ IssuingDistributionPoint pair can result in CRL substitution where an entity knowingly substitutes a known good CRL from a different distribution point for the CRL that is supposed to be used, which would show the entity as revoked. IKE implementations MUST support validating that the contents of CRLDistributionPoints match those of the IssuingDistributionPoint to prevent CRL substitution when the issuing CA is using them. At least one CA is known to default to this type of CRL use. See Section 5.2.2.5 for more information.

CDPs SHOULD be "resolvable". Several non-compliant Certification Authority implementations are well known for including unresolvable CDPs like http://localhost/path_to_CRL and http:///path_to_CRL that are equivalent to failing to include the CDP extension in the certificate.

See the IETF IPR Web page for CRLDistributionPoints intellectual property rights (IPR) information. Note that both the CRLDistributionPoints and IssuingDistributionPoint extensions are RECOMMENDED but not REQUIRED by the PKIX certificate profile, so there is no requirement to license any IPR.

5.1.3.14. InhibitAnyPolicy

Many IKE implementations do not support the InhibitAnyPolicy extension. Since the PKIX certificate profile mandates that this extension be marked critical when present, Certification Authority implementations that are interested in maximal interoperability for IKE SHOULD NOT generate certificates that contain this extension.

5.1.3.15. FreshestCRL

IKE implementations MUST NOT assume that the FreshestCRL extension will exist in peer certificates. Note that most IKE implementations do not support delta CRLs.

5.1.3.16. AuthorityInfoAccess

The PKIX certificate profile defines the AuthorityInfoAccess extension, which is used to indicate "how to access CA information and services for the issuer of the certificate in which the extension appears". Because this document deprecates the sending of CRLs in- band, the use of AuthorityInfoAccess (AIA) becomes very important if OCSP [16] is to be used for revocation checking (as opposed to CRLs). The IPsec peer either needs to have a URI for the OCSP query written into its local configuration, or it needs to learn it from AIA. Therefore, implementations SHOULD support this extension, especially if OCSP will be used.

5.1.3.17. SubjectInfoAccess

The PKIX certificate profile defines the SubjectInfoAccess certificate extension, which is used to indicate "how to access information and services for the subject of the certificate in which the extension appears". This extension has no known use in the context of IPsec. Conformant IKE implementations SHOULD ignore this extension when present.

5.2. X.509 Certificate Revocation Lists

When validating certificates, IKE implementations MUST make use of certificate revocation information, and SHOULD support such revocation information in the form of CRLs, unless non-CRL revocation information is known to be the only method for transmitting this information. Deployments that intend to use CRLs for revocation SHOULD populate the CRLDistributionPoints extension. Therefore, Certification Authority implementations MUST support issuing certificates with this field populated. IKE implementations MAY provide a configuration option to disable use of certain types of revocation information, but that option MUST be off by default. Such an option is often valuable in lab testing environments.

5.2.1. Multiple Sources of Certificate Revocation Information

IKE implementations that support multiple sources of obtaining certificate revocation information MUST act conservatively when the information provided by these sources is inconsistent: when a certificate is reported as revoked by one trusted source, the certificate MUST be considered revoked.

5.2.2. X.509 Certificate Revocation List Extensions

5.2.2.1. AuthorityKeyIdentifier

Certification Authority implementations SHOULD NOT assume that IKE implementations support the AuthorityKeyIdentifier extension, and thus should not generate certificate hierarchies which are overly complex to process in the absence of this extension, such as those that require possibly verifying a signature against a large number of similarly named CA certificates in order to find the CA certificate which contains the key that was used to generate the signature.

5.2.2.2. IssuerAltName

Certification Authority implementations SHOULD NOT assume that IKE implementations support the IssuerAltName extension, and especially should not assume that information contained in this extension will be displayed to end users.

5.2.2.3. CRLNumber

As stated in the PKIX certificate profile, all issuers MUST include this extension in all CRLs.

5.2.2.4. DeltaCRLIndicator

5.2.2.4.1. If Delta CRLs Are Unsupported

IKE implementations that do not support delta CRLs MUST reject CRLs that contain the DeltaCRLIndicator (which MUST be marked critical according to the PKIX certificate profile) and MUST make use of a base CRL if it is available. Such implementations MUST ensure that a delta CRL does not "overwrite" a base CRL, for instance, in the keying material database.

5.2.2.4.2. Delta CRL Recommendations

Since some IKE implementations that do not support delta CRLs may behave incorrectly or insecurely when presented with delta CRLs, administrators and deployers should consider whether issuing delta CRLs increases security before issuing such CRLs. And, if all the elements in the VPN and PKI systems do not adequately support Delta CRLs, then their use should be questioned.

The editors are aware of several implementations that behave in an incorrect or insecure manner when presented with delta CRLs. See Appendix A for a description of the issue. Therefore, this specification RECOMMENDS NOT issuing delta CRLs at this time. On the other hand, failure to issue delta CRLs may expose a larger window of vulnerability if a full CRL is not issued as often as delta CRLs would be. See the Security Considerations section of the PKIX [5] certificate profile for additional discussion. Implementers as well as administrators are encouraged to consider these issues.

5.2.2.5. IssuingDistributionPoint

A CA that is using CRLDistributionPoints may do so to provide many "small" CRLs, each only valid for a particular set of certificates issued by that CA. To associate a CRL with a certificate, the CA places the CRLDistributionPoints extension in the certificate, and places the IssuingDistributionPoint in the CRL. The distributionPointName field in the CRLDistributionPoints extension MUST be identical to the distributionPoint field in the IssuingDistributionPoint extension. At least one CA is known to default to this type of CRL use. See Section 5.1.3.13 for more information.

5.2.2.6. FreshestCRL

Given the recommendations against Certification Authority implementations generating delta CRLs, this specification RECOMMENDS that implementations do not populate CRLs with the FreshestCRL extension, which is used to obtain delta CRLs.

5.3. Strength of Signature Hashing Algorithms

At the time that this document is being written, popular certification authorities and CA software issue certificates using the RSA-with-SHA1 and RSA-with-MD5 signature algorithms. Implementations MUST be able to validate certificates with either of those algorithms.

As described in [17], both the MD5 and SHA-1 hash algorithms are weaker than originally expected with respect to hash collisions. Certificates that use these hash algorithms as part of their signature algorithms could conceivably be subject to an attack where a CA issues a certificate with a particular identity, and the recipient of that certificate can create a different valid certificate with a different identity. So far, such an attack is only theoretical, even with the weaknesses found in the hash algorithms.

Because of the recent attacks, there has been a heightened interest in having widespread deployment of additional signature algorithms. The algorithm that has been mentioned most often is RSA-with-SHA256, two types of which are described in detail in [18]. It is widely expected that this signature algorithm will be much more resilient to collision-based attacks than the current RSA-with-SHA1 and RSA-with- MD5, although no proof of that has been shown. There is active discussion in the cryptographic community of better hash functions that could be used in signature algorithms.

In order to interoperate, all implementations need to be able to validate signatures for all algorithms that the implementations will encounter. Therefore, implementations SHOULD be able to use signatures that use the sha256WithRSAEncryption signature algorithm (PKCS#1 version 1.5) as soon as possible. At the time that this document is being written, there is at least one CA that supports generating certificates with sha256WithRSAEncryption signature algorithm, and it is expected that there will be significant deployment of this algorithm by the end of 2007.

6. Configuration Data Exchange Conventions

Below, we present a common format for exchanging configuration data. Implementations MUST support these formats, MUST support receiving arbitrary whitespace at the beginning and end of any line, MUST support receiving arbitrary line lengths although they SHOULD generate lines less than 76 characters, and MUST support receiving the following three line-termination disciplines: LF (US-ASCII 10), CR (US-ASCII 13), and CRLF.

6.1. Certificates

Certificates MUST be Base64 [19] encoded and appear between the following delimiters:

            -----BEGIN CERTIFICATE-----
            -----END CERTIFICATE-----

6.2. CRLs and ARLs

CRLs and ARLs MUST be Base64 encoded and appear between the following delimiters:

            -----BEGIN CRL-----
            -----END CRL-----

6.3. Public Keys

IKE implementations MUST support two forms of public keys: certificates and so-called "raw" keys. Certificates should be transferred in the same form as Section 6.1. A raw key is only the SubjectPublicKeyInfo portion of the certificate, and MUST be Base64 encoded and appear between the following delimiters:

            -----BEGIN PUBLIC KEY-----
            -----END PUBLIC KEY-----

6.4. PKCS#10 Certificate Signing Requests

A PKCS#10 [9] Certificate Signing Request MUST be Base64 encoded and appear between the following delimiters:

            -----BEGIN CERTIFICATE REQUEST-----
            -----END CERTIFICATE REQUEST-----

7. Security Considerations

7.1. Certificate Request Payload

The Contents of CERTREQ are not encrypted in IKE. In some environments, this may leak private information. Administrators in some environments may wish to use the empty Certification Authority option to prevent such information from leaking, at the cost of performance.

7.2. IKEv1 Main Mode

Certificates may be included in any message, and therefore implementations may wish to respond with CERTs in a message that offers privacy protection in Main Mode messages 5 and 6.

Implementations may not wish to respond with CERTs in the second message, thereby violating the identity protection feature of Main Mode in IKEv1.

7.3. Disabling Certificate Checks

It is important to note that anywhere this document suggests implementers provide users with the configuration option to simplify, modify, or disable a feature or verification step, there may be security consequences for doing so. Deployment experience has shown that such flexibility may be required in some environments, but making use of such flexibility can be inappropriate in others. Such configuration options MUST default to "enabled" and it is appropriate to provide warnings to users when disabling such features.

8. Acknowledgements

The authors would like to acknowledge the expired document "A PKIX Profile for IKE" (July 2000) for providing valuable materials for this document.

The authors would like to especially thank Eric Rescorla, one of its original authors, in addition to Greg Carter, Steve Hanna, Russ Housley, Charlie Kaufman, Tero Kivinen, Pekka Savola, Paul Hoffman, and Gregory Lebovitz for their valuable comments, some of which have been incorporated verbatim into this document. Paul Knight performed the arduous task of converting the text to XML format.

9. References

9.1. Normative References

   [1]   Harkins, D. and D. Carrel, "The Internet Key Exchange (IKE)",
         RFC 2409, November 1998.
   
   [2]   Maughan, D., Schneider, M., and M. Schertler, "Internet
         Security Association and Key Management Protocol (ISAKMP)", RFC
         2408, November 1998.
   
   [3]   Kaufman, C., "Internet Key Exchange (IKEv2) Protocol", RFC
         4306, December 2005.
   
   [4]   Kent, S. and R. Atkinson, "Security Architecture for the
         Internet Protocol", RFC 2401, November 1998.
   
   [5]   Housley, R., Polk, W., Ford, W., and D. Solo, "Internet X.509
         Public Key Infrastructure Certificate and Certificate
         Revocation List (CRL) Profile", RFC 3280, April 2002.
   
   [6]   Piper, D., "The Internet IP Security Domain of Interpretation
         for ISAKMP", RFC 2407, November 1998.
   
   [7]   Bradner, S., "Key words for use in RFCs to Indicate Requirement
         Levels", BCP 14, RFC 2119, March 1997.
   
   [8]   Postel, J., "Internet Protocol", STD 5, RFC 791, September
         1981.
   
   [9]   Nystrom, M. and B. Kaliski, "PKCS #10: Certification Request
         Syntax Specification Version 1.7", RFC 2986, November 2000.

9.2. Informative References

   [10]  Deering, S. and R. Hinden, "Internet Protocol, Version 6 (IPv6)
         Specification", RFC 2460, December 1998.
   
   [11]  Arends, R., Austein, R., Larson, M., Massey, D., and S. Rose,
         "DNS Security Introduction and Requirements", RFC 4033, March
         2005.
   
   [12]  Faltstrom, P., Hoffman, P., and A. Costello,
         "Internationalizing Domain Names in Applications (IDNA)", RFC
         3490, March 2003.
   
   [13]  Lynn, C., Kent, S., and K. Seo, "X.509 Extensions for IP
         Addresses and AS Identifiers", RFC 3779, June 2004.
   
   [14]  Kent, S. and K. Seo, "Security Architecture for the Internet
         Protocol", RFC 4301, December 2005.
   
   [15]  Fuller, V. and T. Li, "Classless Inter-domain Routing (CIDR):
         The Internet Address Assignment and Aggregation Plan", BCP 122,
         RFC 4632, August 2006.
   
   [16]  Myers, M., Ankney, R., Malpani, A., Galperin, S., and C. Adams,
         "X.509 Internet Public Key Infrastructure Online Certificate
         Status Protocol - OCSP", RFC 2560, June 1999.
   
   [17]  Hoffman, P. and B. Schneier, "Attacks on Cryptographic Hashes
         in Internet Protocols", RFC 4270, November 2005.
   
   [18]  Schaad, J., Kaliski, B., and R. Housley, "Additional Algorithms
         and Identifiers for RSA Cryptography for use in the Internet
         X.509 Public Key Infrastructure Certificate and Certificate
         Revocation List (CRL) Profile", RFC 4055, June 2005.
   
   [19]  Josefsson, S., "The Base16, Base32, and Base64 Data Encodings",
         RFC 4648, October 2006.

Appendix A. The Possible Dangers of Delta CRLs

The problem is that the CRL processing algorithm is sometimes written incorrectly with the assumption that all CRLs are base CRLs and it is assumed that CRLs will pass content validity tests. Specifically, such implementations fail to check the certificate against all possible CRLs: if the first CRL that is obtained from the keying material database fails to decode, no further revocation checks are performed for the relevant certificate. This problem is compounded by the fact that implementations that do not understand delta CRLs may fail to decode such CRLs due to the critical DeltaCRLIndicator extension. The algorithm that is implemented in this case is approximately:

fetch newest CRL

check validity of CRL signature

if CRL signature is valid, then

if CRL does not contain unrecognized critical extensions and certificate is on CRL, then set certificate status to revoked

The authors note that a number of PKI toolkits do not even provide a method for obtaining anything but the newest CRL, which in the presence of delta CRLs may in fact be a delta CRL, not a base CRL.

Note that the above algorithm is dangerous in many ways. See the PKIX [5] certificate profile for the correct algorithm.

Appendix B. More on Empty CERTREQs

Sending empty certificate requests is commonly used in implementations, and in the IPsec interop meetings, vendors have generally agreed that it means that send all/any end-entity certificates you have (if multiple end-entity certificates are sent, they must have same public key, as otherwise, the other end does not know which key was used). For 99% of cases, the client has exactly one certificate and public key, so it really doesn't matter, but the server might have multiple; thus, it simply needs to say to the client, use any certificate you have. If we are talking about corporate VPNs, etc., even if the client has multiple certificates or keys, all of them would be usable when authenticating to the server, so the client can simply pick one.

If there is some real difference on which certificate to use (like ones giving different permissions), then the client must be configured anyway, or it might even ask the user which one to use

(the user is the only one who knows whether he needs admin privileges, thus needs to use admin cert, or if the normal email privileges are ok, thus uses email only cert).

In 99% of the cases, the client has exactly one certificate, so it will send it. In 90% of the rest of the cases, any of the certificates is ok, as they are simply different certificates from the same CA, or from different CAs for the same corporate VPN, thus any of them is ok.

Sending empty certificate requests has been agreed there to mean "give me your cert, any cert".

Justification:

Responder first does all it can to send a CERTREQ with a CA, check for IP match in SPD, have a default set of CAs to use in ambiguous cases, etc.

Sending empty CERTREQs is fairly common in implementations today, and is generally accepted to mean "send me a certificate, any certificate that works for you".

Saves responder sending potentially hundreds of certs, the fragmentation problems that follow, etc.

In +90% of use cases, Initiators have exactly one certificate.

In +90% of the remaining use cases, the multiple certificates it has are issued by the same CA.

In the remaining use case(s) -- if not all the others above -- the Initiator will be configured explicitly with which certificate to send, so responding to an empty CERTREQ is easy.

The following example shows why initiators need to have sufficient policy definition to know which certificate to use for a given connection it initiates.

EXAMPLE: Your client (initiator) is configured with VPN policies for gateways A and B (representing perhaps corporate partners).

The policies for the two gateways look something like:

Acme Company policy (gateway A)

            Engineering can access 10.1.1.0
                   Trusted CA: CA-A, Trusted Users: OU=Engineering
            Partners can access 20.1.1.0
                   Trusted CA: CA-B, Trusted Users: OU=AcmePartners

Bizco Company policy (gateway B)

           Sales can access 30.1.1.0
                   Trusted CA: CA-C, Trusted Users: OU=Sales
           Partners can access 40.1.1.0
                   Trusted CA: CA-B, Trusted Users: OU=BizcoPartners

You are an employee of Acme and you are issued the following certificates:

   o  From CA-A: CN=JoeUser,OU=Engineering
   o  From CA-B: CN=JoePartner,OU=BizcoPartners
   
   The client MUST be configured locally to know which CA to use when
   connecting to either gateway.  If your client is not configured to
   know the local credential to use for the remote gateway, this
   scenario will not work either.  If you attempt to connect to Bizco,
   everything will work... as you are presented with responding with a
   certificate signed by CA-B or CA-C... as you only have a certificate
   from CA-B you are OK.  If you attempt to connect to Acme, you have an
   issue because you are presented with an ambiguous policy selection.
   As the initiator, you will be presented with certificate requests
   from both CA-A and CA-B.  You have certificates issued by both CAs,
   but only one of the certificates will be usable.  How does the client
   know which certificate it should present?  It must have sufficiently
   clear local policy specifying which one credential to present for the
   connection it initiates.

Author's Address

   Brian Korver
   Network Resonance, Inc.
   2483 E. Bayshore Rd.
   Palo Alto, CA  94303
   US
   
   Phone: +1 650 812 7705
   EMail: briank@networkresonance.com

Full Copyright Statement

This document is subject to the rights, licenses and restrictions contained in BCP 78, and except as set forth therein, the authors retain all their rights.

This document and the information contained herein are provided on an "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY, THE IETF TRUST AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.

Intellectual Property

The IETF takes no position regarding the validity or scope of any Intellectual Property Rights or other rights that might be claimed to pertain to the implementation or use of the technology described in this document or the extent to which any license under such rights might or might not be available; nor does it represent that it has made any independent effort to identify any such rights. Information on the procedures with respect to rights in RFC documents can be found in BCP 78 and BCP 79.

Copies of IPR disclosures made to the IETF Secretariat and any assurances of licenses to be made available, or the result of an attempt made to obtain a general license or permission for the use of such proprietary rights by implementers or users of this specification can be obtained from the IETF on-line IPR repository at http://www.ietf.org/ipr.

The IETF invites any interested party to bring to its attention any copyrights, patents or patent applications, or other proprietary rights that may cover technology that may be required to implement this standard. Please address the information to the IETF at ietf-ipr@ietf.org.

Acknowledgement

Funding for the RFC Editor function is currently provided by the Internet Society.