I had a voice conversation with Paul Hoffman, and I think that I now understand the things that he would like to see added to the document. Essentially, the Hash Of Root Key certificate extension is a commitment to the next generation public key and algorithm. Recall that the hash covers the SubjectPublicKeyInfo, which is:
SubjectPublicKeyInfo ::= SEQUENCE {
algorithm AlgorithmIdentifier,
subjectPublicKey BIT STRING }
So, a few things can go wrong after making this commitment. (Stephen called it pinning.) The Root CA needs to choose a next generation public key and algorithm that will be secure for the full lifetime. Of course, a cryptoanalytic break through is very difficult to predict, and if one happens before the new key is used, the Root CA remains committed to the now broken key. The remedy is to deploy a new public key and algorithm in the same manner as the initial Root CA self-signed certificate. The benefits of automatic detection of the new public key are lost, but that is a minor concern is the scope of a cryptoanalytic break through.
In addition, from an operational perspective, the Root CA must securely back up the yet-to-be-deployed key pair. If the Root CA stores it in a hardware security module, and that module fails, the Root CA remains committed to the now unavailable key. The remedy is the same as above: deploy a new public key in the same manner as the initial Root CA self-signed certificate.
Russ
> On Jan 2, 2019, at 3:17 PM, Paul Hoffman <paul.hoffman@xxxxxxxx> wrote:
>
> On 2 Jan 2019, at 11:57, Russ Housley wrote:
>
>>> On Jan 2, 2019, at 1:26 PM, Paul Hoffman <paul.hoffman@xxxxxxxx> wrote:
>>>
>>> This extension seems useful to CAs that understand the increased risks that using it incurs, but those risks are not mentioned in the document.
>>>
>>> The document implicitly assumes that the CA will in fact use the key named in the extension next. Using this extension increases the risk of a bad trust anchor rollover at the same time that it makes good rollover more secure. If the key listed in this extension cannot be used when the CA eventually changes the trust anchor, relying parties will mistrust the new trust anchor. There are many reasons why a CA might think they know the next key but cannot use that key when they change trust anchors, such as if the HSM that holds the next key fails or is destroyed. Given the last sentence in Section 2, this could mean that a CA might never be able to issue a new trust anchor, even if the key for its current trust anchor is compromised.
>>>
>>> Given the severity of the new risks of using this extension, they need to be introduced at the beginning of the document and then discussed in more detail in the Security Considerations. Note that this risk affects the last paragraph of the Security Considerations section as well.
>>
>> The point is to facilitate the transition from one Root CA certificate to the next.
>
> To be clear, the transition is from one public key to the next, not from one certificate to the next. But, more importantly, the point of this extension is to facilitate the transition from one Root CA certificate to what is supposed to be the next key. However, if that next key is not seen by every relying party during the transition, the extension prevents the CA from ever making the transition for the relying parties that do not see the new key in a revised trust anchor. This seems like a huge restriction that is only mentioned in the document in exactly one sentence:
>
>> The last sentence in Section 2 says:
>>
>> If either check fails, then potential Root CA
>> certificate is not a valid replacement, and the recipient continues
>> to use the current Root CA certificate.
>>
>> Indeed, these check are necessary to make sure that the relying party makes the transition to the proper replacement. Any failure of the checks leave the trust anchor unchanged, which seems like the right result to me.
>
> It seems right to me as well, but it still seems to be wholly insufficient to not talk about the risks of using the extension early in the document.
>
>>
>> Recall the definition of trust anchor from RFC 5280, Section 6.1.1:
>>
>> (d) trust anchor information, describing a CA that serves as a
>> trust anchor for the certification path. The trust anchor
>> information includes:
>>
>> (1) the trusted issuer name,
>>
>> (2) the trusted public key algorithm,
>>
>> (3) the trusted public key, and
>>
>> (4) optionally, the trusted public key parameters associated
>> with the public key.
>>
>> The checks make sure that the replacement self-signed certificate contains the intended information.
>
> That is all fine, but it does not address the significant risk a CA is undertaking by promising what the next key will be.
>
>>> Editorial:
>>>
>>> The abstract says:
>>> This document specifies the Hash Of Root Key certificate extension.
>>> This certificate extension is carried in the self-signed certificate
>>> for a trust anchor, which is often called a Root Certification
>>> Authority (CA) certificate. This certificate extension unambiguously
>>> identifies the next public key that will be used by the trust anchor
>>> at some point in the future.
>>> The term "trust anchor" is used as the concept, not the bits in the certificate. As such, the last sentence is confusing because the trust anchor will change when the key changes. A possible fix is to replace "will be used by the trust anchor at some point in the future" with "will be used in a trust anchor at some point in the future".
>>
>> I think I understand your point. Does this text resolve the concern?
>>
>> This document specifies the Hash Of Root Key certificate extension.
>> This certificate extension is carried in the self-signed certificate
>> for a trust anchor, which is often called a Root Certification
>> Authority (CA) certificate. This certificate extension unambiguously
>> identifies the next public key that will be used at some point in the
>> future as the next Root CA certificate, replacing the current one.
>
> Not really. A key will not be used as a certificate: it is just a key. A key might be used as the signing key for a certificate, but not as the certificate itself. Maybe instead: "will be used to sign a trust anchor at some point in the future"?
>
>>> The first list in Section 2 would be clearer if the order was R1, R2, H2, C1; this would also then match the order in the second list.
>>
>> Okay. I changed that in my edit buffer.
>>
>>> Later in Section 2, a sentence appears to be missing its subject. "That is, verify the signature on the self-signed certificate..." should probably be "That is, the recipient verifies the signature on the self-signed certificate...".
>>
>> Okay. I changed that in my edit buffer.
>>
>> In addition, I added a bit more detail about the relationship to certification path validation, which I hope adds clarity around your first comment. It now reads:
>>
>> The successor to the Root CA self-signed certificate can be
>> delivered by any means. Whenever a new Root CA certificate is
>> received, the recipient is able to verify that the potential Root CA
>> certificate links back to a previously authenticated Root CA
>> certificate with the hashOfRootKey certificate extension. That is,
>> the recipient verifies the signature on the self-signed certificate
>> and verifies that the hash of the DER-encoded SubjectPublicKeyInfo
>> from the potential Root CA certificate matches the value from the
>> HashOfRootKey certificate extension of the current Root CA
>> certificate. Checking the self-signed certificate signature ensures
>> that the certificate contains the subject name, public key algorithm
>> identifier, and public key algorithm parameters intended by the key
>> owner intends; these are important inputs to certification path
>> validation as defined in Section 6 of [RFC5280]. Checking the hash
>> of the SubjectPublicKeyInfo ensures that the certificate contains the
>> intended public key. If either check fails, then potential Root CA
>> certificate is not a valid replacement, and the recipient continues
>> to use the current Root CA certificate.
>
> Yes, adding this clarifies how all the trust anchor information is linked through the validation process. This is a good addition.
>
> --Paul Hoffman
