Re: [pkix] [Technical Errata Reported] RFC3029 (6444)

[Stefan Santesson <[e-mail]>]

Russ

 

I concede to your superior ASN.1 knowledge here.

 

Can this be fixed without changing the protocol in a way that changes the bits on the wire?

I assume that a change that alters the bits on the wire requires a regular protocol update with a new ASN.1 module.

 

Stefan Santesson 

 

From: Russ Housley <[e-mail]>
Date: Monday, 1 March 2021 at 23:33
To: Stefan Santesson <[e-mail]>
Cc: Carlisle Adams <[e-mail]>, Erwann Abalea <[e-mail]>, "Roman D. Danyliw" <[e-mail]>, Ben Kaduk <[e-mail]>, IETF PKIX <[e-mail]>
Subject: Re: [pkix] [Technical Errata Reported] RFC3029 (6444)

 

Stefan:

 

The addition of this definition solves the "Integer" concern:

 

                Integer ::= INTEGER

 

CRLDistributionPoints compiles without errors or warnings for me.

 

This situation is not two nested CHOICES, so it is not quite the same.  Here, the CHOICE followed by a SEQUENCE followed by a CHOICE is not a problem.

 

For example:

   value DistributionPoint ::= {

     distributionPoint fullName: {

       dNSName "www.example.com",

       uniformResourceIdentifier "http://www.example.com"

     },

     reasons '011111000'B

   }

 

DER-Encodes to:

   3031A02B A029820F 7777772E 6578616D 706C652E 636F6D86 16687474 703A2F2F

   7777772E 6578616D 706C652E 636F6D81 02027C

 

 

DistributionPoint SEQUENCE: tag = [UNIVERSAL 16] constructed; length = 49

  distributionPoint : tag = [0] constructed; length = 43

    DistributionPointName CHOICE

      fullName GeneralNames SEQUENCE OF: tag = [0] constructed; length = 41

        GeneralName CHOICE

          dNSName IA5String: tag = [2] primitive; length = 15

            "www.example.com"

        GeneralName CHOICE

          uniformResourceIdentifier IA5String: tag = [6] primitive; length = 22

            "http://www.example.com"

  reasons ReasonFlags BIT STRING: tag = [1] primitive; length = 2

    0x027c

 

Russ

 



On Mar 1, 2021, at 4:53 PM, Stefan Santesson <[e-mail]> wrote:

 

I think there is a need for clarification and to correct the “Integer -> INTEGER” issue.

 

But I don’t think it is appropriate to change any ASN.1 to attempt to fix neither the implicit tagging or the ambiguity of the “Data” type.

 

Here is my motivation.

On the nested implicit tagged object we have a similar issue with Distribution Point in RFC 5280:

 

   CRLDistributionPoints ::= SEQUENCE SIZE (1..MAX) OF DistributionPoint

 

   DistributionPoint ::= SEQUENCE {

        distributionPoint       [0]     DistributionPointName OPTIONAL,

        reasons                 [1]     ReasonFlags OPTIONAL,

        cRLIssuer               [2]     GeneralNames OPTIONAL }

 

   DistributionPointName ::= CHOICE {

        fullName                [0]     GeneralNames,

        nameRelativeToCRLIssuer [1]     RelativeDistinguishedName }

 

Here we have the Implicitly tagged DistributionPointName containing the implicitly tagged choice of GeneralNames or RelativeDistinguishedName.

I’m not an ASN.1 expert, but this is solved by implementations by converting the outer tag to an explicit tag.

E.g. when GeneralNames is used it is represented by [0][0]{GeneralNames length and value}

 

I assume that the same principle must apply to CertEtcToken.

 

On the “Data” structure it turns out that there are 4 different service types:

 

   - Certification of Possession of Data (cpd),

   - Certification of Claim of Possession of Data (ccpd),

   - Validation of Digitally Signed Document (vsd), and

   - Validation of Public Key Certificates (vpkc).

Where cpd and vsd always use message, ccpd always use messageImprint and vpkc always use sequence of TargetEtcChain

 

So, even if this is not great, it is still possible to implement the protocol, given that you know what service that use the “Data” type.

This is in fact how the implementation pointed out by Russ implements this: https://github.com/a-i-b/jnotary/blob/master/dvcs/src/main/java/org/jnotary/dvcs/Data.java

Se the ”getInstance()” function at line 90:

 

My conclusion is that the protocol has design flaws, but it is not the task of an errata to fix design flaws.

Fixing design flaws is the task of a protocol update. But I don’t think there is a good motivation to put the efforts into releasing a new version of this protocol.

As it is, the protocol is at least implementable, demonstrated by the linked code.

 

 

Stefan Santesson 

 

From: Carlisle Adams <[e-mail]>
Date: Monday, 1 March 2021 at 17:23
To: Russ Housley <[e-mail]>, Erwann Abalea <[e-mail]>
Cc: "Roman D. Danyliw" <[e-mail]>, Ben Kaduk <[e-mail]>, Stefan Santesson <[e-mail]>, IETF PKIX <[e-mail]>
Subject: RE: [pkix] [Technical Errata Reported] RFC3029 (6444)

 

Hi Russ, all,

 

I agree that these errors need fixing because there doesn’t seem to be an easy way to get around them otherwise.  I also agree with Stefan that this is probably an indication that DVCS was never widely implemented…  L

 

Carlisle.

 

 

From: Russ Housley <[e-mail]> 
Sent: February-28-21 5:26 PM
To: Erwann Abalea <[e-mail]>
Cc: Roman D. Danyliw <[e-mail]>; Ben Kaduk <[e-mail]>; Stefan Santesson <[e-mail]>; IETF PKIX <[e-mail]>; Carlisle Adams <[e-mail]>
Subject: Re: [pkix] [Technical Errata Reported] RFC3029 (6444)

 

Attention : courriel externe | external email

Erwann:

 

Le sam. 27 févr. 2021 à 08:45, Russ Housley <[e-mail]> a écrit :

I guess I should have held off on reporting this ASN.1 error.  Once I corrected it, I discovered two errors that I do not know how to fix.

There is an implementation somewhere because Appendix F contains examples.  I do not know how the implementer got around these two problems.


PROBLEM 1:

CertEtcToken ::= CHOICE {
     certificate                  [0] IMPLICIT Certificate ,
     esscertid                    [1] ESSCertId ,
     pkistatus                    [2] IMPLICIT PKIStatusInfo ,
     assertion                    [3] ContentInfo ,
     crl                          [4] IMPLICIT CertificateList,
     ocspcertstatus               [5] IMPLICIT CertStatus,
     oscpcertid                   [6] IMPLICIT CertId ,
     oscpresponse                 [7] IMPLICIT OCSPResponse,
     capabilities                 [8] SMIMECapabilities,
     extension                    Extension{{ExtensionSet}}
}

CertEtcToken is a CHOICE with tags 0 through 8, but CertStatus CHOICE with tags 0 through 2.  You cannot nest a CHOICE in another CHOICE is the IMPLICIT tags overlap.

The use of EXPLICIT tagging would have solved the problem, but the authors clearly preferred IMPLICIT tags.

 

Here, it's easy. The outermost IMPLICIT tag is transformed into an EXPLICIT one by the ASN.1 compiler (it will probably emit a warning, though).

 

Yes, I am aware of that possibility of fixing the specification, but it is unclear what an implementer would do with a ASN.1 module with compilation errors, or a warning that inserts a EXPLICIT.





 

PROBLEM 2

DigestInfo ::= SEQUENCE {
    digestAlgorithm   DigestAlgorithmIdentifier,
    digest            Digest
}

Data ::= CHOICE {
      message           OCTET STRING ,
      messageImprint    DigestInfo,
      certs             SEQUENCE SIZE (1..MAX) OF
                            TargetEtcChain
}

DigestInfo is a SEQUENCE, and certs is a SEQUENCE, so the two have the same tag.  A recipient cannot tell which one the sender intended.

 

For this one, there's clearly no solution. Tagging would have solved it (any kind of tagging mode), but it's missing.

 

Indeed, at lease one of the SEQUENCE needs a tag.

 

Russ