Re: [lamps] EST CSRATTRS specifying the SAN

[David von Oheimb <[e-mail]>]

The ASN.1 declaration of the Attribute type, which is the core of
CsrAttrs type given in RFC 7030 section 4.5.2:

  CsrAttrs ::= SEQUENCE SIZE (0..MAX) OF AttrOrOID

  AttrOrOID ::= CHOICE (oid OBJECT IDENTIFIER, attribute Attribute }

  Attribute { ATTRIBUTE:IOSet } ::= SEQUENCE {
        type   ATTRIBUTE.&id({IOSet}),
        values SET SIZE(1..MAX) OF ATTRIBUTE.&Type({IOSet}{@type}) }

has been copied over from PKCS#10:
https://datatracker.ietf.org/doc/html/rfc2986#section-4.
Since it is pretty incomprehensible for ASN.1 non-experts (including
myself), the reader has to rely in the interpretation following it in
prose and on the example given, but unfortunately both of them are not
very clear. In particular, the meaning of the values field or of a
"value" in the given text remains ambiguous or even misleading. In the
given example,

         Attribute:  type = extensionRequest (1.2.840.113549.1.9.14)
                     value = macAddress (1.3.6.1.1.1.1.22)

the "value" is not a MAC address but the OID of MAC addresses. This
corresponds to what is explained as follows:

   Requests for descriptive information from the client are made by an attribute, to be
   represented as Attributes of the CSR, with a type indicating the
   [RFC2985 <https://datatracker.ietf.org/doc/html/rfc2985>] extensionRequest and the values indicating the particular
   attributes desired to be included in the resulting certificate's extensions.

AFAICS, the key to understanding how to encode a concrete value (such as
a MAC address, a SAN value, etc.) is to appreciate the following
sentence, which provides the key idea for reusing the  Attribute type:

  The structure of the CSR Attributes Response SHOULD, to the greatest
   extent possible, reflect the structure of the CSR it is requesting.

So in essence, the CSR attribute response is to be formed like the
attributes part of a template PCKS#10 CertificationRequestInfo structure
and simply can be given exactly as in a PKCS#10 CSR structure including
any (actual) values, yet with two generalizations:

  * Actual values may be left out, and this way are requested to be
    filled in by the client/EE.
  * Additional OIDs may be given for requesting the use of the
    challengePassword and for specifying algorithms (such as the
    expected type of the public key).

Note that under the (fair) assumption that the order of the AttrOrOID
entries in the CsrAttrs sequence does not play a role, and glossing over
some Attribute subtype peculiarities, the CsrAttrs type is isomorphic to 

  CsrAttrs ::= SEQUENCE {
        oids          SET OF OBEJECT IDENTIFIER
        attributes    SET OF Attribute{{ IOSet }}

    }

  Attribute { ATTRIBUTE:IOSet } ::= SEQUENCE {
        type   ATTRIBUTE.&id({IOSet}),
        values SET SIZE(1..MAX) OF ATTRIBUTE.&Type({IOSet}{@type})
  }

from which the similarity with a classical PKCS#10 structure:

   CertificationRequestInfo ::= SEQUENCE {
        version       INTEGER { v1(0) } (v1,...),
        subject       Name,
        subjectPKInfo SubjectPublicKeyInfo{{ PKInfoAlgorithms }},
        attributes    [0] Attributes{{ CRIAttributes }}
   }

   CRIAttributes  ATTRIBUTE  ::= {
        ... -- add any locally defined attributes here -- }

   Attributes { ATTRIBUTE:IOSet } ::= SET OF Attribute{{ IOSet }}

   Attribute { ATTRIBUTE:IOSet } ::= SEQUENCE {
         type   ATTRIBUTE.&id({IOSet}),
         values SET SIZE(1..MAX) OF ATTRIBUTE.&Type({IOSet}{@type})
   }

would have been more apparent.

Further, RFC 7030 section 4.5.2 does not cover the fact that names and
name components in a CSR may be part of not only various X.509
extensions such as SAN (which are to be encoded via the extra
indirection of the extensionRequest OID) but also of the subject
field/attribute of type (Distinguished)Name.

At least a more comprehensive example would have been very helpful to
clarify the details of the intended encoding, and likewise any serious
reference/example implementation, for instance in Cisco's libEST, which
has remained extremely sketchy regarding the csrattrs topic.

    David



On 21.06.21 19:29, Sean Turner wrote:
> Sorry for just getting to this now. I haven’t read upstream so if this gets answer there oops.
>
>> On Jun 9, 2021, at 16:15, Michael Richardson <[e-mail]> wrote:
>>
>> Signed PGP part
>>
>> Eliot has suggested that:
>>      https://datatracker.ietf.org/doc/html/rfc7030#section-4.5.2
>> does not appear to allow the returned CSR Attributes to contain a value.
>> Please correct me if I'm understanding the problem wrong.
> Oh yeah it does! The whole point of CSRattrs is to allow the client to say “hey RA/CA what do you want my CSR to look like”. The response could just be an OID or it can be an actual attribute with some values:
>
>   Requests for descriptive information from the client are made
>   by an attribute, to be represented as Attributes of the CSR, with
>   a type indicating the [RFC2985] extensionRequest and the
>   values indicating the particular attributes desired to be included
>   in the resulting certificate’s extensions. 
>
> Note the word “values” above.
>
> I also figured that there might be brain dead EEs out there so instead of returning a string of attributes the EST server could just return the entire CSR - see Appendix B of RFC 8295. That way the EE would only need sign the request (for PoP).
>
>> RFC8994 assignment of IPv6 ULA can not, as far as I can understand, work
>> without assigning a value.  The Registrar must allocate a prefix for the new
>> ACP node, and this has to go into the CSR in order for the CA to sign it.
>>
>> (Yes, the Registrar has to *check* that the CSR contains the right value along
>> the way too.  Yes, if the Registrar is ALSO the CA, then it can shortcut, but
>> that's not in general true)
>>
>> My unit testing code is at:
>>  https://github.com/AnimaGUS-minerva/fountain/blob/master/spec/models/csr_attributes_spec.rb
>>
>>
>> I found the ASN.1 code in section 4.5.2 beyond my understanding, so I muddled
>> through, since fortunately, there was an example.
>>
>>   CsrAttrs ::= SEQUENCE SIZE (0..MAX) OF AttrOrOID
>>   AttrOrOID ::= CHOICE (oid OBJECT IDENTIFIER, attribute Attribute }
>>   Attribute { ATTRIBUTE:IOSet } ::= SEQUENCE {
>>        type   ATTRIBUTE.&id({IOSet}),
>>        values SET SIZE(1..MAX) OF ATTRIBUTE.&Type({IOSet}{@type}) }
>>
>> What I think that this says is that one can have:
>>  a) a Sequence of stuff.
> a sequence of stuff that might be empty
>
>>  b) the stuff is either an OID.
>>  c) or the stuff is a SEQUENCE [ type, value ].
>> The complicated &id... is just saying that when you have some attribute type
>> "FOO", then you can have the things that would be valid as values.  I'm
>> actually rather amazed at this level of meta-programming. (Can CDDL do that?)
> So this is the parameterized ASN.1 syntax. What you could so is define an IOSet of attributes. These attributes value values. But, yeah if it’s an attribute it’s an OID and the associated syntax. e.g.,
>
> pKCS7PDU ATTRIBUTE ::= {
>        WITH SYNTAX ContentInfo
>        ID pkcs-9-at-pkcs7PDU
>        }
>
>    pkcs-9-at-pkcs7PDU OBJECT IDENTIFIER ::= {
>        iso(1) member-body(2) us(840) rsadsi(113549) pkcs(1) pkcs9(9)
>        pkcs-9-at(25) 5
>        }
>
> The OID would be 1.2.840.113549.1.9.25.5
>
> and the values (really only one) is:
>
> PKIData ::= SEQUENCE {
>        reqSequence        SEQUENCE SIZE(0..MAX) OF TaggedRequest
>        }
>
>> An example is provided:
>>   MEEGCSqGSIb3DQEJBzASBgcqhkjOPQIBMQcGBSuBBAAiMBYGCSqGSIb3DQEJDjEJ
>>   BgcrBgEBAQEWBggqhkjOPQQDAw==
>>
>> %echo 'MEEGCSqGSIb3DQEJBzASBgcqhkjOPQIBMQcGBSuBBAAiMBYGCSqGSIb3DQEJDjEJBgcrBgEBAQEWBggqhkjOPQQDAw==' | base64 -d | dumpasn1 -
>>  0  65: SEQUENCE {
>>  2   9:   OBJECT IDENTIFIER challengePassword (1 2 840 113549 1 9 7)
>> 13  18:   SEQUENCE {
>> 15   7:     OBJECT IDENTIFIER ecPublicKey (1 2 840 10045 2 1)
>> 24   7:     SET {
>> 26   5:       OBJECT IDENTIFIER secp384r1 (1 3 132 0 34)
>>       :       }
>>       :     }
>> 33  22:   SEQUENCE {
>> 35   9:     OBJECT IDENTIFIER extensionRequest (1 2 840 113549 1 9 14)
>> 46   9:     SET {
>> 48   7:       OBJECT IDENTIFIER '1 3 6 1 1 1 1 22'
>>       :       }
>>       :     }
>> 57   8:   OBJECT IDENTIFIER ecdsaWithSHA384 (1 2 840 10045 4 3 3)
>>       :   }
>>
>>
>> At 2, we have the OID "challengePassword", which tells the client that this
>> attribute should be included (that's [b]).
>> At 13-26, we have an attributed, ecPublicKey, with a *value* secp384r1.
>> This tells the client what kind of things to sign with.  (that's [c])
>> At 33, we have extensionRequest with oid value 1.3.6.1.1.1.1.22. (also [c])
>> Finally, at 57, we have just the attribute ecdsaWithSHA384, which is telling
>> the client to use SHA2-384.
>>
>> In order to get this all right, I used the example above, made sure I could
>> decode it, and then made sure that I could encode exactly that.
> Yep!
>
>> I have an example building the subjectAltName rfc822Name (yes, my code hasn't
>> caught up to otherName as specified in RFC8994 yet) which is:
>>
>> obiwan-[projects/pandora/fountain](2.6.6) mcr 10263 %dumpasn1 tmp/hellobulb0.der
>>  0  32: SEQUENCE {
>>  2  30:   SEQUENCE {
>>  4   3:     OBJECT IDENTIFIER subjectAltName (2 5 29 17)
>>  9  23:     SET {
>> 11  21:       SEQUENCE {
>> 13  19:         [1] {
>> 15  17:           UTF8String '[e-mail]'
>>       :           }
>>       :         }
>>       :       }
>>       :     }
>>       :   }
>>
>> I don't know what the [1] is actually, but possibly that's the rfc822Name.
>> Yup, it is:
>>  #        rfc822Name                      [1]     IA5String,   <-- this one
> The one is the tag.
>
>> So, I think that I'm entirely within spec for CSR Attributes, and I'd like to
>> understand more from Owen and Eliot as to what problems they are having with
>> CSR Attributes.
>>
>> Eliot has suggested we do it all in JSON (CBOR?), and I would certainly be
>> happy with that.
>>
>> --
>> Michael Richardson <[e-mail]>   . o O ( IPv6 IøT consulting )
>>           Sandelman Software Works Inc, Ottawa and Worldwide
>>