Re: [smime] PKCS#7 v1.5 vs. CMS / ContentInfo vs. EncapsulatedContentInfo based on version

[[e-mail] (Martin Rex)]

Hi Jim,

Thanks for the reply, but this leaves me even more confused now.

Admittedly my personal implementors experience is tiny.  I once patched
support for processing rfc3161 TimeStamps into an existing PKCS#7 v1.5
implementation, and I needed a few tweaks for the ASN.1 encoder & decoder
-- but rfc3161 uses id-ct-TSTInfo content type and version 3 rather than
id-data and version 1, so I needed tweaks for processing of
EncapsulatedContentInfo.


Jim Schaad <[e-mail]> wrote:
>
> To begin with, this is only a problem if you are looking at wrapping
> contents other than id-data, for id-data there is no difference.  In both
> cases there is an OCTET wrapper.

This statement looks like a self-contradiction.  Either there is _no_
difference between PKCS#7 v1.5 SignedData for id-data, then there is
no wrapper.  Or there is a difference, and EncapsulatedContentInfo is used.

> 
> For things which are not id-data, there is going to be a difference between
> the two encodings in that for one an octet wrapper is there and for the
> other case it is not.  I would say that you need to look at the content type
> and the type of the field and then make a decision about what you are doing.

Do you mean that while the PDU encoding for SignedData with id-data
ContentInfo is the same for PKCS#7 v1.5 and CMS, the actual signature
(or more precisely the hash over that id-data) is computed _differently_
for PKCS#7 v1.5 and CMS (covering the 0x04 plus ASN.1 length field for
CMS, and omitting this for PKCS#7 v1.5) ?

I wouldn't like a heuristic on decoding because it results in needlessly
complex code and seems to have an ambituity for certain id-data that
conicidentally matches the beginning of an ASN.1 DER OctetString.

But requiring a heuristic on SignedData would be magnitudes worse,
because of significantly higher CPU cycles impact for computing and
verifying two different hashes.


> I will note that there is a security problem with the PKCS#7 encoding where
> the content and length bytes are not correctly protected.  This is one of
> the reasons that CMS added the OCTET wrapper in all cases rather than just
> in the case of id-data.  

But "in all cases rather than just in the case of id-data" is a contradiction
to the above (with respect to the encoding).  Or is this comment _not_
about the encoding, but rather about the data which gets signed (hashed) ?


>
> There was never any intent that a version number of one would indicate that
> this was PKCS#7 rather than CMS.

That sounds wrong to me.  At least my copy of PKCS#7 v1.5
(rfc2315) is _explicit_ that this version indicates PDU/protocol, and
version==1 therefore implies PKCS#7 (rfc2315) syntax/encoding
**AND** processing rules (semantics).

https://tools.ietf.org/html/rfc2315#section-9.1

   The fields of type SignedData have the following meanings:

        o    version is the syntax version number. It shall be
             1 for this version of the document.


The PKCS#7 v1.5 PDU was *ALWAYS* supposed to be self-describing,
and later revisions of it to identify different syntax as well as
different processing rules by using a different version in the PDU.


For the particular (governmentally mandated) data exchange scenario in
Germany, they're currently using PKCS#7 v1.5 with RSA PKCS#1 v1.5,
and they want to transition to using RSA-PSS (signatures on certs
and PKCS#7/CMS SignedData) and RSA-OAEP (EvelopedData), with
EndEntity certs that carry rsaEncryption keys (so that the keys
can be used for both, signature and encryption).


-Martin