RE: comments on draft-ietf-smime-cms-auth-enveloped-02

["Jim Schaad" <ietf@xxxxxxxxxxxxxxxxx>]

Russ,

> 
> 
> Jim:
> 
> >Here is the first round of comments on this draft.
> 
> Thanks for your review.
> 
> >1.  I would like to see all information about padding out to a block
> size
> >moved from this document to the algorithm documents.  This allows for
> each
> >algorithm to specify if padding is needed and what type of padding is
> >required.
> >
> >As an example, the CCM algorithm already as a padding technique built
> in and
> >therefore the padding is not needed, or is it?  I cannot tell from the
> >current documents if the data is supposed to be padded before it is
> passed
> >in or not.  Padding is properly a part of the algorithm processing and
> not
> >the superstructure processing.
> 
> The point is that some ciphers require the input to be a multiple of
> a block size.  AES-CBC is one such cipher.  We require that all such
> ciphers use the same padding mechanism in EnvelopedData.  I thin that
> we want similar consistency in AuthEnvelopedDtata.
> 

I understand completely that some ciphers require padding.  However I
disagree that we want all ciphers to use the same padding mechanism.  I
strongly feel that we should remove the padding from EnvelopedData and place
it in the algorithm definitions.  I lost that argument solely on the grounds
of backwards compatibility.  My code has all of the padding done in the low
level encryption code, not in the CMS EnvelopedData code.  I think that this
means that the specification of the padding needs to be specified by the
algorithm not CMS.  

In any event you still need to specify if the AD data needs to be padded for
CCM since the algorithm definition has a padding algorithm and thus does not
need to be padded?

> >2.  If I implement a "stream" version of the decoder for authenticated
> data,
> >am I permitted to pass the decrypted data along prior to doing the
> >authentication step or not?   I raise the question because the CCM
> document
> >states
> >
> >    The recipient MUST verify the CBC-MAC before releasing any
> >    information such as the plaintext.  If the CBC-MAC verification
> >    fails, the receiver MUST destroy all information, except for the
> fact
> >    that the CBC-MAC verification failed.
> >
> >Which implies to me that until the validation is finished the
> decryption
> >code should not pass on any data but buffer it.
> 
> You read it correctly.  Are you suggesting a change?  I think that
> the current document is consistent with the AES-CCM and AES-GCM
> cipher descriptions.

I think that this needs to be explicitly stated in this document, otherwise
I would not implement this buffering required for this.

> 
> >3.  Just for clarity I think you need to state that you are not doing
> the
> >"change the '1' to Set" mess before passing the authenticated
> attributes
> >into the authentication algorithm.
> 
> I assume that this comment is about the tag.  If so, please propose
> text for a "Note that ..." paragraph.

I guess the first question is are you going to do this in the same way that
SignedAttributes are done for compatibility or not?

> 
> >4.  Need to add text to state that each authenticated attribute must
> be
> >internally DER encoded as well as the set of attributes.
> 
> Since attributes doe not use the "octet hole" syntax, I am not sure
> why this is needed.  Please explain.

Same reason that it is needed in the SignedData case.  If you place an
attribute in that I do not have a coder/decoder for, it needs to already be
DER encoded so that I can validate it.

> 
> >5.  There is actually a downside to including the message-digest
> attribute.
> >It should be stated that this attribute MUST NOT be included.
> 
> These may be a performance downside, but I do not see a security
> one.  Do you?  If not, then I do not want to include a MUST NOT
> statement.

I think that if you place the message-digest of the un-encrypted data in
then I can track the un-encrypted data being sent.  I can also do candidate
looking based on the hash value.

> 
> >6.  I am having a problem understanding the text ins section 3 para 4.
> When
> >using the symmetric key-encryption keys (and passwords) the actual AEK
> is
> >wrapped by the original key and thus would be a "one-time-use" key
> with just
> >as much certainty as for the key-transport method.  A better
> description of
> >the problem is needed here in order to fully justify what you are
> saying.
> 
> Yep, the current text is muddled.  I propose:
> 
> Fortunately, the CMS authenticated-enveloped-data content type
> provides all of the tools needed to avoid misuse of counter
> mode.  All of the existing key management techniques permit a fresh
> content-encryption key to be generated for each content.  In
> addition, existing authenticated encryption algorithms that make use
> of counter mode support the use of an unpredictable nonce value in
> the counter block.  This unpredictable nonce value (sometimes called
> a "salt") should be carried in an algorithm identifier parameter.

Yes looks better.


Jim