Jim:
By this argument are you saying that RSA Labs has incorrectly defined
RSA-OAEP? Remember this method of transporting keys does not make any
statements about their indended usage. Additionally, if this is to be a
check that needs to be made by programs it needs to be placed in the
documents as as explicit step. I.e. part of the key unwrap algorithm
should be that the CEK algorithm and the KEK algorithm identifiers are
consistant with each other. I don't believe that the Microsoft
implementations do this check, I can't say if others do or do not, but I
would not be supprised if they do not.
No. Neither PKCS#1 v1.5 nor OAEP provides key type information inside the
wrapping. My understanding is that this was done to avoid structure that
would be helpful to in a cryptanalytic attack. This does not bean that the
unwrapped key should not be checked for consistency with the expected
outcome. I was suggesting that key size and parity (for Triple-DES) are
reasonable checks.
Perhaps I was confusing in my comments about the HMAC key wrap
algorithm. Right now, no other key wrap algorithm has the same structure,
so if the unwrap is successful, the recipient cane be sure that the
originator intended that the wrapped key be used with HMAC.
Also please address the question I raised in #3 - should the CEK (hmac)
or KEK bulk algorithm (3DES, AES...) be the one used in the key
derivation process.
I do not know what the working group consensus will be yet. Once the
consensus in clear, then we can figure out what documents need to address it.
This whole issue surfaced when trying to address AuthenticatedData with
Static-Static Diffie-Hellman (S-S D-H) key agreement. S-S D-H is used
since it provides data origin authentication when the originator's public
keys is certified. Let me summarize the steps (assuming one recipient for
simplicity). These steps come from rfc2630bis, section 9.
1. Originator generates a random HMAC key.
2. Originator generates a random value. It will be transferred to the
recipient in the ukm field. Originator uses the random value, her private
D-H key and the Recipient's D-H public key to compute a pairwise
key-encryption key (KEK). The random value ensures that unique KEK is
generated for this operation. In this case, the KEK is a Triple-DES
key. Next, the originator wraps the HMAC key generated in step 1 with the KEK.
3. The originator uses the recipient's certificate identification
information, the wrapped HMAC key, the ukm, the appropriate algorithm
identifiers, and so on are collected into a RecipientInfo value.
4. Using the HMAC key, the originator computes a MAC value on the content
or the authenticated attributes. If the HMAC is computed on authenticated
attributes, then one of them must be a message digest of the content.
Now, given this context, all of the issues raised by this thread are
associated with step 2 (and perhaps the algorithm identifiers that are
selected for step 3.
Jim, you seem to be looking for parallel situations in the RSA and D-H
cases. However, they do not exist in the AuthenticatedData
context. Neither RSA nor Ephemeral-Static D-H (E-S D-H) provides data
origin authentication. Anyone can wrap an HMAC key with the recipient'
public key. The whole point of AuthenticatedData is to know the origin of
the content, and you cannot know the origin of the content unless you know
the origin of the HMAC key.
I hope this helps clarify the situation.
> Sent: Wednesday, March 06, 2002 2:09 PM
> To: jimsch(_at_)exmsft(_dot_)com
> Cc: pgut001(_at_)cs(_dot_)aucKland(_dot_)ac(_dot_)nz;
ietf-smime(_at_)imc(_dot_)org
> Subject: RE: I-D ACTION:draft-ietf-smime-hmac-key-wrap-00.txt
>
>
>
> Jim:
>
> Once the key is unwrapped, we want the recipient to know that it is
> appropriate for use with the intended algorithm. Some checks are
> obvious. In Triple-DES, the key size must be either 128 bits or 192
> bits. In AES, the the key size must be either 128 bits, 192 bits, or 256
> bits. Such checks are not so obvious with algorithms that allow a
variable
> key size, like RC2 and HMAC.
>
> The key wrap algorithm provides integrity of the keying material, so the
> recipient knows that the received key has not been modified. The way that
> the HMAC key wrap algorithms are defined, the recipient should know that
> the wrapped key is intended for use with HMAC. I suppose that this
feature
> could be lost if the structure that we have defined is copied for many
> other purposes.
>
> Russ
>
>
> >After long discussions with Russ and Peter over the last week, I have
> >come up with the following list of items that need clarification.
> >
> >1. How OIDs are assigned for key wrap algorithms. Under the current
> >naming scheme the following items are identified by the assigned OID:
> >1) The exact key wrap algorithm to be used (including padding done to
> >the data), 2) The block encryption cipher applied during the key wrap
> >algorithm and 3) The purpose for which the embedded key is to be used.
> >
> >2. Should a common key algorithm (including padding) be used for all
> >128-bit block cipher algorithms. (i.e. should the AES and the HMAC key
> >wrap algorithm be identical including padding).
> >
> >3. Should we update any current or existing documents based on any new
> >understandings that we arrive at.
> >
> >COMMENTARY:
> >
> >1. I find that the third item in assign a key OID to be at present
> >insecure and and not really enforcable. It is a feature which is not
> >currently present in PKCS#1 v1.5 and is far better covered for the DH
> >case by the fact that the CEK (i.e. destination) algorithm is used for
> >the KEK derivation process. I was not really aware until I spoke to
> >Russ that the third item was even present in the OID definition and have
> >therefore never even thought to enforce it in my code in any way.
> >
> >2. Russ, please address the following for me. Is it really the CEK not
> >the KEK block encryption algorithm that should be protected during the
> >DH key derivation process? If they are the same as is presently true it
> >is not an issue, but is of interest if you use an HMAC algorithm rather
> >than a CEK algorithm.
> >
> >3. I believe that there is an element of simplicity gained from having
> >a common key wrap algorithm for the AES key sizes. I do not currently
> >believe that there is a benefit to breaking backwards compatablility to
> >achive the same effect with the 64-bit block size algorithms.
> >
> >Conclusions:
> >
> >I think that changing the OID assignment method should be looked at for
> >the 128-bit block algorithms, but we should not change our current
> >course for 64-bit block algorithms. This does mean that we need to
> >change both AES-keywrap and HMAC-keywrap, but need input from the list,
> >from Russ and Peter and from NIST (the current AES key wrap algorithm
> >definers).
> >
> >jim
> >
[SNIP]