Another suggestion is to use the kek to wrap the hmac just like you wrapped the
CEK.
From: owner-ietf-smime(_at_)mail(_dot_)imc(_dot_)org
[mailto:owner-ietf-smime(_at_)mail(_dot_)imc(_dot_)org] On Behalf Of Trevor
Freeman
Sent: Monday, May 05, 2008 1:40 PM
To: Jim Schaad; ietf-smime(_at_)imc(_dot_)org
Subject: RE: weak authentication issue with rfc5083
Hi Jim,
You can demonstrate knowledge of both HMAC and KEK fairly simply per recipient
e.g. by concatenating the KEK with the message HMAC and hashing the result.
Since Mallet does not know the KEK they cannot compute the new hash. This fix
works for both 3852 and 5083.
Trevor
From: Jim Schaad [mailto:ietf(_at_)augustcellars(_dot_)com]
Sent: Monday, May 05, 2008 1:23 PM
To: Trevor Freeman; ietf-smime(_at_)imc(_dot_)org
Subject: RE: weak authentication issue with rfc5083
Trevor,
The same issue also occurs for AuthenticatedData in RFC 3852. This has been
known for a long time.
I am not sure how a demonstration of the KEK would solve this issue however.
The only way that I know to actually deal with this problem is to place the
computed MAC under encryption for each recipient.
Jim
From: owner-ietf-smime(_at_)mail(_dot_)imc(_dot_)org
[mailto:owner-ietf-smime(_at_)mail(_dot_)imc(_dot_)org] On Behalf Of Trevor
Freeman
Sent: Monday, May 05, 2008 12:52 PM
To: ietf-smime(_at_)imc(_dot_)org
Subject: weak authentication issue with rfc5083
We have identified a potential for spoofing when using rfc5083.
Rfc5083 computes a message HMAC over the content using the authenticated
encryption algorithms. The message HMAC is calculated based on the content and
the content encryption key (CEK).
A number of allowed key management techniques derive a key encryption key (KEK)
which can be used to authenticate the sender e.g. Static-Static Diffie Hellman,
Kerberos shared secret, shared password. The KEK is used to encrypt the CEK per
recipient.
However when the integrity of the message is validated via the message HMAC,
the sender only demonstrates knowledge of the CEK to the recipient.
Any other recipient knows the CEK so can construct another message using the
same CEK and compute the new HMAC and reuse the encrypted KEK data from the
original message. For example:
Alice sends a message to Bob and Mallet. Alice uses different KEK's for both
Bob and Mallet to encrypt the CEK. Bob decrypts the CEK using his KEK. The
message decrypts and the HMAC checks out so Bob think the message is from Alice
because only Bob and Alice know the KEK.
Mallet is the recipient of the original message so he can use his KEK to learn
the message CEK. Mallet then constructs a new message and encrypts it using
the same CEK as the original message. Mallet copies Bobs KEK recipient Info
from Alice's message which contains the same encrypted CEK. Because Mallet
knows the CEK he can also compute the correct HMAC for the new message. Mallet
sends the new message to Bob. Bob again uses his KEK to discover the CEK and
successfully decrypts the message and the HMAC check out so Bob thinks this new
message is also from Alice.
This issue only applies when there are multiple recipients. The motivation to
use the new authenticated encryption is to improve performance so solutions
like bifurcating messages to use different CEK or signing the message are
counterproductive.
We need an additional structure to allow the sender to demonstrate knowledge of
both the KEK and CEK per recipient when sending to multiple recipients.
What is the consensus to resolve this?
We can version 5083 to document the new structure or we can do document the new
structure via a standalone document.