Re: [openpgp] [Cfrg] streamable AEAD construct for stored data?

[Bryan Ford <[e-mail]>]

On Feb 10, 2016, at 1:52 PM, Nils Durner <[e-mail]> wrote:
> Hi,
> 
>> To be clear, there are two separate use-cases, each of which make
>> sense without the other and require different technical solutions (but
>> could also make sense together):
>> 
>> 1. Streaming-mode integrity protection:
>> 
>> [...]
>> 
>> To achieve goal #1 properly, it appears that what we need is not only
>> a MAC per chunk but a signature per chunk.
> 
> Different ideas:
> 
> 1. asymmetrically encrypt and sign the MAC key, make this a new packet
>    type to be prepended to the symmetrically encrypted data

By this, do you mean just write one asymmetrically encrypted-and-signed MAC key at the beginning of the stream, followed by a bunch of records that are only MAC-authenticated with that symmetric key?  

This would appear insecure to me, at least in the case the stream is encrypted to two or more recipients.  Say Alice signs-and-encrypts a stream to Bob and Charlie.  Bob takes Alice’s encrypted-and-signed MAC key record, then uses the same MAC key to construct a completely different stream of actual content (all of whose MAC records verify just fine) and sends it to Charlie, claiming that it’s from Alice.

Maybe this is only a problem in the two-or-more-receivers case, but even if so it makes me nervous.  If PGP had a reputable, non-signing sender-authentication mode for 2-party communication only, then it might make sense for an asymmetric “repudiable authentication record” to be followed by a stream of MAC-authenticated records.  But that seems like a fairly different protocol (or at least a fairly different mode).

> 2. derive the MAC key from the symmetric encryption key, sign it (but
>    do not store it) and make this a new packet type to be prepended
>    (thus saving the asymmetric encryption from #1)
> 3. use an authenticating sym cipher mode with intermediate
>    authentication tags, with the symmetric key asymmetrically signed
>    (like #2)

Assuming I’m correctly understanding that in cases #2 and #3 also just have one asymmetric record at the beginning of the stream, it seems like the same considerations apply as with #1.  Perhaps OK for 2-party repudiable authentication, but not if we need to retain the signed-message semantics that PGP currently provides especially in the multiple-receiver case.

Cheers
Bryan