On Wed, February 20, 2019 2:23 pm, Wiktor Kwapisiewicz wrote:
1) Allow a PGP Key Certificate without a signature key. The idea here
is that some small devices may only have a key-agreement key
available so may not be able to create a signature. However, 4880
did not allow this configuration and required a top-level signature
key and relegated other keys to sub-usage.
By "signature key" do you mean "certification key"?
Because as far as I can see RFC 4880 definitely allows having the
primary key being certification only:
"In a V4 key, the primary key MUST be a key capable of certification" 
(I'm using C-only primary key, it's available via WKD).
I think you misunderstood my issue. My issue is exactly what you state
here, that the primary key MUST be a key capable of certification. That's
exactly what I DO NOT want. The keys I'm using are NOT capable of
certification, but I still want them to be the primary key. Hence, I
needed to modify that requirement.
2) The reason for the User ID Attribute subpacket was that we wanted to
have multiple Attribute subpackets included in the certificate in a
primary signature, but this was not possible with 4880. My memory
hazy on what the exact issue was, but IIRC you could EITHER have a
UserID packet OR a set of Attribute packets, but not both. Because
wanted both a UserID *AND* additional attributes in a single
signature, this seemed the best way to do it.
What would you store in these User ID Attributes that would not be
possible in regular User IDs?
This was from 3-4 years ago, so like I said, memory hazy, and project long
turned over to others.. But my recollection was not just about what was
being stored in the UserID,, but what ELSE was being stored in the same
signature. Like I said above, my recollection is that I could not store
both a UserID *AND* additional attributes at the same time, so the best
workaround was to add a UserID Attribute so I could.
In short, the specification did not allow us to do what we wanted to do,
which is why I suggested this change. I do not recall exactly what it
was, and frankly I don't have the time to go dig up the project and see
how we were using it.
3) Yes, we could have just used FQDN-based notations, but then we're
litterally adding at least 13 bytes PER NOTATION. Given the number
of notations in use, that was adding on the order of 65-130 BYTES to
the certificate, or about 10-15%!
This depends on the FQDN, e.g. using '@ihtfp.com' adds only 10 bytes,
and there are shorter domains. This can be coupled with shortening of
the keys (e.g. 'prodid' -> 'p') although I admit they look shortened
This is not for ihtfp.com -- The 13 bytes is the actual number. I chose
the registrations to be as short as reasonable. I didn't want to use 'p'
because I knew in 5-10 years someone would come along and go "what the F
is 'p' for?". Whereas "prodid" is a bit more self-descriptive.
Are all of them needed and used at the same time?
Yes. Hence my 10-15% increase in size! Indeed, some of them are even
used more than once in a single certificate.
Having said that I wasn't around when it was conceived so probably
there is some rationale for its inclusion.
Indeed. Honestly, at the time the intent was to progress the
device-certs draft on its own and register those notations that way.
But then the WG stalled, and Werner kindly incorporated those
definitions into 4880bis.
Still, adding these to 4880bis does not add significant complexity to
the draft but DOES make a marked difference in its usability. Please do
not remove these improvements.
I don't know what are the criteria for inclusion but when reading the
rest of the RFC device-certs strike me as something with a narrow focus.
A stark contrast to otherwise generic and versatile constructs.
True -- which is why it was originally proposed as a separate document.
I'm fine with it continuing to be a separate document if that's the desire
of the group, but I DO want to make these IANA registrations and I DO want
to make the change to allow for a non-certification primary key. I'm fine
either way, either the registrations being in 4880bis or resurrecting the
device-certs draft and progressing that alongside 4880bis.
The process just got muddied when the WG was shut down.
Could you share what's the rationale for keys being minimal? From what I
have seen it has something to do with devices of limited memory but I'm
eager to know details, are there any implementations of device-certs in
This is exactly it -- we are storing a certificate on an NFC (or even UHF
RFID) chip that would be presented to a verifier -- and then the chip
would perform a key exchange using the key in the certificate for a
proof-of-possession authentication. Every byte matters, which is exactly
why we went with OpenPGP and not X509!
For example what would the User ID Attribute look like?
See above; I don't recall -- I'd have to go dig it up from this 3-4 year
old project. However, it wasn't the content but the co-existence that was
the issue IIRC.
 I did experiment with putting URIs in User IDs for extended info
Cool. But not useful to me ;)
Hard to recommend something if I don't know what you're after :)
Sure, but I'm not looking for you to recommend something. :)
Derek Atkins 617-623-3745
Computer and Internet Security Consultant
openpgp mailing list