Re: [Ace] IETF 108 tentative agenda and presentations (Daniel Migault)

["Panos Kampanakis (pkampana)" <[e-mail]>]

Hi Hendrik, 

Thank you. Understood about the end-to-end protection of CMP and POP. 

I would argue that establishing the end-to-end keys to offer the application
level protection functionality in a scalable way does not come easily. On
the other hand, even CMP allows for an RA trust model instead of end-to-end
POP like EST-coaps does. 

> I am uncertain if I understand your question correctly. Est-over-coaps
described EST transport and not CMP transport on top of CoAP.

I meant why do we need two enrollment protocols to run over COAP?
est-over-coaps took a long time and a lot of work. The reason we pursued it
is because we were getting requests from vendors that wanted to enroll certs
in constrained environments in the energy sector and industrial automation
and EST was standardized in IEC 62351. Is the cmp-over-coap argument that
you could run it over plan UDP and use out-of-band established, end-to-end
protection the sole motivating reason? 

Rgs, 
Panos


-----Original Message-----
From: Ace <[e-mail]> On Behalf Of Brockhaus, Hendrik
Sent: Wednesday, July 22, 2020 9:42 AM
To: Panos Kampanakis (pkampana) <[e-mail]>; Benjamin Kaduk
<[e-mail]>; Michael Richardson <[e-mail]>
Cc: Mohit Sahni <[e-mail]>; [e-mail];
[e-mail]
Subject: Re: [Ace] IETF 108 tentative agenda and presentations (Daniel
Migault)

Hi Panos,

thnaks for you feedback.

> Von: Panos Kampanakis (pkampana) <[e-mail]>
> 
> Hi,
> 
> > Looking into Mohits draft, cmp-over-coap is much simpler than
> est-over-coaps, as CMP does not need any binding to an underlying 
> (D)TLS handshake.
> 
> Not sure that is accurate. And EST does not bind to the tunnel 
> protocol either unless proof of possession is used. For now the 
> cmp-over-coap draft says
> 
>    When the end to end secrecy is desired for CoAP transport, CoAP over
>    DTLS [RFC6347] as a transport medium SHOULD be used.
> 
> COAP can run over DTLS or plain UDP and in rare cases TCP, TLS and 
> maybe Websockets. I am not sure someone would run cmp-over-coap over 
> TCP because then he could just run CMP natively without COAP in the 
> middle. Any application layer protocol (CMP etc) can run over any 
> transport but I am
not
> sure there are more transports than the usual ones for cmp-over-coap
anyway.

It is not needed to run CMP over CoAP over TCP. UDP as transport protocol is
fine. Actually CMP over CoAP also does not need DTLS underneath. But it also
does not hinder to have a second line of defense.
As I understand EST, proof-of-possession is purely provided by the
self-signature in PKCS#10. But EST provides the proof-of-identity of the
requesting party by the (D)TLS client authentication bound to the PKCS#10
(tls-unique copied in the P10 password filed). Is this correct?
Such binding is not required for CMP. CMP does not have any requirements in
this regard and provides prove-of-identity by signing the CMP messages. The
advantage is that this prove-of-identity can be end-to-end and not only on
the first hop to the (D)TLS server, like with EST.

> 
> 
> I agree that if this gets picked up it should be by ACE.
> 
> I would like to understand what gaps it is filling compared to 
> est-over-coaps which took a lot of work and where it will be used and 
> implemented in.

I am uncertain if I understand your question correctly. Est-over-coaps
described EST transport and not CMP transport on top of CoAP.
One prototypic implementation can be found on
github.com/siemens/embeddedCMP.

- Hendrik