Hi Jim,

the random value in the DTLS/TLS handshake needs to be random.

In the OAuth-ACE framework we assume some initial keying material
provided to the device (for example using some commissioning tool or
during manufacturing). That key is the long-term key.

Generating further keys that are bound to access tokens is then
something provided by the OAuth-ACE protocol and also there randomness
is needed when deriving the keys (except for the PSK case at the moment
where that key is generated on the server-side).

Note that some ciphers in TLS also utilize ECDHE and therefore also need
to generate those keys.

Ciao
Hannes

PS: Assuming wall-time support is yet another challenge. It is easier to
find micro-controllers with a hardware-based random number generator
than it is to integrate time into the solution.

On 06/09/2016 02:09 AM, Jim Schaad wrote:
> The client random value does not need the same quality of randomness
> that would be desired for a private key generation process – especially
> a long term key.  I would need to sit down and think about it, but it is
> quite possible that a counter would work for the client random value. 
> The main thing you want with this is non-repeating so that the final
> secret is always uniquely generated.  There have been cases where time
> has been part of this random value and I don’t believe that it makes the
> final key any less secure.
> 
>  
> 
> Jim
> 
>  
> 
>  
> 
> *From:*Ace [mailto:[email protected]] *On Behalf Of *Samuel Erdtman
> *Sent:* Monday, June 06, 2016 9:58 PM
> *To:* Julien Vermillard <[email protected]>
> *Cc:* [email protected]
> *Subject:* Re: [Ace] Constrained Environment PKI enrollment
> 
>  
> 
> Okay so there needs to be some sort of random number generation for the
> TLS-PSK handshake, I was not aware of that. (is the requirements on it
> as rigor as for the key generation)
> 
> The other reason I could think of that is kind of mentioned previously,
> it is the "quality" of the generated key, in some deployments, FIPS
> certified key generation is required and that could be easier to
> guarantee server-side.
> 
> But it might be that there are no absolute situations where server-side
> key-generation but rather a design option.
> 
> //Samuel
> 
>  
> 
>  
> 
> On Mon, Jun 6, 2016 at 5:05 PM, Julien Vermillard <[email protected]
> <mailto:[email protected]>> wrote:
> 
>     Yes, but since you need some random numbers for doing a TLS-PSK
>     handshake (ClientHello random) why a TLS-PSK client should no be
>     able to generate ECDSA keys?
> 
> 
>     --
>     Julien Vermillard
> 
>      
> 
>     On Mon, Jun 6, 2016 at 4:30 PM, Samuel Erdtman <[email protected]
>     <mailto:[email protected]>> wrote:
> 
>         Hi Julien,
> 
>         The first one that comes to mind would be where Pre-Shared-keys
>         are used i.e. symmetric keys. These keys could be factory keys
>         used only to setup the PKI-keys.
> 
>         If you look at EST the server generated keys is also recommended
>         to be wrapped in an extra layer of encryption in addition to the
>         transport layer security (DTLS/TLS). So if the transport layer
>         security is not absolute no keys should be leaked.
> 
>          
> 
>         //Samuel
> 
> 
> 
> 
>          
> 
>         On Mon, Jun 6, 2016 at 3:18 PM, Julien Vermillard
>         <[email protected] <mailto:[email protected]>> wrote:
> 
>             Hi Samuel,
> 
>             I wonder in which scenario a RNG is safe enough for running
>             a DTLS stack but not good enough for generating a ECDSA key
>             couple?
> 
> 
>             --
>             Julien Vermillard
> 
>              
> 
>             On Fri, Jun 3, 2016 at 5:08 PM, Samuel Erdtman
>             <[email protected] <mailto:[email protected]>> wrote:
> 
>                 The company I previously worked for where looking into
>                 adopting EST for this purpose, the benefit of EST
>                 compared to cmp or scep was that it defined the process
>                 for server side generated keys, which could be
>                 beneficial if key generation would be to cumbersome for
>                 the device or if you don't trust the device to generate
>                 a "good" key.
> 
>                  
> 
>                 Maybe Shahid could give sold more updates since he was
>                 helping us with this project
> 
> 
> 
>                 On Thursday, 2 June 2016, Julien Vermillard
>                 <[email protected] <mailto:[email protected]>>
>                 wrote:
> 
>                     Hi,
> 
>                     In industrial or enterprise M2M/IoT application we
>                     often use PSK for authentication, but more and more
>                     user want to enroll the device on their public key
>                     infrastructure like they does with some routers
>                     using SCEP/CMP.
> 
>                     I wonder if it was explored to enroll devices, and
>                     renew certificates on PKI only using CoAP and not HTTP?
> 
> 
>                     --
>                     Julien Vermillard
> 
>              
> 
>          
> 
>      
> 
>  
> 
> 
> 
> _______________________________________________
> Ace mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/ace
> 

Attachment: signature.asc
Description: OpenPGP digital signature

_______________________________________________
Ace mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/ace

Reply via email to