Method to verify that a device is a genuine Tillitis TKey

[mchack-work]

Goal: Let a user verify that the TKey is a genuine Tillitis.

We need new provisioning and verification software, mostly (only?) on the host side, possibly the same program with different arguments.

Provisioning:

• Get UDI from TKey.
• Run an app, probably signerapp from a tagged version.
• Get signerapp's public key.
• hash(udi,pubkey)
• Sign hash with our own secret key.
• Store signed hash somewhere (print a QR code? publish behind a web form?), possibly publish the signed hash in Sigsum.

Verification by user:

• Somehow get hold of our public key. Perhaps in the software repo and/or our web site.
• Get the published signed hash. Perhaps a web page? Perhaps a QR code? (But see below about Sigsum.)
• Get UDI from TKey.
• Run the same TKey app we did (signerapp?)
• Get signerapp's public key.
• hash(udi, pubkey)
• Get the published signed hash. Perhaps a web page? Perhaps a QR code? Perhaps in Sigsum?
• Verify signature on the signed hash.
• Compare published hash with computed hash. If it's the same, the we ran the same software on the same stick in provisioning.

Software needs:

• One provisioning host program that extracts UDI, runs the signerapp, gets the stick's pub key, and outputs a signed hash(udi, pubkey). Possibly storing in Sigsum or somewhere else.

• One user host program (same as provisioning?) that extracts UDI, runs the signerapp, extracts the pub key, does hash(udi, pubkey) and checks against a provided hash.

[mchack-work]

We start with generating signatures and storing them in a file named as the hash. Figure out where to actually store them later.

[secworks]

Or store with the UDI as the name. UDI-file that contains the hash, tag and signature.

[quite]

I think we wanted to use the key (file) named after the hash H to store the signature S. If only because H realistically can't be enumerated.

[secworks]

I think we want to use the UID to allow users to easily look up the hash. It is a public ledger. Anybody could download the hashes for all ID:s We can do, should do a threat analysis. But to me, using the UDI in this instance does not really leak any info about a user, bind a user to a specific device.

What is the benefit of not being able to do enumeration in this case?

[quite]

I'm closing this as done. Let's open new issues for anything arising when deploying the current implementation.