Greetings again. draft-ietf-dnsop-domain-verification-techniques talks in many 
places about random tokens. Section 5.1.1 mixes up "unique tokens" and 
cryptographically secure random tokens, including having MUST-level 
requirements that are inappropriate.

If an ASP wants to be sure that a particular user has control of a domain, it 
may be better for them to tell the user to put that user's ASP-associated name 
in the token because then the ASP doesn't have to store a long-term association 
between the user and the random value. Thus, the token is really not random, 
but is in fact unique for the site.

Worse, Section 5.1.1 says that the random value must have 128 of entropy, even 
though the use case for DCV described in Section 3 has absolutely nothing to do 
with randomness. To be unique, the token can have a 32-bit number (assuming 
that the service doesn't have more than 4 billion users...).

The implications of "random" will make the other good proposals in the document 
less likely to be implemented correctly. I propose that the document 
consistently use the term "unique token" throughout, that section 5.1.1 be 
rewritten to be about uniqueness, and the text in that section be amended to 
say "if you want, a good way to generate unique tokens is to pick random values 
and then encode them this way". This way, we will be describing a best practice 
(using a token) and telling the reader multiple ways to do so.

I'm happy to propose specific text changes in the repo if the WG agrees.

--Paul Hoffman

_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to