[Top] [All Lists]

Re: [ietf-smtp] [pkix] key identities, why you shouldn't even try to canonicalize local parts

2016-03-22 13:44:02
For encrypted mail, this is clearly right.  For signed mail, it's not 
but it's also less clear.  The whole issue of binding real world identities 
to e-
mail addresses is a swamp, not one that I think we are any better at draining
than anyone else.  For PGP, there's the web of trust which is supposed to
help you decide whether a key matches a person, but doesn't really say
anything about the e-mail addresses attached to the key.  For S/MIME the
CA does what it does, which these days is rarely any more than a challenge
message to the e-mail address.

So I mostly agree with you, with the caveat that you have to be careful not to
misinterpret assurances about an e-mail address as assurances about the
person or entity allegedly associated with that address.

What I loved about SPKI/SDSI was the idea of namespaces and namespace chaining. 
 "The key that Alice nicknamed 'Bob's key'" is unambiguous and implementable.  
Once I've named a key, a computer can track that name across multiple different 
identifiers effortlessly.  Once I tell the computer that *this specific key* 
represents "Alice from the Home Office," Alice can use any address--indeed, any 
protocol that allows her to sign messages--and my computer can tell me it's 
Alice from the Home Office.

No complex web of trust, no CAs.  If I were implementing, I'd be straight up 
stealing this.

Yes, a priori discovery in a morass of SPKI/SDSI certificates is a right royal 
PITA, but you can't have everything.  But then again, maybe we don't actually 
have that problem.  It's all in how you think about the issue.

ietf-smtp mailing list

<Prev in Thread] Current Thread [Next in Thread>