Historically, the approach to "secure" mail has been to provide mechs that
let a receiver validate the identity of the sender, but I am really
wondering if this isn't inappropriate in the majority of the cases.
In the paper-based world, we "trust" the delivery system, and do not
normally attempt to validate the supposed sender of the materials. We
don't demand that people physically bring contracts to us so that we can
verify that they are the authors. We don't demand thumbprints for
comparison to a notarized set of fingerprints on file. Instead, we "trust"
the letterhead and the postmarks (or the fax caller-ID), and then use
blind assumptions for the rest of it. Only in special cases do we require
validation of sender identity.
Similarly, with protocols like HTTP, we trust the transfer protocol, and
do not demand that each HTML document be independently signed. We assume
that the documents have not been forged and rely on the delivery system to
provide the minimum comfort level.
So with mail, for 99% of the cases, we could probably provide "secure
mail" by adding integrity to the delivery system. We can still use
certificates and signatures for the remaining cases where we need to
"prove" identity, but for the vast majority of cases, just knowing that
"this host is authorized to send mail for this domain, so I will trust
that this sender is legitimate" will probably be good enough. It would
certainly be good enough to crush anonymous spam, and would also be
sufficient for the overwhelming day-to-day communications.
Also, the SMTP part of the transfers is the only time when the parties are
guaranteed to be connected, so providing the necessary credentials at this
phase is somewhat feasible, whereas user certs and the like may not be
workable due to asynchronous connectivity problems.
The questions then, is it possible to provide this level of security
within the protocol, such that it is easy to implement on a wide basis,
but it is secure enough not to be easily forged? Forget the users, is
there a practical way to ensure that public SMTP agents can always be
positively identifed?
--
Eric A. Hall http://www.ehsco.com/
Internet Core Protocols http://www.oreilly.com/catalog/coreprot/