On Sat, 2005-01-15 at 18:01 -0500, John R Levine wrote:
I have not seen an argument that these proposals scale to a key per
message. I'm not saying they don't; I'm saying that's not how people
seem to be thinking about the problem or writing software.
I don't think that's likely either, but assuming you had a specialized DNS
server that could handle the stream of key updates, I don't see any reason
to rule it out.
The domains, where per-user-keys becomes a valuable approach for
preventing abuse of the signature, the number of such accounts will
likely be very large. The ability to scale DNS deployment is one issue,
expecting everyone to cache this influx of public-keys is another.
Scaling concerns are raised when the amount of data sent over DNS
requires specialized servers.
If there is a revocation scheme using records just to revoke specific
accounts (identified within the header with u=xyz01234 as example), then
acceptance of the message may conditioned upon an address lookup of:
xyz01234._arl.<domain>.
Any record answering the query indicates the account is no longer valid.
The TTL on the domain key could be long, but the ability to revoke use
of their key for a specific account could be very rapid. This would be
using the same technique employed for blackhole lists. The difference
from using a blackhole list would be this would be managed by the domain
signing the messages. Their incentive would be defensive. Otherwise,
wide spread abuse would put their DNS servers at risk and reduce
benefits of having deployed signatures as a means to ensure message
acceptance.
Should the domain monitor address queries for *._arl.<domain>, they
would be able to quickly identify when there is likely a gross level of
abuse of their signatures occurring. Rapidly curtailing abuse would be
a means to discourage it in the first place. A broken window approach
to the problem so to speak. A typical lifespan for a mail delivery key
may not seem long, but it would allow billions of messages to taint the
signature. The short lifespan means a revocation list should remain
small and easy to manage however.
Most domains would not need to deploy this mechanism provided they do
not see their domain signature abused. This arrow in the quiver could
be deployed rapidly by simpling including the u= information. The
efforts to handle abuse on a small scale could be done manually and
require only a few DNS entries, even for domains with thousands of
accounts.
-Doug