spf-discuss
[Top] [All Lists]

Re: RCPT TO: rejecting

2004-05-26 18:40:44
On May 26, 2004, at 8:32 PM, Seth Goodman wrote:
I take that to mean that one client was maliciously knocked off the net by a concerted attack by Verizon. I take the second instance as meaning that AOL intentionally targeted that server and saturated its network bandwidth to
the point where it had no effective internet connectivity left.  If,
instead, you are saying that the first site got more CBV's than it wanted,
the second site received lots of forged bounce spam and both were very
annoyed, that's life on the uncontrolled internet, not a DoS attack.

No. It is pretty clear that Verizon and AOL wouldn't intentionally do that. They were used by a 3rd party as a weapon do that. And yes the services were taken offline in both cases due to service saturation.

"more than they wanted" and "lots of forged bounce spam"? If either of those occupy enough of my resources to degrade my services and thus deny service to any clients, then yes it is absolutely a denial of service attack. Denial of Service attacks do not need to completely saturate anything, they just need to affect them and denies legitimate traffic (not necessarily *all* legitimate traffic). http://www.cert.org/tech_tips/denial_of_service.html

Here I'd like to ask you a question. Isn't it possible to engineer an MTA so that a CBV is not as burdensome as you describe? Perhaps it isn't, but here's the basis for my question. You can't tell after RCPT TO: whether the SMTP-client is doing a CBV or sending a DSN. If the response to RCPT TO: is a 250, a CBV request will terminate shortly thereafter but a DSN will go on
to a DATA command.  Would it be possible to delay the evaluation of the
user's quota and anything else besides local address validity until after you get the DATA command? If for some reason the account cannot accept the DSN, isn't it possible to respond to the DATA command with a 503 or 554 to
terminate the transaction instead of a 354?

It's very awkward with multi recipient mail. It can be done with single recipient mail, only your don't have forward knowledge of that.

The amusing thing here is that I don't particularly like the idea of CBV's, either. If you can think of a more efficient way to validate MAIL FROM: after an arbitrary number of forwards, I'd be very interested. I would much rather that a DNS server could provide the validation, since it is based on UDP, but that doesn't seem to be terribly feasible. A full PKI system is
far more expensive than CBV's, and shouldn't be necessary to solve this
particular problem. Looking at the big picture, I feel that a cooperative
model, which may include CBV's (or some other mechanism), gets us all
further than a system where we never "bother" each other and continue to
suffer individually.  I'm sure that game theory has something to offer
concerning this scenario, but unfortunately, I'm unqualified to comment.

I'm practical person too.

Full PKI isn't really that expensive and the beauty of it is that is doesn't victimize anyone. As for signing and verification, the cost for reasonable performance is really minimal (as in most people won't need hardware acceleration). With a $1000 crypto card in a server you can do 4,000 or so signs or validates per second. 4,000 messages/second is above our test lab results for Ecelerity. We can really only push about 2000 message/second over SMTP in our lab (about 7.2 million message/hour). As people are striving for these lab results in real life and 300 message/second on a single server seems to satisfy most people with performance concerns, PKI just doesn't seem that intimidating.

We'll see how the DomainKeys stuff plays out, that stuff looks exciting too.

// Theo Schlossnagle
// Principal Engineer -- http://www.omniti.com/~jesus/
// OmniTI Computer Consulting, Inc. -- http://www.omniti.com/
// Ecelerity: fastest MTA on Earth


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