a note to those who would automate their rejection notices
bruns at 2mbit.com
Sat Dec 27 22:01:47 UTC 2003
On Saturday, December 27, 2003 3:23 PM [GMT-5=EST], Paul Vixie <paul at vix.com>
> Anyway, I hope folks will stop sending automated rejection notices to
> domains who were not involved, other than by forgery, in the transmission
> of a virus or spam. In other words, there's relevant operational content
> in this thread, and when "fighting" spam it would be reasonable to avoid
> hurting uninvolved third parties. AOL, please listen.
Cox in particular was doing this until recently (we got their attention rather
quickly after blacklisting their main mail servers). We were being joe jobbed
badly, and cox's mail servers were generating massive amounts of bounces per
minute, and out of all the bounces, cox was generating the most (at least 3/4
The result was that each one of their mail servers (more then a dozen) was
sending one bounce per connection, and launching anywhere between 5-12
connections at a time, then reconnecting right away after sending the single
bounce and disconnecting. We quickly ran out of connection slots on both the
primary and secondary mail spoolers, leaving us unable to get incoming mail
until we firewalled out cox's mail servers.
One would think, if your going to run a cluster of mail servers to handle your
mail, that you would rate limit your bounces so that people (like myself) who
can't afford to have a dozen or more heavy duty mail servers don't end up
getting DoS'd by your mail server's ability to pump out millions of messages
Someone said on one of the newsgroups, "Well, maybe they setup their system
correctly, and don't see a need to change something that works." The problem
is, theres a difference between properly configuring a mail server and
responsibly configuring a mail server. When you responsibly configure a mail
server, you take into account OTHER people's systems and how THEY will be able
to deal with your server.
Part of the issue comes with when you accept a mail, then bounce afterwards,
instead of just bouncing after RCPT TO: or DATA. When you delay the bounce,
you will generate a bounce to the From: address, even if it is forged. When
you outright reject the message, you pretty much reduce the risk of that
happening by far, as the sending server will see that the message was
rejected, and hopefully move on. Now, this works with open proxies, but not
with open relays. Do spammers use open relays much anymore? No, not really.
Why leave a trail back to yourself when you can hide completely?
AOL has _not_ done this to us though, we've seen maybe one or two bounces from
AOL's servers, but nothing even remotely close to what Cox is doing.
Just my thoughts, flame away :)
The Summit Open Source Development Group
Open Solutions For A Closed World / Anti-Spam Resources
The AHBL - http://www.ahbl.org
More information about the NANOG