Greylisting is a hoary technique for rejecting spam sent by botnets and other poorly written spamware. When a mail server receives an attempt to deliver mail from a hitherto unseen sending host IP address, it rejects the message with a “soft fail” error which tells the sender to try again later. Real mail software does try again, at which point you note that the host knows how to retry and you don’t greylist mail from that IP again. The theory is that spamware doesn’t retry, so you won’t get that spam. I wrote a paper on it for the 2005 CEAS conference, and concluded that conservative greylisters worked well.

We’ve now been using greylisting for close to a decade, and some people have argued that it’s no longer useful, since the bad guys could easily fix their spamware to retry, or since bots are so cheap, they could just send everything twice. So does it still work?

I recently went through my greylister’s logs and collected some statistics for both a recent week, and the past year, about hosts that I greylisted:

The first row is the number of hosts that got a soft fail and never came back. The second row is the number that retried the message that failed, but never sent anything again, and the third row is the number that retried and sent more messages after that.

As you can see, for the week, about half of the greylisted hosts didn’t retry, and over a year, about 2/3 didn’t. That’s still a lot of mail my mail server didn’t have to filter. I attribute the different ratios to the shutdown of several botnets over the past year, evidently botnets that didn’t retry.

So it’s certainly not a magic bullet (what is?) but greylisting still is an effective way to deter a lot of spam cheaply.

Next, Greylisting Still Works - Part II