Anecdotal evidence seems to indicate that Google and possibly other mail service providers are either quite ignorant of history when it comes to email and spam, or are applying unsavoury tactics to capture market dominance.
The first inklings that Google had reservations about delivering mail coming from my bsdly.net domain came earlier this year, when I was contacted by friends who have left their email service in the hands of Google, and it turned out that my replies to their messages did not reach their recipients, even when my logs showed that the Google mail servers had accepted the messages for delivery.
Contacting Google about matters like these means you first need to navigate some web forums. In this particular case (I won’t give a direct reference, but a search on the likely keywords will likely turn up the relevant exchange), the denizens of that web forum appeared to be more interested in demonstrating their BOFH ishness than actually providing input on debugging and resolving an apparent misconfiguration that was making valid mail disappear without a trace after it had entered Google’s systems.
The forum is primarily intended as a support channel for people who host their mail at Google (this becomes very clear when you try out some of the web accessible tools to check domains not hosted by Google), so the only practial result was that I finally set up DKIM signing for outgoing mail from the domain, in addition to the SPF records that were already in place. I’m in fact less than fond of either of these SMTP addons, but there were anyway other channels for contact with my friends, and I let the matter rest there for a while.
If you’ve read earlier instalments in this column, you will know that I’ve operated bsdly.net with an email service since 2004 and a handful of other domains from some years before the bsdly.net was set up, sharing to varying extents the same infrastructure. One feature of the bsdly.net and associated domains setup is that in 2006, we startedpublishing a list of known bad addresses in our domains, that we used as spamtrap addresses as well as publising the blacklist that the greytrapping generates.
Over the years the list of spamtrap addresses — harvested almost exclusively from records in our logs and greylists of apparent bounces of messages sent with forged From: addresses in our domains – has grown to a total of 29757 spamtraps, a full 7387 in the bsdly.net domain alone. At the time I’m writing this 31162 hosts have attempted to deliver mail to one of those spamtrap addresses. The exact numbers will likely change by the time you read this — blacklisted addresses expire 24 hours after last contact, and new spamtrap addresses generally turn up a few more each week. With some simple scriptery, we pick them out of logs and greylists as they appear, and sometimes entire days pass without new candidates appearing. For a more general overview of how I run the blacklist, see this post from 2013.
In addition to the spamtrap addresses, the bsdly.net domain has some valid addresses including my own, and I’ve set up a few addresses for specific purposes (actually aliases), mainly set up so I can filter them into relevant mailboxes at the receiving end. Despite all our efforts to stop spam, occasionally spam is delivered to those aliases too (see eg the ethics of running the traplist page for some amusing examples).
Then this morning a piece of possibly well intended but actually quite clearly unwanted commercial email turned up, addressed to one of those aliases. For no actually good reason, I decided to send an answer to the message, telling them that whoever sold them the address list they were using were ripping them off.
That message bounced , and it turns out that the domain was hosted at Google.
The page, which again is intended mainly for Google’s own customers, specifies that you should set up SPF and DKIM for domains. But looking at the headers, the message they reject passes both those criteria:
Received-SPF: pass (google.com: domain of email@example.com designates 2001:16d8:ff00:1a9::2 as permitted sender) client-ip=2001:16d8:ff00:1a9::2; Authentication-Results: mx.google.com; dkim=pass (test mode) firstname.lastname@example.org; spf=pass (google.com: domain of email@example.com designates 2001:16d8:ff00:1a9::2 as permitted sender) firstname.lastname@example.org DKIM-Signature: v=1; a=rsa-sha256; q=dns/txt; c=relaxed/relaxed; d=bsdly.net; s=x; h=Content-Transfer-Encoding:Content-Type:In-Reply-To:MIME-Version:Date:Message-ID:From:References:To:Subject; bh=OonsF8beQz17wcKmu+EJl34N5bW6uUouWw4JVE5FJV8=; b=hGgolFeqxOOD/UdGXbsrbwf8WuMoe1vCnYJSTo5M9W2k2yy7wtpkMZOmwkEqZR0XQyj6qoCSriC6Hjh0WxWuMWv5BDZPkOEE3Wuag9+KuNGd7RL51BFcltcfyepBVLxY8aeJrjRXLjXS11TIyWenpMbtAf1yiNPKT1weIX3IYSw=;
Then for reasons known only to themselves, or most likely due to the weight they assign to some unknown data source, they reject the message anyway.
We do not know what that data source is. But with more than seven thousand bogus addresses that have generated bounces we’ve registered it’s likely that the number of invalid bsdly.net From: addresses Google’s systems has seen is far larger than the number of valid ones. The actual number of bogus addresses is likely higher, though: in the early days the collection process had enough manual steps that we’re bound to have missed some. Valid bsdly.net addresses that do not eventually resolve to a mailbox I read are rare if not entirely non-existent. But the ‘bulk mail’ classification is bizarre if you even consider checking Received: headers.
The reason Google’s systems most likely has seen more bogus bsdly.net From: addresses than valid ones is that by historical accident faking sender email addresses in SMTP dialogues is trivial.
Anecdotal evidence indicates that if a domain exists it will sooner or later be used in the from: field of some spam campaign where the messages originate somewhere else completely, and for that very reason the SPF and DKIM mechanisms were specified. I find both mechanisms slightly painful and inelegant, but used in their proper context, they do have their uses.
For the domains I’ve administered, we started seeing saw log entries, and in the cases where the addresses were actually deliverable, actual bounce messages for messages that definitely did not originate at our site and never went through our infrastructure a long time before bsdly.net was created. We didn’t even think about recording those addresses until a practical use for them suddenly appeared with the greytrapping feature in OpenBSD 3.3 in 2003.
A little while after upgrading the relevant systems to OpenBSD 3.3, we had a functional greytrapping system going, at some point before the 2007 blog post I started publishing the generated blacklist. The rest is, well, what got us to where we are today.
From the data we see here, mail sent with faked sender addresses happens continuously and most likely to all domains, sooner or later. Joejobs that actually hit deliverable addresses happen too. Raw data from a campaign in late 2014 that used my main address as the purported sender is preserved here , collected with a mind to writing an article about the incident and comparing to a similar batch from 2008. That article could still be written at some point, and in the meantime the messages and specifically their headers are worth looking into if you’re a bit like me. (That is, if you get some enjoyment out of such things as discovering the mindbogglingly bizarre and plain wrong mail configurations some people have apparently chosen to live with.)
Anyone who runs a mail service and bothers even occasionally to read mail server logs will know that joejobs and spam campaigns with fake and undeliverable return addresses happen all the time. If Google’s mail admins are not aware of that fact, well, I’ll stop right there and refuse to believe that they can be that incompentent.
The question then becomes, why are they doing this? Are they giving other independent operators the same treatment? If this is part of some kind of intimidation campaign (think "sign up for our service and we’ll get your mail delivered, but if you don’t, delivering to domains that we host becomes your problem" ). I would think a campaign of intimiation would be a less than useful strategy when there are alread antitrust probes underway, these things can change direction as discoveries dictate.
Normally I would put oddities like the ones I saw in this case down to a silly misconfiguration, some combination of incompetence and arrogance and, quite possibly, some out of control automation thrown in. But here we are seeing clearly wrong behavior from a company that prides itself in hiring only the smartest people they can find. That doesn’t totally rule out incompetence or plain bad luck, but it makes for a very strange episode. (And lest we forget, here is some data on a previous episode involving a large US corporation, spam and general silliness.)
One other interesting question is whether other operators, big and small behave in any similar ways. If you have seen phenomena like this involving Google or other operators, I would like to hear from you by email (easily found in this article) or in comments.
[Opinions offered here are my own and may or may not reflect the views of my employer.]
I will be giving a PF tutorial at BSDCan 2016 , and I welcome your questions now that I’m revising the material for that session. Seethis blog post for some ideas (note that I’m only giving the PF tutorial this time around).