Skip to content
Seven mail settings along the path between two servers, with DNSSEC and IPv6 underneath

Why Every Mail Setting Has To Be Right

Every layer of email security is designed to fall back to working rather than fail safely. Strip the STARTTLS line out of a greeting and an opportunistic sender carries on in the clear, which RFC 3207 both describes as an attack and, on a public server, forbids you to defend against. This post captures that man-in-the-middle in a lab, then walks the seven settings that protect a message in transit, showing for each one what it protects, what it does not, and how it fails open when it is wrong. From there to DANE: why a TLSA record is worthless without DNSSEC under it, why 3 1 1 and key reuse matter, and how a stale record is the one failure that closes rather than opens. Then a read-only census of UK banks, the big mail providers and the certificate authorities, counting how many actually do any of it.

22nd October 2026 · 20 min · 4350 words · Damien Dye
An SMTP session as six stages, each with the rule that rejects there

Postfix, DANE And Every Setting Done Properly

Out of the box Postfix sends in the clear when the far end offers no TLS and takes mail from anybody claiming to be anybody. This is the other build: Postfix 3.10.13 with Dovecot for logins, Rspamd for SPF, DKIM and DMARC, and postfix-tlspol for DANE and MTA-STS. No plaintext in or out, port 25 never relays, a login owns its addresses in the envelope and the From header, sender domains that do not exist are refused, and a tampered or spoofed message never gets past DATA. Every reply was captured in a lab over IPv4 and IPv6, along with the six settings that silently do nothing until something else is set too, the certbot default that breaks DANE on renewal, and what all of it costs you in forwarded mail.

22nd October 2026 · 22 min · 4858 words · Damien Dye
A split between the settings Microsoft 365 controls for you and the few you are left to configure, with the rules it cannot meet marked

Microsoft 365: The Fisher-Price IT Edition

Most organisations do not run their own mail server. They pay Microsoft, and Exchange Online does the work. So this post holds Microsoft 365 to exactly the house rules the Postfix and Stalwart builds met, using nothing but Microsoft’s own documentation and read-only DNS against real tenants. Some of the rules it meets well: outbound DANE and MTA-STS are on by default, and inbound DANE with DNSSEC finally went GA. Several it cannot meet at all. There is no documented way to require inbound TLS from every sender, no way to reject an SPF hard fail at the door, nothing that rejects on a failed DKIM signature alone, and no control that drops mail from a sender domain that does not exist. DNSSEC is explicitly not planned for the mail host every tenant actually uses. The IPv6 rollout that was meant to take six months has left the NCSC’s own mailbox unreachable over IPv6 two years on. And the documentation contradicts itself on wildcards, on SPF, and on DNSSEC. This is the platform most of the country’s mail runs on, and this is what it will and will not let you do.

22nd October 2026 · 12 min · 2800 words · Damien Dye
Stalwart as one binary containing the whole mail stack, against the four-package Postfix build

Stalwart: DANE And The Rest, Built In

Stalwart is a whole mail server in one binary: SMTP and IMAP and JMAP, SPF, DKIM, DMARC, ARC, DANE, MTA-STS, TLS reporting, a spam filter and the mail store, with no milter to wire up and no policy daemon to run beside it. This post holds it to the same house rules as the Postfix build and captures every reply in the lab. The rules all go on, but the story is in the detail: the settings live in a database rather than a file, the defaults as shipped accept plaintext and a tampered invoice, each rule lands in a different one of four configuration mechanisms, DKIM-fail rejection needs a Sieve script because the built-in check reports a tampered body as neutral, the unknown-sender check runs only after the whole message is in and fails open on a DNS error, and the MTA-STS sender bounces on the first attempt where the standard says it should defer. Built in is genuinely less to wire together. It is not the same as safe by default.

22nd October 2026 · 12 min · 2654 words · Damien Dye
A chain of trust running from the signed root through a signed TLD and stopping at an unsigned second-level domain, with the counted figures beside each step

DNSSEC: Protecting Your Traffic From Forgery

The hard part of DNSSEC was finished years ago. Counted from the live root zone on 27th September 2026, 1,351 of 1,438 top level domains carry a DS record and every one of the 1,038 gTLDs is signed. Then it stops dead. A census of every gov.uk domain in the official register finds 39 signed out of 2,390 that still resolve, nine of them parish councils, while HMRC, the NHS, GCHQ and the National Cyber Security Centre are not among them. One certificate authority in nine has signed. So has one Linux or BSD distribution in five. windowsupdate.com has no DS at all. This is what a forged answer actually costs, what signing does about it, why the usual excuses do not survive contact with the numbers, and whether forty-two years after Mockapetris the real problem is that almost nobody understands what DNS actually promises.

27th September 2026 · 48 min · 10979 words · Damien Dye
A name lookup passing through systemd-resolved's stub listener into a cache, a validator and an encrypted transport, with the validator and the encryption switched off

Resolved: The Resolver You Are Already Running

systemd-resolved is running on most Linux desktops right now, caching every lookup and validating none of them. This walks the resolution path from nss-resolve to the two stub listeners, measures what the cache is worth on a real machine, turns DNSSEC on and shows the three verdicts it can return, and explains why validation is off by default when upstream ships it on. Then a census of how little of the web is actually signed, DNS over TLS and its strict-versus-opportunistic trap, and the DNS over HTTPS support that has been requested since 2018 and still does not exist. Ends with what Fedora, Ubuntu, Debian and the RHEL family each ship, how to wire it into each one, and why the features your distro disabled are defaults rather than missing code.

27th September 2026 · 38 min · 8443 words · Damien Dye

Who Actually Controls DNS

The root of the internet is a 1.5 MB text file that one American company edits and signs. Who really controls DNS, what the 2016 IANA transition did and did not change, and the documented record of how ICANN has used that control.

25th August 2026 · 50 min · 10443 words · Damien Dye
A delegation chain from the root down to an internal Active Directory zone, with client writes confined to a separate sub-zone

Samba4 and Securing AD Records Using DNSSEC

Every domain-joined machine finds its domain controller by asking DNS for an SRV record, so the _msdcs locators are the most security-critical records you own. This is how to publish and sign them properly from a Samba4 DC: BIND with dlz_bind9 reading the directory, inline signing, a hidden primary that clients never reach, and client dynamic updates kept out of the zone that holds the locators. Then how to force Windows and Linux clients to actually check the signatures, because a signed zone nobody validates behaves exactly like an unsigned one.

24th August 2026 · 65 min · 13681 words · Damien Dye