go.pennock.tech/smtpdane

SMTP DANE testing tool


smtpdane

Continuous Integration Coverage Status

Early Beta Software

There are not yet enough tests to be sure that this fails when it should, against bad certificates or DNS, but in real-world usage it has so far failed when it should.

Go 1.8+ : go get go.pennock.tech/smtpdane (but see helpers below)

This is an SMTP client which can connect to an SMTP server, issue STARTTLS and verify the certificate using DANE (TLSA records signed with DNSSEC). Validity of the certificate is checked, including date validity periods, but not PKIX CA anchoring.

Per RFC7672 we only support DANE-TA(2) and DANE-EE(3); PKIX-TA(0) and PKIX-EE(1) are explicitly unsupported.

This relies upon a validating DNS resolver; we do not yet validate internally. (Most tools should not validate themselves, but perhaps a monitoring tool should?)

Optionally this client can speak TLS-on-connect instead of STARTTLS, for RFC8314 submissions service (historically called smtps or ssmtp); this is port 465 mail service for clients to submit mail.

The tool will connect to each SMTP server specified, in parallel. If there are multiple IP addresses, then each will be connected to, in parallel.

Flags may be used to request looking up MX records or SRV records for a domain.

Installation

Go 1.8 or greater is required. We use the crypto/tls.Config.VerifyPeerCertificate callback introduced in that release.

$ mkdir ~/go
$ go get go.pennock.tech/smtpdane

With those install steps, the binary can be found in ~/go/bin/smtpdane. The go get command will fetch this repo, any dependent repos and perform the build. This assumes that $GOPATH and other Golang-controlling environment variables have not been set; as of Go 1.8, ~/go is the default solitary entry in the $GOPATH list.

Use ~/go/src/go.pennock.tech/smtpdane/.compile to build with embedded version information from various repositories.

To build as a static binary for deployment into a lib-less environment:

# simple
~/go/src/go.pennock.tech/smtpdane/.compile static
# manual:
cd /go/src/go.pennock.tech/smtpdane
go build -ldflags "-linkmode external -extldflags -static"

At this time there is no vendoring of dependencies. If this matters in your environment, capture them for your use-cases. If our dependency list grows to include packages with unstable APIs then this decision will be revisited.

Optionally, use ./.compile instead of go build to embed extra repository information into the binary.

Invoking

Invoke with -help to see help output listing known flags and defaults.

Most commonly: smtpdane -mx my-domain.example.org

The host to connect to is provided as a list of one or more hosts after any options.

Use -port to specify a different port to speak on, for each host which doesn’t specify a specific port. Note that -port specifies a default; if looking up SRV records, ports from SRV override the -port option. However, port overrides on the host (see below) override SRV.

Use -tls-on-connect to immediately start TLS instead of negotiating.
Use -mx to indicate that names supplied are domain-names and MX records should be looked up.
Use -submission to do the same but look up service submission SRV records, typically used for port 587 service.
Use -submissions to do the same, looking up for submissions though and forcing on the -tls-on-connect option.

The port can be included with the host in the usual :1234 suffix notation; if the host is an IPv6 address, either do not include a port or use the otherwise-optional square-brackets, thus [2001:db8::25]:1234.

By default, the EHLO command will supply a hostname of smtpdane.invalid; use the -helo flag to override that value.

Use -quiet (or -q) to not emit any messages unless there’s a failure.
Use -terse to shorten the amount of output text.
Use -nagios to use Nagios exit codes (and be -terse & -nocolor).

The -quiet approach is suitable for cron jobs which should only emit when there’s a problem. The -nagios approach is better for less ad-hoc monitoring. We’re open to supporting other output formats for other monitoring systems.

Examples

# Regular lookup of a host; check every address-record:
smtpdane mx1.example.org

# Regular lookup of a domain; check every MX, every address:
smtpdane -mx example.org

# Regular lookup of SMTP Submission for a domain:
smtpdane -submission example.org

# Regular lookup of SMTP Submissions TLS-on-connect for a domain:
smtpdane -submissions example.org

# Connect to port 26 for a server, IPv4-only:
smtpdane -4 -port 26 mx1.example.org

# Check if there is a Submissions (TLS-on-connect, 465) service on
# each IP found for Submission service (587) to confirm that you're
# good to add the newer _submissions._tcp SRV records too:
smtpdane -tls-on-connect -submission example.org:465

# Also try checking another hostname
smtpdane -aka mail.example.net mail.example.org

# See much more information about the certs
smtpdane -show-cert-info -mx example.org

# See expiring certificates much sooner; alas, Golang duration parsing
# maxes out in units of hours, so extend in shell;
# 3 months of 31 days each, 24 hours per day, don't forget 'h' unit
smtpdane -expiration-warning $((3*31*24))h -mx example.org

# Turn missing OCSP stapling information into an error
smtpdane -expect-ocsp -mx example.org

# Be invoked for Nagios monitoring, with terse output, no color codes,
# avoiding stderr, but checking for OCSP (& DANE) on all MX servers
smtpdane -nagios -expect-ocsp -mx example.org

Note that the -aka names are added to the list of “acceptable” names; you’ll see each success/failure if you pay attention to the output, but as long as one name succeeds, the probe of that host:ip will be deemed a success.

The expiration time of all certificates in the validated chain is checked for validity, unless -expiration-warning 0s is passed. This examines the NotAfter time. NotBefore is ignored. Only the validated chains are examined, so multiple-chain presentations require more care to check each thoroughly (suggestions welcome). While a normal TLS client only checks the current time, smtpdane checks two times: it checks for outright expired certificates, treating those as errors, and it checks for “expiring soon” certificates, treating those as warnings. To effectively only check for outright expiry, use -expiration-warning 1ns to shift the warning to be enabled with a 1 nanosecond warning period; this leaves warnings as technically possible, albeit somewhat unlikely.

OCSP status is only reported if either -show-cert-info or -expect-ocsp is passed. The latter will cause missing OCSP information to be treated as an error, and present/good OCSP information to be shown in green. Note that a TryLater response-code is treated as a warning.

A simple invocation for a crontab(5) might be:

17 */3 * * * /home/myname/go/bin/smtpdane -q -expect-ocsp -mx example.org

That will check every 3 hours, at 17 minutes past the hour, and check every IP for every hostname returned by the MX records for the domain, checking certificate validity with default notification periods, and declaring an absence of OCSP information to be an error. No output will be produced as long as everything is fine, but there will be output if there are problems, and cron will send an email.

Access needed

You should be able to write a security sandbox profile to constrain this tool, based upon the information here. If it’s not listed but is needed, then that’s a documentation bug, please report it.

  1. Network connectivity, outbound on port 53, UDP and TCP
  2. Outbound TCP, on port 25 and any other ports required for monitoring SMTP. (587 and 465 are common choices).
  3. Stdio, ability to write to stdout/stderr.
  4. /etc/resolv.conf
  5. Read-only access to $SSL_CERT_FILE and $SSL_CERT_DIR locations, and if neither of those is set then to a set of common locations for those files.
  6. Read-only access to /etc/services; on many OSes also /etc/nsswitch.conf to handle indirection to that, and then if that’s not just the file, then wherever else services are read from. Sometimes other /etc files used for DNS resolution.
  7. Usually some source of system entropy (/dev/urandom) if not available via a system call.
  8. Any other common OS start-up files used even for statically linked files.
  9. No other filesystem access should be required, if statically linked.