Skip to content

MarcelWaldvogel/dnstemple

Repository files navigation

DNStemple logo

DNS Temple • A DNS TEMPLating Engine

Maintaining several domains ("zones") can be a hassle. Often, a lot of basic information is shared between domains (name, mail or web servers, anti-spam configuration etc.). CNAMEs can help to keep information in a single location, but are not compatible e.g. with information stored at the apex (the domain name itelf), and thus are of limited or no use for the above-mentioned shared information. When creating a new zone, a previous zone can be copied; however, if later things have to change for these domains, a lot of manual labor is required, prone to errors. DNS Temple tries to simplify this.

Motivating example

Traditional setup

Main site:

$ORIGIN		example.com.
$TTL		86400
@		SOA	(ns1.example.com. hostmaster.example.com. 2020121901 3600 1800 2419200 300)
		NS	ns1.example.com.
		NS	ns2.example.com.
		MX	10 mail.example.com.
		TXT	"v=spf1 mx ~all"
		TXT	"google-site-verification=zeePiegahyeiVoh4SheiK5ootees2Uy4DaiKawoh"
		A	192.0.2.80
		AAAA	2001:db8:1234:5678::80
		CAA	0 issue "letsencrypt.org"
ns1		A	203.0.113.53
		AAAA	2001:db8:1:2::53
ns2		A	198.51.100.53
		AAAA	2001:db8:3:4::53
# Web
www		A	192.0.2.80
		AAAA	2001:db8:1234:5678::80
# Mail
autodiscover	CNAME	mail
autoconfig	CNAME	mail
mail		A	192.0.2.25
		AAAA	2001:db8:1234:5678::25
smtp		CNAME	mail
imap		CNAME	mail
dkim._domainkey	TXT	"v=DKIM1;k=rsa;t=s;s=email;p=MII..."
_dmarc		TXT	"v=DMARC1; p=none; rua=mailto:dkim@example.com; fo=1:d:s"
# Cloud
cloud		A	192.0.2.90
		AAAA	2001:db8:1234:5678::90

One of many secondary sites:

$ORIGIN		example.ch.
$TTL		86400
@		SOA	(ns1.example.com. hostmaster.example.com. 2020121901 3600 1800 2419200 300)
		NS	ns1.example.com.
		NS	ns2.example.com.
		MX	10 mail.example.com.
		TXT	"v=spf1 mx ~all"
		TXT	"google-site-verification=isei8oox1gahc7oox1ezith9eith2ki8aigh9aiD"
		A	192.0.2.80
		AAAA	2001:db8:1234:5678::80
		CAA	0 issue "letsencrypt.org"
# Web
www		A	192.0.2.80
		AAAA	2001:db8:1234:5678::80
# Mail
autodiscover	CNAME	mail
autoconfig	CNAME	mail
mail		A	192.0.2.25
		AAAA	2001:db8:1234:5678::25
smtp		CNAME	mail
imap		CNAME	mail
dkim._domainkey	TXT	"v=DKIM1;k=rsa;t=s;s=email;p=MII..."
_dmarc		TXT	"v=DMARC1; p=none; rua=mailto:dkim@example.com; fo=1:d:s"

The mail, _dkim, and www entries could be changed to CNAMEs and the SPF entry could use include:, but many problems would still remain, such as:

  1. Changing the address of the web server would require changing all the files anyway.
  2. Changing the list of name servers, certificate authorities, or mail servers cause modification of every file.
  3. SOA serials need to be incremented carefully to avoid hard-to-diagnose problems.
  4. Adding an additional service, such as a secondary mail server, CardDAV, CalDAV, or XMPP would require touching every file.
  5. Over time, files will diverge, due to some services (such as cloud) appearing in some zones only or authentication entries (such as google-site-verification, maybe _domainkey) differing between domains.
  6. You will lose oversight because of all the clutter and differences.

Configuration with DNS Temple

Create a central configuration where shared information is collected:

addresses:
  ns1:     203.0.113.53  2001:db8:1:2::53
  ns2:     198.51.100.53 2001:db8:3:4::53
  mail:    192.0.2.25    2001:db8:1234:5678::25
  web:     192.0.2.80    2001:db8:1234:5678::80
  cloud:   192.0.2.90    2001:db8:1234:5678::90
variables:
  ca:      letsencrypt.org
  dkimkey: |
  "MII..."

Create a set of templates using {variable} references, $ADDRESS explosions (expanding dynamically to a set of A and AAAA records), $INCLUDEs, and $DEFAULT values for some variables using magic {_domain} and {_serial} variables:

  • header.t:
$ORIGIN		{_domain}
$TTL		86400
@		SOA	(ns1.example.com. hostmaster.example.com. {_serial} 3600 1800 2419200 300)
		NS	ns1.example.com.
		NS	ns2.example.com.
		CAA	0 issue "{ca}"
  • mail.t:
$DEFAULT			domain	{_domain}.
$ADDRESS			mail	mail

{domain}			MX	10 mail.example.com.
				TXT	"v=spf1 mx ~all"
autodiscover.{domain}		CNAME	mail
autoconfig.{domain}		CNAME	mail
smtp.{domain}			CNAME	mail
imap.{domain}			CNAME	mail
dkim._domainkey.{domain}	TXT	"v=DKIM1;k=rsa;t=s;s=email;p={dkimkey}"
_dmarc.{domain}			TXT	"v=DMARC1; p=none; rua=mailto:dkim@example.com; fo=1:d:s"
  • web.t:
$ADDRESS	web	@
$ADDRESS	web	www
  • common.t:
$INCLUDE	header.t
$INCLUDE	mail.t
$INCLUDE	web.t

These configuration templates will be shared among the zones, resulting in much more compact files:

$INCLUDE	common.t
$INCLUDE	mail.t	domain=noreply
$ADDRESS	ns1	ns1
$ADDRESS	ns2	ns2
$ADDRESS	cloud	cloud
@		TXT	"google-site-verification=zeePiegahyeiVoh4SheiK5ootees2Uy4DaiKawoh"
$INCLUDE	common.t
$ADDRESS	ns1	ns1
$ADDRESS	ns2	ns2
$ADDRESS	cloud	cloud
@		TXT	"google-site-verification=isei8oox1gahc7oox1ezith9eith2ki8aigh9aiD"

Syntax

The BIND syntax is augmented to achieve the power of variables, modules, and functions. The changes, however, are minimal, to retain familiarity.

Behind the scenes, scoped variables and include parameters do all the work: Each $INCLUDE creates a new context, initialized with the variables of the parent file and the parameters passed in the $INCLUDE statement.

The new or extended statements are:

  • $DEFAULT: Setting default values for variables,
  • $INCLUDE: Create a child context and include the file, and
  • $ADDRESS: Simplify the handling of addresses, whether IPv4, IPv6 or groups thereof.

Configuration file

The configuration file is structured as follows:

config:
  serial: <`online`|`dateserial`|`unixtime`|any integer>
  skip: <comma-separated list of `empty`, `comment`>
extensions:
  in: <extension on input file names>
  out: <extension for output file names>
addresses:
  <name1>: <addresses1…>
  <name2>: <addresses2…>
  ...
variables:
  <name1>: <value1>
  <name2>: <value2>
  ...

skip suppresses the output of empty and/or comment lines (starting with ;).

The in extension is removed from the input file names, if present, and the out extension is appended. The basename is also used to set the _domain variable, below.

Serial number modes are as follows:

  • online: Queries the name servers and increments; but uses at least the value that dateserial would produce. For this, existing SOA serials are obtained by querying the local resolver and an authoritative name server. Highest priority for authoritative name server is given to the master name (MNAME) extracted from the SOA record returned by the local resolver, with the remaining NS entries as fallbacks. The minimum SOA corresponds to what date -u +%Y%m%d00 would return.
  • dateserial: Uses the YYYYMMDD00 format
  • unixtime (default): Uses the current second since start of the epoch
  • An integer, e.g. 1: Use this integer as the serial number

Variables

Variable names in curly brackets ({}) are expanded directly when reading the line from the input file (the entire Python format() specification is supported). Only after that, lines are parsed; i.e., whitespace or quotes have no special significance to dnstemple and cannot be escaped (at least as of now).

The following special variables are automatically set:

  • _config: The file name the YAML configuration has been read from.
  • _domain: The domain name, as determined by the basename of the top-level file, i.e., the file specified on the command line, after removing the in extension.
  • _serial: A serial number usable for the SOA record.

All variable names starting with _ are reserved.

Addresses

One of the important reasons for using dnstemple is to make addressing more dynamic. This is similar to a variable, but with extra possibilities:

  • Be able to change addresses in multiple zone files, even if CNAMEs are not possible (e.g., at the apex)
  • Reduce inter-zone dependencies: Have the same address in multiple zones, without causing multiple lookups (i.e., follow the CNAME) when resolving or requiring the other zone to be ready.
  • Handle IPv4 and IPv6 addresses using the same syntax (no deciding between A and AAAA)
  • Support multiple addresses for a name (IPv4+IPv6, but also multiple of each) without having to add/drop lines in the zone file

Addresses are defined similar to variables, but with the following distinctions:

  • They are defined in the address block

  • They are used with the $ADDRESS directive, which expands them into any number of A and AAAA records, as appropriate

  • They can be defined recursively, e.g. for fail-over/load-balancing setups:

    addresses:
      server1: 192.0.2.11 2001:db8:1234:5678::11
      server2: 192.0.2.12
      farm: server1 server2

    There is no need to use variable notation. Anything not an IPv4 or IPv6 address will be recursively looked up as an address, up to 5 levels deep.

    Of course, variables could also be included using the standard {} variable notation, but then the reference is to the variables section.

Directives

$INCLUDE <file> [<var>=<value>…]

Include the given file. The set of variables available to the included file is the set of variables available to the parent, plus the variables passed as parameters. This is similar to how parameters would be passed to functions.

Separation is done at all whitespace, after variable substitution, without consideration of quotes or escape characters. So do not try to include whitespace into variable values.

$ADDRESS <address> <prefix>

Create one line for each of the addresses listed for the named address parameter specified in the configuration file. Both IPv4 and IPv6 addresses can be mixed arbitrarily and will be prefixed with A or AAAA, as appropriate. Typically, the prefix will just be a name, but can also contain TTL information; in fact, anything that could precede the A or AAAA.

$DEFAULT <variable> <value>

If the specified variable is undefined or empty, set it to the value given.