dnscontrol/documentation/opinions.md
Jeffrey Cafferata 2b70f24068
DOCS: [REVCOMPAT] Fixed broken URL's (#2970)
Co-authored-by: Tom Limoncelli <tlimoncelli@stackoverflow.com>
2024-05-21 15:00:55 -04:00

8.5 KiB

DNSControl is an opinionated system

DNSControl is an opinionated system. That means that we have certain opinions about how things should work.

This page documents those opinions.

Opinion #1: DNS should be treated like code

Code is written in a high-level language, version controlled, commented, tested, and reviewed by a third party... and all of that happens before it goes into production.

DNS information should be stored in a version control system, like Git or Mercurial, and receive all the benefits of using VCS. Changes should be in the form of PRs that are approved by someone-other-than-you.

DNS information should be tested for syntax, pass unit tests and policy tests, all in an automated CI system that assures all changes are made the same way. (We don't provide a CI system, but DNSControl makes it easy to use one; and not use one when an emergency update is needed.)

Pushing the changes into production should be effortless, not requiring people to know which domains are on which providers, or that certain providers do things differently that others. The credentials for updates should be controlled such that anyone can write a PR, but not everyone has access to the credentials.

Opinion #2: Non-experts should be able to safely make DNS changes

The goal of DNSControl is to create a system that is set up by DNS experts like you, but updates and changes can be made by your coworkers who aren't DNS experts.

Things your coworkers should not have to know:

  • Your coworkers should not have to know obscure DNS technical knowledge. That's your job.

  • Your coworkers should not have to know what happens in ambiguous situations. That's your job.

  • Your coworkers should be able to submit PRs to dnsconfig.js for you to approve; preferably via a CI system that does rudimentary checks before you even have to see the PR.

  • Your coworkers should be able to figure out the language without much training. The system should block them from doing dangerous things (even if they are technically legal).

Opinion #3: dnsconfig.js are not zonefiles

A zonefile can list any kind of DNS record. It has no judgement and no morals. It will let you do bad practices as long as the bits are RFC-compliant.

dnsconfig.js is a high-level description of your DNS zone data. Being high-level permits the code to understand intent, and stop bad behavior.

Opinion #4: All DNS is lowercase for languages that have such a concept

DNSControl downcases all DNS names (domains, labels, and targets). #sorrynotsorry

When the system reads dnsconfig.js or receives data from DNS providers, the DNS names are downcased.

This reduces code complexity, reduces the number of edge-cases that must be tested, and makes the system safer to operate.

Yes, we know that DNS is case insensitive. See Opinion #3.

Opinion #5: Users should state what they want, and DNSControl should do the rest

When possible, dnsconfig.js lists a high-level description of what is desired and the compiler does the hard work for you.

Some examples:

  • Macros and iterators permit you to state something once, correctly, and repeat it many places.
  • TXT strings are expressed as JavaScript strings, with no weird DNS-required special escape characters. DNSControl does the escaping for you.
  • Domain names with Unicode are listed as real Unicode. Punycode translation is done for you.
  • IP addresses are expressed as IP addresses; and reversing them to in-addr.arpa addresses is done for you.
  • SPF records are stated in the most verbose way; DNSControl optimizes it for you in a safe, opt-in way.

Opinion #6: If it is ambiguous in DNS, it is forbidden in DNSControl

When there is ambiguity an expert knows what the system will do. Your coworkers should not be expected to be experts. (See Opinion #2).

We would rather DNSControl error out than require users to be DNS experts.

For example:

We know that "bar.com." is a FQDN because it ends with a dot.

Is "bar.com" a FQDN? Well, obviously it is, because it already ends with ".com" and we all know that "bar.com.bar.com" is probably not what the user intended.

We know that "bar" is not an FQDN because it doesn't contain any dots.

Is "meta.xyz" a FQDN?

That's ambiguous. If the user knows that "xyz" is a top level domain (TLD) then it is obvious that it is a FQDN. However, can anyone really memorize all the TLDSs? There used to be just gov/edu/com/mil/org/net and everyone could memorize them easily. As of 2000, there are many, many, more. You can't memorize them all. In fact, even before 2000 you couldn't memorize them all. (In fact, you didn't even realize that we left out "int"!)

"xyz" became a TLD in June 2014. Thus, after 2014 a system like DNSControl would have to act differently. We don't want to be surprised by changes like that.

Therefore, we require all CNAME, MX, and NS targets to be FQDNs (they must end with a "."), or to be a shortname (no dots at all). Everything else is ambiguous and therefore an error.

Opinion #7: Hostnames don't have underscores

DNSControl prints warnings if a hostname includes an underscore (_) because underscores are not permitted in hostnames.

We want to prevent a naive user from including an underscore when they meant to use a hyphen (-).

Hostnames are more restrictive than general DNS labels.

"While a hostname may not contain other characters, such as the underscore character (_), other DNS names may contain the underscore. Systems such as DomainKeys and service records use the underscore as a means to assure that their special character is not confused with hostnames. For example, _http._sctp.www.example.com specifies a service pointer for an SCTP capable webserver host (www) in the domain example.com." — the Wikipedia entry on hostnames

However that leads to an interesting problem. When is a DNS label a hostname and when it it just a DNS label? There is no way to know for sure because code can't guess intention.

Therefore we print a warning if a label has an underscore in it, unless the rtype is SRV, TLSA, TXT, or if the name starts with certain prefixes such as _dmarc. We're always willing to add more exceptions.

Opinion #8: TXT Records are one long string

  • TXT records are a single string with a length of 0 to 65,280 bytes (the maximum possible TXT record size).

It is the provider's responsibility to split, join, quote, parse, encode, or decoded the string as needed by the provider's API. This should be invisible to the user.

The user may represent the string any way that JavaScript permits strings to be represented (usually double-quotes). For backwards compatibility they may also provide a list of strings which will be concatenated.

You may be wondering: Isn't a TXT record really a series of 255-octet segments? Yes, TXT record's wire-format is a series of strings, each no longer than 255-octets. However that kind of detail should be hidden from users. The user should represent the string they want and DNSControl should magically do the right thing behind the scenes. The same with quoting and escaping required by APIs.

You may be wondering: Are there any higher-level applications which ascribe semantic value to the TXT string boundaries? I believe that the answer is "no". My proof is not based on reading RFCs, but instead based on (a) observing that I've never seen a DNS provider's control panel let you specify the boundaries, (b) I've never seen a FAQ or reddit post asking how to specify those boundaries. Therefore, there is no need for this. I also assert that there will be no such need in the future.

Opinion #9: RFC 4183 is better than RFC 2317

There is no standard for how to do reverse lookup zones (in-addr.arpa) for CIDR blocks that are not /8, /16, or /24. There are only recommendations.

RFC 2317 is a good recommendation, but it only covers /25 to /32. It also uses / in zone names, which many DNS providers do not support.

RFC 4183 covers /8 through /32 and uses hyphens, which are supported universally.

Originally DNSControl implemented RFC 2317.

In v5.0 we will adopt RFC 4183 as the default. A new function, REVCOMPAT(), will be provided to enable backwards compatibility. v4.x users can use the function to adopt the new behavior early.

See REVCOMPAT() for details.