Our second article looks into the new built-in DNS resolver feature of ORF 5.4.
Let’s begin with an excerpt from the ORF Deployment Guide about the requirements for DNS servers used with ORF:
- The server must support recursion
- The server should be on the local network or on the ORF computer. Using ISP DNS servers and third-party DNS resolution services (such as OpenDNS or Google Public DNS) is discouraged.
- The server should not use forwarders (e.g. ISP DNS servers)
- The server should not be the same DNS server that supports your Active Directory
As odd the last three requirements might sound, we have good reasons for them.
- ISP DNS servers and third-party DNS resolution services and accumulate traffic from lots of clients, so they can get occassionally banned by DNSBLs and SURBLs for violating their fair use policy. This results in diminished spam filtering performance, or worse, false positives. Note that using these services as forwarders poses the very same problem, but now with one more server involved.
- When your AD domain is the same as the public name (e.g. both are example.com instead of example.local and example.com), your AD-supporting DNS server will act authoritative for example.com, resulting in all sorts of funny split-horizon DNS problems, like when your domain actually has an MX record, but your AD DNS doesn’t think so.
Issues arising from not meeting the above requirements are more common than you’d think. And it’s for a simple reason: the DNS server on your network is likely configured almost the opposite way than ORF needs it to be. It makes sense, because you may only have the AD DNS server. It may use the ISP DNS server as forwarder, so you can enjoy resilient, low-latency DNS access with a decent sized DNS cache to speed up lookups. It may use a third-party service and you get all that + security by blocking access to botnets, malware sites, phishing sites + more resilience + enormous cache + super-low latency anycast DNS access, so thank you, ORF does not get a dedicated server for its weird and unusual requirements.
This is where 5.4’s new built-in DNS resolver enters to save the day. To understand how it helps, we need to take a brief glimpse at how DNS data is resolved.
DNS name resolution is an iterative process. When looking up the DNS “A” record of example.com, a DNS resolver first contacts one of the well-known root DNS servers which provide a starting point for all DNS lookups. The root DNS server will not have the DNS data for example.com, but it knows which DNS servers service the .com zone and responds with a referral to these servers. The DNS resolver then contacts one of these referred servers, which in turn provides a referral to example.com DNS servers. This cycle of referrals continues until the DNS resolver reaches a name server in the hierarchy which has the answer. This latest server in the chain is called an authoritative name server.
Now if you take a look at that process, it looks quite complicated. Also, you might have noticed that your DNS server at 192.168.251.6 is curiously missing from the picture. You may even have started developing a vague respect for programmers, going through all that pain for turning example.com into 126.96.36.199. Nah, worry not, we’re a lazy bunch and that’s not how we do it. Almost always, we use something called a “stub resolver”, which asks your DNS server at 192.168.251.6 to do all the above and pretty please send the result back to us. This is called recursion, which is effectively delegating the task of obtaining DNS data to an external DNS server.
So as demonstrated above, the custom is that programs rely on a recursive DNS server for name resolution, but there’s nothing really that stops them from cutting the middleman. The new resolver in ORF does exactly this and by doing its own stunts, the whole issue around the DNS server configuration is eliminated and the day is saved.
From ORF 5.4, the new resolver will be the default way to perform lookups in ORF, but you will be free to continue using external DNS servers (in fact, when you are upgrading, you will need to manually switch resolvers, because by policy ORF does not change your earlier settings on upgrade). If you choose the built-in resolver, you are in good hands: we know that properly implementing a recursive resolver is no small feat, so ORF 5.4 relies on libunbound, a part of NLnet Lab’s excellent Unbound resolver for this purpose. The library is very reliable and offers extensive caching.
We recommend using the built-in resolver in all cases when you don’t have a good reason to use external servers. Some of those reasons might be:
- You have multiple servers handling a high volume of emails and you want to take advantage of the shared DNS data caching provided by a central DNS server,
- You can allow DNS traffic only to specific DNS servers (why would you do that?).
If you need those external servers, just please be absolutely, positively, double definitely, cross-my-heart sure that all servers meet the requirements. Otherwise, stick to the new resolver.
Ready to give it a try? The first public preview of ORF 5.4 is coming next week, so stay tuned.