The Domain Name System is a very central part of how the Internet we use today works. Before the introduction of DNS, networked computers were referenced solely by IP address. After a while, this became confusing to remember and use on a daily name, thus DNS was born.
As the popularity of the Internet grew, and more networked computers came online, there was an increasing need to be able to reference remote machines in a less-confusing way than solely by IP address. With a small enough network, referencing machines by IP address alone can work absolutely fine. The addition of descriptive names, however, makes referencing machines much easier.
The first example of DNS was a HOSTS.TXT file created by staff running ARPANET. ARPANET staff would amend this global HOSTS.TXT file on a regular basis, and it was distributed to anyone on the Internet who wanted to use it to reference machines by name rather than by number. Eventually, as the Internet grew, it was realized that a more automated system for mapping descriptive names to IP addresses was needed. To that end, HOSTS.TXT can be seen as the immediate forerunner to the Domain Name System we use today.
In 1983, Paul Mockapetris authored RFC 882, which describes how a system mapping memorable names to unmemorable IP addresses could work. A team of students at UC Berkeley created the first implementation of Mockapetris’ ideas in 1984, naming their creation Berkeley Internet Name Domain (BIND) server. Today, BIND is still the most widely-used nameserver software on the Internet with over 70% of domains using it, according to the ISC and Don Moore’s survey. Since then, a number of RFC documents have been published which have continued to improve how DNS works and runs.
A domain name is likely the way you interface with DNS most often when browsing the Internet. Examples are, quite literally, everywhere - a very limited example set includes google.com and wikipedia.org.
A top-level domain is an important, but rather generic, part of a domain name. Examples include com, net, gov and org - they were originally defined in RFC 920. [ICANN](https://www.icann.org) controls the TLDs, and delegate responsibility for the registration and maintenance of specific domains to registrars.
A fully-qualified domain name is equivalent to the absolute name. An absolute is one in which the full location of an item is specified. For instance, in HTML, we might use <a href=http://www.google.com/>Google</a> to link to Google. But we might use <a href=page.html>Page</a> to link to a specific page relatively. The difference here is that relative names are relative to the current location. Domain names can also be relative to one another and therefore have a tendency to become ambiguous at times. Specifying an FQDN relative to the root ensures that you have specified exactly which domain you are interested in. Examples of FQDNs include www.google.com. and www.gov.uk..
An IP address is used to uniquely address a machine on a network in numerical (IPv4) or alphanumerical (IPv6) form. It is important that we understand the concept of “network” used here to be relative to what we are trying to achieve. For instance, trying to contact another computer inside your home or office network means that the IP address of the machine you are trying to reach must be unique within your home or office. In terms of websites and publicly-accessible information available via the Internet, the “network” is - in fact - the Internet.
There are two types of IP addresses: one is becoming increasingly popular as we get close to running out of available IPv4 addresses.
An IPv4 address referenced everything in four sets of three period-separated digits. For instance, 184.108.40.206 and 220.127.116.11 are examples of IPv4 addresses. As more devices and people across the world come online, the demand for IPv4 addresses hit a peak, and ICANN are now very close to running out of available addresses. This is where IPv6 comes in.
IPv6 follows similar principles to IPv4 - it allows for machines to be uniquely referenced on the network on which they reside, but the addressing syntax incorporates alphanumeric characters to increase the number of available addresses by a significant base. They are written as 2001:0db8:85a3:0042:1000:8a2e:0370:7334, although short-hand notations do exist (for instance, ::1 to refer to the local machine at any time).
A zonefile is simply a text file made of a variety of different records for an individual domain. Each line of a zonefile contains the name of a particular domain, and then the value and type associated with it. For instance, in google.com‘s zonefile, there may exist a line which denotes www translates, via an A record, to 18.104.22.168.
A DNS record is a single mapping between a domain and relevant data - for instance, an IP address in the case of an A record, or a mail server’s domain name, in the case of an MX record. Many records make up a zonefile.
At the very top of the DNS tree are root servers. These are controlled by the Internet Corporation for Assigned Names and Numbers. As of writing, there are thirteen unique root servers - however, an interesting caveat applies in that each of these root servers is actually a pool of root servers, acting in a load-balanced fashion in order to deal with the huge number of requests they get from the billions of Internet users daily. Their purpose is to handle requests for information about Top-Level Domains, such as .com and .net, where lower level nameservers cannot handle the request sufficiently. The root servers don’t hold any records of real use, insofar as they cannot respond with an answer to a query on their own. Instead, they respond to the request with details of which nameserver is best advised to proceed further.
For example, let’s assume that a request for www.google.com came straight in to a root server. The root server would look at its records for www.google.com, but won’t be able to find it. The best it will be able to produce is a partial match for .com. It sends this information back in a response to the original request.
Once the request for www.google.com has been replied to, the requesting machine will instead ask the nameserver it received in reply to the original request to the root server where www.google.com is. At this stage, it knows that this server handles .com, so at least it is able to get some way further in mapping the address to an IP address. The TLD server will try to find www.google.com in its records, but it will only be able to reply with details about google.com.
By this stage, the original request for www.google.com has been responded to twice: once by the root server to tell it that it doesn’t handle any records, but knows where .com is handled, and once by the TLD server which says that it handles .com, and knows where google is. We’ve still got one more stage to get to, though - that’s the www stage. For this, the request is played against the server responsible for google.com, which duly looks up www.google.com in its records and responds with an IP address (or more, depending on the configuration).
We’ve finally got to the end of a full request! In reality, DNS queries take place in seconds, and there are measures in place which we’ll come on to in these DNS chapters about how DNS can be made faster.
Whilst at it’s most basic, DNS is responsible for mapping easily-remembered domain names to IP addresses, it is also used as a form of key/value database for the Internet. DNS can hold details on which mail servers are responsible for a domain’s mail and arbitrary human-readable text which is best placed in DNS for whatever reason.
The most common types you’ll see are:
|A||Responsible for mapping individual hosts to an IP address. For instance, www in the google. com syntax.|
|AAAA||The IPv6 equivalent of an A record (see above)|
|CNAME||Canonical name. Used to alias one record to another. For example, foo.example.com could be aliased to bar. example.com.|
|MX||Specifies mail servers responsible for handling mail for the domain. A priority is also assigned to denote an order of responsibility.|
|PTR||Resolves an IP address to an FQDN. In practice, this is the reverse of an A record.|
|SOA||Specifies authoritative details about a zonefile, including the zonemaster’s email address, the serial number (for revision purposes) and primary nameserver.|
|SRV||A semi-generic record used to specify a location. Used by newer services instead of creating protocol-specific records such as MX.|
|TXT||Arbitrary human-readable information that needs to be stored in DNS. Examples include verification codes and SPF records.|
There’s a good in-depth list of every record type, the description of its use and the related RFC in which it is defined in this Wikipedia article.
$TTL 86400; // specified in seconds, but could be 24h or 1d $ORIGIN example.com @ 1D IN SOA ns1.example.com. hostmaster.example.com. ( 123456 ; // serial 3H ; // refresh 15 ; // retry 1w ; // example 3h ; // minimum ) IN NS ns1.example.com IN NS ns2.example.com // Good practice to specify multiple nameservers for fault-tolerance IN NS ns1.foo.com // Using external nameservers for fault-tolerance is even better IN NS ns1.bar.com // And multiple external nameservers is better still! IN MX 10 mail.example.com // Here, 10 is the highest priority mail server, so is the first to be used IN MX 20 mail.foo.com // If the highest priority mail server is unavailable, fall back to this one ns1 IN A 22.214.171.124 ns1 IN AAAA 1234:5678:a1234::12 // A and AAAA records can co-exist happily. Useful for supporting early IPv6 adopters. ns2 IN A 126.96.36.199 ns2 IN A 1234:5678:a1234::89 mail IN A 188.8.131.52 www IN A 184.108.40.206 sip IN CNAME www.example.com. ftp IN CNAME www.example.com. mail IN TXT "v=spf1 a -all" _sip._tcp.example.com. IN SRV 0 5 5060 sip.example.com.
If you are administering systems, specifically Unix systems, you should be aware of two pieces of host-side configuration which allow your machines to interface with DNS:
The /etc/hosts file has the purpose of acting as a local alternative to DNS. You might use this when you want to override the record in place in DNS on a particular machine only, without impacting that record and its use for others - therefore, DNS can be overridden using /etc/hosts. Alternatively, it can be used as a back-up to DNS: if you specify the hosts that are mission-critical in your infrastructure inside /etc/hosts, then they can still be addressed by name even if the nameserver(s) holding your zonefile are down.
However, /etc/hosts is not a replacement for DNS - in fact, it is far from it: DNS has a much richer set of records that it can hold, whereas /etc/hosts can only hold the equivalent of A records. An /etc/hosts file might, therefore, look like:
127.0.0.1 localhost 255.255.255.255 broadcasthost ::1 localhost fe80::1%lo0 localhost 192.168.2.2 sql01 192.168.2.3 sql02 192.168.1.10 puppetmaster puppet pm01
The first four lines of /etc/hosts are created automatically on a Unix machine and are used at boot: they shouldn’t be changed unless you really know what you’re doing! In fact, the last two lines of this section are the IPv6 equivalents of the first line. After these first four lines, though, we can specify a name and map it an IP address. In the above example, we’ve mapped sql01 to 192.168.2.2, which means that on a host with the above /etc/hosts configuration, we could refer to sql01 alone and get to the machine responding as 192.168.2.2. You’ll see a similar example for sql02, too. However, there is a slightly odd example for the box named puppetmaster in that multiple friendly names exist for the one box living at 10.0.0.2. When referenced in this way - with multiple space-separated names against each IP address - the box at 10.0.0.2 can be reached at any of the specified names. In effect, puppetmaster, puppet, and pm01 are all valid ways to address 10.0.0.2.
/etc/resolv.conf exists on Unix machines to allow system administrators to set the nameservers which the machine should use. A DNS domain can also be referenced in this file, too. An example /etc/resolv.conf might look like:
domain opsschool nameserver 192.168.1.1 nameserver 192.168.1.2 nameserver 192.168.1.3
In this example, we would be specifying that any of 192.168.1.1, 192.168.1.2 and 192.168.1.3 can be used by the host with the above configuration to query DNS. We are actually telling the host that it is allowed to use any of the nameservers in this file when it resolves (i.e.: makes a request for an entry and waits for a response) a host in DNS.
Setting the domain directive - as in the above example, where we specified it as opsschool - allows users to specify hosts by address relative the domain. For instance, a user could reference sql01, and a query would be sent to nameservers specified asking for records for both sql01 and sql01.home. In most cases, the responses should match - just be careful if they don’t, as you’ll end up with some very confused machines when DNS has split-brained like this!
By itself, DNS doesn’t scale very well. Imagine having a machine that needed to make many millions of DNS queries per day in order to perform its function - it would need to perform well and be constantly available. In order to cut the cost of hardware somewhat, to reduce pressure on networks, and to speed up receiving responses to common queries, many client machines will cache DNS records. The SOA record at the start of each zonefile on the nameservers specifies an expiry value, which tells clients for how long they can keep the zonefile in its current state before they must re-request it. This rather crude but effective updating method works well in the case of DNS.
Generally speaking, caching of DNS records (at least on Unix-based machines) is managed by individual applications. In a Windows environment, however, it is more centralised. To that end, whilst you cannot easily view the cache as it exists on an individual machine all in one place in Unix, you can using Windows - the ipconfig /displaydns command will print the cache as it stands. In Windows, you’ll be presented with the record name as a number - this is a binary representation of the record type itself. Conversion charts can be found online, for example at Wikipedia.
Caching links directly to a phenomenon called propagation. Propagation is the process by which records that have previously existed and have been updated begin to get updated in other machines’ caches. If the SOA record for a zonefile tells hosts to check back with the DNS server every 24 hours, then it should take - at most - 24 hours for machines to update their caches with the new record.
TTLs, or ‘time to live’ values, are a useful feature in DNS which allows you to force the expiry of individual records, thus bypassing the expiry time referenced in the SOA record on a per-record basis. For instance, let’s say that opsschool.org has moved to a new web host but it needs to ensure that the service is available as much as possible. By reducing the TTL for the www and * records in the opsschool.org zonefile, the switch between previous and new vendor should be relatively pain-free. TTLs and caching (see above) work well together - with a suitably high TTL and suitable caching in place, the time for a request to be responded to and the time for updated records to exist on caches are both dramatically reduced.
By this point, we’ve covered many of the basic concepts of DNS - we’ve looked at what exactly DNS is, how the DNS tree works (in the forms of nameserver hierarchies and record types), and we’ve looked at host-side configuration using /etc/resolv.conf and /etc/hosts. There is, however, one further concept we need to cover: forward and reverse DNS.
Forward DNS is, in essence, simply DNS as described above. When ftp.example.com is requested, the root nameserver will reply with details of the nameserver responsible for .com, which will reply with the address of the nameserer responsible for example.com, which will then look in the example.com zonefile for the ftp record and reply appropriately. In fact, the terms ‘forward DNS’ and ‘DNS’ are pretty interchangeable: when talking about DNS, if you don’t otherwise specify, most ops engineers will assume you’re talking about forward DNS as it’s the most often used direction.
However, whilst forward DNS is the type you’re likely to run in to most often, it’s also very important to know how reverse DNS works. If forward DNS maps hostnames to IP addresses, then reverse DNS does exactly the opposite: it maps IP addresses to hostnames. To do this, the zonefile in question must have a PTR record set for the record you’re interested in. Getting used to PTR records and reverse DNS can be tricky, so it might take a few attempts until it catches on.
Domain names follow a specific syntax - foo.tld, where .tld is set by ICANN and chosen by the registrant when they register their domain. For instance, people can choose to register .aero, .com and .tv domains wherever they live in the world, subject to a fee. With reverse DNS, a similar syntax exists. Let’s assume that we want to know which hostname responds at 220.127.116.11. We do this as follows:
- Reverse the octets of the IP address - 18.104.22.168 becomes 22.214.171.124, for instance
- Add in-addr.arpa to the end of the reversed address - we now have 126.96.36.199.in-addr.arpa
- The root nameserver tells queries to find the arpa nameserver
- The arpa nameserver directs the query to in-addr.arpa‘s nameserver
- The in-addr.arpa nameserver then responds with details of 22.in-addr.arpa, and so on...
- In the zonefile, the IP address matching the query is then found and the relevant hostname is returned
There are a number of very useful tools for querying DNS. A list of the most common and some example commands can be found below - for further instructions, see each tool’s man page (found, in Unix, by typing man $toolname at a prompt, and in Windows by appending -h to the command.
ipconfig is a useful tool for diagnosing and configuring TCP/IP networks. Among its many switches, it allows the use of /displaydns which will dump the output of the DNS cache to the console for you. You can then use the /flushdns entry to clear your DNS cache on a Windows machine.
nslookup, however, might be more useful in your day-to-day use of DNS. It allows you to look up an entry on any nameserver that you know the public IP address or hostname for. In many respects, therefore, it acts much like dig on Unix systems does.
dig can be used to query a nameserver to see what values it holds for a specific record. For instance, you could run dig opsschool.org will produce the entire query and entire response, which whilst useful, is often not the information you are looking for. Running the same command, but specifying the +short switch just provides you with relevant detail - in the case of looking up an IP address for a hostname by way of A record, then the output from dig will just be the relevant IP address. Dig can also be used to query external nameservers, such as 188.8.131.52, to see what values they hold.
For instance, dig can be invoked as follows:
In place of dig, you may also see host used in its place. Essentially, both tools perform approximately the same action - given a DNS server (or not, as not doing queries the one you have specified in /etc/resolv.conf), host also allows you to query a record and its value.
For further details on the usage of each tool, have a look at the relevant manual pages - type man dig and man host to find the man pages on any Unix system. You might choose to stick with one tool, or get used to both.