Purchase  Copyright © 2002 Paul Sheer. Click here for copying permissions.  Home 

next up previous contents
Next: 28. Network File System, Up: rute Previous: 26. TCP and UDP   Contents

Subsections

27. DNS and Name Resolution

We know that each computer on the Internet has its own IP address. Although this address is sufficient to identify a computer for purposes of transmitting packets, it is not particularly accommodating to people. Also, if a computer were to be relocated, we would like to still identify it by the same name.

Hence, each computer is given a descriptive textual name. The basic textual name of a machine is called the unqualified host name [This is my own terminology.] and is usually less than eight characters and contains only lowercase letters and numbers (and especially no dots). Groups of computers have a domain name. The full name of machine is unqualified_host_name .domain_name and is called the fully qualified host name [Standard terminology.]or the qualified host name. [My terminology.] For example, my computer is cericon. The domain name of my company is cranzgot.co.za, and hence the qualified host name of my computer is cericon.cranzgot.co.za, although the IP address might be 160.123.76.9.

Often the word domain is synonymous with domain name, and the host name on its own can mean either the qualified or unqualified host name.

This system of naming computers is called the Domain Name System (DNS)

27.1 Top-Level Domains (TLDs)

Domains always end in a standard set of things. Here is a complete list of things that the last bit of a domain can be.

.com
A U.S. or international company proper. In fact, any organization might have a .com domain.
.gov
A U.S. government organization.
.edu
A U.S. university.
.mil
A U.S. military department.
.int
An organization established by international treaties.
.org
A U.S. or nonprofit organization. In fact, anyone can have a .org domain.
.net
An Internet service provider (ISP). In fact, any bandwidth reseller, IT company, or any company at all might have a .net domain.

Besides the above, the domain could end in a two-letter country code.

The complete list of country codes is given in Table 27.1. The .us domain is rarely used, since in the United States .com, .edu, .org, .mil, .gov, .int, or .net are mostly used.


Table 27.1: ISO country codes
\begin{table}{\scriptsize\begin{tabularx}{1.0\textwidth}{\vert c @{ } X \vert c ...
...c}}} & St. Vincent and the Grenadines & & \\
\hline
\end{tabularx}}
\end{table}


Within each country, a domain may have things before it for better description. Each country may implement a different structure. Some examples are:

.co.za
A South African company. ( za = Zuid Afrika, from Dutch.)
.org.za
A South African nonprofit organization.
.ac.za
A South African academic university.
.edu.au
An australian tertiary educational institution.
.gov.za
A South African government organization.

Note that a South African company might choose a .com domain instead of a .co.za domain. The Internet has become more commercialized than organized, meaning that anyone can pretty much register any domain that is not already taken.

27.2 Resolving DNS Names to IP Addresses

In practice, a user will type a host name (say, www.cranzgot.co.za) into some application like a web browser. The application has to then try find the IP address associated with that name, in order to send packets to it. This section describes the query structure used on the Internet so that everyone can find out anyone else's IP address.

An obvious lookup infrastructure might involve distributing a long table of host name vs. IP numbers to every machine on the Internet. But as soon as you have more than a few thousand machines, this approach becomes impossible.

Another imaginary infrastructure might have one huge computer on the Internet somewhere whose IP address is known by everyone. This computer would be responsible for servicing requests for IP numbers, and the said application running on your local machine would just query this big machine. Of course, with billions of machines out there, this approach will obviously create far too much network traffic. [Actually, some Microsoft LANs kind of work this way--that is, not very well.]

27.2.1 The Internet DNS infrastructure

The DNS structure on the Internet actually works like this.

There are computers that service requests for IP numbers--millions of them. They are called name servers (or DNS servers), and a request is called a DNS lookup (or just a lookup). However, each name server only has information about a specific part of the Internet, and they constantly query each other.

There are 13 root name servers on the Internet. [This list can be gotten from ftp://ftp.rs.internic.net/domain/named.root.]

 
 
 
 
5 
 
 
 
 
10 
 
 
 
a.root-servers.net  198.41.0.4
b.root-servers.net  128.9.0.107
c.root-servers.net  192.33.4.12
d.root-servers.net  128.8.10.90
e.root-servers.net  192.203.230.10
f.root-servers.net  192.5.5.241
g.root-servers.net  192.112.36.4
h.root-servers.net  128.63.2.53
i.root-servers.net  192.36.148.17
j.root-servers.net  198.41.0.10
k.root-servers.net  193.0.14.129 
l.root-servers.net  198.32.64.12
m.root-servers.net  202.12.27.33

Each country also has a name server, and in turn each organization has a name server. Each name server only has information about machines in its own domain, as well as information about other name servers. The root name servers only have information on the IP addresses of the name servers of .com, .edu, .za, etc. The .za name server only has information on the IP addresses of the name servers of .org.za, .ac.za, .co.za, etc. The .co.za name server only has information on the name servers of all South African companies, like .cranzgot.co.za, .icon.co.za, .mweb.co.za, etc. The .cranzgot.co.za, name server only has info on the machines at Cranzgot Systems, like www.cranzgot.co.za.

Your own machine will defined in its configuration files a name server that is geographically close to it. The responsibilities of this name server will be to directly answer any queries about its own domain that it has information about and to answer any other queries by querying as many other name servers on the Internet as is necessary.

27.2.2 The name resolution process

Now our application is presented with www.cranzgot.co.za. The following sequence of lookups takes place to resolve this name into an IP address. This procedure is called host name resolution and the algorithm that performs this operation is called the resolver.

  1. The application checks certain special databases on the local machine. If it can get an answer directly from them, it proceeds no further.
  2. The application looks up a geographically close name server from the local machine's configuration file. Let's say this machine is called ns.
  3. The application queries ns with `` www.cranzgot.co.za?''.
  4. ns determines whether that IP has been recently looked up. If it has, there is no need to ask further, since the result would be stored in a local cache.
  5. ns checks whether the domain is local. That is, whether it is a computer about which it has direct information. In this case, this would only be true if the ns were cranzgot.co.za's very own name server.
  6. ns strips out the TLD (top level domain) .za. It queries a root name server, asking what name server is responsible for .za. The answer will be ucthpx.uct.ac.za of IP address 137.158.128.1.
  7. ns strips out the next highest domain co.za It queries 137.158.128.1, asking what name server is responsible for .co.za. The answer will be secdns1.posix.co.za of IP address 160.124.112.10.
  8. ns strips out the next highest domain cranzgot.co.za. It queries 160.124.112.10, asking what name server is responsible for cranzgot.co.za. The answer will be pizza.cranzgot.co.za of IP address 196.28.123.1.
  9. ns queries 196.28.123.1 asking for the IP address of www.cranzgot.co.za. The answer will be 160.123.176.1.
  10. ns returns the result to the application.
  11. ns stores each of these results in a local cache with an expiration date, to avoid having to look them up a second time.

27.3 Configuring Your Local Machine

We referred to ``configuration files'' above. These are actually the files: /etc/host.conf, /etc/hosts, and /etc/resolv.conf. These are the three and only files that specify how all applications are going to look up IP numbers; and have nothing to do with the configuration files of the name server daemon itself, even though a name server daemon might be running on the local machine.

When an application needs to look up a host name, it goes through the following procedure. [What is actually happening is that the application is making a C library call to the function gethostbyname(), hence all these configuration files really belong to the C library packages glibc or libc. However, this is a detail you need not be concerned about.] The following are equivalent to steps 1, 2, and 3 above, with the details of the configuration files filled in. The configuration files that follow are taken from an actual installation.

  1. The application checks the file /etc/host.conf. This file will usually have a line order hosts,bind in it, specifying that it should first ( hosts) check the local database file /etc/hosts, and then ( bind) query the name server specified in /etc/resolv.conf. The file /etc/hosts contains a plain text list of IP addresses and names. An example is given below. If the application can get an answer directly from /etc/hosts, it proceeds no further.
  2. The application checks in the file /etc/resolv.conf for a line nameserver <nameserver>. There can actually be three of these lines so that if one name server fails, the application can try the next in turn.
  3. The application sends to the name server a query with the host name. If the host name is unqualified, then the application, before trying the query, appends to the host name a local domain name. A line search <domain1> <domain2> ... <domainN> may appear in the configuration file to facilitate this. A query is made with each of <domain1>, <domain2> etc. appended in turn until the query successfully returns an IP. This just saves you having to type in the full host name for computers within your own organization.
  4. The name server proceeds with the hierarchical queries described from step 4 onward.

The /etc/hosts file should look something like this:

 
 
 
 
127.0.0.1            localhost.localdomain         localhost
192.168.3.9          cericon.cranzgot.co.za        cericon
192.168.3.10         pepper.cranzgot.co.za         pepper
192.168.2.1          onion.cranzgot.co.za          onion

The hosts pepper, cericon, and onion are the hosts that this machine has the most communication with, and hence are listed here. cericon is the local machine and must be listed. You can list any hosts to which you want fast lookups, or hosts that might need to be known in spite of name servers being down.

The /etc/host.conf might look like this. All of the lines are optional:

 
 
 
 
5 
 
order    hosts, bind, nis
trim     some.domain
spoofalert
nospoof
multi    on
reorder

order
The order in which lookups are done. Don't try fiddling with this value. It never seems to have any effect. You should leave it as order hosts,bind (or order hosts,bind,nis if you are using NIS--search for the NIS-HOWTO on the web.) Once again, bind means to then go and check the /etc/resolv.conf which holds the name server query options.
trim
Strip the domain some.domain from the end of a host name before trying a lookup. You will probably never require this feature.
spoofalert
Try reverse lookups on a host name after looking up the IP (i.e., do a query to find the name from the IP). If this query does not return the correct result, it could mean that some machine is trying to make it look like it is someone it really isn't. This is a hacker's trick called spoofing. spoofalert warns you of such attempts in your log file /var/log/messages.
nospoof
Disallow results that fail the spoof test.
multi on
Return more than one result if there are aliases. Actually, a host can have several IP numbers, and an IP number can have several host names. Consider a computer that might want more than one name ( ftp.cranzgot.co.za and www.cranzgot.co.za are the same machine.) Or a machine that has several networking cards and an IP address for each. This option should always be turned on. multi off is the alternative. Most applications use only the first value returned.
reorder
If more than one IP is returned by a lookup, then sort that list according to the IP that has the most convenient network route.

Despite this array of options, an /etc/host.conf file almost always looks simply like

 
 
order    hosts, bind
multi    on

The /etc/resolv.conf file could look something like this:

 
 
 
 
5 
 
nameserver 192.168.2.1
nameserver 160.123.76.1
nameserver 196.41.0.131
search cranzgot.co.za ct.cranzgot.co.za uct.ac.za
sortlist 192.168.3.0/255.255.255.0 192.168.2.0/255.255.255.0
options ndots:1 timeout:30 attempts:2 rotate no-check-names inet6

nameserver
Specifies a name server to query. No more than three may be listed. The point of having more than one is to safeguard against a name server being down; the next in the list will then be queried.
search
If given a host name with less than ndots dots (i.e., 1 in this case), add each of the domains in turn to the host name, trying a lookup with each. This option allows you to type in an unqualified host name and the application work out what organization it is belongs to from the search list. You can have up to six domains, but then queries would be time consuming.
domain
The line `` domain ct.cranzgot.co.za'' is the same as `` search ct.cranzgot.co.za cranzgot.co.za co.za''. Always use search explicitly instead of domain to reduce the number of queries to a minimum.
sortlist
If more than one host is returned, sort them according to the following network /masks.
options
Various additional parameters can be specified in this one line:
ndots
Explained under search above. The default is 1.
timeout
How long to wait before considering a query to have failed. The default is 30 seconds.
attempts
Number of attempts to make before failing. The default is 2. This means that a down name server will cause your application to wait 1 full minute before deciding that it can't resolve the IP.
rotate
Try the name servers in round robin fashion. This distributes load across name servers.
no-check-names
Don't check for invalid characters in host names.
inet6
The man page for resolv.conf ( resolver(5)) says:
      inet6     sets RES_USE_INET6 in _res.options .  This has the ef-
                fect of trying a AAAA query before an A query inside
                the gethostbyname function, and of mapping IPv4 re-
                sponses in IPv6 ``tunnelled form'' if no AAAA records
                are found but an A record set exists.
An AAAA query is a 128-bit ``next generation,'' or ``IPV6'' Internet address.

Despite this array of options, an /etc/resolv.conf file almost always looks simply like:

 
 
nameserver 192.168.2.254
search cranzgot.co.za

27.4 Reverse Lookups

A reverse lookup, mentioned under nospoof, is the determining of the host name from the IP address. The course of queries is similar to forward lookups using part of the IP address to find out what machines are responsible for what ranges of IP address.

A forward lookup is an ordinary lookup of the IP address from the host name.

27.5 Authoritative for a Domain

I have emphasized that name servers only hold information for their own domains. Any other information they may have about another domain is cached, temporary data that has an expiration date attached to it.

The domain that a name server has information about is said to be the domain that a name server is authoritative for. Alternatively we say: ``a name server is authoritative for the domain.'' For instance, the server ns2.cranzgot.co.za is authoritative for the domain cranzgot.co.za. Hence, lookups from anywhere on the Internet having the domain cranzgot.co.za ultimately are the responsibility of ns2.cranzgot.co.za, and originate (albeit through a long series of caches) from the host ns2.cranzgot.co.za.

27.6 The host, ping, and whois Command

The command host looks up a host name or an IP address, by doing a name server query. Try

 
host www.cnn.com

for an example of a host with lots of IP address. Keep typing host over and over. Notice that the order of the hosts keeps changing randomly. This reordering distributes load among the many cnn.com servers.

Now, pick one of the IP addresses and type

 
host <ip-address>

This command will return the host name cnn.com.

Note that the host command is not available on all UNIX systems.

The ping command has nothing directly to do with DNS but is a quick way of getting an IP address and at the same time checking whether a host is responding. It is often used as the acid test for network and DNS connectivity. See Section 25.10.1.

Now enter:

 
whois cnn.com@rs.internic.net

(Note that original BSD whois worked like whois -h <host> <user>.) You will get a response like this:

 
 
 
 
5 
 
 
 
 
10 
 
 
 
 
15 
 
 
 
 
20 
 
 
 
[rs.internic.net]
 
Whois Server Version 1.1
 
Domain names in the .com, .net, and .org domains can now be registered
with many different competing registrars. Go to http://www.internic.net
for detailed information.
 
   Domain Name: CNN.COM
   Registrar: NETWORK SOLUTIONS, INC.
   Whois Server: whois.networksolutions.com
   Referral URL: www.networksolutions.com
   Name Server: NS-01A.ANS.NET
   Name Server: NS-01B.ANS.NET
   Name Server: NS-02A.ANS.NET
   Name Server: NS-02B.ANS.NET
   Updated Date: 22-sep-1999
 
 
>>> Last update of whois database: Thu, 20 Jan 00 01:39:07 EST <<<
 
The Registry database contains ONLY .COM, .NET, .ORG, .EDU domains and
Registrars.

(Internic happens to have this database of .com, .net, .org, and .edu domains.)

27.7 The nslookup Command

nslookup is a program to interactively query a name server. If you run

 
nslookup

you will get a > prompt at which you can type commands. If you type in a host name, nslookup will return its IP address(s), and visa versa. Also, typing

 
help

any time will return a complete list of commands. By default, nslookup uses the first name server listed in /etc/resolv.conf for all its queries. However, the command

 
server <nameserver>

will force nslookup to connect to a name server of your choice.

27.7.1 NS, MX, PTR, A and CNAME records

The word record is a piece of DNS information.

Now enter the command:

 
set type=NS

This tells nslookup to return the second type of information that a DNS can deliver: the authoritative name server for a domain or the NS record of the domain. You can enter any domain here. For instance, if you enter

 
 
set type=NS
cnn.com

nslookup returns

 
 
 
 
5 
 
 
 
 
10 
 
Non-authoritative answer:
cnn.com nameserver = NS-02B.ANS.NET
cnn.com nameserver = NS-02A.ANS.NET
cnn.com nameserver = NS-01B.ANS.NET
cnn.com nameserver = NS-01A.ANS.NET
 
Authoritative answers can be found from:
NS-02B.ANS.NET  internet address = 207.24.245.178
NS-02A.ANS.NET  internet address = 207.24.245.179
NS-01B.ANS.NET  internet address = 199.221.47.8
NS-01A.ANS.NET  internet address = 199.221.47.7

This output tells us that four name servers are authoritative for the domain cnn.com (one plus three backups). It also tells us that it did not get this answer from an authoritative source, but through a cached source. It also tells us what name servers are authoritative for this very information.

Now, switch to a name server that is authoritative for cnn.com:

 
server NS-02B.ANS.NET

and run the same query:

 
cnn.com

The new result is somewhat more emphatic, but no different.

There are only a few other kinds of records that you can get from a name server. Try

 
 
set type=MX
cnn.com

to get the so-called MX record for that domain. The MX record is the server responsible for handling mail destined to that domain. MX records also have a priority (usually 10 or 20). This tells any mail server to try the 20 one should the 10 one fail, and so on. There are usually only one or two MX records. Mail is actually the only Internet service handled by DNS. (For instance, there is no such thing as a NEWSX record for news, or a WX record for web pages, whatever kind of information we may like such records to hold.)

Also try

 
 
 
 
5 
 
set type=PTR
<ip-address>
set type=A
<hostname>
set type=CNAME
<hostname>

So-called PTR records are reverse lookups, or Poin Te Rs to host names. So-called A records are forward lookups (the default type of lookup when you first invoke nslookup and the type of lookup the first half of this chapter was most concerned with), or Address lookups. So-called CNAME records are lookups of Canonical NAMEs. DNS allows you to alias a computer to many different names, even though each has one real name (called the canonical name). CNAME lookups returns the machine name proper.

27.8 The dig Command

dig stands for domain information groper. It sends single requests to a DNS server for testing or scripting purposes (it is similar to nslookup, but non-interactive).

It is usually used like,

 
dig @<server> <domain> <query-type>

where <server> is the machine running the DNS daemon to query, <domain> is the domain of interest and <query-type> is one of A, ANY, MX, NS, SOA, HINFO, or AXFR--of these, you can read about the non-obvious ones in dig(1). dig can also be used to test an Internet connection. See Section 20.7.4.

Useful is the AXFR record. For instance

 
dig @dns.dial-up.net icon.co.za AXFR

lists the entire domain of one of our local ISPs.


next up previous contents
Next: 28. Network File System, Up: rute Previous: 26. TCP and UDP   Contents