DNS Config Types
From: https://www.zytrax.com/books/dns/ch4/
Contents
4. DNS Configuration Types
Most DNS servers are schizophrenic - they may be masters (authoritative) for
some zones, slaves for others and provide resolver or forwarding services for
others. Many observers object to the concept of DNS types partly because of the
schizophrenic behaviour of most DNS servers (they are frequently of more than
one type) and partly to avoid confusion with the named.conf zone parameter
'type' which only allows master, slave, stub, forward, hint. Nevertheless, the
following terms are commonly used to describe the primary functionality of DNS
servers.
Notes:- Running any DNS server that supports recursive queries received from any or
all users (an Open DNS) is an Extremely Bad Idea ™. While an Open DNS may look
like a friendly and neighbourly thing to do such a server may be used in DDoS
attacks and carries a significantly increased risk of cache poisoning. It is
always possible to define the range of IP addresses that are allowed or
permitted to use the recursive feature of any DNS server. Such a server is
termed Closed. The various configurations have been modified to ensure that the
DNS stays Closed to non-permitted users.
- One of the basic rules of security is that only the minimum services
necessary to meet the objectives should be deployed. This means that a secure
DNS server should provide only a single function, for instance, authoritative
only, or caching only, not both capabilities in the same server. This is a
correct but idealistic position, generally possible only in larger organizations.
In practice many of us run mixed mode DNS servers. While much can be done to
mitigate any security implications it must always be accepted that, in mixed
configurations, increased risk is the downside of flexibility.
4.1 Master (Primary) Name Servers
A Master DNS defines one or more zone files for which this DNS is Authoritative
('type master'). The zone has been delegated (via an NS Resource Record) to this
DNS.
The term master was introduced with BIND 8.x and replaced the term 'primary'.
Diagram 1 DNS Master
Master status is defined in BIND by including 'type master' in the zone declaration
section of the named.conf file as shown by the following fragment.
// example.com fragment from named.conf
// defines this server as a zone master
zone "example.com" in
{
type master;
file "pri.example.com";
};
Notes:- The terms Primary and Secondary DNS entries in Windows TCP/IP network
properties mean nothing, they may reflect the 'master' and 'slave' name-server
or they may not - you decide this based on operational need, not BIND configuration.
- It is important to understand that a zone 'master' is simply a server which
gets its zone data from a local source as opposed to a 'slave' which gets its
zone data from an external (networked) source (typically the 'master' but not
always). This apparently trivial point means that you can have any number of
'master' servers for any zone if it makes operational sense. You have to ensure
(by a manual or other process) that the zone files are synchronised but apart
from this there is nothing to prevent it.
- Just to confuse things still further you may run across the term 'Primary
Master' this has a special meaning in the context of dynamic DNS updates and is
defined to be the name server that appears in the SOA RR record.
When a master DNS receives Queries for a zone for which it is authoritative then
it will respond as 'Authoritative' (AA bit is set in a query response).
If a DNS server receives a query for a zone for which it is neither a Master nor
a Slave then it will act as configured (in BIND this behaviour is defined in
the named.conf file):
- If caching behaviour is permitted and recursive queries are allowed the
server will completely answer the request or return an error.
- If caching behaviour is permitted and Iterative (non-recursive) queries are
allowed the server can respond with the complete answer (if it is already in the
cache because of another request), a referral or return an error.
- If caching behaviour is NOT permitted (an 'Authoritative Only' DNS server)
the server will return a referral or an error.
A master DNS server can NOTIFY zone changes to defined (typically slave) servers
- this is the default behaviour. NOTIFY messages ensure zone changes are rapidly
propagated to the slaves (interrupt driven) rather than rely on the slave server
periodically polling for changes. The BIND default is to notify the servers
defined in NS records for the zone - except itself, obviously.
A zone master can be 'hidden' (only one or more of the slaves know of its
existence). There is no requirement in such a configuration for the master
server to appear in an NS RR for the domain. The only requirement is that two
(or more) name servers support the zone. Both servers could be any combination
of master-slave, slave-slave or even master-master.
If you are running Stealth Servers and wish them to be notified you will have
to add an also-notify parameter as shown in the BIND named.conf file fragment
below:
// example.com FRAGMENT FROM named.conf
// DEFINES THIS SERVER AS A ZONE MASTER
// 192.168.0.2 IS A STEALTH SERVER NOT LISTED IN A NS RECORD
zone "example.com" in
{
type master;
also-notify {192.168.0.2;};
file "pri/pri.example.com";
};
You can turn off all NOTIFY operations by specifying 'notify no' in the zone
declaration.
Example configuration files for a master DNS are provided.
4.2 Slave Name Servers
A Slave DNS gets its zone data using a zone transfer operation (typically from a
zone master) and it will respond as authoritative for those zones for which it
is defined to be a 'slave' and for which it has a currently valid zone
configuration. It is impossible to determine from a query result that it came
from a zone master or slave.
DNS Master - Slave
Diagram 2 DNS Slave Server
The term 'slave' was introduced with BIND 8.x and replaced the term
'secondary'.
There can be any number of slave DNS's for any given zone.
Slave status is defined in BIND by including 'type slave' in the zone
declaration section of the named.conf file as shown by the following
fragment.
// example.com FRAGMENT FROM named.conf
// DEFINES THIS SERVER AS A ZONE SLAVE
zone "example.com" in
{
type slave;
file "sec/sec.example.com";
masters {192.168.23.17;};
};
Notes:- The master DNS for each zone is defined in the 'masters' statement of the zone
clause and allows slaves to refresh their zone record when the 'expiry' parameter
of the SOA Record is reached. If a slave cannot reach the master DNS when the
'expiry' time has been reached it will stop responding to requests for the zone.
It will not use time-expired data.
- The file parameter is optional and allows the slave to write the transferred
zone to disc and hence if BIND is restarted before the 'expiry' time the Slave
server will use the saved data. In large DNS systems this can save a
considerable amount of network traffic.
Assuming NOTIFY is allowed in the master DNS for the zone (the default behaviour)
then zone changes are propagated to all the servers defined with NS Records in
the zone file. Other acceptable NOTIFY sources can be defined using the
also-notify parameter in named.conf.
Example configuration files for a slave DNS are provided.
4.2.1 But Slaves can also be Masters
Oh, stop this pain. This section can get a bit confusing. Read it only when
accompanied by your favorite keep-me-awake-cos-I-can't-take-anymore-of-this-stuff
beverage.
The definition of a slave server is simply that it gets its zone data via zone
transfer, whereas a master gets its zone data from a local file system. The
source of the zone transfer could just as easily be another slave as a master.
So what sane human would want to do that?
- Assume you want to hide your master servers in, say, a stealth
configuration then at least one slave server will sit on the public side of a
firewall, or similar configuration, providing perimeter defence. To provide
resilience you would need two or more such public slaves. The second slave can
be updated from the same master as the first or it could be updated from the
slave server - we'll call it the 'boss' slave to avoid getting into tortuous
terminology (is it a master-slave or a slave-master?). To configure this
miracle the second slave server would define the 'boss' slave's IP in its
masters statement. When the 'boss' slave has sucessfully transfered a zone file
(from the master) it will send out NOTIFY messages (the default) unless
configured not to do so. This type of configuration will marginally increase
latency for updating the zone on the second slave - but that may be more than
offset by increased stealth.
- In a DNSSEC environment the master will likely have all kinds of whizzo
dodads concerned with keeping keys secure. Whereas DNSSEC slaves simply send
the data in the zone file in response to queries and have no requirements
for secure key maintenance. Hidden master configurations will become
increasingly the norm in this environment.
4.3 Resolver or Caching Name Servers
A DNS Reolver (frequently known as a caching or even recursive Name Server)
obtains information from another server (a Authoritative Name Server) in
response to a host query and then saves (caches) the data locally. On a
second or subsequent request for the same data the Resolver (Caching Name
Server) will respond with its locally stored data (from the cache) until the
time-to-live (TTL) value of the response expires, at which time the server
will refresh the data from the zone master.
If the resolver (caching name server) obtains its data directly from a zone
master it will respond as 'authoritative', if the data is supplied from its
cache the response is 'non-authoritative'.
The default BIND behaviour is to cache and this is associated with the
recursion parameter (the default is 'recursion yes'). There are many
configuration examples which show caching behaviour being defined using a
type hint statement in a zone declaration. These configurations confuse two
distinct but related functions. If a server is going to provide caching
services then it must support recursive queries and recursive queries need
access to the root servers which is provided via the 'type hint' statement.
DNS Resolver (caching server)
Diagram 3 DNS Resolver (Recursive Server)
A Resolver (caching name server) will typically have a named.conf file which
includes the following fragment:
// options section fragment of named.conf
// recursion yes is the default and may be omitted
options
{
directory "/var/named";
version "not currently available";
recursion yes;
};
// zone section
....
// THE DOT INDICATES THE ROOT domain = all domains
zone "." IN
{
type hint;
file "root.servers";
};
Notes:- BIND defaults to recursive queries which by definition provides caching
behaviour. The named.conf recursion parameter controls this behaviour.
- The zone '.' is shorthand for the root domain which translates to 'any
domain not defined as either a master or slave in this named.conf file'.
- cache data is discarded when BIND is restarted.
The most common DNS server caching configurations are:
- A DNS server acting as master or slave for one or more zones (domains) and
as resolver (caching name server) for all other requests. This could be
termed a general purpose DNS server.
- A Resolver only (caching only) server - typically used to minimise external
access or to compensate for slow external links. This is sometimes called a
Proxy server though we prefer to associate the term with a Forwarding
server
To cache or not is a crucial question in the world of DNS. BIND is regarded
as the reference implementation of the DNS specification. As such it
provides excellent - if complex to configure - functionality. The down side
of generality is suboptimal performance on any single function - in
particular caching involves a non-trivial performance overhead.
For general usage the breadth of BIND functionality typically offsets any
performance concerns. However, if the DNS is being 'hit' thousands of times
per second performance is a major factor. There are now a number of
alternate Open Source DNS servers some of which stress performance. These
servers typically do NOT provide caching services (they are said to be
'Authoritative only' servers).
Example configuration files for a caching DNS are provided.
Note: The response to a query is Authoritative under three conditions:
- The response is received from a Zone master.
- The response is received from a Zone slave with non time-expired zone data.
- The response is received by a caching server directly from either a Zone
master or slave. If the response is supplied from the cache it is not
authoritative.
4.4 Forwarding (a.k.a Proxy) Name Servers
A forwarding (a.k.a. Proxy, Client, Remote) server is one which simply forwards
requests to another DNS and caches the results. On its face this looks like a
pretty pointless exercise. It is, however, a frequently undervalued and
extremely useful configuration in a number of situations:- Where access to the external network is slow or expensive:
- Local DNS caching - results are cached in the forwarding server so that
frequently requested domains will provide fast results from the cache.
- The Remote (forwarded to) DNS server provides recursive query support
resulting in a single query across the network (from the forwarding DNS to
the 'forwared to' DNS) thus reducing traffic congestion (on busy networks),
traffic volume (on expensive networks) and increasing performance (on slow
networks).
- Forwarding servers also can be used to ease the burden of local
administration by providing a single point at which changes to remote name
servers may be managed, rather than having to update all hosts. Thus, all
hosts in a particular network section or area can be configured to point to
a fixed forwarding DNS which can be configured to stream DNS traffic as
desired and changed over time with minimal effort.
- Sanitizing traffic. Especially in larger private networks it may be
sensible to stream DNS traffic for local domain access by forwarding to the
local DNS servers while forwarding external DNS requests to a dirty or
hardened caching DNS (or resolver).
- Forwarding can also be used as part of a Split Server configuration for
perimeter defence.
DNS Forwarding Server
Diagram 4 - DNS Forvarding Server
BIND allows configuration of forwarding using the forward and forwarders
parameters either at a 'global' level (in an options section) or on a per
-zone basis in a zone section of the named.conf file. Both configurations
are shown in the examples below:
Global Forwarding - All Requests
// OPTIONS SECTION FRAGMENT OF named.conf
// FORWARDERS CAN HAVE MULTIPLE CHOICES
options
{
directory "/var/named";
version "not currently available";
forwarders {10.0.0.1; 10.0.0.2;};
forward only;
};
// zone file sections
....
Per Domain Forwarding
// ZONE SECTION FRAGMENT OF named.conf
zone "example.com" IN
{
type forward;
forwarders {10.0.0.1; 10.0.0.2;};
};
Where dial-up links are used with DNS forwarding servers BIND's general
purpose nature and strict standards adherence may not make it an optimal
solution. A number of the Alternate DNS solutions specifically target
support for such links. BIND provides two parameters dialup and heartbeat
-interval (neither of which is currently supported by BIND 9) as well as a
number of others which can be used to minimise connection time.
Example configuration files for a forwarding DNS are provided.
4.5 Stealth (a.k.a. DMZ or Hidden Master) Name Server
A stealth server is defined as being a name server which does not appear in
any publicly visible NS Records for the domain. The stealth server can be
roughly defined as having the following characteristics:
- The organisation needs a public DNS to enable access to its public services
e.g. web, mail ftp etc..
- The organisation does not want the world to see any of its internal hosts
either by interrogation (query or zone transfer) or should the DNS service
be compromised.
A Stealth configuration is shown in Figure 4-5.
Split (Stealth) Server configuration
Figure 4-5 Stealth Server Topology
The external server(s) is(are) configured to provide Authoritative Only
responses and no caching (no recursive queries accepted). The zone file for
this server would be unique and would contain ONLY those systems or services
that are publicly visible e.g. SOA, NS records for the public (not stealth)
name servers, MX record(s) for mail servers and www and ftp service A
records. Zone transfers can be allowed between the public servers as
required but they MUST NOT transfer or accept transfers from the Stealth
server. While this may seem to create more work, the concern is that should
the host running the external service be compromised then inspection of the
named.conf or zone files must provide no more information than is already
publically visible. If 'master', 'allow-notify','allow-transfer' options are
present in named.conf (each of which will contain a private IP) then the
attacker has gained more knowledge about the organisation - they have
penetrated the 'veil of privacy'.
There are a number of articles which suggest that the view statement may be
used to provide similar functionality using a single server but this does
not address the problem of the DNS host system being compromised and by
simple inspection of the named.conf file additional data about the
organisation could be discovered. In our opinion 'view' does not provide
adequate security in a 'Split DNS' solution.
A minimal public zone file is shown below:
; public zone master file
; provides minimal public visibility of external services
example.com. IN SOA ns.example.com. root.example.com.
(
2003080800 ; se = serial number
3h ; ref = refresh
15m ; ret = update retry
3w ; ex = expiry
3h ; min = minimum
)
IN NS ns1.example.com.
IN NS ns2.example.com.
IN MX 10 mail.example.com.
ns1 IN A 192.168.254.1
ns2 IN A 192.168.254.2
mail IN A 192.168.254.3
www IN A 192.168.254.4
ftp IN A 192.168.254.5
The internal server (the Stealth Server) can be configured to make visible
internal and external services, provide recursive queries and all manner of
other services. This server would use a private zone master file which could
look like this:
; private zone master file used by stealth server(s)
; provides public and private services and hosts
example.com. IN SOA ns.example.com. root.example.com.
(
2003080800 ; se = serial number
3h ; ref = refresh
15m ; ret = update retry
3w ; ex = expiry
3h ; min = minimum
)
IN NS ns1.example.com.
IN NS ns2.example.com.
IN MX 10 mail.example.com.
; public hosts
ns1 IN A 192.168.254.1
ns2 IN A 192.168.254.2
mail IN A 192.168.254.3
www IN A 192.168.254.4
ftp IN A 192.168.254.5
; private hosts
joe IN A 192.168.254.6
bill IN A 192.168.254.7
fred IN A 192.168.254.8
....
accounting IN A 192.168.254.28
payroll IN A 192.168.254.29
Using BIND 9's view statement can provide different services to internal and
external requests can reduce further the Stealth server's visibility e.g.
forwarding all DNS internal requests to the external server.
Example configuration files for a stealth DNS are provided.
4.6 Authoritative Only Server
The term Authoritative Only is normally used to describe two concepts:
- The server will deliver Authoritative Responses - it is a zone master or
slave for one or more domains.
- The server will NOT cache.
There are two configurations in which Authoritative Only servers are
typically used:
- As the public or external server in a Stealth (a.k.a. DMZ or Hidden Master)
DNS used to provide perimeter security.
- High Performance DNS servers. In this context general purpose DNS servers
such as BIND may not provide an ideal solution and there are a number of Open
Source Alternatives some of which specialise in high performance Authoritative
only solutions.
You cannot completely turn off caching in BIND but you can control it and
provide the functionality described above by simply turning off recursion in
the 'option' section of named.conf as shown in the example below.
// OPTIONS SECTION FRAGMENT OF named.conf
// recursion no = limits caching
options
{
directory "/var/named";
version "not currently available";
recursion no;
};
// ZONE FILE SECTIONs
....
BIND provides three more parameters to control caching, max-cache-size and
max-cache-ttl neither of which will have much effect on performance in this
particular case and allow-recursion which uses a list of hosts that are
permitted to use recursion (all others are not).
Example configuration files for a authoritative-only DNS are provided.
4.7 Split Horizon DNS Server
This section was introduced at the suggestion of Maren Leizaola - many thanks
for both taking the time to write and for providing interesting usage examples.
The term Split Horizon is normally used to describe a DNS server that will
give different responses (IP addresses) based on the source address, or some
other characteristic, of the query. While it has similar configuration
properties to the Stealth Server it can also be used in a varity of unique
situations such as:
- Geographic Mapping: Assume that, for example, a web service is replicated
in a number of locations (for either performance or access latency reasons) then
a specific IP address may be returned based on the source address of the query
to ensure the shortest possible path from the user to the service. For those
familiar with anycast you could consider this as a poor man's anycast service.
- Naming Consistency: Assume that you have, say, a corporate in-house LDAP
service and that you want to keep certain highly secure data on one server only
accessible to certain individuals or organizational sections, which have unique
or identifiable IP addresses or address ranges, but for reasons of consistency
(scripts, configuration files etc) you want both the secure and insecure LDAP
services to be named, say, ldap.example.com.
- Load Balancing: Assume that an analysis of incoming service users shows
that their source-ip addresses can be separated into contiguous ranges: 50%
from a to b, 50% from b to c. In this case rather than simply provide
multiple A/AAAA RRs (where load balancing is essentially random) it may be
more effective to use a split-horizon strategy.
Other possibilities may strike imaginative readers. The unifying element is
that some characteristic of the incoming query will cause the DNS to
generate a query-dependent result.
BIND's view clause provides a method that can be used to build such
configurations and example files are provided .