DNS and Internet Firewalls (DNS and BIND, 4th Edition)

11.3. DNS and Internet Firewalls

The Domain Name System wasn't designed to work with Internet firewalls. It's a testimony to the flexibility of DNS and of its BIND implementation that you can configure DNS to work with, or even through, an Internet firewall.

That said, configuring BIND to work in a firewalled environment, although not difficult, takes a good, complete understanding of DNS and a few of BIND's more obscure features. Describing it also requires a large portion of this chapter, so here's a roadmap.

We'll start by describing the two major families of Internet firewall software -- packet filters and application gateways. The capabilities of each family have a bearing on how you'll need to configure BIND to work through the firewall. Next, we'll detail the two most common DNS architectures used with firewalls, forwarders and internal roots, and describe the advantages and disadvantages of each. We'll then introduce a solution using a new feature, forward zones, which combines the best of internal roots and forwarders. Finally, we'll discuss split namespaces and the configuration of the bastion host, the host at the core of your firewall system.

11.3.1. Types of Firewall Software

Before you start configuring BIND to work with your firewall, it's important to understand what your firewall is capable of. Your firewall's capabilities will influence your choice of DNS architecture and determine how you implement it. If you don't know the answers to the questions in this section, track down someone in your organization who does know and ask. Better yet, work with your firewall's administrator when designing your DNS architecture to ensure it will coexist with the firewall.

Note that this is far from a complete explanation of Internet firewalls. These few paragraphs describe only the two most common types of Internet firewalls and only in enough detail to show how the differences in their capabilities affect name servers. For a comprehensive treatment of Internet firewalls, see Elizabeth D. Zwicky, Simon Cooper, and D. Brent Chapman's Building Internet Firewalls (O'Reilly).

11.3.1.1. Packet filters

The first type of firewall we'll cover is the packet-filtering firewall. Packet-filtering firewalls operate largely at the transport and network levels of the TCP/IP stack (layers three and four of the OSI reference model, if you dig that). They decide whether to route a packet based on packet-level criteria like the transport protocol (e.g., whether it's TCP or UDP), the source and destination IP address, and the source and destination port (see Figure 11-1).

Figure 11-1. Packet filters operate at the network and transport layers of the stack

What's most important to us about packet-filtering firewalls is that you can typically configure them to allow DNS traffic selectively between hosts on the Internet and your internal hosts. That is, you can let an arbitrary set of internal hosts communicate with Internet name servers. Some packet-filtering firewalls can even permit your name servers to query name servers on the Internet, but not vice versa. All router-based Internet firewalls are packet-filtering firewalls. Checkpoint's FireWall-1, Cisco's PIX, and Sun's SunScreen are popular commercial packet-filtering firewalls.

A Gotcha with BIND 8 or 9 and Packet-Filtering Firewalls
BIND 4 name servers always send queries from port 53, the well-known port for DNS servers, to port 53. Resolvers, on the other hand, usually send queries from high-numbered ports (above 1023) to port 53. Though name servers clearly have to send their queries to the DNS port on a remote host, there's no reason they have to send the queries from the DNS port. And, wouldn't you know it, BIND 8 and 9 name servers don't send queries from port 53 by default. Instead, they send queries from high-numbered ports, the same as resolvers do.
This can cause problems with packet-filtering firewalls that are configured to allow name server-to-name server traffic but not resolver-to-name server traffic, because they typically expect name server-to-name server traffic to originate from port 53 and terminate at port 53.
There are two solutions to this problem:

Reconfigure the firewall to allow your name server to send and receive queries from ports other than 53 (assuming this doesn't compromise the security of the firewall by allowing packets from Internet hosts to high-numbered ports on internal name servers).

Configure BIND to revert to its old behavior with the query-source substatement.
query-source takes as arguments an address specification and an optional port number. For example, the statement:
options { query-source address * port 53; };
tells BIND to use port 53 as the source port for queries sent from all local network interfaces. You can use a nonwildcard address specification to limit the addresses that BIND will send queries from. For example, on wormhole.movie.edu, the statement:
options { query-source address 192.249.249.1 port *; };
tells BIND to send all queries from the 192.249.249.1 address (i.e., not from 192.253.253.1) and to use dynamic, high-numbered ports.
The use of query-source with a wildcard address is broken in BIND 9 before 9.1.0, though you can tell an early BIND 9 name server to send all queries from a particular address's port 53.

11.3.1.2. Application gateways

Application gateways operate at the application protocol level, several layers higher in the OSI reference model than most packet filters (see Figure 11-2). In a sense, they "understand" the application protocol in the same way a server for that particular application would. An FTP application gateway, for example, can make the decision to allow or deny a particular FTP operation, such as a RETR (a get) or a STOR (a put).

Figure 11-2. Application gateways operate at the application layer of the stack

The bad news, and what's important for our purposes, is that most application gateway-based firewalls handle only TCP-based application protocols. DNS, of course, is largely UDP-based, and we know of no application gateways for DNS. This implies that if you run an application gateway-based firewall, your internal hosts will likely not be able to communicate directly with name servers on the Internet.

Note that these two categories of firewall are really just generalizations. The state of the art in firewalls changes very quickly, and by the time you read this, you may have a firewall that includes an application gateway for DNS. Which family your firewall falls into is important only because it suggests what that firewall is capable of; what's more important is whether your particular firewall will let you permit DNS traffic between arbitrary internal hosts and the Internet .

11.3.2. A Bad Example

The simplest configuration is to allow DNS traffic to pass freely through your firewall (assuming you can configure your firewall to do that). That way, any internal name server can query any name server on the Internet, and any Internet name server can query any of your internal name servers. You don't need any special configuration.

Unfortunately, this is a really bad idea, for a number of reasons:

Version control: The developers of BIND are constantly finding and fixing security-related bugs in the BIND code. Consequently, it's important to run a recent version of BIND, especially on name servers directly exposed to the Internet. If one or just a few of your name servers communicate directly with name servers on the Internet, upgrading them to a new version is easy. If any of the name servers on your network can communicate directly with name servers on the Internet, upgrading all of them is vastly more difficult.

Possible vector for attack: Even if you're not running a name server on a particular host, a hacker might be able to take advantage of your allowing DNS traffic through your firewall and attack that host. For example, a co-conspirator working on the inside could set up a Telnet daemon listening on the host's DNS port, allowing the hacker to telnet right in.

For the rest of this chapter, we'll try to set a good example.

Figure 11-5 illustrates a common forwarding setup, with internal name servers forwarding queries to a name server running on a bastion host.

Figure 11-5. Using forwarders

At Movie U., we put in a firewall to protect ourselves from the Big Bad Internet several years ago. Ours is a packet-filtering firewall, and we negotiated with our firewall administrator to allow DNS traffic between Internet name servers and two of our name servers, terminator.movie.edu and wormhole.movie.edu. Here's how we configured the other internal name servers at the university. For our BIND 8 and 9 name servers, we used the following:

options {
	forwarders { 192.249.249.1; 192.249.249.3; };
	forward only;
};

and for our BIND 4 name servers, we used:

forwarders 192.249.249.3 192.249.249.1
options forward-only

We vary the order in which the forwarders appear to help spread the load between them, though that's not necessary with BIND 8.2.3 name servers, which choose which forwarder to query according to roundtrip time.

When an internal name server receives a query for a name it can't resolve locally, such as an Internet domain name, it forwards that query to one of our forwarders, which can resolve the name using name servers on the Internet. Simple!

11.3.3.1. The trouble with forwarding

Unfortunately, it's a little too simple. Forwarding starts to get in the way once you delegate subdomains or build an extensive network. To explain what we mean, take a look at part of the configuration file on zardoz.movie.edu:

options {
	directory "/var/named";
	forwarders { 192.249.249.1; 192.253.253.3; };
};

zone "movie.edu" {
	type slave;
	masters { 192.249.249.3; };
	file "bak.movie.edu";
};

zardoz.movie.edu is a slave for movie.edu and uses our two forwarders. What happens when zardoz.movie.edu receives a query for a name in fx.movie.edu ? As an authoritative movie.edu name server, zardoz.movie.edu has the NS records that delegate fx.movie.edu to its authoritative name servers. But it's also been configured to forward queries it can't resolve locally to terminator.movie.edu and wormhole.movie.edu. Which will it do?

It turns out that zardoz.movie.edu ignores the delegation information and forwards the query to terminator.movie.edu. That works since terminator.movie.edureceives the recursive query and asks an fx.movie.edu name server on zardoz.movie.edu's behalf. But it's not particularly efficient since zardoz.movie.edu could easily have sent the query directly.

Now imagine that the scale of the network is much larger: a corporate network that spans continents, with tens of thousands of hosts and hundreds or thousands of name servers. All the internal name servers that don't have direct Internet connectivity -- the vast majority of them -- use a small set of forwarders. What's wrong with this picture?

Single point of failure: If the forwarders fail, your name servers lose the ability to resolve both Internet domain names and internal domain names that they don't have cached or stored as authoritative data.

Concentration of load: The forwarders have an enormous query load placed on them. This is both because of the large number of internal name servers that use them, and because the queries are recursive and require a good deal of work to answer.

Inefficient resolution

Imagine two internal name servers, authoritative for west.acmebw.com and east.acmebw.com, respectively, both on the same network segment in Boulder, Colorado. Both are configured to use the company's forwarder in Bethesda, Maryland. For the west.acmebw.com name server to resolve a name in east.acmebw.com, it sends a query to the forwarder in Bethesda. The forwarder in Bethesda then sends a query back to Boulder to the east.acmebw.comname server, the original querier's neighbor. The east.acmebw.com name server replies by sending a response back to Bethesda, which the forwarder sends back to Boulder.

In a traditional configuration with root name servers, the west.acmebw.com name server would have learned quickly that an east.acmebw.com name server was next door and would favor it (because of its low roundtrip time). Using forwarders "short-circuits" the normally efficient resolution process.

The upshot is that forwarding is fine for small networks and simple namespaces, but probably inadequate for large networks and complex namespaces. We found this out the hard way at Movie U., as our network grew and we were forced to find an alternative.

11.3.3.2. Using forward zones

We can solve this problem by using the forward zones introduced in BIND 8.2. We change zardoz.movie.edu's configuration to this:

options {
	directory "/var/named";
	forwarders { 192.249.249.1; 192.253.253.3; };
};

zone "movie.edu" {
	type slave;
	masters { 192.249.249.3; };
	file "bak.movie.edu";
	forwarders {};
};

Now, if zardoz.movie.edu receives a query for a domain name ending in movie.edu but outside the movie.edu zone (e.g., in fx.movie.edu), it ignores the forwarders and sends iterative queries.

With this configuration, zardoz.movie.edu still sends queries for domain names in our reverse-mapping zones to our forwarders. To relieve that load, we can add a few zone statements to named.conf:

zone "249.249.192.in-addr.arpa" {
	type stub;
	masters { 192.249.249.3; };
	file "stub.192.249.249";
	forwarders {};
};

zone "253.253.192.in-addr.arpa" {
	type stub;
	masters { 192.249.249.3; };
	file "stub.192.253.253";
	forwarders {};
};

zone "254.253.192.in-addr.arpa" {
	type stub;
	masters { 192.253.254.2; };
	file "stub.192.253.254";
	forwarders {};
};

zone "20.254.192.in-addr.arpa" {
	type stub;
	masters { 192.253.254.2; };
	file "stub.192.254.20";
	forwarders {};
};

These new zone statements bear some explaining: first of all, they configure Movie U.'s reverse-mapping zones as stubs. That makes our name server track the NS records for those zones by periodically querying the master name servers for those zones. The forwarders substatement then turns off forwarding for domain names in the reverse-mapping domains. Now, instead of querying the forwarders for, say, the PTR record for 2.254.253.192.in-addr.arpa, zardoz.movie.edu will query one of the 254.253.192.in-addr.arpaname servers directly.

We'll need zone statements like these on all of our internal name servers, which also implies that we'll need all of our name servers to run some version of BIND 8 after 8.2.[84]

[84]As we mentioned in the last chapter, BIND 9 doesn't support forward zones until BIND 9.1.0.

This gives us a fairly robust resolution architecture that minimizes our exposure to the Internet: it uses efficient, robust iterative name resolution to resolve internal domain names, and forwarders only when necessary to resolve Internet domain names. If our forwarders fail or we lose our connection to the Internet, we lose only our ability to resolve Internet domain names.

11.3.4. Internal Roots

If you want to avoid the scalability problems of forwarding, you can set up your own root name servers. These internal roots will serve only the name servers in your organization. They'll know about only the portions of the namespace relevant to your organization.

What good are they? By using an architecture based on root name servers, you gain the scalability of the Internet's namespace (which should be good enough for most companies), plus redundancy, distributed load, and efficient resolution. You can have as many internal roots as the Internet has roots -- 13 or so -- whereas having that many forwarders may be an undue security exposure and a configuration burden. Most of all, the internal roots don't get used frivolously. Name servers need to consult an internal root only when they time out the NS records for your top-level zones. Using forwarders, name servers may have to query a forwarder once per resolution.

The moral of our story is that if you have, or intend to have, a large namespace and lots of internal name servers, internal root name servers will scale better than any other solution.

11.3.4.1. Where to put internal root name servers

Since name servers "lock on" to the closest root name server by favoring the one with the lowest roundtrip time, it pays to pepper your network with internal root name servers. If your organization's network spans the U.S., Europe, and the Pacific Rim, consider locating at least one internal root name server on each continent. If you have three major sites in Europe, give each of them an internal root.

11.3.4.2. Forward-mapping delegation

Here's how an internal root name server is configured. An internal root delegates directly to any zones you administer. For example, on the movie.edunetwork, the root zone's data file would contain:

movie.edu.  86400  IN  NS  terminator.movie.edu.
            86400  IN  NS  wormhole.movie.edu.
            86400  IN  NS  zardoz.movie.edu.
terminator.movie.edu.  86400  IN  A  192.249.249.3
wormhole.movie.edu.    86400  IN  A  192.249.249.1
                       86400  IN  A  192.253.253.1
zardoz.movie.edu.      86400  IN  A  192.249.249.9
                       86400  IN  A  192.253.253.9

On the Internet, this information would appear in the edu name servers' zone data files. On the movie.edu network, of course, there aren't any eduname servers, so you delegate directly to movie.edu from the root.

Notice that this doesn't contain delegation to fx.movie.edu or to any other subdomain of movie.edu. The movie.edu name servers know which name servers are authoritative for all movie.edu subdomains, and all queries for information in those subdomains pass through the movie.edu name servers, so there's no need to delegate them here.

11.3.4.3. in-addr.arpa delegation

We also need to delegate from the internal roots to the in-addr.arpa zones that correspond to the networks at the university:

249.249.192.in-addr.arpa.  86400  IN  NS  terminator.movie.edu.
                           86400  IN  NS  wormhole.movie.edu.
                           86400  IN  NS  zardoz.movie.edu.
253.253.192.in-addr.arpa.  86400  IN  NS  terminator.movie.edu.
                           86400  IN  NS  wormhole.movie.edu.
                           86400  IN  NS  zardoz.movie.edu.
254.253.192.in-addr.arpa.  86400  IN  NS  bladerunner.fx.movie.edu.
                           86400  IN  NS  outland.fx.movie.edu.
                           86400  IN  NS  alien.fx.movie.edu.
20.254.192.in-addr.arpa.   86400  IN  NS  bladerunner.fx.movie.edu.
                           86400  IN  NS  outland.fx.movie.edu.
                           86400  IN  NS  alien.fx.movie.edu.

Notice that we did include delegation for the 254.253.192.in-addr.arpa and the 20.254.192.in-addr.arpazones, even though they correspond to the fx.movie.edu zone. We don't need to delegate to fx.movie.edu because we'd already delegated to its parent, movie.edu. The movie.edu name servers delegate to fx.movie.edu, so by transitivity the roots delegate to fx.movie.edu. Since neither of the other in-addr.arpa zones is a parent of 254.253.192.in-addr.arpaor 20.254.192.in-addr.arpa, we need to delegate both zones from the root. As we explained earlier, we don't need to add address records for the three Special Effects name servers, bladerunner.fx.movie.edu, outland.fx.movie.edu, and alien.fx.movie.edu, because a remote name server can already find their addresses by following delegation from movie.edu.

11.3.4.4. The db.root file

All that's left is to add an SOA record for the root zone and NS records for this internal root name server and any others:

$TTL 1d
.  IN  SOA  rainman.movie.edu.  hostmaster.movie.edu.  (
            1    ; serial
            3h   ; refresh
            1h   ; retry
            1w   ; expire
            1h ) ; negative caching TTL

   IN  NS  rainman.movie.edu.
   IN  NS  awakenings.movie.edu.

rainman.movie.edu.    IN  A  192.249.249.254
awakenings.movie.edu. IN  A  192.253.253.254

rainman.movie.edu and awakenings.movie.edu are the hosts running the internal root name servers. We shouldn't run an internal root on a bastion host, because if a name server on the Internet accidentally queries it for data it's not authoritative for, the internal root will respond with its list of roots -- all internal!

So the whole db.root file (by convention, we call the root zone's data file db.root) looks like this:

$TTL 1d
.  IN  SOA  rainman.movie.edu.  hostmaster.movie.edu.  (
            1    ; serial
            3h   ; refresh
            1h   ; retry
            1w   ; expire
            1h ) ; negative caching TTL

   IN  NS  rainman.movie.edu.
   IN  NS  awakenings.movie.edu.

rainman.movie.edu.    IN  A  192.249.249.254
awakenings.movie.edu. IN  A  192.253.253.254

movie.edu.  IN  NS  terminator.movie.edu.
            IN  NS  wormhole.movie.edu.
            IN  NS  zardoz.movie.edu.

terminator.movie.edu.  IN  A  192.249.249.3
wormhole.movie.edu.    IN  A  192.249.249.1
                       IN  A  192.253.253.1
zardoz.movie.edu.      IN  A  192.249.249.9
                       IN  A  192.253.253.9

249.249.192.in-addr.arpa.  IN  NS  terminator.movie.edu.
                           IN  NS  wormhole.movie.edu.
                           IN  NS  zardoz.movie.edu.
253.253.192.in-addr.arpa.  IN  NS  terminator.movie.edu.
                           IN  NS  wormhole.movie.edu.
                           IN  NS  zardoz.movie.edu.
254.253.192.in-addr.arpa.  IN  NS  bladerunner.fx.movie.edu.
                           IN  NS  outland.fx.movie.edu.
                           IN  NS  alien.fx.movie.edu.
20.254.192.in-addr.arpa.   IN  NS  bladerunner.fx.movie.edu.
                           IN  NS  outland.fx.movie.edu.
                           IN  NS  alien.fx.movie.edu.

The named.conf file on both the internal root name servers, rainman.movie.eduand awakenings.movie.edu, contains the lines:

zone "." {
	type master;
	file "db.root";
};

Or, for a BIND 4 server's named.boot file:

primary    .    db.root

This replaces a zone statement of type hint or a cache directive -- a root name server doesn't need a root hints file to tell it where the other roots are; it can find that in db.root. Did we really mean that eachroot name server is a primary master for the root zone? Not unless you're running an ancient version of BIND. All BIND versions after 4.9 let you declare a server as a slave for the root zone, but BIND 4.8.3 and earlier insist that all root name servers load the root zone as primaries.

If you don't have a lot of idle hosts sitting around that you can turn into internal roots, don't despair! Any internal name server (i.e., one that's not running on a bastion host or outside your firewall) can serve double duty as an internal root and as an authoritative name server for whatever other zones you need it to load. Remember, a single name server can be authoritative for many, many zones, including the root zone.

11.3.4.5. Configuring other internal name servers

Once you've set up internal root name servers, configure all your name servers on hosts anywhere on your internal network to use them. Any name server running on a host without direct Internet connectivity (i.e., behind the firewall) should list the internal roots in its root hints file:

; Internal root hints file, for Movie U. hosts without direct
; Internet connectivity
;
; Don't use this file on a host with Internet connectivity!
;

.  99999999  IN  NS  rainman.movie.edu.
   99999999  IN  NS  awakenings.movie.edu.

rainman.movie.edu.     99999999  IN  A  192.249.249.254
awakenings.movie.edu.  99999999  IN  A  192.253.253.254

Name servers running on hosts using this root hints file will be able to resolve domain names in movie.edu and in Movie U.'s in-addr.arpa domains, but not outside those domains.

11.3.4.6. How internal name servers use internal roots

To tie together how this whole scheme works, let's go through an example of name resolution on an internal caching-only name server using these internal root name servers. First, the internal name server receives a query for a domain name in movie.edu, say the address of gump.fx.movie.edu. If the internal name server doesn't have any "better" information cached, it starts by querying an internal root name server. If it has communicated with the internal roots before, it has a roundtrip time associated with each, telling it which of the internal roots is responding to it most quickly. It then sends a nonrecursive query to that internal root for gump.fx.movie.edu's address. The internal root answers with a referral to the movie.edu name servers on terminator.movie.edu, wormhole.movie.edu, and zardoz.movie.edu. The caching-only name server follows up by sending another nonrecursive query to one of the movie.edu name servers for gump.fx.movie.edu's address. The movie.edu name server responds with a referral to the fx.movie.edu name servers. The caching-only name server sends the same nonrecursive query for gump.fx.movie.edu's address to one of the fx.movie.eduname servers and finally receives a response.

Contrast this with the way a forwarding setup would work. Let's imagine that instead of using internal root name servers, our caching-only name server were configured to forward queries first to terminator.movie.edu and then to wormhole.movie.edu. In that case, the caching-only name server would check its cache for the address of gump.fx.movie.eduand, not finding it, would forward the query to terminator.movie.edu. Then, terminator.movie.edu would query an fx.movie.edu name server on the caching-only name server's behalf and return the answer. Should the caching-only name server need to look up another name in fx.movie.edu, it would still ask the forwarder, even though the forwarder's response to the query for gump.fx.movie.edu's address probably contains the names and addresses of the fx.movie.edu name servers.

11.3.4.7. Mail from internal hosts to the Internet

But wait! That's not all internal roots will do for you. We talked about getting mail to the Internet without changing sendmail 's configuration all over the network.

Wildcard records are the key to getting mail to work -- specifically, wildcard MX records. Let's say that we want mail to the Internet to be forwarded through postmanrings2x.movie.edu, the Movie U. bastion host, which has direct Internet connectivity. Adding the following records to db.root will get the job done:

*        IN    MX    5 postmanrings2x.movie.edu.
*.edu.   IN    MX    10 postmanrings2x.movie.edu.

We need the *.edu MX record in addition to the * record because of wildcard production rules, which you can read more about in Section 16.2, "Wildcards" of Chapter 16, "Miscellaneous". Basically, since there is explicit data for movie.edu in the zone, the first wildcard won't match movie.edu or any other subdomains of edu. We need another, explicit wildcard record for eduto match subdomains of edubesides movie.edu.

Now mailers on our internal movie.eduhosts will send mail addressed to Internet domain names to postmanrings2x.movie.edufor forwarding. For example, mail addressed to nic.ddn.mil will match the first wildcard MX record:

% nslookup -type=mx nic.ddn.mil.  -- Matches the MX record for *
Server:  rainman.movie.edu
Address:  192.249.249.19

nic.ddn.mil
     preference = 5, mail exchanger = postmanrings2x.movie.edu
postmanrings2x.movie.edu    internet address = 192.249.249.20

Mail addressed to vangogh.cs.berkeley.eduwill match the second MX record:

% nslookup -type=mx vangogh.cs.berkeley.edu.  -- Matches the MX record for *.edu
Server:  rainman.movie.edu
Address:  192.249.249.19

vangogh.cs.berkeley.edu
     preference = 10, mail exchanger = postmanrings2x.movie.edu
postmanrings2x.movie.edu    internet address = 192.249.249.20

Once the mail has reached postmanrings2x.movie.edu, our bastion host, postmanrings2x.movie.edu's mailer will look up the MX records for these addresses itself. Since postmanrings2x.movie.edu will resolve the destination's domain name using the Internet's namespace instead of the internal namespace, it will find the real MX records for the domain name and deliver the mail. No changes to sendmail 's configuration are necessary.

11.3.4.8. Mail to specific Internet domain names

Another nice perk of this internal root scheme is that it gives you the ability to forward mail addressed to certain Internet domain names through particular bastion hosts, if you have more than one. We can choose, for example, to send all mail addressed to recipients in the uk domain to our bastion host in London first and then out onto the Internet. This can be very useful if we want our mail to travel across our own network as far as possible or if we're billed for our usage of some network in the U.K.

Movie U. has a private network connection to our sister university in London near Pinewood Studios. For security reasons, we'd like to send mail addressed to correspondents in the U.K. across our private link and then through the Pinewood host. So we add the following wildcard records to db.root:

; holygrail.movie.ac.uk is at the other end of our U.K. Internet link
*.uk.    IN    MX    10 holygrail.movie.ac.uk.
holygrail.movie.ac.uk.    IN   A    192.168.76.4

Now, mail addressed to users in subdomains of uk will be forwarded to the host holygrail.movie.ac.uk at our sister university, which presumably has facilities to forward that mail to other points in the U.K.

11.3.4.9. The trouble with internal roots

Unfortunately, just as forwarding has its problems, internal root architectures have their limitations. Chief among these is the fact that your internal hosts can't see the Internet namespace. On some networks, this isn't an issue because most internal hosts don't have any direct Internet connectivity. The few that do can have their resolvers configured to use a name server on the bastion host. Some of these hosts will probably need to run proxy servers to allow other internal hosts access to services on the Internet.

On other networks, however, the Internet firewall or other software may require that all internal hosts have the ability to resolve names in the Internet's namespace. For these networks, an internal root architecture won't work.

11.3.5. A Split Namespace

Many organizations would like to advertise different zone data to the Internet than they advertise internally. In most cases, much of the internal zone data is irrelevant to the Internet because of the organization's Internet firewall. The firewall may not allow direct access to most internal hosts, and may also translate internal, unregistered IP addresses into a range of IP addresses registered to the organization. Therefore, the organization might need to trim out irrelevant information from the external view of the zone or change internal addresses to their external equivalents.

Unfortunately, BIND doesn't support automatic filtering and translation of zone data. Consequently, many organizations manually create what have become known as "split namespaces." In a split namespace, the real namespace is available only internally, while a pared-down, translated version of it called the shadow namespace is visible to the Internet.

The shadow namespace contains the name-to-address and address-to-name mappings of only those hosts accessible from the Internet through the firewall. The addresses advertised may be the translated equivalents of internal addresses. The shadow namespace may also contain one or more MX records to direct mail from the Internet through the firewall to a mail server.

Since Movie U. has an Internet firewall that greatly limits access from the Internet to the internal network, we elected to create a shadow namespace. For the zone movie.edu, the only information we need to give out is about the domain name movie.edu (an SOA record and a few NS records), the bastion host (postmanrings2x.movie.edu), and our new external name server, ns.movie.edu, which also functions as an external web server, www.movie.edu. The address of the external interface on the bastion host is 200.1.4.2, and the address of the name/web server is 200.1.4.3. The shadow movie.edu zone data file looks like this:

$TTL 1d
@    IN    SOA    ns.movie.edu.    hostmaster.movie.edu. (
                           1    ; Serial
                           3h   ; Refresh
                           1h   ; Retry
                           1w   ; Expire
                           1h ) ; Negative caching TTL

    IN    NS    ns.movie.edu.
    IN    NS    ns1.isp.net.        ; our ISP's name server is a movie.edu slave

    IN    A     200.1.4.3
    IN    MX    10 postmanrings2x.movie.edu.
    IN    MX    100 mail.isp.net.

www             IN    CNAME movie.edu.

postmanrings2x  IN    A     200.1.4.2
                IN    MX    10 postmanrings2x.movie.edu.
                IN    MX    100 mail.isp.net.

;postmanrings2x.movie.edu handles mail addressed to ns.movie.edu
ns              IN    A     200.1.4.3
                IN    MX    10 postmanrings2x.movie.edu.
                IN    MX    100 mail.isp.net.

*               IN    MX    10 postmanrings2x.movie.edu.
                IN    MX    100 mail.isp.net.

Note that there's no mention of any of the subdomains of movie.edu, including any delegation to the name servers for those subdomains. The information simply isn't necessary since there's nothing in any of the subdomains that you can get to from the Internet, and inbound mail addressed to hosts in the subdomains is caught by the wildcard.

The db.200.1.4 file, which we need in order to reverse map the two Movie U. IP addresses that hosts on the Internet might see, looks like this:

$TTL 1d
@    IN    SOA    ns.movie.edu.    hostmaster.movie.edu. (
                           1    ; Serial
                           3h   ; Refresh
                           1h   ; Retry
                           1w   ; Expire
                           1h ) ; Negative caching TTL

    IN    NS    ns.movie.edu.
    IN    NS    ns.isp.net.

2    IN    PTR    postmanrings2x.movie.edu.
3    IN    PTR    ns1.movie.edu.

One precaution we have to take is to make sure that the resolver on our bastion host isn't configured to use the server on ns.movie.edu. Since that server can't see the real, internal movie.edu, using it would render postmanrings2x.movie.edu unable to map internal domain names to addresses or internal addresses to names.

11.3.5.1. Configuring the bastion host

The bastion host is a special case in a split namespace configuration. It has a foot in each environment: one network interface connects it to the Internet and another connects it to the internal network. Now that we have split our namespace in two, how can our bastion host see both the Internet namespace and our real internal namespace? If we configure it with the Internet's root name servers in its root hints file, it will follow delegation from the Internet's edu name servers to an external movie.edu name server with shadow zone data. It would be blind to our internal namespace, which it needs to see to log connections, deliver inbound mail, and more. On the other hand, if we configure it with our internal roots, then it won't see the Internet's namespace, which it clearly needs to do in order to function as a bastion host. What to do?

If we have internal name servers that can resolve both internal and Internet domain names -- using forward zones per the configuration earlier in this chapter, for example -- we can simply configure the bastion host's resolver to query those name servers. But if we use forwarding internally, depending on the type of firewall we're running, we may also need to run a forwarder on the bastion host itself. If the firewall won't pass DNS traffic, we'll need to run at least a caching-only name server, configured with the Internet roots, on the bastion host so that our internal name servers will have somewhere to forward their unresolved queries.

If our internal name servers don't support forward zones, the name server on our bastion host must be configured as a slave for movie.edu and any in-addr.arpa zones in which it needs to resolve addresses. This way, if it receives a query for a domain name in movie.edu, it uses its local authoritative data to resolve the name. (If our internal name servers support forward zones and are configured correctly, the name server on our bastion host will never receive queries for names in movie.edu.) If the domain name is in a delegated subdomain of movie.edu, it follows NS records in the zone data to query an internal name server for the name. Therefore, it doesn't need to be configured as a slave for any movie.edu subdomains, such as fx.movie.edu, just the "topmost" zone (see Figure 11-6).

Figure 11-6. A split DNS solution

The named.conf file on our bastion host looks like this:

options {
	directory "/var/named";
};

zone "movie.edu" {
	type slave;
	masters { 192.249.249.3; };
	file "bak.movie.edu";
};

zone "249.249.192.in-addr.arpa" {
	type slave;
	masters { 192.249.249.3; };
	file "bak.192.249.249";
};

zone "253.253.192.in-addr.arpa" {
	type slave;
	masters { 192.249.249.3; };
	file "bak.192.253.253";
};

zone "254.253.192.in-addr.arpa" {
	type slave;
	masters { 192.253.254.2; };
	file "bak.192.253.254";
};

zone "20.254.192.in-addr.arpa" {
	type slave;
	masters { 192.253.254.2; };
	file "bak.192.254.20";
};

zone "." {
	type hint;
	file "db.cache";
};

An equivalent named.boot file would look like this:

directory    /var/named
secondary    movie.edu    192.249.249.3   bak.movie.edu
secondary    249.249.192.in-addr.arpa    192.249.249.3    bak.192.249.249
secondary    253.253.192.in-addr.arpa    192.249.249.3    bak.192.253.253
secondary    254.253.192.in-addr.arpa    192.253.254.2    bak.192.253.254
secondary    20.254.192.in-addr.arpa     192.253.254.2    bak.192.254.20
cache    .    db.cache    ; lists Internet roots

11.3.5.2. Protecting zone data on the bastion host

Unfortunately, loading these zones on the bastion host also exposes them to the possibility of disclosure on the Internet, which we were trying to avoid by splitting the namespace in the first place. But as long as we're running BIND 4.9 or better, we can protect the zone data using the secure_zone TXT record or the allow-query substatement, both discussed earlier in the chapter. With allow-query, we can place a global access list on our zone data. Here's the new options statement from our named.conf file:

options {
	directory "/var/named";
	allow-query { 127/8; 192.249.249/24; 192.253.253/24;
		192.253.254/24; 192.254.20/24; };
};

With BIND 4.9's secure_zone feature, we can turn off all external access to our zone data by including these TXT records in each zone data file:

secure_zone     IN    TXT    "192.249.249.0:255.255.255.0"
                IN    TXT    "192.253.253.0:255.255.255.0"
                IN    TXT    "192.253.254.0:255.255.255.0"
                IN    TXT    "192.254.20.0:255.255.255.0"
                IN    TXT    "127.0.0.1:H"

Don't forget to include the loopback address in the list, or the bastion host's resolver may not get answers from its own name server!

11.3.5.3. The final configuration

Finally, we need to apply the other security precautions we discussed earlier to our bastion host's name server. In particular, we should:

Restrict zone transfers
Use the ID pool feature (on BIND 8.2 or newer name servers but not BIND 9)
(Optionally) Run BIND chroot ed and with least privilege

In the end, our named.conf file ends up looking like this:

acl "internal" {
	127/8; 192.249.249/24; 192.253.253/24;
	192.253.254/24; 192.254.20/24;
};

options {
	directory "/var/named";
	allow-query { "internal"; };
	allow-transfer { none; };
	use-id-pool yes;
};

zone "movie.edu" {
	type slave;
	masters { 192.249.249.3; };
	file "bak.movie.edu";
};

zone "249.249.192.in-addr.arpa" {
	type slave;
	masters { 192.249.249.3; };
	file "bak.192.249.249";
};

zone "253.253.192.in-addr.arpa" {
	type slave;
	masters { 192.249.249.3; };
	file "bak.192.253.253";
};

zone "254.253.192.in-addr.arpa" {
	type slave;
	masters { 192.253.254.2; };
	file "bak.192.253.254";
};

zone "20.254.192.in-addr.arpa" {
	type slave;
	masters { 192.253.254.2; };
	file "bak.192.254.20";
};

zone "." {
	type hint;
	file "db.cache";
};

11.3.5.4. Using views on the bastion host

If we're running BIND 9 on our bastion host, we can use views to safely present the shadow movie.edu to the outside world on the same name server that resolves Internet domain names. That may obviate the need to run an external name server on the same host as our web server, www.movie.edu. If not, it'll give us two name servers to advertise the external movie.edu.

This configuration is very similar to one shown in Section 10.6, "Views":

options {
	directory "/var/named";
};

acl "internal" {
	127/8; 192.249.249/24; 192.253.253/24; 192.253.254/24; 192.254.20/24;
};

view "internal" {
	match-clients { "internal"; };
	recursion yes;

	zone "movie.edu" {
		type slave;
		masters { 192.249.249.3; };
		file "bak.movie.edu";
	 };

	zone "249.249.192.in-addr.arpa" {
		type slave;
		masters { 192.249.249.3; };
		file "bak.192.249.249";
	};

	zone "253.253.192.in-addr.arpa" {
		type slave;
		masters { 192.249.249.3; };
		file "bak.192.253.253";
	};

	zone "254.253.192.in-addr.arpa" {
		type slave;
		masters { 192.253.254.2; };
		file "bak.192.253.254";
	};

	zone "20.254.192.in-addr.arpa" {
		type slave;
		masters { 192.253.254.2; };
		file "bak.192.254.20";
	};

	zone "." {
		type hint;
		file "db.cache";
	};
};

view "external" {
	match-clients { any; };
	recursion no;

	acl "ns1.isp.net" { 199.11.28.12; };

	zone "movie.edu" {
		type master;
		file "db.movie.edu.external";
		allow-transfer { "ns1.isp.net"; };
	};

	zone "4.1.200.in-addr.arpa" {
		type master;
		file "db.200.1.4";
		allow-transfer { "ns1.isp.net"; };
	};

	zone "." {
		type hint;
		file "db.cache";
	};
};

Notice that the internal and external views present different versions of movie.edu: one loaded from the zone data file db.movie.edu, and one loaded from db.movie.edu.external. If there were more than a few zones in our external view, we probably would have used a different subdirectory for our external zone data files than we used for the internal zone data files.