Packages Used |
|
Net::DNS modules CPAN BIND http://www.isc.org/bind.html |
This article talks about the Net::DNS::Update module, and how you can use it to remotely update information on a DNS server. You might do this if you are a DNS administrator wanting to balance load between a range of machines, or if you are the owner of a domain and want to programmatically update the information in that domain even though you don't have access to the DNS server's configuration files.
Whenever you send mail or visit a web site, your computer has to know how to reach a remote computer. In particular, it has to translate a name like "perl.com" to its Internet address, 199.45.135.9. Only when that address is known can your mail program send mail to webmaster@perl.com and your web browser display http://www.perl.com/CPAN/.
The act of translating hostnames to Internet addresses is called name resolution, and the infrastructure supporting it is called the Domain Name System, or DNS. In this article, I'll demonstrate how to access some of the new features of The Berkeley Internet Name Daemon (BIND) Version 8 with the Net::DNS modules. In particular, I'll focus on Net::DNS::Update, which allows you to change the information stored in a nameserver on the fly. You could use this technique to take a crashed box in a server farm out of rotation, manage PPP or DHCP clients, or perform load balancing.
To get the most out of this article you should have a decent knowledge of the workings of DNS and BIND, a BIND version 8 nameserver, and a zone to play around with. A zone is similar to a domain such as perl.com, but it could also refer to a subdomain such as test.perl.com. While it is possible that the code examples will work with other RFC 2136-compliant nameservers, I haven't tested anything other than BIND 8 running on Unix.
A full description of DNS is beyond the scope of this article. For those who want the full story, I highly recommend O'Reilly's DNS & Bind book by Albitz & Liu.
Without doubt, the most popular nameserver is BIND (http://www.isc.org/bind.html). The current BIND implementation, BIND 8, is maintained by the Internet Software Consortium (ISC), and supports a feature called dynamic update; see http://www.isi.edu/in-notes/rfc2136.txt for the low-level description of the dynamic update protocol. BIND 8 has been in production use since May 1997, so you shouldn't fear upgrading.
One way to tell if your system is running BIND 8 is to look at how your named gets started. If it uses a file called named.conf, it is version 8. The older model, version 4, which is still quite common, uses named.boot. You could also look at the named log messages when it starts up. Mine looks like this:
Dec 16 11:41:09 hudsucker named[14993]: starting. named
8.1.2 Thu Jan 8 12:47:42 PST 1998
jsd@hudsucker.gamespot.com:/usr/home/jsd/build/bind/src/bin/named
If your system uses version 4, you should upgrade to version 8, since that's the only version being actively maintained and developed. BIND 4 has been deprecated by the ISC.
Originally, zone files (the files that map IP addresses to names and vice versa) were fairly static. You'd set up a machine, give it an address, and that was the end of the story. Most machines on the net were mainframes then, so there was no compelling need for a flexible, dynamic protocol.
In this day of dialup PPP, DHCP, and server farms, things are clearly different. We need a way to dynamically update entries in a zone file on the fly. Enter RFC 2136 and Net::DNS::Update.
Net::DNS is a collection of object-oriented modules designed to interface with nameservers. The author, Michael Fuhr, claims they're slow, but I've never encountered any debilitating speed problems. Usually you have more than enough latency in your network to offset any problems with script speed!
Before you can start writing code, you must configure your nameserver to accept dynamic updates. I recommend creating a separate zone for dynamic updates because BIND writes out the zone file when it changes, and it writes it out in a compact form without any formatting or comments you might have put in the original.
So, say your zone is test.com and you want to create dyn.test.com to hold the dynamic updates. Put the following lines in your named.conf file:
zone "dyn.test.com" {
type master;
file "db.dyn.test.com";
allow-update { 10.0.0.0/24; };
};
There are several salient features in this entry. First, the zone name, which is self-evident. The type is either master or slave. In this case we are authoritative for test.com and all its subdomains, so we are the master.
The filename can be whatever you want; it is created in the directory specified in the options section of your configuration file. The really important feature is the allow-update directive. Without this, the nameserver rejects all updates for the zone. I have specified the hosts in the IP/Netmask form. The number after the slash is the number of "one" bits in the netmask, counted from the left hand side. So, rendering the above in binary would give us:
00001010 00000000 00000000 00000000 host address
11111111 11111111 11111111 00000000 netmask
In English, this means that any IP address starting with 10.0.0 is allowed to update the zone.
If this discussion of host addresses and netmasks has you dizzy, don't worry. You aren't the first to be perplexed. However, I recommend that you spend some time learning about them, because if you deal with DNS or other Internet protocols on a regular basis, you will undoubtedly encounter them over and over again.
Now that you have created a new zone, you need to tell your nameserver where to get information on it. In the test.com zone file, you need to add the following line:
dyn IN NS nameserver.test.com
This assumes that test.com is already being served by nameserver.test.com and that you are creating dyn.test.com on the same machine. Now when people ask nameserver.test.com about dyn.test.com, it sends them a response which points right back at the same machine. Obviously that's redundant in this case, but remember that one of the key features of DNS is its distributed nature -- you could just have easily directed queries about dyn.test.com to a machine on the other side of the world. This process is known as delegation.
Incidentally, you can declare yourself authoritative for any zone, regardless of whether or not this is actually the case. What happens in this situation is that any clients using your nameserver for resolution will take whatever it says as gospel, whereas the rest of the net will get different information from the real authoritative nameservers. This could be extremely confusing, so if you are unsure whether you are authoritative for a zone, double check with the appropriate registry, or use dig to find the answer. dig is an excellent tool bundled with the BIND package. It is invaluable for diagnosing DNS problems.
$ dig test.com ; <<>> DiG 2.2 <<>> test.com ;; res options: init recurs defnam dnsrch ;; got answer: ;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 6 ;; flags: qr rd ra; Ques: 1, Ans: 1, Auth: 2, Addit: 2 ;; QUESTIONS: ;; test.com, type = A, class = IN ;; ANSWERS: test.com. 3568 A 207.206.9.99 ;; AUTHORITY RECORDS: test.com. 3571 NS mercury.xa.com. test.com. 3571 NS grail.BOOKS.com.
It turns out that the real world test.com is served by mercury.xa.com and grail.books.com. Replies coming from any other sources are non-authoritative. The reverse of this situation, where a server is designated authoritative but is not responding authoritatively for a request, is called a lame delegation. If you run named with the default logging levels, you will see plenty of lame delegation warning messages!
We're still not done setting up. In the example below, we are going to dynamically change the address of ftp.test.com. Except that to avoid messing up the test.com zone, we are really going to change ftp.dyn.test.com. We need to make an alias that maps ftp.dyn.test.com to ftp.test.com. In DNS jargon, such an alias is called a CNAME, which means "canonical name."
So put the following line in the test.com zone file:
ftp IN CNAME ftp.dyn.test.com.
It's time to restart the nameserver. I recommend using the ndc restart command. See your BIND documentation for details.
Now that we've configured a nameserver and delegated the zone, we can actually get down to writing some Perl code!
For this example, let's imagine that you have two FTP servers, ftp1.test.com and ftp2.test.com, both containing the same content. Their IP addresses are 10.0.0.1 and 10.0.0.2, respectively. You want users who come to ftp.test.com to end up on either of those machines, and you want the load distributed evenly between them. (Note that you don't need a script like this to achieve a round-robin distribution. That functionality is already built into BIND; this example is meant to be instructive rather than practical.) Here's the Perl script:
1 use strict;
2 use Net::DNS;
3
4 while (1) {
5 set_dns('10.0.0.1');
6 sleep 300;
7 set_dns('10.0.0.2');
8 sleep 300;
9 }
10
11 sub set_dns {
12 my $new_ip=shift;
13 my $update = new Net::DNS::Update('dyn.test.com');
14 $update->push('update', rr_del('ftp.dyn.test.com'));
15 $update->push('update',
rr_add("ftp.dyn.test.com 1800 A $new_ip"));
16
17 my $res = new Net::DNS::Resolver;
18 $res->nameservers('nameserver.test.com');
19 my $reply = $res->send($update);
20 if (defined $reply) {
21 if ($reply->header->rcode ne "NOERROR") {
22 print $reply->header->rcode,"\n";
23 } else {
24 print "Update to $new_ip succeeded!\n";
25 }
26 } else {
27 print 'Update failed: ', $res->errorstring, "\n";
28 }
29 }
In lines 4-8, we set up an infinite loop that changes the IP address and sleeps for five minutes. The interesting stuff happens in the set_dns() subroutine. In line 12, we grab the IP address that we're going to associate with ftp.dyn.test.com, and create a new Net::DNS::Update object in line 13. Lines 14 and 15 push data into the update packet.
Update packets are processed atomically -- once the nameserver determines that you are allowed to initiate your change, all of the requests in your packet are serviced in order, with no possibility of any other requestor changing data out from under you while your changes are being made. So, in our packet, we have a request to delete the current RR (resource record) set for ftp.dyn.test.com, followed by an immediate request to insert a new RR for ftp.dyn.test.com. The format of the string passed to rr_add() is a little odd, but should be familiar to anyone who's worked with zone files. The first field is the hostname, of course. The second field is the time-to-live for the entry, expressed in seconds. This one says that this record is valid for 300 seconds (5 minutes), at which point any nameserver caching it should throw it out and request a new one from another nameserver. (Unfortunately the DNS resolver clients in MacOS and Windows are broken to various degrees, and often ignore the timeouts. Sometimes it appears that your machine is fixated on a given IP address; the only way to get it to look up the new information is to reboot.)
Now we have a nice packet of data. Net::DNS::Update doesn't actually do anything with the data besides put it in a format that a nameserver recognizes. Now we have to bring in Net::DNS::Resolver. This is a somewhat misleading name, with roots in history. As I mentioned earlier, it used to be that nameservers were largely read-only. Clients that talked to DNS servers were known as resolvers, since resolving names to IP addresses was their exclusive function. Nowadays even though DNS clients can read and write, they are still called "resolvers."
In line 17, we create the resolver object, and in line 18, we tell it that the nameserver it will talk to is nameserver.test.com. Line 19 is where all the magic happens. The send() method sends our carefully crafted update packet off to the nameserver specified in the resolver object.
At this point, you should check your nameserver logfiles to see if there are any messages about your action. A common mistake, such as not setting the permissions in named.conf properly, results in an error message like this:
Jan 7 10:29:38 ns1 named[12691]: unapproved update from [205.216.163.199].3251 for dyn.test.com
If you don't see any errors, check the zone file directory for a file named dyn.test.com.log. This contains a log of your update requests, in a format like this:
[DYNAMIC_UPDATE] id 804 from [205.216.163.199].1541 at
915837961 (named pid 84):
zone: origin dyn.test.com class IN serial 199817937
update: {delete} ftp.dyn.test.com. IN
update: {add} ftp.dyn.test.com. 1800 IN A 10.0.0.1
Lines 20-28 do some error-checking on the response. If the update fails because the nameserver is unavailable or some other network-related error occurred, $reply is undefined, and an error message is available in $res->errorstring. On the other hand, if the nameserver accepts the packet, it puts a cryptic shorthand message indicating the status of the request in $reply->header->rcode. For example, NOERROR means the update was accepted and performed. Some other common return codes are listed below.
Common Nameserver Return Codes Return Code Meaning NOTAUTH This nameserver is not authoritative for the zone you requested. REFUSED The nameserver refused your request for security reasons. (Your IP address is probably not listed in named.conf.) NOTZONE A name in the update packet does not exist in the target zone.
At this point, your IP change has taken effect. Remember that due to the distributed nature of DNS, your change may not be instantly visible everywhere. Any client that previously looked up ftp.test.com will hang on to the old response until it expires, at which point it will request a new one.
As I mentioned, the above script is merely an example, in that each IP address is active for only five minutes. Although round-robin is quite useful as a basic method of load balancing, what happens when ftp1 crashes, or is taken down for routine maintenance? When you put it back up, it has zero users, but even though ftp2 has many users, new connections are still assigned in a round-robin fashion. You need a way to intelligently know how many connections are active and reassign the IP address to the least-loaded machine. If you alter the while loop at the top of the example program to query each machine and determine its load, you can then intelligently decide whether to change the IP address to ftp1.
As you've seen, the basics of dynamic update are easy to learn and implement. Using this framework you could write an application which queries several machines to find out which one is best able to service new requests, or monitor a pool of machines and take a crashed box out of rotation. Some other ideas include: having name.mydomain.com be your laptop no matter how it's connected, providing aliases for servers on a network with free resources say, (freedisk.domain.com, freecpu.domain.com, freeram.domain.com), or using mod_perl to configure Apache and the DNS server from one virtual host configuration file.
__END__