The pdnsd FAQ

There are complete and well-tested name servers around, such as the BIND. These do also perform caching. Why should I use pdnsd?

pdnsd does not aim to be a complete name server implementation, such as the BIND. It is optimized for caching, and you can only specify a small subset of all dns record types pdnsd knows in your local "zone" definitions. This of course reduces the code size drastically, and such the memory footprint. There are some features especially interesting for dialin networks, ordinary (non-server) internet hosts and computers that are often not connected to to their network, e.g. notebooks (I originally wrote this program for use with my notebook). These features are:

permanent disk cache (useful for frequent power-offs/reboots)
usually smaller memory footprint (depends on cache size) (see next question)
offline-detection prevents hangs (e.g. the typical hang on startup of some Netscape Navigator versions if not dialled in)
better control about timeouts (also to prevent hangs)
better control over the cache
better run-time control

When I look at the process size with ps, top, gtop, or a similar tool, I see some processes with a total size well above 3.5 MB. This is much more than e.g. BIND named (about 1.4 MB). Why?

Really, it is not. pdnsd uses multithreading, not multiprocessing. That means that the processes share most of their process space. In the LinuxThreads library or NPTL (Native Posix Thread Libary), which are used by pdnsd on Linux, in fact the total process address space is shared (although the processes have different stacks, these are in one process address space). You may check this by looking at the at the process sizes of the pdnsd threads: all should be the same. The effective size that pdnsd occupies is thus the size of any of the processes, not the sum of those. So, pdnsd with empty cache occupies about 800 kB, and the maximum size should be about the cache size plus this size (in fact, ca 5-10% more).

What do I need the status control (option -s) for?

It enables you to do some things you might or might not need. With it, you can:

query pdnsd's settings at runtime to debug configuration files and see which servers are regarded to be available
mark servers as available or unavailable, or force a status retest - very handy if you want to control which servers pdnsd queries, e.g for muliple dial-up accounts
delete, invalidate or add DNS records - useful e.g. when you want to build records for dynamically assigned IP addresses or domain names
reload pdnsd's configuration file without restarting pdnsd
print information about the contents of pdnsd's cache.

Q: What do I need local records (rr- and source-sections in the config file) for?

A: Some resolver programs, e.g. nslookup, want to look up the name of the server they are using before doing anything else. This option is for defining a PTR record for your IP such that those programs get an answer even if the name server you are caching is not available or does not offer these records. By extension, you may also define A and SOA records. This allows you to build very small zones without having to use a "big" name server. It is NOT intended to replace such a complete server in anything but VERY small networks. Alternatively, you may start a named on another host or on the same host on another port and cache it with pdnsd in addition to other (more distant) name servers.
The source section allows you to let pdnsd read in your /etc/hosts file on startup and serve its contents. This file is used by your local resolver before it even tries the name servers and usually contains fully-qualified domain names (FQDNs) for all of the internet addresses your host has. If you source this file, you usually won't need any additional rr sections. Sourcing it also allows other hosts (eg. in your local network) to access the names defined in your hosts file. You can of course just add other hosts in your local network to the servers hosts file, thus making them known to your server's resolver and pdnsd (if you sourced that file).
If you don't know what this answer was all about, you should just take the source section in the sample config file that comes with pdnsd, copy it into your config file and forget about it.

Q: When compiling, I get an error message like
Please define __BYTE_ORDER to be __LITTLE_ENDIAN or __BIG_ENDIAN
What's up?

A: Normally, this macros should be defined in your C library's header files. There are two different methods, most C libraries support both (and pdnsd honors both): either __BYTE_ORDER is set to __LITTLE_ENDIAN or __BIG_ENDIAN, or __LITTLE_ENDIAN or __BIG_ENDIAN are directly defined as macros.
Linux glibc, for example, does set those macros correctly. Never mind. You just have to know whether your machine is little-endian or big-endian, this means wheter your machine saves the least significant byte of a word or double-word first in memory (little-endian) or the most significant first (big-endian). All intel x86 and Alpha machines are little-endian, for example, while SPARC and PowerPC architectures are big-endian. If your machine is little-endian, add the following line to your config.h:
#define __BYTE_ORDER __LITTLE_ENDIAN
Likewise, if your machines byte order is big-endian:
#define __BYTE_ORDER __BIG_ENDIAN
Pathological byte orders like pdp-endian are not yet supported really; However, for the place the endianess is needed, __LITTLE_ENDIAN should do (it deals only with 16 bits; for all other occurances, ntoh[sl]/hton[sl] is used).

Q: At startup, I get a warning saying:
Uptest command [...] will implicitly be executed as root
What does that mean?

A: This warning only occurs if you use the uptest=exec option in your configuration. It means that the uptest command is run as root because pdnsd is running as root, and this was not explicitly specified. The idea is that it may introduce security holes (in the programs being run) when they run as root, and so they shouldn't do that if possible. You can specify the user that shall run the command by appending its name comma-separated as string to the uptest_cmd line:
uptest_cmd="<your command>","<user>";
If it is correctly running as root, just append the user string "root" to the command and the warning will not occur again.

Q:	I cannot run my `uptest_cmd` command as root (it says permission denied), although the pdnsd executable is setuid root. Why?
A:	pdnsd will drop privileges gained through setuid/setgid before executing the uptest commands (you shouldn't set the pdnsd executable setuid/setgid anyway). The reason is clear: if you install the pdnsd executable as setuid root and this wouln't be done, any user could execute shellcode with root privileges using that option!

Q: At startup, I get an error saying:
Bad config file permissions: the file must be only writeable by the user
Why is that?

A: pdnsd has an option (uptest=exec) that allows the execution of arbitrary shell code (for testing whether an interface is up). This must be of course secured against unauthorized use. One of these protection is the one that produces the error message: if you routinely run pdnsd, e.g. at system startup, and your config file is editable for others, someone could change it and insert shell code that is executed in the next pdnsd run -- with your user privileges! To prevent this, pdnsd will exit if the config file is writeable by others than the owner. To get rid of this message, just do
chmod go-w <filename>
on your config file (for the default file: chmod go-w /etc/pdnsd.conf). You should also check that the ownership is set correct.

Q:	`serve_aliases` does not seem to work.
A:	Some resolvers (e.g. of the glibc 2.1) seem sometimes not to look up unmodified names, but the names with an entry of the search path already appended. Since pdnsd will serve short names with this option anyway, you can delete the search an domain options from your /etc/resolv.conf. This is reported to work in some cases.

Q:	Some queries for domains that have many records (e.g. www.gmx.de) fail mysteriously.
A:	pdnsd versions prior to 1.1.0 had the tcp server thread disabled by default. Most resolvers repeat their query using tcp when they receive a truncated answer (the answer is truncated when it exceeds a length of 512 bytes). You need to recompile pdnsd with the option `--enable-tcp-server` to fix this.

Q: I am behind some kind of firewall. In the configuration file I have only listed addresses of name servers on the local (ISP's) network, but pdnsd is slow and DNS queries frequently time out.

A: In some cases pdnsd will not consider the answer of the local name server authoritative enough, and will try to get answers from the name servers listed in the authority section of the reply message. If pdnsd is behind a firewall that blocks the UDP reply packets from remote name servers, pdnsd will wait in vain for a reply. One solution is to set proxy_only=on in the servers sections of the configuration file. This will prevent pdnsd from querying name servers that are not listed in the configuration file. Another solution that can be tried is specifying query_method=tcp_only in the global section of the configuration file, because a firewall that blocks UDP packets from outside might still allow outgoing TCP connections to port 53.

Q: Is pdnsd vulnerable to DNS cache poisoning as described in CERT vulnerability note VU#800113?

A: Short answer: Yes.
Somewhat longer answer: The problem is not so much that pdnsd's implementation is flawed but rather that the DNS protocol currently being used is fundamentally flawed from a security viewpoint. As long as a more secure protocol is not in place, all that the developers of pdnsd can do is to try to tweak the current implementation to make it as difficult as possible for an attacker to succeed.
From version 1.2.7 onwards, the default for the query_port_start option is 1024, which means that the pdnsd resolver will randomly select source ports in the range 1024-65535. (In previous versions the default was to let the kernel select the source ports, which will often result in a more or less predictable sequence of ports.) It also helps to use a good quality source of random numbers. On platforms where this is supported, it is preferable to configure with --with-random-device=/dev/urandom. There is still more that can be done to make pdnsd less vulnerable, but this remains (as of this writing) a work in progress.
Please note that pdnsd was designed for small (private) networks, and that it is generally not recommended to let untrusted users access pdnsd.

Thomas Moestl and Paul Rombouts

Last revised: 18 August 2008 by Paul Rombouts