12.1. The inetd Super Server

Programs that provide application services via the network are called network daemons. A daemon is a program that opens a port, most commonly a well-known service port, and waits for incoming connections on it. If one occurs, the daemon creates a child process that accepts the connection, while the parent continues to listen for further requests. This mechanism works well, but has a few disadvantages; at least one instance of every possible service you wish to provide must be active in memory at all times. In addition, the software routines that do the listening and port handling must be replicated in every network daemon.

To overcome these inefficiencies, most Unix installations run a special network daemon, what you might consider a “super server.” This daemon creates sockets on behalf of a number of services and listens on all of them simultaneously. When an incoming connection is received on any of these sockets, the super server accepts the connection and spawns the server specified for this port, passing the socket across to the child to manage. The server then returns to listening.

The most common super server is called inetd, the Internet Daemon. It is started at system boot time and takes the list of services it is to manage from a startup file named /etc/inetd.conf. In addition to those servers, there are a number of trivial services performed by inetd itself called internal services. They include chargen, which simply generates a string of characters, and daytime, which returns the system's idea of the time of day.

An entry in this file consists of a single line made up of the following fields:
service type protocol wait user server cmdline

Each of the fields is described in the following list:

service

Gives the service name. The service name has to be translated to a port number by looking it up in the /etc/services file. This file will be described later in this chapter in the section Section 12.3.”

type

Specifies a socket type, either stream (for connection-oriented protocols) or dgram (for datagram protocols). TCP-based services should therefore always use stream, while UDP-based services should always use dgram.

protocol

Names the transport protocol used by the service. This must be a valid protocol name found in the protocols file, explained later.

wait

This option applies only to dgram sockets. It can be either wait or nowait. If wait is specified, inetd executes only one server for the specified port at any time. Otherwise, it immediately continues to listen on the port after executing the server.

This is useful for “single-threaded” servers that read all incoming datagrams until no more arrive, and then exit. Most RPC servers are of this type and should therefore specify wait. The opposite type, “multi-threaded” servers, allow an unlimited number of instances to run concurrently. These servers should specify nowait.

stream sockets should always use nowait.

user

This is the login ID of the user who will own the process when it is executing. This will frequently be the root user, but some services may use different accounts. It is a very good idea to apply the principle of least privilege here, which states that you shouldn't run a command under a privileged account if the program doesn't require this for proper functioning. For example, the NNTP news server runs as news, while services that may pose a security risk (such as tftp or finger) are often run as nobody.

server

Gives the full pathname of the server program to be executed. Internal services are marked by the keyword internal.

cmdline

This is the command line to be passed to the server. It starts with the name of the server to be executed and can include any arguments that need to be passed to it. If you are using the TCP wrapper, you specify the full pathname to the server here. If not, then you just specify the server name as you'd like it to appear in a process list. We'll talk about the TCP wrapper shortly.

This field is empty for internal services.

A sample inetd.conf file is shown in Example 12-1. The finger service is commented out so that it is not available. This is often done for security reasons, because it can be used by attackers to obtain names and other details of users on your system.

Example 12-1. A Sample /etc/inetd.conf File

#
# inetd services
ftp      stream tcp nowait root  /usr/sbin/ftpd    in.ftpd -l
telnet   stream tcp nowait root  /usr/sbin/telnetd in.telnetd -b/etc/issue
#finger  stream tcp nowait bin   /usr/sbin/fingerd in.fingerd
#tftp    dgram  udp wait  nobody /usr/sbin/tftpd   in.tftpd
#tftp    dgram  udp wait  nobody /usr/sbin/tftpd   in.tftpd /boot/diskless
#login   stream tcp nowait root  /usr/sbin/rlogind in.rlogind
#shell   stream tcp nowait root  /usr/sbin/rshd    in.rshd
#exec    stream tcp nowait root  /usr/sbin/rexecd  in.rexecd
#
#       inetd internal services
#
daytime  stream tcp nowait root internal
daytime  dgram  udp nowait root internal
time     stream tcp nowait root internal
time     dgram  udp nowait root internal
echo     stream tcp nowait root internal
echo     dgram  udp nowait root internal
discard  stream tcp nowait root internal
discard  dgram  udp nowait root internal
chargen  stream tcp nowait root internal
chargen  dgram  udp nowait root internal

The tftp daemon is shown commented out as well. tftp implements the Trivial File Transfer Protocol (TFTP), which allows someone to transfer any world-readable files from your system without password checking. This is especially harmful with the /etc/passwd file, and even more so when you don't use shadow passwords.

TFTP is commonly used by diskless clients and Xterminals to download their code from a boot server. If you need to run tftpd for this reason, make sure to limit its scope to those directories from which clients will retrieve files; you will need to add those directory names to tftpd's command line. This is shown in the second tftp line in the example.