xinetd

ArticleCategory:

System Administration

AuthorImage:

TranslationInfo:

Original in fr Frédéric Raynal

fr to en Georges Tarbouriech 

AboutTheAuthor:

Fr�d�ric Raynal is preparing a thesis about watermarking of digital images at the INRIA (Institut National de Recherche en Informatique et Automatique).

Abstract

xinetd - extended Internet services daemon - provides a good security against intrusion and reduces the risks of Deny of Services (DoS) attacks. Like the well known couple (inetd+tcpd), it allows to fix the access rights for a given machine, but it can do much more. In this article we will discover its many features.

ArticleIllustration

security

ArticleBody

What's that, xinetd ?

The classical inetd helps to control network connections to a computer. When a request comes to a port managed by inetd, then inetd forwards it to a program called tcpd. Tcpd decides according to the rules contained in the hosts.{allow, deny} files whether or not the request is allowed. If the request is allowed the the corresponding server process can be started (e.g ftp). This mechanism is also referred to as tcp_wrapper.

xinetd provides access control capabilities similar to the ones provided by tcp_wrapper. However, its capabilities extend much further :

The main drawback, as already mentioned, concerns not weakly managed RPC calls. However, portmap and xinetd work fine .

The first part of this article explains how xinetd works. We'll spend some time on a service configuration, on some specific options (binding to an interface, redirection) and demonstrate this with the help of examples. The second part shows xinetd working, the logs it generates and finishes with a useful tip.

Compilation & Installation

You can get xinetd from www.xinetd.org. For this article we use version 2.1.8.9pre10.
Compilation and installation is done in the classical way: the usual commands ./configure; make; make install do it all :) configure supports the usual options. Three specific options are available at compile time:
  1. --with-libwrap : with this option, xinetd checks the configuration files for tcpd (/etc/hosts.{allow, deny}) and then if access is accepted, it uses its own control routines. For this option to work, tcp_wrapper and its libraries have to be installed on the machine (Author's note: what can be done with the wrapper can also be done with xinetd. Allowing this compatibility leads to multiplying the config files, and makes the administration heavier... in short, I don't recommend it);
  2. --with-loadavg : this option allows xinetd to handle the max_load configuration option. This allows to deactivate some services when the machine is overloaded. An options essential to prevent some DoS attacks (check the attribute max_load in table 1 ;
  3. --with-inet6 : if you feel like using IPv6, this option allows to support it. The IPv4 and IPv6 connections are managed, but IPv4 addresses are changed into IPv6 format.
Before starting xinetd, you don't have to stop inetd. Nevertheless, not to do so may lead to an unpredictable behavior of both daemons!

Some signals can be used to modify xinetd behavior:

There are a few others (let's mention a mistake in the documentation and the man pages: SIGHUP writes its dump in the file /var/run/xinetd.dump and not in /tmp/xinetd.dump), but the three above mentioned can be easily managed with a small script containing the start, stop, restart, soft, hard options (the last two respectively correspond to SIGUSR1 and SIGUSR2).

Configuration

The /etc/xinetd.conf file is the default configuration file for the xinetd daemon (a command line option allows to provide another one). If xinetd configuration is not very complex, it's a long work and the syntax is quite different from the one of its predecessor inetd.

Two utilities (itox and xconv.pl) are provided with xinetd and allow to convert the /etc/inetd.conf file into a configuration file for xinetd. Obviously, that's not enough since the rules specified in the wrapper configuration are ignored. The itox program is no longer developed if it's still maintained. xconv.pl is a better solution, even if the result has to be modified, since xinetd offers more features than inetd:

>>/usr/local/sbin/xconv.pl < /etc/inetd.conf > /etc/xinetd.conf
The configuration file begins with a default section. Attributes in this section will be used by every service xinetd manages. After that, you find as many sections as services, everyone of them being able to re-define specific options in relation to the default ones.

The default values section looks like:

defaults
{
    attribute operator value(s)
    ...
}
Each attribute defined in this section keeps the provided value(s) for all the next described services. Thus, the only_from attribute, allows to give a list of authorized addresses that should be able to connect to servers:
only_from = 192.168.1.0/24 192.168.5.0/24 192.168.10.17
Next, every declared service allows access to machines having an address contained in the list. However, these default values can be modified for each service (check the operators, explained a bit further down). Nevertheless, this process is a bit risky. As a matter of fact, to keep things simple and secure, it's much better not to define default values and change them later on within a service. For instance, talking about access rights, the simplest policy consists in denying access to everyone and next allowing access to each service to those who really need it (with tcp_wrapper, this is done from an hosts.deny file containing ALL:ALL@ALL, and an hosts.allow file only providing authorized services and addresses).

Each section describing a service in the config file looks like:

serviceservice_name
{
    attribute operator value(s)
    ...
}
Three operators are available: '=', '+=' and '-='. Most of the attributes only support the '=' operator. It gives a fix value to an attribute. The '+=' operator adds an item to a list of values, while the '-=' operator removes this item.

The table 1 briefly describes some of these attributes. We'll see how to use them with a few examples. Reading the xinetd.conf man page provides more information.
 
   

Attribute Values and description
flags Only the most current values are mentioned here, check the documentation to find new ones:
  • IDONLY : only accepts connexions from clients having an identification server;
  • NORETRY : avoids a new process to be forked again in case of failure;
  • NAMEINARGS : the first argument of the server_args attribute is used as argv[0] for the server. This allows to use tcpd by putting it in the server attribute, next writing the server name and its arguments such as server_args, as you would do with inetd.
log_type xinetd uses syslogd and the daemon.info selector by default.
  • SYSLOG selector [level] : allows to choose among daemon, auth, user or local0-7 from syslogd ;
  • FILE [max_size [absolute_max_size]] : the specified file receives information. The two options set the file size limit. When the size is reached, the first one sends a message to syslogd, the second one stops the logging for this service (if it's a common file - or fixed by default - then various services can be concerned).
log_on_success Different information can be logged when a server starts:
  • PID : the server's PID (if it's an internal xinetd service, the PID has then a value of 0) ;
  • HOST : the client address ;
  • USERID : the identity of the remote user, according to RFC1413 defining identification protocol;
  • EXIT : the process exit status;
  • DURATION : the session duration.
log_on_failure Here again, xinetd can log a lot of information when a server can't start, either by lack of resources or because of access rules:
  • HOST, USERID : like above mentioned ;
  • ATTEMPT : logs an access attempt. This an automatic option as soon as another value is provided;
  • RECORD : logs every information available on the client.
nice Changes the server priority like the nice command does.
no_access List of clients not having access to this service.
only_from List of authorized clients. If this attribute has no value, the access to the service is denied.
port The port associated to the service. If it's also defined in the /etc/services file, the 2 port numbers must match.
protocol The specified protocol must exist in the /etc/protocols file. If no protocol is given, the service's default one is used instead.
server The path to the server.
server_args Arguments to be given to the server.
socket_type stream (TCP), dgram (UDP), raw (IP direct access) or seqpacket ().
type xinetd can manage 3 types of services :
  1. RPC : for those defined in the /etc/rpc file... but doesn't work very well;
  2. INTERNAL : for services directly managed by xinetd (echo, time, daytime, chargen and discard) ;
  3. UNLISTED : for services not defined either in the /etc/rpc file, or in the /etc/services file ;
Let's note it's possible to combine various values, as we'll see with servers, services and xadmin internal services.
wait Defines the service behavior towards threads. Two values are acceptable:
  • yes : the service is mono-thread, only one connexion of this type can be managed by the service;
  • no : a new server is started by xinetd for each new service request according to the defined maximum limit (Warning, by default this limit is infinite).
cps Limits the number of incoming connexions. The first argument is this number itself. When the threshold is exceeded, the service is deactivated for a given time, expressed in seconds, provided with the second argument.
instances Defines the maximum number of servers of a same type able to work at the same time.
max_load This gives really the maximum load for a server (for example, 2 or 2.5). Beyond this limit, requests on this server are rejected.
per_source Either an integer, or UNLIMITED, to restrict the number of connexion from a same origin  to a server 
Tab. 1 : a few attributes for xinetd

The four last attributes shown in table1 allow to control the resources depending on a server. This is efficient to protect from Deny of Service (DoS) attacks (freezing a machine by using all of its resources)

This section presented a few xinetd features. The next sections show how to use it and give some rules to make it work properly.

Service defaults

The defaults section allows to set values for an number of attributes (check the documentation for the whole list). Some of these attributes (only_from, no_access, log_on_success, log_on_failure, ...) hold simultaneously the values allocated in this section and the ones provided in the services.

By default denying access to a machine,  is the first step of a reliable security policy. Next, it'll be enough to allow access according to the services. We've seen two attributes allowing to control access to a machine, based on IP addresses: only_from and no_access. Selecting the second one we write:

no_access = 0.0.0.0/0
fully blocks services access. However, if you wish to allow everyone to access echo (ping) for instance, you then should write in the echo service:
only_from = 0.0.0.0/0
Here is the logging you get with this configuration:
Sep 17 15:11:12 charly xinetd[26686]: Service=echo-stream: only_from list and no_access list match equally the address 192.168.1.1
As a matter of fact, the access control is done comparing the lists of addresses contained in both attributes. When the client address matches the 2 lists, the least general one is preferred. In case of equality, like in our example, xinetd is unable to choose and refuses the connexion.  To get rid of this ambiguity, you should have written:
only_from = 192.0.0.0/8
An easier solution is to only control the access with the attribute:
only_from =
Not giving a value makes every connexion fail :) Then, every service allows access by means of this same attribute.

Important, not to say essential: in case of no access rules at all (i.e. neither only_from, nor no_access) for a given service (allocated either directly or with the default) section, the access to the service is allowed!

Here is an example of defaults :

defaults
{
  instances       = 15
  log_type        = FILE /var/log/servicelog
  log_on_success  = HOST PID USERID DURATION EXIT
  log_on_failure  = HOST USERID RECORD
  only_from       =
  per_source      = 5

  disabled = shell login exec comsat
  disabled = telnet ftp
  disabled = name uucp tftp
  disabled = finger systat netstat

  #INTERNAL
  disabled = time daytime chargen servers services xadmin

  #RPC
  disabled = rstatd rquotad rusersd sprayd walld
}

among internal services, servers, services, and  xadmin allow to manage xinetd. More on this later.

Configuring a service

To configure a service, we need ...nothing :) In fact, everything works like it does with defaults values: you just have to precise the attributes and their value(s) to manage the service. This implies either a change in the defaults values or another attribute for this service.

Some attributes must be present according to the type of service (INTERNAL, UNLISTED ou RPC) :
 
 

Attribute Comment
socket-type Every service.
user Only for non INTERNAL services
server Only for non INTERNAL services
wait Every service.
protocol Every RPC service and the ones not contained in /etc/services.
rpc_version Every RPC service.
rpc_number Every RPC service, not contained in /etc/rpc.
port Every non RPC service, not contained in /etc/services.
Tab. 2: required attributes

This example shows how to define services:

service ntalk
{
  socket_type   = dgram
  wait          = yes
  user          = nobody
  server        = /usr/sbin/in.ntalkd
  only_from     = 192.168.1.0/24
}

service ftp
{
  socket_type  = stream
  wait         = no
  user         = root
  server       = /usr/sbin/in.ftpd
  server_args  = -l
  instances    = 4
  access_times = 7:00-12:30 13:30-21:00
  nice         = 10
  only_from    = 192.168.1.0/24
}

 Let's note these services are only allowed on the local network (192.168.1.0/24). Concerning FTP, some more restrictions are expected: only 4 instances are allowed and it'll be available only during segments of time.

Port binding: the bind attribute

It allows to bind a service to an IP address. If a machine has only one address, this is useless. On the other hand, a computer beeing part of a local network and connected to Internet has at least two addresses.

For instance, a company wishes to install an FTP server for its employees (to access and read internal documentation). This company wants to provide its clients with an FTP access towards its products: bind has been made for this company :) Let's define two FTP services. However, xinetd must be able to differentiate them: the solution is the id attribute. It defines a service in a unique way (when not defined within a service, its value defaults to the name of the service).

service ftp
{
  id           = ftp-public
  wait         = no
  user         = root
  server       = /usr/sbin/in.ftpd
  server_args  = -l
  instances    = 4
  nice         = 10
  only_from    = 0.0.0.0/0 #allows every client
  bind         = 212.198.253.142 #public IP address for this server
}

service ftp
{
  id           = ftp-private
  socket_type  = stream
  wait         = no
  user         = root
  server       = /usr/sbin/in.ftpd
  server_args  = -l
  only_from    = 192.168.1.0/24 #only for internal use
  bind         = 192.168.1.1  #local IP address for this server (charly)
}

The use of bind will allow to call the corresponding daemon, according to the destination of the packets. Thus, with this configuration, a client on the local network must give the local address (or the associated name) to access internal data. In the log file, you can read:
00/9/17@16:47:46: START: ftp-public pid=26861 from=212.198.253.142
00/9/17@16:47:46: EXIT: ftp-public status=0 pid=26861 duration=30(sec)
00/9/17@16:48:19: START: ftp-internal pid=26864 from=192.168.1.1
00/9/17@16:48:19: EXIT: ftp-internal status=0 pid=26864 duration=15(sec)
The first part comes from the command ftp 212.198.253.142, while the second part is about the command from charly to itself: ftp 192.168.1.1.

Obviously, there's a problem: what happens if a machine doesn't have two static IP addresses? This can happen with ppp connections or when using the dhcp protocol. It seems it would be much better to bind services to interfaces than to addresses. However, this is still not expected in xinetd and is a real problem (for instance, writing a C module to access an interface or address depends on the OS, and since xinetd is supported on many OSes...). Using a script allows to solve the problem:

#!/bin/sh

PUBLIC_ADDRESS=`/sbin/ifconfig $1 | grep "inet addr" | awk '{print $2}'| awk -F: '{print $2}'`
sed s/PUBLIC_ADDRESS/"$PUBLIC_ADDRESS"/g /etc/xinetd.base > /etc/xinetd.conf

This script takes the /etc/xinetd.base file, containing the desired configuration with  PUBLIC_ADDRESS as a replacement for the dynamic address, and changes it in /etc/xinetd.conf, modifying the PUBLIC_ADDRESS string with the address associated to the interface passed as an argument to the script. Next, the call to this script depends on the type of connection: the simplest is to add the call into the right ifup-* file and to restart xinetd.

Service redirection towards an other machine: the redirect attribute

xinetd can be used as a transparent proxy, sort of (well, almost ... as we'll see it later) with the redirect attribute. It allows to send a service request towards an other machine to the desired port.
telnet service
{
  flags  = REUSE
  socket_type = stream
  wait  = no
  user  = root
  server = /usr/sbin/in.telnetd
  only_from = 192.168.1.0/24
  redirect = 192.168.1.15 23
}
Let's watch what's going on now:
>>telnet charly
Trying 192.168.1.1...
Connected to charly.
Escape character is '^]'.
 

Digital UNIX (sabrina) (ttyp1)

login:

As a matter of fact, the connection seems to be established on charly, but the following shows that sabrina (an alpha machine, hence "Digital UNIX") took over. This mecanism can be both useful and dangerous. When setting it up, logging must be done on both ends of the connection. Furthermore, for this type of service, the use of DMZ and firewall is strongly recommended;-)

Special services

Three services only belong to xinetd. Since these services can't be found in /etc/rpc or /etc/services, they must have the UNLISTED flag ( besides the INTERNAL flag informing they are xinetd services)
  1. servers: informs about servers in use ;
  2. services: informs about available services, their protocol and their port ;
  3. xadmin: mixes the functions of the two previous ones.
Obviously, these services make your computer more vulnerable.. They provide important information. Presently, their access is not protected (password protected, for instance). You should use them only at configuration time. Next, in the defaults section, you must deny their use:
defaults {
  ...
  disabled = servers services xadmin
  ...
}
Before activating them, you should take some precautions:
  1. The machine running xinetd must be the only one able to connect to these services
  2. Limit the number of instances to one
  3. Allow access only from the machine running the server.
Let's take the example of the xadmin service (the two others can be configured in the same way, apart from the port number ;-) :
service xadmin
{
  type  = INTERNAL UNLISTED
  port  = 9100
  protocol = tcp
  socket_type = stream
  wait  = no
  instances = 1
  only_from = 192.168.1.1  #charly
}
The xadmin service has 5 commands :
  1. help ...
  2. show run : like the servers service, informs about the presently running servers
  3. show avail : like the services service, informs about the available services (and a bit more)
  4. bye or exit ...
Now, you know they exist: forget them ;-) You can test without these services. Commands such as (netstat, fuser, lsof, ... allow you to know what's going on on your machine, without making it vulnerable as you would when using these services!

Let's play a bit...

Starting with a riddle

Here is a small exercise for the ones who survived ;-) First I will explain configuration used in this exercise and then we will try to find out what happens and why it does not work.

We only need the finger service :

finger service
{
  flags  = REUSE NAMEINARGS
  server = /usr/sbin/tcpd
  server_args = in.fingerd
  socket_type = stream
  wait  = no
  user  = nobody
  only_from = 192.168.1.1  #charly
}
xinetd wasn't compiled with the --with-libwrap option (check the attribute server). The defaults section is of the same kind of the one previously provided: every access to charly is denied wherever the connexion comes from. The finger service is not deactivated, nevertheless:
pappy@charly >> finger pappy@charly
[charly]
pappy@charly >>

pappy@bosley >>  finger pappy@charly
[charly]

pappy@bosley >>


It seems the request didn't work properly, neither from charly (192.168.1.1), an authorized machine, nor from bosley (192.168.1.10). Let's have a look at the log files:

/var/log/servicelog :
00/9/18@17:15:42: START: finger pid=28857 from=192.168.1.1
00/9/18@17:15:47: EXIT: finger status=0 pid=28857 duration=5(sec)
00/9/18@17:15:55: FAIL: finger address from=192.168.1.10
The request from charly (the two first lines) works properly according to xinetd: the access is allowed and the request takes 5 seconds. On the other hand, the request from bosley is rejected (FAIL).
If we look at the configuration of the finger service, the server used is not really in.fingerd, but the tcp_wrapper tcpd service. The wrapper log says:
/var/log/services :
Sep 18 17:15:42 charly in.fingerd[28857]: refused connect from 192.168.1.1
We see that there's only one line matching our two queries! The one from bosley (the second one) was intercepted by xinetd, so it's quite normal not to find it in that log. The selected line really corresponds to the request xinetd allowed, sent from charly to charly (the first one): time and PID are identical.

Let's summarize what we have:

  1. xinetd allowed the request;
  2. the finger request goes through tcpd ;
  3. in.fingerd rejected this request.
What's going on, then? Since the request is accepted by xinetd, it's sent to the specified server (here tcpd). Nevertheless, tcpd rejects this connection. Then, we must have a look at hosts.{allow,deny}. The /etc/hosts.deny file only contains ALL:ALL@ALL, what explains why the request has been rejected by the wrapper!

According to the way the server and server_args service lines have been defined, the wrapper features are still accessible (banner - there's a banner attribute in xinetd-, spawn, twist, ...). Remember that the --with-libwrap compilation option only adds access rights control (with the help of hosts.{allow,deny} files), before xinetd process starts. In this example we saw that this configuration allows us to continue using the tcp wrapper features.

This overlapping of features, if it can work, may as well lead to stange behaviors. To use xinetd together with inetd and portmap, it's much better to manage a service with only one of these "super-daemons".

chroot a service

It's often suggested to restrict the fields of some services, or to create a new environment. The chroot command allows to change the root directory for a command (or a script):
chroot [options] new_root
This is often used to protect services such as bind/DNS or ftp. To duplicate this behavior while benefiting from xinetd features, you have to declare chroot as a server. Then, you just have to pass other arguments via the server_args attribute :)
service ftp
{
  id           = ftp
  socket_type  = stream
  wait         = no
  user         = root
  server       = /usr/sbin/chroot
  server_args  = /var/servers/ftp /usr/sbin/in.ftpd -l
}
Thus, when a request is sent to this service, the first instruction used is chroot. Next, the first argument passed to it is the first one on the server_args line, that is the new root. Last, the server is started.

Conclusion

You could now ask which daemon should I choose xinetd or inetd. As a matter of fact, xinetd requires a bit more administration, especially as long as it won't be included into distributions (it is in Red Hat 7.0). The most secure solution is to use xinetd on machines with public access (like Internet) since it offers a better defense. For machines within a local network inetd should be enough.