David Mertz, Ph.D.
Professional Neophyte
February, 2006
Welcome to "System Security", the sixth of seven tutorials covering intermediate network administration on Linux. This tutorial touches on a wide array of topics related to using Linux as a security-conscious network server, of necessity each somewhat cursorily. General issues of routing, firewalls and NAT translation are discussed and the relevant tools presented. Setting security policies for FTP and SSH are addressed also. General access control withtcpd,hosts.allowand friends is reviewed. Finally, some basic scurity monitoring tools are presented, as well as where to find security resources.
The Linux Professional Institute (LPI) certifies Linux system administrators at junior and intermediate levels. There are two exams at each certification level. This series of seven tutorials helps you prepare for the second of the two LPI intermediate level system administrator exams--LPI exam 202. A companion series of tutorials is available for the other intermediate level exam--LPI exam 201. Both exam 201 and exam 202 are required for intermediate level certification. Intermediate level certification is also known as certification level 2.
Each exam covers several or topics and each topic has a weight. The weight indicate the relative importance of each topic. Very roughly, expect more questions on the exam for topics with higher weight. The topics and their weights for LPI exam 202 are:
Topic 205: Network Configuration (8) Topic 206: Mail and News (9) Topic 207: Domain Name System (DNS) (8) Topic 208: Web Services (6) Topic 210: Network Client Management (6) Topic 212: System Security (10) * Topic 214: Network Troubleshooting (1)
Welcome to "System Security", the sixth of seven tutorials covering
intermediate network administration on Linux. This tutorial touches on
a wide array of topics related to using Linux as a security-conscious
network server, of necessity each somewhat cursorily. General issues
of routing, firewalls and NAT translation are discussed and the
relevant tools presented. Setting security policies for FTP and SSH
are addressed also. General access control with tcpd, hosts.allow
and friends is reviewed (reiterating the discussion in LPI 201, Topic
209: File Sharing Servers). Finally, some basic scurity monitoring
tools are presented, as well as where to find security resources.
To get the most from this tutorial, you should already have a basic knowledge of Linux and a working Linux system on which you can practice the commands covered in this tutorial.
As with most Linux tools, it is always useful to examine the manpages for any utilities discussed. Versions and switches might change between utility or kernel version, or with different Linux distributions. For more in depth information, the Linux Documentation Project has a variety of useful documents, especially its HOWTOs. See http://www.tldp.org/. A variety of books on Linux networking have been published, consult your local bookseller or library for latest titles on Linux security issues.
The Linux kernel includes "netfilter" infrastructure, which gives you
the the capability of filtering network packages. Usually this
capability is compiled into the base kernel, but a kernel module may
be needed for the capability. However, module loading should be
seamless: i.e. running iptables will load iptables_filter.o it it
needs it. Packet filtering is controlled with the utility iptables
in modern Linux systems; older systems used ipchains, and before
that ipfwadm. While you can still use ipchains in conjunction
with recent kernels if backward compatibility is needed, you will
almost always prefer to use the enhanced capabilities and improved
syntax in iptables. That said, most of the concepts and switches in
iptables are compatible enhancements to ipchains.
Depending on the exact scenario of filtering (firewall, NAT, etc.)
filtering and address translation may occur either before or after
routing itself. The same ipchains tool is used in either case, but
different rule sets ("chains") are used for the cases: at base, INPUT
and OUTPUT. Filtering, however, can also affect the routing decision
by filtering on the FORWARD chain; this may lead to dropping packets
rather than routing them.
As well as filtering with iptables (or legacy ipchains), the Linux
kernel performs routing of IP packets it receives. Routing is a
simpler process than filtering, though the two are conceptually
related. A host, during routing, simply looks at a destination IP
address, and decides whether it knows how deliver to a packet directly
to that address, or whether a gateway is available that knows how to
deliver to that address. If a host can neither deliver a packet
itself nor knows what gateway to forward it to, the packet is dropped.
However, typical configurations include a "default gateway" that
handles every otherwise unspecified address.
Configuration and display of routing information is peformed with the
utility route. However, routing may either be static or dynamic.
With static routing, delivery is determined by a routing table that is
explicitly configured by invocations of the route command and its
add or del commands. However, often more useful is configuring
dynamic routing using the routed or gated daemons that broadcast
routing information to adjacent routing daemons. The routed daemon
supports the Routing Information Protocol (RIP); the gated daemon
adds support for a number of other protocols (and can use multiple
protocols at once): Routing Information Protocol Next
Generation (RIPng), Exterior Gateway Protocol (EGP), Border Gateway
Protocol (BGP) and BGP4+, Defense Communications Network Local-Network
Protocol (HELLO), Open Shortest Path First (OSPF), Intermediate System
to Intermediate System (IS-IS), and Internet Control Message Protocol
(ICMP and ICMPv6)/Router Discovery.
Let us take a look at a fairly typical static routing table:
% /sbin/route Kernel IP routing table Destination Gateway Genmask Flags Metric Ref Use Iface 66.98.217.0 * 255.255.255.0 U 0 0 0 eth0 10.10.12.0 * 255.255.254.0 U 0 0 0 eth1 66.98.216.0 * 255.255.254.0 U 0 0 0 eth0 169.254.0.0 * 255.255.0.0 U 0 0 0 eth1 default ev1s-66-98-216- 0.0.0.0 UG 0 0 0 eth0
What this means is that addresses in the 66.98.217/24 and 66.98.216/23
ranges will be directly delivered over eth0. Address ranges
10.10.12/23 and 169.254/16 will be delivered on eth1. Anything left
over will be sent to the gateway ev1s-66-98-216-1.ev1servers.net
(the name is cut off in the route display; you could also use route
-n to see that name was IP address 66.98.216.1). If you wanted to
add a different gateway for some other address ranges, you might run
something like:
% route add -net 192.168.2.0 netmask 255.255.255.0 gw 192.168.2.1 dev eth0
For a machine that serves as a gateway itself, you will generally want
to run dynamic routing, using the routed or gated daemons, which
may supplement a smaller number of static routes. The routed daemon
is configured by the contents of /etc/gateways. The gated daemon
is more modern, and has more capabilities as indicated, and is
configured by /etc/gated.conf. Generally if you use either of
these, you will want to launch them in your startup scripts. You
must not run both routed and gated on the same machine, results
will be unpredictable, and almost certainly not desirable.
iptables
The Linux kernel stores a table of filter rules for IP packets that form a sort of state-machine. Sets of rules that are processed in sequence are know as "(firewall) chains". When one chain meets a condition, one of the possible actions is to shift control to processing another chain, as in a state-machine. Before you have added any rules or states, three chains are automatically present: INPUT, OUTPUT and FORWARD. The INPUT chain is where packets addressed to the host machine passes, and potentially from there to a local application process. The FORWARD chain is where a packet addressed to a different machine passes, assuming forwarding is enabled and the routing system knows how to forward that packet. A packet generated on the local host is sent into the OUTPUT chain for filtering--if it passes the filters in the OUTPUT chain (or any linked chains), it is routed out over its network interface.
One action that a rule can take is to DROP a packet; in that case,
naturally, no further rule processing or state transition is taken for
that packet. But if a packet is not dropped, the next rule in a chain
is examined to see if it matches the packet. In some cases,
satisfaction of a rule will branch processing to a different chain,
and its set of rules. Creation, deletion or modification of rules, and
of chains in which rules live is performed with the tool iptables.
In older Linux systems, the same function was done using ipchains
instead. The concepts behind both tools, and even for the ancient
ipfwadm are similar, but iptables syntax is discussed here.
A rule specifices a set of conditions that a packet might meet, and
what action to take if the packet does meet that condition. As
mentioned, one common action is to DROP packets. For example, suppose
you wanted (for some reason) to disable ping on the loopback
interace (i.e. the ICMP interface). You could enable this with:
% iptables -A INPUT -s 127.0.0.1 -p icmp -j DROP
Of course, that is a silly rule, and we probably want to remove it after we test it, e.g.:
% iptables -D INPUT -s 127.0.0.1 -p icmp -j DROP
Deleting a rule with the -D option requires either exactly the same
options as specified when it was added, or specification by rule
number (which you, therefore, must determine first), e.g.:
% iptables -D INPUT 1
A more interesting rule might look at source and destination addresses in packets. For example, suppose that a problem remote network is trying to utilize services on a particular subnet of your network. You might block this on your gateway/firewall machine with, e.g.:
% iptables -A INPUT -s 66.98.216/24 -d 64.41.64/24 -j DROP
Doing this will stop anything from the 66.98.216.* IP block from
communicating with anything in the local 64.41.64.* subnet. Of
course, singling out a specific IP block for blacklisting is fairly
limited as protection. A more likely scenario might be to allow
only a specific IP block to access a local subnet, e.g.:
% iptables -A INPUT -s ! 66.98.216/24 -d 64.41.64/24 -j DROP
In this case, only the 66.98.216.* IP block can access the
specified subnet. Moreover, you can use a symbolic name for a address,
and can specify a particular protocol to be filtered. You can also
select a specific network interface (e.g. eth0) to filter, but that
is less commonly useful. For example, to let only a specific remote
network access a local web server, you might use:
% iptables -A INPUT -s ! example.com -d 64.41.64.124 -p TCP -sport 80 -j DROP
There are a number of other options you can specify with iptables,
including for example rate limits on the number of packets that will
be allowed, or filtering on TCP flags. See the manpage for
iptables for more details.
We have seen the basics of adding rules to the automatic chains. But
much of the configurability in iptables comes with adding user
defined chains, and branching to them if patterns are matched. New
chains are defined with the -N option, and branching we have already
seen, using the special target DROP. ACCEPT is also a special target
with the obivous meaning. Also, special targets RETURN and QUEUE are
available. The first means to stop processing a given chain, and
return to its parent/caller. The QUEUE handler lets you pass packets
to a user space process for further processing (which might be
logging, modification of the packet, or more elaborate filtering than
iptables supports). The simple example in Rusty Russell's "Linux
2.4 Packet Filtering HOWTO" is a good example of adding a user defined
chain:
# Create chain to block new connections, except established locally % iptables -N block % iptables -A block -m state --state ESTABLISHED,RELATED -j ACCEPT % iptables -A block -m state --state NEW -i ! ppp0 -j ACCEPT % iptables -A block -j DROP # DROP everything else not ACCEPT'd # Jump to that chain from INPUT and FORWARD chains % iptables -A INPUT -j block % iptables -A FORWARD -j block
Notice that the "block" chain ACCEPTs in a limited class of cases, then the final rule DROPs everything not previously ACCEPT'd.
Once you have established some chains, whether adding rules to the
automatic chains or adding user defined chains, you may use the -L
option to view the current rules.
The examples we have looked at are basically in the class of firewall
rules. But Network Address Translation (NAT) is also configured by
iptables. Basically, NAT is a way of using connection tracking to
masquerade packets coming from a local subnet address as the external
WAN address before sending them out "over the wire" (on the OUTPUT
chain). The gateway/router that performs NAT needs to remember which
local host connected to which remote host, and reverse the address
translation if packets arrive back from the remote host. From a
filtering perspective though, you simply pretend that NAT did not
exist. The rules you specify should simply use the "real" local
addresses, regardless of how NAT might masquerade them to the outside
world. Enabling masquerading, i.e. basic NAT, just uses the below
iptables command. To use this you will need to make sure the kernel
module iptables_nat is loaded, and also turn on IP forwarding.
E.g.:
% modprobe iptables_nat # Load the kernel module % iptables -t nat -A POSTROUTING -o eth0 -j MASQUERADE % echo 1 > /proc/sys/net/piv4/ip_forward # Turn on IP forwarding
This capability is what is actually called "Source NAT". That is, the address of the outgoing packet is modified. "Destination NAT" (DNAT) also exists to enable port forwarding, load sharing, and transparent proxying. In those cases, incoming packets are modified to get to the relevent local host or subnet. But most of the time when users, or even administrators, talk about NAT, they mean source NAT. If you mean to configure destination NAT, you would specify "PREROUTING" rather than "POSTROUTING". That is for DNAT, the packets are transformed before they are routed.
Many different FTP servers are available for Linux, and different
distributions distribute different servers. Naturally, configuration
of different servers vary, though most tend to follow similar
configuration directives. A popular FTP server is vsftpd (Very
Secure FTP daemon). ProFTP is also in wide use, as are wu-ftpd and
ncftpd. For many purposes, FTP is not really needed at all. For
example, secure transfers for users who have accounts on a server
machine can often be accomplished using scp (secure copy), which
relies on the underlying SSH installation, but otherwise mostly mimics
the familiar cp command.
The configuration file for vsftpd is /etc/vsftpd.conf. Other FTP
servers use similar files.
A few options to keep in mind in /etc/vsftpd.conf (and probably in
your server if you use a different one) are listed here:
anonymous_enabled: Let anonymous users login using the usernames "anonymous" or "ftp".
anon_mkdir_write_enable: Let anonymous users create directories (within world writable parent directories).
anon_upload_enable: Let anonymous users upload files.
anon_world_readable_only: YES by default, and rarely a good idea to change. Only lets anonymous FTP access world-readable files.
chroot_list_enable: Specify a set of users (listed in
/etc/vsftpd.chroot-list) in a "chroot jail" in their home
directory upon login.
ssl_enable: Support SSL encrypted connections.
Read the manpages for your FTP server for more complete options. Generally, running an FTP server is as simple as tweaking a configuration file and running the server within your intilization scripts.
Most every Linux machine, and most other operating systems, should have a secure shell (SSH) client. Often, the OpenSSH version is used, but a variety of compatible SSH clients are sometimes used. While an SSH client is essential to connect to a host, the larger security issues arise in properly configuring an SSH server. Since a client initiates a connection to a server, the client is actively choosing to trust the server. Just having an SSH client does not allow any kind of access into a machine, and hence does not expose vulnerabilities. Configuring a server is also not particularly complex, and the server daemon is designed to enable and enforce good security practices. But clearly it is a server that is sharing resources with clients, based on requests from the clients the server decides to honor.
The SSH protocol has two versions, version 1 and version 2. In modern
systems, using protocol version 2 is always preferred, but generally
both clients and servers maintain backward compatibility with version
1 (unless this capability is disabled with configuration options);
this lets you connect to increasingly uncommon version 1-only systems.
Somewhat different configuration files are used between version 1 and
version 2 protocols. For protocol version 1, a client first creates an
RSA key pair using ssh-keygen, and stores the private key in
$HOME/.ssh/identity and the public key in $HOME/.ssh/identity.pub.
This same identity.pub should be appended to the remote
$HOME/.ssh/authorized_keys files. Obviously, there is a
chicken-and-egg problem here: how can you copy a file to a remote
system before you have access? Fortunately, SSH also supports a
fallback authentication method of sending encrypted-on-the-wire
passwords that are evaluated through the usual remote-system login
tests (i.e. the user account must exist, and the right password must
be provided).
Protocol 2 supports both RSA and DSA keys, but RSA authentication is
somewhat enhanced rather than identical to that in protocol 1. For
protocol 2, private keys are stored in $HOME/.ssh/id_rsa and
$HOME/.ssh/id_dsa. Protocol 2 also supports a number of extra
confidentiality and integrity algorithms: AES, 3DES, Blowfish,
CAST128, HMAC-MD5, HMAC-SHA1, and so on. The server can be configured
as to preferred algorithms, and order of fallbacks.
For general configuration options, rather than key information, the
client stores its in /etc/ssh/ssh_config or if available
/$HOME/.ssh/config. Client options can also be configured with the
-o switch; a particularly common switch is the -X or -x to
enable or disable X11 forwarding. If enabled, the X11 port is
tunnelled through SSH to enable encrypted X11 connections. Tools like
scp also use similar port forwarding over SSH. For example, to on
the local machine I am working on I can launch onto the local display
an X11 application that only exists remotely (on my local subnet in
this case), e.g.:
$ which gedit # not on local system $ ssh -X [email protected] Password: Linux averatec 2.6.10-5-386 #1 Mon Oct 10 11:15:41 UTC 2005 i686 GNU/Linux No mail. Last login: Thu Feb 23 03:51:15 2006 from 192.168.2.101 dqm@averatec:~$ gedit &
The sshd daemon, specifically the OpenSSH version available from
<http://www.openssh.org/>, enables secure encrypted communications
between two untrusted hosts over an insecure network. The base sshd
server is normally started during initialization, and listens for
cleitn connections forking a new daemon for each client connection.
The forked daemons handle key exchange, encryption, authentication,
command execution, and data exchange.
As with the client tool, the sshd server accepts a variety of
options on the command-line, but is normally configured by the file
/etc/ssh/sshd_config. A number of other configuration files are used
also. For example, the access controls /etc/hosts.allow and
/etc/hosts.deny are honored. Keys are stored in a similar fashion to
the client side, in /etc/ssh/ssh_host_key (protocol 1),
/etc/ssh/ssh_host_dsa_key, /etc/ssh/ssh_host_rsa_key, and public
keys in /etc/ssh/ssh_host_dsa_key.pub and friends. Also as with the
client, you will use ssh-keygen to generate keys in the first place.
See the manpage for sshd and ssh-keygen for details on
configuration files and copying generated keys to appropriate files.
Theere are a large number of configuration options in
/etc/ssh/sshd_config, and the default values are generally sensible
(and sensibly secure). A few options are particularly notable though.
AllowTcpForwarding enables or disables port forwarding (tunneling),
and is "yes" by default. Ciphers controls the list and order of
encryption algorithms to be utilized. AllowUsers and AllowGroups
accept wildcard patterns and allow you to control which users may even
attempt further authentication. DenyGroups and DenyUsers act
symmetrically, as you would expect. PermitRootLogin lets the "root"
user SSH into a machine. Protocol lets you specify whether both
protocol versions are accepted (and if not, which one is).
TCPKeepAlive is good to look at if you are loosing SSH connections.
A "keepalive" message is sent to check connections if this is enabled,
but this can cause disconnection if transient errors occur in the
route.
OpenSSH lets you create a tunnel to encapsulate another protocol
within an encrypted SSH channel. This capability is enabled on the
sshd server by default, but could have been disabled with
command-line or configuration file options. Assuming the capability is
enabled, a client can easily emulate whatever port/protocol they wish
to use for a connection. For example, to create a tunnel for telnet:
% ssh -2 -N -f -L 5023:localhost:23 [email protected] % telnet localhost 5023
This example, of course, if fairly pointless since a SSH command shell
does the same thing as a telnet shell. But you could create a POP3,
HTTP, SMTP, FTP, X11, or other protocol connection in the analogous
manner. The basic concept is that a particular localhost port acts as
if it were the remote service, with actual communication packets
travelling over the SSH connection in encrypted form.
The options we used in the example are: -2 (use protocol 2); -N
(no command/tunnel only), -f (SSH in background); -L (describe
tunnel as "localport:remotehost:remoteport". The server (with
username) are also specified.
tcp_wrappers?
The first thing to know about TCP_wrappers is that you should not
use it, and it is not actively maintained. However, you might find the
tcpd daemon from TCP_wrappers still running on a legacy system. In
its time, this was a good application, but its functionality has been
superceded by iptables and other tools. The general purpose of
TCP_wrappers is to monitor and filter incoming requests for the
SYSTAT, FINGER, FTP, TELNET, RLOGIN, RSH, EXEC, TFTP, TALK, and other
network services.
TCP_wrappers can be configured in a couple manners. One is to
substitute tcpd for other servers, providing arguments to pass
control on to the particular server once tcpd has done its logging
and filtering. Another method leave the network daemons alone and
modifies the inetd configuration file. For example, an entry such
as:
tftp dgram udp wait root /usr/etc/tcpd in.tftpd -s /tftpboot
Causes an incoming tftp request to run through the wrapper program
(tcpd) with a process name in.tftpd.
A few tools and sites are worth keeping in mind for a Linux administrator minding security. Websites worth monitoring for security issues and patches include:
* Security Focus news <http://www.securityfocus.com/>. The Security Focus website is one of the best sites for reporting and discussion of security issues and specific vulnerabilities. The site includes a number of newsletters and alertes you can subscribe to, as well as general columns and searchable bug reports.
* The Bugtraq mailing list <http://www.securityfocus.com/archive/1> is a full disclosure moderated mailing list for the detailed discussion and announcement of computer security vulnerabilities: what they are, how to exploit them, and how to fix them.
* CERT Coordination Center <http://www.cert.org/>. Hosted by Carnegie Mellon University, CERT has a similar range of advisories as Security Focus site, with a bit more emphasis on tutorials and guidelines. Keeping track of multiple such sites is a good way to make sure you are current on all the security incident affecting your OS, distribution, and specific tools or servers.
* Computer Incident Adisory Capability <http://www.ciac.org/ciac/index.html>. CIAC Information Bulletins are distributed to the Department of Energy community to notify sites of computer security vulnerabilities and recommended actions. Similarly, CIAC Advisory Notices serve to alert sites to severe, time-critical vulnerabilities and solutions to be applied as soon as is possible. CIAC Technical Bulletins cover technical security issues and analyses of a less time sensitive nature.
* Information on securing open mail relays <http://www.ordb.org/faq/>. A common vulnerability on systems with mail servers is failure to properly secure systems against malicious use by spammer and fraudulent mailers. The Open Relay Database provides both tutorials on security particular mail tools, open relay testing online tools, and a database of known problem servers that can be used to configure blacklists if site administrators so desire.
Tools to monitor security you might consider running are:
* Open Source Tripwire <http://sourceforge.net/projects/tripwire/>: A security and data integrity tool for monitoring and alerting on specific file changes.
* scanlogd <http://www.openwall.com/scanlogd/>: A TCP port scan
detection tool.
* Snort <http://www.snort.org/>: Network intrusion prevention and detection, utilizing a rule-driven language. Uses signature, protocol and anomaly based inspection methods.