This article is a practical step-by-step guide for securing Linux production systems. It discusses basic Linux Security
requirements for systems that need to pass various audits in an enterprise environment.
If you work on a corporate Linux Security Standard, or if you do
Sarbanes-Oxley Act (SOX) or
Statement on Auditing Standards No. 70 (SAS 70) related work, then
this article should provide you a good baseline.
This Linux Security HOWTO is intended for a technical audience, Linux system administrators, and security people in corporations and organizations
that have to use commercial Linux distributions for their production environment.
If you are a Linux expert then you may find familiar material here, but you will have difficulties to find documentation
on various topics like restricting su access to system and shared accounts only as covered in this article, see
Restricting su Access to System and Shared Accounts.
If you need to make Linux production systems compliant with various audit requirements, then this article should
offer a good baseline and starting point.
The main objective of this Linux Security guide is to discuss basic Linux security requirements including account policies for production systems that are being
This document covers various system services like SSH which are usually enabled and required on all Linux production servers. But it does not cover
services or applications like Apache, Samba etc., since these applications/services are usually not needed across all Linux
servers and should therefore not be installed on all systems. In fact, these applications warrant their own security HOWTO.
Also, this article does not cover security features that require kernel patching. This is not an option
for most companies due to vendor support issues.
The steps in this guide have been tested on RHEL 3 and 4.
This document comes without warranty of any kind. But every effort has been made to provide the information as accurate as possible.
I welcome emails from any readers with comments, suggestions, and corrections at webmaster_at_puschitz.com.
If you believe that I did not address a basic and important Linux security topic, please drop me an email.
Physical security should be of the utmost concern. Linux production servers should be in locked datacenters
where only people with passed security checks have access. But physical security is out of scopy for this article.
Depending on the environment and circumstances, you may want to consider boot loader passwords.
It is strongly recommended to have scripts available which verify that all security action items have
been executed. Even the best sysadmins can make mistakes and miss steps. If you have a larger
Linux environment, it would be a good investment to write scripts for checking Linux security action items.
To retire servers with sensitive data, it is important to ensure that data cannot be
recovered from the hard disks. To ensure that all traces of data are removed, the DiskSanitizer
tool can be used. This tool can be operated from a floppy disk and it removes data according with the
U.S. Department of Defense (DoD) standards.
DiskSanitizer is available at
If your system gets compromised, your backups become invaluable. But also in cases like bugs, accidents etc.
backups can be used to compare you current system against your backed-up system. For production systems it is
very important to take some Backups offsite for cases like disasters.
For legal reasons, some firms and organizations must be careful about backing up too much information and
holding it too long. If your environment has a policy regarding the destruction of old paper files, you
might have to extend this policy to Linux backup tapes as well.
Servers should have separate partitions for at least /, /boot, /usr, /var,
/tmp, and /home. You don't want that e.g. logging and temporary space under /var
and /tmp fill up the root partition. Third party applications should be on separate
filesystems as well, e.g. under /opt.
I will not cover iptables in this paper. Most companies use hardware based
firewalls to protect their servers in a production network, which is strongly recommended for such environments.
If you are interested in a Linux Stateful Firewall using iptables, check out my HOWTO for
Stateful Firewall and Masquerading on Linux.
For lots of iptables tutorials and examples, see
SELinux is an advanced technology for securing Linux systems. Hardening Linux using SELinux technology,
on its own, warrants it's own security HOWTO and is out of scope for this guide.
I highly recommend the book
SELinux: NSA's Open Source Security Enhanced Linux.
FTP, telnet, and rlogin (rsh) are vulnerable to eavesdropping, which is one of the reasons why SSH/SCP/SFTP should be used instead.
It is highly recommended not to run these services. Due to the high risk, this guide does not cover these services.
It would also be a good idea not to have FTP and Telnet server RPMs installed on the system.
A very important step in securing a Linux system is to determine the primary function or role of the Linux server.
You should have a detailed knowledge of what is on your system. Otherwise you will have a difficult time
to understand what needs to be secured and hence securing your Linux systems proactively won't be that effective.
Therefore, it is very critical to look at the default list of software packages and remove unneeded packages or packages
that don't comply with your security policy. If you do that you will have less packages to update and to maintain
when security alerts and patches are released.
For example, you should not have Apache or Samba installed on your system if you don't use them.
Also, it is a good practice not to have development packages, desktop software packages (e.g. X Server) etc.
installed on production servers. Other packages like FTP and Telnet daemons should not be installed as well
unless there is a justified business reason for it (SSH/SCP/SFTP should be used instead).
One of the first action items should be to create a Linux image that only contains RPMs
needed by the applications, and needed for maintenance and troubleshooting purposes.
A good approach is to start with a minimum list of RPMs and then add packages as needed.
It may be time-consuming but worth the efforts.
To get a list of all installed RPMs you can use the following command:
If you want to know more about a particular RPM, run:
rpm -qi <package_name>
To check for and report potential conflicts and dependencies for deleting a RPM, run:
rpm -e --test <package_name>
For information on performing Kickstart installations and how to build an image, see
for more information.
Building an infrastructure for patch management is another very important step to proactively secure Linux
It is recommended to have a written security policy and procedure to handle Linux security updates
For example, a security policy should detail the timeframe for assessment, testing, and rollout of patches.
Network related security vulnerabilities should get the highest priority and
should be addressed immediately within a short timeframe.
For example, a security procedure should detail the process for assesment, testing, and rollout of
patches. The assessment phase should occur within a testing lab, and initial rollout should
occur on development systems first.
A separate security log should detail what Linux security notices have been received, when patches have
been researched and assessed, when patches have been applied etc.
One of the most important tasks is to detect and close network ports that are not needed.
To get a list of listening network ports (TCP and UDP sockets), you can run the following command:
# netstat -tulp
Active Internet connections (only servers)
Proto Recv-Q Send-Q Local Address Foreign Address State PID/Program name
tcp 0 0 *:auth *:* LISTEN 2328/xinetd
tcp 0 0 localhost.localdomain:smtp *:* LISTEN 2360/sendmail: acce
tcp 0 0 *:ssh *:* LISTEN 2317/sshd
From the output you can see that xinetd, sendmail, and sshd are listening.
On all newer Red Hat Linux distributions sendmail is configured to listen for local connections only.
Sendmail should not listen for incoming network connections unless the server is a mail or relay server.
Running a port scan from another server will confirm that (make sure that you have permissions to probe a machine):
# nmap -sTU <remote_host>
Starting nmap 3.70 ( http://www.insecure.org/nmap/ ) at 2004-12-10 22:51 CST
Interesting ports on jupitor (172.16.0.1):
(The 3131 ports scanned but not shown below are in state: closed)
PORT STATE SERVICE
22/tcp open ssh
113/tcp open auth
Nmap run completed -- 1 IP address (1 host up) scanned in 221.669 seconds
Note that the above nmap command can take a while. If you remove the UDP port scan (without the option "-U"),
then nmap will finish the port scan immediately. If you run it on the local machine it will also complete very fast.
Also note that nmap might not show all listening network sockets if a firewall is being used to block ports.
From the output above you can see that the xinetd daemon is listening on port auth (port 113) for IDENT
(for more information on this service, see below). You can also
see that sendmail is not listening for remote incoming network connections, see also
Another method to list all of the TCP and UDP sockets to which programs are listening is lsof:
One of the most important tasks is to remove any network services from the system startup process
that are not needed.
On Red Hat systems you can list all services which are started at bootup using the following command:
chkconfig --list |grep on
You will notice that there are quite a lot of services enabled on your system. But many runlevel services (Stand-Alone Services)
are not network related services like kudzu which is responsible for
detecting and configuring new and/or changed hardware on your system. This service is only run during
the boot process. Ensure not to disable runlevel services that are needed by the system to run smoothly.
Here are examples of Red Hat Runlevel System Services which you may or may not want to enable:
gpm needed if you want to use the mouse at the console
kudzu important for detecting new hardware
syslog important for syslog services
netfs needed only if there are NFS shares that should be mounted at boot time
network important for starting network interfaces (e.g. eth0, eth1, bonding,...)
random used for the system entropy pool
atd needed if the at(1) service is used instead of cron
apmd Advanced Power Management (APM) daemon is used for laptops and some desktops
isdn needed if ISDN is being used
iptables needed if Netfilter (iptables) Firewall is being used
ip6tables needed if ip6tables Firewall is being used
pcmcia not needed on servers - needed for laptops
irqbalance important for distributing interrupts across all CPUs
sendmail needed if Sendmail is used - Procmail should be used which is more secure
autofs needed if automounter is used - production applications should not be dependent on automounter
sshd important for logins via SSH
portmap needed if e.g. NFS is being used
nfslock needed if NFS shares are mounted
nfs needed if server runs the NFS server
mdmonitor needed only if software RAID is being used
crond important for running cron jobs
xinetd needed if xinetd services are being used, see /etc/xinetd.d/ for list of services
cups needed if CUPS is used for the printing system
rhnsd needed if server should connect to RHN to check for software updates etc.
sysstat needed to reset system statistics logs
audit needed only if Linux Audit Subsystem (LAuS) should run for collecting system call audit records
psacct needed only if kernel process accounting information is needed
smartd important for monitoring disk problems if hard disks support SMART technology
netdump important if kernel oops data and memory dumps should be sent to a Netdump server for server crashes
The start/stop scripts of all runlevel services can be found in the /etc/init.d directory.
For example, if you don't know what the atd service does, go to /etc/init.d
and open the file atd. And in the script look for lines that start programs. In the
atd script the "daemon /usr/sbin/atd" line starts the binary atd. Now having
the name of the program that is started by this service, you can check the online pages
of atd by running man atd. This will help you to find out more about a system service.
To permanently disable e.g. the runlevel service nfs, run:
chkconfig nfs off
To immediately disable the runlevel service nfs, run:
The xinetd daemon is a replacement for inetd, the internet services daemon. It monitors the ports
for all network services configured in /etc/xinetd.d, and starts the services in response to
To check if xinetd is enabled and running, execute:
If xinetd is active, it is important to check which Unix services are active and controlled by xinetd.
The following command shows all services configured in /etc/xinetd.d and wheter xinetd monitors the
ports for these services:
# chkconfig --list | awk '/xinetd based services/,/""/'
xinetd based services:
To get a list of only active services for which xinetd monitors the ports, you could run:
# chkconfig --list | awk '/xinetd based services/,/""/' | grep -v off
xinetd based services:
In the above example you can see that the telnet-server RPM is not installed on the system.
If the Telnet Server package telnet-server would be installed, it would show up on the list
whether it's active or not.
Here is an example how to disable a service. Assuming the
telnet service is active, run the following commands to disable it and to see
how the telnet service entries are being updated:
# chkconfig --list telnet
# cat /etc/xinetd.d/telnet | grep disable
disable = no
# chkconfig telnet off
# chkconfig --list telnet
# cat /etc/xinetd.d/telnet | grep disable
disable = yes
For the telnet service it would be better to remove the package from the system since SSH should be used instead:
# rpm -e telnet-server
It is important to investigate all active xinetd services and to disable them if they are not needed.
Here is an example how to find out what a service does.
Assuming you don't know what the auth service does which is listed as active in the list above, run the following
# grep " server" /etc/xinetd.d/auth
server = /usr/sbin/in.authd
server_args = -t60 --xerror --os -E
# man in.auth
No manual entry for in.auth
# rpm -qf /usr/sbin/in.authd
# rpm -qi authd-1.4.1-1.rhel3 | awk '/Description/,/""/'
authd is a small and fast RFC 1413 ident protocol daemon
with both xinetd server and interactive modes that
supports IPv6 and IPv4 as well as the more popular features
# rpm -ql authd-1.4.1-1.rhel3
This example shows what can be done if there exists no online manuals for the binary in.authd that is started by xinetd.
The steps above should be helpful for finding out more about services.
The auth service (aka IDENT, see RFC 1413) allows remote daemons to query information about users
establishing TCP connections on the local server. In a trusted environment it helps a server
to identify who is trying to use it. For example, it can provide vital information for troubleshooting
and who has done what. IDENT requests are needed by some applications like IRC. However, IDENT can
be a security risk.
To disable the auth service, run the following command:
# chkconfig auth off
The xinetd daemon is quite flexible and has many features. Here are just a few functionalities of Xinetd:
- Acces control for TCP, UDP, and RPC services
- Acess limitations based on time
- Provides mechanisms to prevent DoS attacks
The inittab file /etc/inittab also describes which processes are started
at bootup and during normal operation. For example, Oracle uses it to start cluster services at bootup.
Therefore, it is recommended to ensure that all entries in /etc/inittab are legitimate in your environment.
I would at least remove the CTRL-ALT-DELETE trap entry to prevent accidental reboots:
# sed -i 's/ca::ctrlaltdel:/#ca::ctrlaltdel:/g'
The default runlevel should be set to 3 since in my opinion X11 (X Windows System) should not be running on a production server.
In fact, it shouldn't even be installed.
And depending on your environment you might want to comment out the UPS entries as well.
To have changes in /etc/inittab become effective immediately, you can run:
# init q
The /etc/rc.local script is used for commands or startup scripts which are pertinent only to a
specific server. (/etc/rc.local is a link to /etc/rc.d/rc.local).
Ensure that all startup scripts in /etc/rc.d/rc.local are legitimate.
Usually a firewall is used to protect a server from other servers and networks. However, in some cases you may also want to protect
a server within a network by using a TCP Wrapper.
The Xinetd super server that comes with most Linux distributions includes a built-in TCP wrapper. It can be used to explicitly define network services
to accept incoming connections from specified servers and networks. The TCP wrappers implements access control
through the use of two files, /etc/hosts.allow and /etc/hosts.deny.
Note that the hosts.allow file takes precedence over the hosts.deny file.
And you may want to change the permissions on the two configuration files since they are both world readable.
A recommended security-strategy is to block all incoming requests by default, but allow specific hosts
or networks to connect. This is the strategy I will describe here.
To deny everything by default, add the following line to
To accept incoming SSH connections from e.g. nodes rac1cluster, rac2cluster and rac3cluster, add the following line to
sshd: rac1cluster rac2cluster rac3cluster
To accept incoming SSH connections from all servers from a specific network, add the name of the subnet to
/etc/hosts.allow. For example:
The version of TCP wrapper that comes with Red Hat also supports the extended options documented in the
hosts_options(5) man page. Here is an example how an additional program can be spawned in e.g. the
sshd: ALL : spawn echo "Login from %c to %s" | mail -s "Login Info for %s" log@loghost
For information on the % expansions, see "man 5 hosts_access".
The TCP wrapper is quite flexible. And xinetd provides its own set of host-based and time-based access
control functions. You can even tell xinetd to limit the rate of incoming connections.
I recommend reading various documentations about the Xinetd super daemon on the Internet.
Many network services like telnet, rlogin, and rsh are vulnerable
to eavesdropping which is one of several reasons why SSH should be used instead.
Red Hat's default configuration for SSH meets the security requirements for many environments.
However, there are a few parameters in /etc/ssh/sshd_config
that you may want to change on RHEL and other Linux systems.
Also ensure to have privilege separation enabled where the daemon is split into two parts. With privilege separation a
small part of the code runs as root and the rest of the code runs in a chroot jail
environment. Note that on older RHEL systems this feature can break some functionality, for example see
Preventing Accidental Denial of Service.
Since SSH protocol version 1 is not as secure you may want to limit the protocol to version 2 only:
You may also want to prevent SSH from setting up TCP port and X11 forwarding if you don't need it:
Ensure the StrictModes directive is enabled which checks file permissions and ownerships
of some important files in the user's home directory like ~/.ssh, ~/.ssh/authorized_keys etc.
If any checks fail, the user won't be able to login.
Ensure that all host-based authentications are disabled. These methods should be avoided
as primary authentication.
Disable sftp if it's not needed:
#Subsystem sftp /usr/lib/misc/sftp-server
After changing any directives make sure to restart the sshd daemon:
Postfix is a replacement for Sendmail which has several security advantages over Sendmail.
Postfix consists of several small programs that perform their own
small task. And almost all programs run in a chroot jail. These are just a few examples why
Postfix is recommended over Sendmail.
For more information on chroot jail, see
Using Chroot Securely.
Linux servers that are not dedicated mail or relay servers should not accept external emails.
However, it is important for production servers to send local emails to a relay server.
Before you continue on a Red Hat system, make sure Postfix is activated using the following command:
# alternatives --set mta /usr/sbin/sendmail.postfix
The following parameters in /etc/postfix/main.cf
should be set to ensure that Postfix accepts only local emails for delivery:
This article focuses on security issues that pertain to most Linux servers in a production environment. Therefore, securing a mail
or relay server is out of scope for this article since not all Linux servers in a production environment are mail or relay servers.
However, Sendmail or Postfix is usually required for local mail delivery. Note that it is recommended to use Postfix over Sendmail
for various security reasons, see
for more information.
On newer Linux systems Sendmail is configured to run
in the background for local mail delivery and not to accept incoming network connections.
If your server is not a mail or relay server, then it is important that Sendmail
is not accepting incoming network connections from any host other than the local server.
The default sendmail.cf configuration file on RedHat does not allow Sendmail to accept incoming network
The following setting in /etc/mail/sendmail.cf tells Sendmail
not to accept incoming network connections from servers other than the local node:
O DaemonPortOptions=Port=smtp,Addr=127.0.0.1, Name=MTA
If that's not the case on your system, you can change it by setting or uncommenting the DAEMON_OPTIONS parameter in the
Uncomment the DAEMON_OPTIONS line in
/etc/mail/sendmail.mc to read:
NFS (Network File System) allows servers to share files over a network.
But like all network services using NFS involves risks.
Here are some basic rules:
- NFS should not be enabled if not needed.
- If you must use NFS, use TCP wrapper to restrict remote access.
- Make sure you export to only those machines that you really need to.
- Use fully qualified domain names to diminish spoofing attempts.
- Export only directories you need to export.
- Export read-only wherever possible.
- Use NFS over TCP.
If you don't have shared directories to export, ensure that the NFS service
is NOT enabled and running:
# service nfs status
rpc.mountd is stopped
nfsd is stopped
rpc.rquotad is stopped
# chkconfig --list nfs
nfs 0:off 1:off 2:off 3:off 4:off 5:off 6:off
You probably don't need the portmap service as well which is used by NFS
(the portmap daemon registers rpc-based services for services like NFS, NIS, etc.):
# service portmap status
portmap is stopped
# chkconfig --list portmap
portmap 0:off 1:off 2:off 3:off 4:off 5:off 6:off
If you must use NFS, it can be activated using the following commands on Red Hat Linux:
chkconfig portmap on
chkconfig nfs on
service portmap start
service nfs start
The "portmap" service starts the portmap daemon.
Depending on the Linux OS distribution and version, the "nfs" service starts the rpc.rquotad, nfsd, lockd,
rpciod, rpc.mountd and rpc.idmapd daemons on Red Hat Fedora Core 3 which is the OS that
Red Hat Advanced Server 4 is based on. On Red Hat Advanced Server 3, the NFS service starts
rpc.mountd, nfsd and rpc.rquotad.
To probe the portmapper for all registered NFS related RPC programs, you can run rpcinfo.
On Red Hat Advanced Server 3, the output will look like this:
If hosts.deny and hosts.allow are displayed, or if libwrap is displayed,
then the program includes a built-in TCP wrapper. If none of these strings are displayed, then adding the
program name to /etc/hosts.deny and /etc/hosts.allow will most probably have no effect.
To block all incoming requests by default, add the following line to /etc/hosts.deny
if you have not done so yet:
Verify from a remote server that portmapper does not list any registered RPC programs:
# rpcinfo -p <server>
No remote programs registered.
To allow NFS mounts from trusted servers and networks only, you have to configure the servers and networks for
portmap and various NFS related programs in /etc/hosts.allow.
To find out which NFS programs are invoked when you start the NFS server, you can check the
/etc/init.d/nfs service script.
Then run the "strings <program> | egrep "hosts.deny|hosts.allow|libwrap"" command I described above.
On Red Hat Advanced Server 3, the /usr/sbin/rpc.rquotad program includes a built-in TCP wrapper. On Red Hat Fedora Core 3,
/usr/sbin/rpc.mountd now also includes a built-in TCP wrapper.
To allow NFS requests from e.g. servers rac1pub.example.com, rac2pub.example.com, rac3pub.example.com and from the .subnet.puschitz.com network, the configuration
in /etc/hosts.allow would look like as follows:
Note that options MUST NOT be separated from hostnames or networks with whitespace(s). And use fully qualified domain names to diminish spoofing attempts.
All entries in /etc/exports are exported with the root_squash option
('root squashing') by default. This means that a root user on a client machine does not have root privileges (root access)
to root-owned files on exported NFS filesystems/directories. It is not recommended to turn 'root squashing" off using
the no_root_squash option!
After you've made all your entries in /etc/exports, you can export
all filesystems/directories using the following command:
# exportfs -a
To unexport all shared filesystems/directories, run:
# exportfs -ua
To see all shared filesystems/directories, run:
# showmount -e localhost
Export list for localhost:
If you need NFS, it is recommended to use NFS over TCP since NFS over UDP is not very secure.
All 2.4 and 2.6 kernels support NFS over TCP on the client side.
Server support for TCP appears in later 2.4 kernels, and in all 2.6 kernels.
To verify whether your server supports NFS over TCP, use the wire-test
command (/usr/sbin/wire-test is part of the am-utils package).
If your server supports NFS over TCP, the output looks like this:
# wire-test localhost
Network 1: wire="172.16.1.0" (netnumber=172.16.1).
Network 2: wire="172.16.1.1" (netnumber=172.16.1).
My IP address is 0xac100101.
NFS Version and protocol tests to host "localhost"...
testing vers=2, proto="udp" -> found version 2.
testing vers=3, proto="udp" -> found version 3.
testing vers=2, proto="tcp" -> found version 2.
testing vers=3, proto="tcp" -> found version 3.
If the server does not support NFS over TCP, the output will look like this:
# wire-test localhost
Network 1: wire="172.16.1.0" (netnumber=172.16.1).
Network 2: wire="172.16.1.1" (netnumber=172.16.1).
My IP address is 0xac100101.
NFS Version and protocol tests to host "localhost"...
testing vers=2, proto="udp" -> found version 2.
testing vers=3, proto="udp" -> found version 3.
testing vers=2, proto="tcp" -> failed!
testing vers=3, proto="tcp" -> failed!
To mount a shared directory using NFS over TCP, use the "proto=tcp" mount option:
# mount -o proto=tcp <nfs_server_name>:/pub /usr/local/pub
Make sure the target directory, in this example /usr/local/pub, exists on the client.
You can verify the NFS over TCP mount using the mount command:
nfsserver:/pub on /usr/local/pub type nfs (rw,proto=tcp,addr=172.16.10.8)
To have the shared directory mounted on the client at boot time, use the /etc/fstab
For the above example, the /etc/fstab entry could look like this:
The following example is a suggestion you may want to use in some cases.
It shows how files can be copied over the network using ssh without providing an interactive login prompt.
SSH allows you to do a forced command using the "command" option. When you use this option you can disable scp and enforce every
passed ssh command to be ignored.
On the server side where you want to retrieve the file from, add the following entry to the beginning of the SSH key
in the ~/.ssh/authorized_keys2 file:
The following list shows tunable kernel parameters you can use to secure your Linux server against attacks.
For each tunable kernel parameters I will show the entry that needs to be added to the
/etc/sysctl.conf configuration file to make the change permanent
after reboots. To activate the configured kernel parameters immediately at runtime, use:
Source Routing is used to specify a path or route through the network from source to destination.
This feature can be used by network people for diagnosing problems.
However, if an intruder was able to send a source routed packet into the network, then he
could intercept the replies and your server might not know that it's not communicating with a trusted server.
To disable Source Route Verification, edit the /etc/sysctl.conf
file and add the following line:
ICMP redirects are used by routers to tell the server that there is a better path to other networks than
the one chosen by the server.
However, an intruder could potentially use ICMP redirect packets to alter the hosts's routing table by causing traffic to
use a path you didn't intend.
To disable ICMP Redirect Acceptance, edit the /etc/sysctl.conf
file and add the following line:
IP spoofing is a technique where an intruder sends out packets which claim to be from another host
by manipulating the source address. IP spoofing is very often used for denial of service attacks.
For more information on IP Spoofing, I recommend the article
IP Spoofing: Understanding the basics.
To enable IP Spoofing Protection, turn on Source Address Verification. Edit the
file and add the following line:
The umask (user file-creation mode mask) command is a shell built-in command which
determines the default file permissions for newly created files. This can be overwritten
by system calls but many programs and utilities make use of umask.
By default, Red Hat sets umask to 022 or 002 which is fine. If the name of the
user account and the group account is the same and the UID is 100 or larger, then umask is set to 002,
otherwise it's set to 022, see /etc/bashrc for bash shells.
uid=509(test) gid=510(test) groups=100(users),510(test) context=user_u:system_r:unconfined_t
uid=0(root) gid=0(root) groups=0(root),1(bin),2(daemon),3(sys),4(adm),6(disk),10(wheel) context=root:system_r:unconfined_t
Here is an example how umask works:
$ umask 000
$ touch file1
$ ls -l file1
-rw-rw-rw- 1 oracle oinstall 0 Dec 26 19:24 file1
$ umask 002
$ touch file2
$ ls -l file2
-rw-rw-r-- 1 oracle oinstall 0 Dec 26 19:24 file2
$ umask 022
$ touch file3
$ ls -l file3
-rw-r--r-- 1 oracle oinstall 0 Dec 26 19:25 file3
For the bash shell you can find the setting of umask in /etc/bashrc.
The /etc/bashrc file is for system-wide aliases and functions and is invoked by ~/.bashrc.
When the SUID (set user ID) or SGID (set group ID) bits are set on an executable, it executes
with the UID or GID of the owner of the executable rather than that of the person executing it.
This means that e.g. all executables that have the SUID bit set and are owned by root are
executed with the UID of root. A good example is the passwd command that allows ordinary users
to update the password field in the /etc/shadow file which is owned by root.
But SUID/SGID bits can be misused when the SUID/SGID executable has a security hole.
Therefore, you might want to search the entire system for SUID/SGID executables and
document it. For example, ensure that code developers don't set SUID/SGID bits on their programs
if it's not an absolute requirement. Very often you can use workarounds like removing just the executable
bit for world/others. However, a better approach is to change the design of the software if possible.
To search the entire system for SUID or SGID files, you can run the following command:
The "! -type l" parameter skips all symbolic links since symbolic links are always
world-writable. However, this is not a problem as long as the target of the link is
not world-writable, which is checked by the above find command.
World-Writable directories with sticky bit such as the /tmp directory do not
allow anyone except the owner of a file to delete or modify it in this directory. The sticky bit makes files
stick to the user who created it and it prevents other users from deleting and renaming
the files. Therefore, depending on the purpose of the directory world-writable directories
with sticky are usually not an issue.
An example is the /tmp directory:
$ ls -ld /tmp
drwxrwxrwt 18 root root 16384 Dec 23 22:20 /tmp
The "t" mode, which denotes the sticky bit, allows files to be deleted and renamed
only if the user is the owner of this file or the owner of the directory.
Files not owned by any user or group might not necessarily be a security problem in itself. However,
unowned files could pose a security problem in the future. For example, if a new user
is created and the new users happens to get the same UID as the unowned files have, then this new user
will automatically become the owner of these files.
To locate files not owned by any user or group, use the following command:
It is important that all system and vendor accounts that are not used for logins are locked.
To get a list of unlocked accounts on your system, you can check for
accounts that do NOT have an encrypted password string starting with "!" or "*" in the /etc/shadow file.
If you lock an account using passwd -l, it will put a '!!' in front of the encrypted password,
effectively disabling the password. If you lock an account using usermod -L, it will put
a '!' in front of the encrypted password. Many system and shared accounts are usually locked by default
by having a '*' or '!!' in the password field which renders the encrypted password into an invalid string.
Hence, to get a list of all unlocked (encryptable) accounts, run:
Also make sure all accounts have a 'x' in the password field in /etc/passwd. The following command
lists all accounts that do not have a 'x' in the password field:
# grep -v ':x:' /etc/passwd
A 'x' in the password fields means that the password has been shadowed, i.e. the encrypted
password has to be looked up in the /etc/shadow file. If the password field in
/etc/passwd is empty, then the system will not lookup the shadow file and it will not prompt
the user for a password at the login prompt.
All system or vendor accounts that are not being used by users, applications, by the system or by daemons should be removed from the system.
You can use the following command to find out if there are any files owned by a specific account:
Some admins suggest to add the following line to the /etc/inittab file to ensure that a root password
is required for Single User Mode logons:
This works but can be circumvented easily! At the GRUB or LILO prompt you can tell
the boot loader to alternate the init program by using the boot params "init=/bin/bash". This
will place you at a root shell prompt without a password.
In general I do not recommend that the system enforces password expiration for system and shared accounts.
This could lead to outages if an application's account expires:
# su oracle -c id
You are required to change your password immediately (password aged)
Changing password for test
(current) UNIX password:
Rather a corporate policy should govern password changes for system and shared accounts.
But for individual user accounts the system should expire the passwords
automatically. The following example shows how password expiration can be setup for individual user accounts.
The following files and parameters in the table are used when a new account is created with the useradd command. These settings are recorded for each user account in the /etc/shadow file.
Therefore, make sure to configure the following parameters before you create any user accounts using the
Maximum number of days a password is valid.
Minimum number of days before a user can change the password since the last change.
On an audited system it is important to restrict people from using simple passwords that can be cracked too easily.
However, if the passwords being enforced are too strong, people will write them down. Strong passwords that are written down
are not much safer than weak passwords.
Some will argue that strong passwords protect you against e.g. Dictionary Attacks and you can defeat it by locking
the accounts after a few failed attempts. However, this is not always an option. As I will show at
Locking User Accounts After Too Many Login Failures,
locked system accounts could bring down your applications and systems which would be nothing short of a denial of service attack.
Undoubtedly, it is important to practise safe password management. In my opinion, a password should have
at least one digit number, one other character, and one upper case letter. But keep in mind not to make it
The pam_cracklib module checks the password against dictionary words and other constraints.
Unfortunately, however, the original Linux PAM module pam_cracklib uses a credit mechanism. E.g. if you
define password length minlen=10, then you will get 1 credit for e.g. using a single digit number in your password
if you defined dredit=1.
This means that pam_cracklib will accept a password of the length of minlen-credit.
If you don't use a digit number, then the minimum length of the password would be minlen.
There was no way to tell the module that a password _must_ include a digit number.
Back in 2000 I wrote a patch for the pam_cracklib module where you can assign negative values to
the pam_cracklib parameters lcredit, ucredit, dcredit, and ocredit. Using negative values
will disable the credit mechanism. For example, if you define dredit=-1, then the module will
only accept a password if it includes at least one digit number and if the password has a length of at least minlen.
Red Hat has finally applied my pam_cracklib patch and you don't have to patch
the pam_cracklib module any more. The new pam_cracklib feature works in Red Hat Enterprise Linux 4
and Red Hat Fedora Core 3. This feature is now also included with the Red Hat Enterprise Linux 3 Update 4 and
Red Hat Enterprise Linux 2.1 Update 6 release.
If the Linux distribution you are using does not use the patched pam_cracklib module yet, you can find the procedure
for patching pam_cracklibhere.
In the following example I'll assume that you are using the new pam_cracklib module, or that you patched
the module if your Linux distribution doesn't include the patched version yet.
The following example shows how to enforce the following password rules:
- Minimum length of password must be 8
- Minimum number of lower case letters must be 1
- Minimum number of upper case letters must be 1
- Minimum number of digits must be 1
- Minimum number of other characters must be 1
Minimum length of password is 8
Minimum number of lower case letters is 1
Minimum number of upper case letters is 1
Minimum number of digits is 1
Minimum number of other characters is 1
To setup these password restrictions, edit the /etc/pam.d/system-auth file
and add/change the following pam_cracklib arguments highlighted in blue:
Now verify that the new password restrictions work for new passwords. Simply login to a non-root account and change
the password using the passwd command. Note that the above requirements are not enforced if you run the
passwd command under root.
NOTE: The /etc/pam.d/system-auth PAM configuration file is auto-generated and contains records
which dictate a generic authentication scheme.
Keep in mind that authconfig might clobber some changes you made.
Since I never run authconfig I usually make changes to this file because it's used by many PAM aware
applications. Otherwise I'd have to make changes to many configuration files. Changing system-auth
is usually the preferred method.
You might even want to disable all execution bits from the /usr/sbin/authconfig binary to prevent
authconfig from clobbering your changes.
The pam_unix module parameter remember can be used to configure the number of previous passwords that
cannot be reused. And the pam_cracklib module parameter difok can be used to specify the number of
characters hat must be different between the old and the new password.
In the following example I will show how to tell the system that a password cannot be reused for at least 6 months
and that at least 3 characters must be different between the old and new password.
Remember that in the chapter
Enabling Password Aging
we set PASS_MIN_DAYS to 7, which specifies the minimum number of days
allowed between password changes. Hence, if we tell pam_unix to remember 26 passwords, then the previously used
passwords cannot be reused for at least 6 months (26*7 days).
Here is an example. Edit the /etc/pam.d/system-auth file
and add/change the following pam_cracklib and pam_unix arguments:
I do not recommend that the system automatically locks system and shared accounts after too many failed login or su attempts.
This could lead to outages if the application's account gets locked due to too many login failures like in this example:
The first added line counts failed login and failed su attempts for each user. The default location for attempted accesses
is recorded in /var/log/faillog.
The second added line specifies to lock accounts automatically after 5 failed login or su attempts (deny=5).
The counter will be reset to 0 (reset) on successful entry if deny=n was not exceeded.
But you don't want system or shared accounts to be locked after too many login failures (denial of service attack).
To exempt system and shared accounts from the deny=n parameter, I added the per_user parameter to the module.
The per_user parameter instructs the module NOT to use the deny=n limit for accounts where
the maximum number of login failures is set explicitly. For example:
# faillog -u oracle -m -1
# faillog -u oracle
Username Failures Maximum Latest
oracle 0 -1 Fri Dec 10 23:57:55 -0600 2005 on unknown
The faillog command with the option "-m -1" has the effect of not placing a limit on the number of failed logins.
To instruct the module to activate the deny=n limit for this account again, run:
# faillog -u oracle -m 0
By default, the maximum number of login failures for each account is set to 0 which instructs pam_tally to
use the deny=n parameter.
The faillog manual page on my Red Hat system says that selecting maximum number of login failures of 0
will deactivate deny=n for the account.
The PAM documentation, however, says that per_user will only work if the .fail_max field contains a
non-zero value. After testing both values setting it to -1 worked. Maybe because it's read as a high unsigned
To see failed login attempts, run:
To unlock an account after too many login failures, run:
# faillog -u <user> -r
Make sure to test these changes thoroughly on your system using e.g. ssh and
su, and make sure root does not get locked!
To lock/unlock accounts manually, you can run one of the following commands:
Since the /var/log/faillog is owned by root and only root can write to the /var/log/faillog file, xscreensaver
and vlock won't work correctly. Each time xscreensaver or vlock is executed as a non-root
user, you won't be able to do an unlock since these programs can't write to /var/log/faillog.
I don't have a good solution for that. I can only think of setting the SUID bits on these programs.
On an audited production system it is very important to know who switched to which system or shared account.
Therefore it is prudent to restrict direct logins for all system and shared account where more than
one individual knows the password. All users should do a direct login using their own account and then switch to the
system or shared account. (If you are just interested in restricting direct root SSH logins, see
However, there are situations where you have to allow direct logins for system or shared accounts.
For example, within an Oracle RAC cluster you have to enable direct ssh logins for oracle. But in such an
environment you have to protect the whole cluster as a single entity against incoming ssh connection,
i.e. direct oracle logins should not
work if you come from a node that is not part of the cluster. In the following example I will show how to achieve
this goal as well.
Usually all system and shared accounts have one thing in common, that is they are not in the "users" group.
The following example assumes that all individual user accounts are in the "users" group but system and
shared accounts like root and oracle are not. If you want to go a step further, a good solution would be to
implement a new 'logingroup' users group which would require users to be given explicit access.
In this example I will show how to restrict direct logins for:
- SSH (/etc/pam.d/sshd)
- Console Login (/etc/pam.d/login)
- Graphical Gnome Login (/etc/pam.d/gdm
- or for all logins (/etc/pam.d/system-auth)
To accomplish this goal I will add the pam_access module to the PAM configuration files listed above.
This module provides logdaemon-style login access control based on login names, host names, IP addresses, etc.
The PAM module type that has to be used in the configuration files is account.
This module type does the authorization, i.e. is the user allowed to login (e.g. time, day)?
Don't confuse the PAM module type account with auth which does the authentication, for example
checking the password.
And the control flag I will use is required.
It specifies that Success is required, Failure means that it will still call the remaining modules, but the result is already determined.
For SSH Logins add the pam_access module to /etc/pam.d/sshd as follows:
Now add the following line to the /etc/security/access.conf configuration file:
-:ALL EXCEPT users :ALL
The /etc/security/access.conf configuration file is read by the pam_access module. This entry specifies
that no users are accepted except users that are in the "users" group.
Since the pam_access module has been configured for "Authorization" (account) in the above
PAM configuration files, it denies direct logins for all accounts except the ones that are in the "users" group.
Now on some systems like Oracle RAC clusters you have to enable direct ssh logins for oracle within the cluster.
On such systems you can enable direct ssh logins for oracle within the cluster by adding/changing the
following lines in /etc/security/access.conf:
The first line has been edited to include the oracle account which will allow general direct logins.
However, the second line specifies that direct logins for oracle are only allowed from Oracle RAC nodes
(rac1cluster, rac2cluster, and rac3cluster) that are part of the cluster.
In RHEL4 pam_access is already configured for crond:
This means that the above entries in /etc/security/access.conf will stop cron from working.
Note that it is very prudent to always check
whether pam_access is configured for any other service on the system!
To ensure that all users on the system can still run cron jobs you can add the following argument to pam_access
This ensures that the /etc/security/access.conf configuration file is not invoked by crond.
Since pam_cracklib does not grant permissions if the configuration file does not exist, execute the
following command to create an empty file:
# touch /etc/security/access-cron.conf
Now verify that cron jobs can be launched by any user on the system.
The above example will only work if there exists no "users" account in the /etc/passwd file on the
system, which is usually the case. Otherwise you have to either delete the "users" account or you have to designate or create another group name.
This chapter shows how to restrict people from su-ing to system and shared accounts even if they know the passwords.
Users usually don't share the passwords of their own accounts but are less hesitant to share it for shared accounts. This chapter helps to mitigate this problem.
The following example shows how to restrict su access to the root, oracle, and postgres account
to a specific set of users.
NOTE: The documentation about the pam_wheel module included in many Linux distributions
is wrong. For instance, in Red Hat Advanced Server 2.1 the pam_wheel module does not only restrict people
from su-ing to the root account like it used to be. It restricts people from su-ing to any account.
Add all users who are allowed to su to the root, oracle,
and postgres account to the new member groups created above.
The following requirement will be configured:
- Only admin1 should be able to su to root, oracle, and postgres.
- Only oracledba1 should be able to su to oracle.
- Only postgresdba1 should be able to su to postgres.
- No one else on the system should be able to su to any account.
As you probably noted, I did not add adminuser1 to the other member groups. Instead,
I will show how to give people in the rootmembers group automatically su access to the oracle and postgres
account without adding them to the oraclemembers and postgresmembers groups. I consider root admins an exception.
They should not be added to all member groups on the system.
Next add the three authentication lines highlighted in blue to the /etc/pam.d/su
file as shown below:
These additional authentication lines specify that nobody should be able to su to any account
unless at least one of the PAM services, su-root-members or su-other-members, returns Success.
The control flag sufficient means that a Success will bypass the remaining authentication modules and overall
Success is returned for the authentication part.
Failure means that the failed authentication PAM service is ignored. If both authentication PAM services fail,
then the last authentication module pam_deny is invoked which will deny all requests for any available
authentication module. This will cause the authentication part to fail for the su command.
Next the new authentication PAM service configuration files /etc/pam.d/su-root-members and
/etc/pam.d/su-other-members need to be created.
The file /etc/pam.d/su-root-members referenced in
/etc/pam.d/su should read like:
The file /etc/security/su-rootmembers-access referenced in
/etc/pam.d/su-root-members should read like:
The control flag required which is specified for both modules means that both modules have to return Success. Otherwise
this PAM service will return Failure to the "su" PAM service configured in /etc/pam.d/su.
The first line returns Success only if the user is in the rootmembers groups.
The second line allows only access (sense=allow) to those users specified in /etc/security/rootusername, which is
root, oracle, and postgres - these are the only users that will be accepted as a user argument to su.
The item=user argument instructs pam_listfile that the entries in /etc/security/rootusername are usernames.
If an error occurs, such as an unreadable configuration file, access is denied (onerr=fail).
NOTE: Once su access to root is working for users in the rootmembers, I recommend to avoid making any changes
to the /etc/pam.d/su-root-members file in the future. Making a mistake in this file could revoke access to root for all users
on the system.
That's the reason why I created two PAM service files, /etc/pam.d/su-root-members for people in
the rootmembers group, and /etc/pam.d/su-other-members (see below) for all other member groups since you will most probably
add more member groups to this file in the future.
Next the file /etc/pam.d/su-other-members referenced in
/etc/pam.d/su should be created and read like:
If one of the two PAM services returns Success, it will return Success to the "su" PAM service configured in /etc/pam.d/su.
Otherwise the last module will be invoked which will deny all further requests and the authentication fails.
Next the PAM services "su-oracle-members" and "su-postgres-members" have to be created.
The file /etc/pam.d/su-oracle-members referenced in
/etc/pam.d/su-other-members should read like:
The file /etc/security/su-postgresmembers-access referenced in
/etc/pam.d/su-postgres-members should read like:
Now verify that adminuser1 can su to root, oracle, and postgres.
No one else should be able to su to root.
oracleuser1 should be able to su to oracle only, and postgresuser1 should be able to su to postgres only.
No one else on the system should be able su to any of these accounts even if they know the password.
Linux allows you to set limits on the amount of system resources that users and groups can use. This is also very
handy if bugs in programs accidentally use up too much resources, slow down the machine, or even render the system unusable. I've seen systems
where incorrect settings have allowed programs to use up too much resources which made the server unresponsible for
new connections or local logins (e.g. a program uses up all file handles on the system). This could become a security issue
if someone is allowed to use up all resources and causes a denial of service attack.
Depending on your environment you may want to review resource limits for user accounts and groups.
The following example shows a practical use of setting or restricting system resources for an Oracle user account.
For a list of system resource settings, see /etc/security/limits.conf.
It would be a good idea to review the default settings of system resource.
Most shells like Bash provide control over various resources like the maximum allowable number of open
file descriptors or the maximum number of processes available to a user.
To see all shell limits, run:
For more information on ulimit for the Bash shell, see man bash and search for ulimit.
Setting "hard" and "soft" limits in the following examples might not work properly when you login
to oracle using a SSH session. It should work if you log in as root and su
Resource limits should also work if the application is started automatically during the boot process.
But if you experience the problem that the changed resource limits in /etc/security/limits.conf are not
applied when logging in through SSH,
then you may have to try to set UsePrivilegeSeparation
in /etc/ssh/sshd_config to "no" and restart the SSH daemon by executing
/etc/init.d/sshd restart. Unfortunately, privilege separation does not work properly with PAM on some Linux distributions.
But also note that turning off privilege separation is not really recommended since it's a
valuable security feature that has already prevented exploitation of SSH vulnerabilities.
For example, to change the number of file handles or open files that the Oracle user can use,
you have to edit the file /etc/security/limits.conf as root and make the
following changes or add the following lines, respectively:
oracle soft nofile 4096
oracle hard nofile 63536
The "soft limit" in the first line defines the number of file handles or open files that the Oracle user will
have after login. If the Oracle user gets error messages about running out of file handles,
then the Oracle user can increase the number of file handles like in this example up to 63536 ("hard limit") by running
the following command:
ulimit -n 63536
You can set the "soft" and "hard" limits higher if necessary.
Note that I do not recommend to set the "hard" limit for nofile
for the oracle user equal to /proc/sys/fs/file-max.
If you do that and the oracle user uses up all the file handles, then the whole system will be out of file handles.
This could mean that you won't be able to
initiate new remote logins any more since the system won't be able to open any PAM modules which
are required for performing a login.
You also need to ensure that pam_limits is configured in the file
/etc/pam.d/system-auth, or in
/etc/pam.d/sshd (for SSH),
/etc/pam.d/su (for su),
or /etc/pam.d/login (local logins and telnet) if you
don't want to enable it for all logins, or if
/etc/pam.d/system-auth does not exist like on SUSE.
This is the PAM module that will read the /etc/security/limits.conf file.
The entry should read like:
session required /lib/security/pam_limits.so
Here are the two "session" entries I have in my /etc/pam.d/system-auth file:
To make this change permanent, add "ulimit -n 63536" (for Bash) to the
which is the user startup file for the Bash shell on Red Hat Linux (to verify your shell run:
To do this you could simply copy/paste the following commands for the oracle's Bash shell:
It is prudent to place a legal banner on login screens on all servers for legal reasons and to potentially deter intruders among other things.
Consult legal counsel for the content of the banner.
If you want to print a legal banner after a user logs in using ssh, local console etc., you can use the /etc/motd file.
Create the file if it doesn't exist and type in the banner that applies to your organization.
# cat /etc/motd
This system is classified...
Use of this system constitutes consent to official monitoring.
For SSH you can edit the Banner parameter in the
/etc/ssh/sshd_config file which will display the banner before
the login prompt.
For local console logins you can edit the /etc/issue
which will display the banner before the login prompt.
For GDM you could make the following changes to require a user to acknowledge the legal banner by
selecting 'Yes' or 'No'.
Edit the /etc/X11/gdm/PreSession/Default file and add the following lines
at the beginning of the script:
if ! gdialog --yesno '\nThis system is classified...\n' 10 10; then
You have to add a sleep of 10 seconds, otherwise GDM will believe that X crashed if the session lasted less than
10 seconds. Unfortunately, at the time of this writing the 10 seconds timeout is hardcoded and there is no configuration
parameter to change it - I checked the source code.
Before you put a server into production or better, before you put a server on the network,
you should have an integrity checker installed on your system that lets you check if unauthorized changes
have been made. In this way, if an intruder compromises your system you will know what changed
on your server. You should also have an Intrusion Detection Software (IDS) solution in place
that alarms you about intrusions, as well as Intrusion Prevention software.
It is outside the scope of this article to cover Linux Monitoring and Intrusion Detection
solutions. There are lots of interesting articles out there to read and there are several good products
available on the market including open source solutions.
For a powerful open source network-intrusion prevention and detection system, see
Securing your system with Snort.
Here is a list of commands you can use to get data about user logins:
Shows a listing of currently logged-in users.
Shows who is logged on and what they are doing.
Shows a list of last logged-in users, including login time, logout time, login IP address, etc.
Same as last, except that by default it shows a log of the file /var/log/btmp,
which contains all the bad login attempts.
This command reports data maintained in /var/log/lastlog, which is a record of the
last time a user logged in.
Prints out the connect time in hours on a per-user basis or daily
basis etc. This command reads /var/log/wtmp.
Converts raw data from /var/run/utmp or /var/log/wtmp into ASCII-parsable format.
The information provided on this website comes without warranty of any kind and is distributed AS IS.
Every effort has been made to provide the information as accurate as possible, but no warranty or fitness is implied.
The information may be incomplete, may contain errors or may have become out of date.
The use of this information described herein is your responsibility, and to use it in your own
environments do so at your own risk.