[ previous ] [ Contents ] [ 1 ] [ 2 ] [ 3 ] [ 4 ] [ 5 ] [ 6 ] [ 7 ] [ 8 ] [ 9 ] [ 10 ] [ 11 ] [ 12 ] [ A ] [ B ] [ C ] [ D ] [ E ] [ F ] [ G ] [ H ] [ next ]

Securing Debian Manual
Chapter 4 - After installation

Once the system is installed you can still do more to secure the system; some of the steps described in this chapter can be taken. Of course this really depends on your setup but for physical access prevention you should read Change the BIOS (again), Section 4.3,Set a LILO or GRUB password, Section 4.4,Remove root prompt on the kernel, Section 4.6, Restricting console login access, Section 4.7, and Restricting system reboots through the console, Section 4.8.

Before connecting to any network, especially if it's a public one you should, at the very least, execute a security update (see Execute a security update, Section 4.2). Optionally, you could take a snapshot of your system (see Taking a snapshot of the system, Section 4.18).

4.1 Subscribe to the Debian Security Announce mailing list

In order to receive information on available security updates you should subscribe yourself to the debian-security-announce mailing list in order to receive the Debian Security Advisories (DSAs). See The Debian Security Team, Section 7.1 for more information on how the Debian security team works. For information on how to subscribe to the Debian mailing lists read http://lists.debian.org.

DSAs are signed with the Debian Security Team's signature which can be retrieved from http://security.debian.org.

You should consider, also, subscribing to the debian-security mailing list for general discussion on security issues in the Debian operating system. You will be able to contact other fellow system administrators in the list as well as Debian developers and upstream developers of security tools who can answer your questions and offer advice.

FIXME: Add the key here too?

4.2 Execute a security update

As soon as new security bugs are detected in packages, Debian maintainers and upstream authors generally patch them within days or even hours. After the bug is fixed, a new package is provided on http://security.debian.org.

If you are installing a Debian release you must take into account that since the release was made there might have been security updates after it has been determined that a given package is vulnerable. Also, there might have been minor releases (there have been four for the Debian 3.0 sarge release) which include these package updates.

You need to note down the date the removable media (if you are using it) was made and check the security site in order to see if there are security updates. If there are and you cannot download the packages from the security site on another system (you are not connected to the Internet yet? are you?) before connecting to the network you could consider (if not protected by a firewall for example) adding firewall rules so that your system could only connect to security.debian.org and then run the update. A sample configuration is shown in Security update protected by a firewall, Appendix F.

Note: Since Debian woody 3.0, after installation you are given the opportunity to add security updates to the system. If you say 'yes' to this, the installation system will take the appropriate steps to add the source for security updates to your package sources and your system, if you have an Internet connection, will download and install any security updates that might have been produced after your media was created. If you are upgrading a previous version of Debian, or you asked the installation system not to do this, you should take the steps described here.

To manually update the system, put the following line in your sources.list and you will get security updates automatically, whenever you update your system.

deb http://security.debian.org/ stable/updates main contrib non-free

Note: If you are using the testing branch use the security testing mirror sources as described in Security support for the testing branch, Section 10.1.4.

Once you've done this you can use multiple tools to upgrade your system. If you are running a desktop system you will have[9] an application called update-notifier that will make it easy to check if new updates are available, by selecting it you can make a system upgrade from the desktop (using update-manager). For more information see Checking for updates at the Desktop, Section In desktop environments you can also use synaptic (GNOME), kpackage or adept (KDE) for more advanced interfaces. If you are running a text-only terminal you can use aptitude, apt or dselect (deprecated) to upgrade:

If you like, you can add the deb-src lines to /etc/apt/sources.list as well. See apt(8) for further details.

Note: You do not need to add the following line:

deb http://security.debian.org/debian-non-US stable/non-US main contrib non-free

this is because security.debian.org is hosted in a non-US location and doesn't have a separate non-US archive.

4.2.1 Security update of libraries

Once you have executed a security update you might need to restart some of the system services. If you do not do this, some services might still be vulnerable after a security upgrade. The reason for this is that daemons that are running before an upgrade might still be using the old libraries before the upgrade [10]. In order to detect which daemons might need to be restarted you can use the checkrestart program (available in the debian-goodies package) or use this one liner[11] (as root):

# lsof | grep <the_upgraded_library> | awk '{print $1, $9}' | uniq | sort -k 1

Some packages (like libc6) will do this check in the postinst phase for a limited set of services specially since an upgrade of essential libraries might break some applications (until restarted)[12].

Bringing the system to run level 1 (single user) and then back to run level 3 (multi user) should take care of the restart of most (if not all) system services. But this is not an option if you are executing the security upgrade from a remote connection (like ssh) since it will be severed.

Excercise caution when dealing with security upgrades if you are doing them over a remote connection like ssh. A suggested procedure for a security upgrade that involves a service restart is to restart the SSH daemon and then, inmediately, attempt a new ssh connection without breaking the previous one. If the connection fails, revert the upgrade and investigate the issue.

4.2.2 Security update of the kernel

First, make sure your kernel is being managed through the packaging system. If you have installed using the installation system from Debian 3.0 or previous releases, your kernel is not integrated into the packaging system and might be out of date. You can easily confirm this by running:

$ dpkg -S `readlink -f /vmlinuz` linux-image-2.6.18-4-686: /boot/vmlinuz-2.6.18-4-686

If your kernel is not being managed you will see a message saying that the package manager did not find the file associated to any package instead of the message above, which says that the file associated to the current running kernel is being provided by the linux-image-2.6.18-4-686. So first, you will need to manually install a kernel image package. The exact kernel image you need to install depends on your architecture and your prefered kernel version. Once this is done, you will be able to manage the security updates of the kernel just like those of any other package. In any case, notice that the kernel updates will only be done for kernel updates of the same kernel version you are using, that is, apt will not automatically upgrade your kernel from the 2.4 release to the 2.6 release (or from the 2.4.26 release to the 2.4.27 release[13]).

The installation system of recent Debian releases will handle the selected kernel as part of the package system. You can review which kernels you have installed by running:

$ COLUMNS=150 dpkg -l 'linux-image*' | awk '$1 ~ /ii/ { print $0 }'

To see if your kernel needs to be updated run:

$ kernfile=`readlink -f /vmlinuz` $ kernel=`dpkg -S $kernfile | awk -F : '{print $1}'` $ apt-cache policy $kernel linux-image-2.6.18-4-686: Installed: 2.6.18.dfsg.1-12 Candidate: 2.6.18.dfsg.1-12 Version table: *** 2.6.18.dfsg.1-12 0 100 /var/lib/dpkg/status

If you are doing a security update which includes the kernel image you need to reboot the system in order for the security update to be useful. Otherwise, you will still be running the old (and vulnerable) kernel image.

If you need to do a system reboot (because of a kernel upgrade) you should make sure that the kernel will boot up correctly and network connectivity will be restored, specially if the security upgrade is done over a remote connection like ssh. For the former you can configure your boot loader to reboot to the original kernel in the event of a failure (for more detailed information read Remotely rebooting Debian GNU/Linux machines). For the latter you have to introduce a network connectivity test script that will check if the kernel has started up the network subsystem properly and reboot the system if it did not[14]. This should prevent nasty surprises like updating the kernel and then realizing, after a reboot, that it did not detect or configure the network hardware properly and you need to travel a long distance to bring the system up again. Of course, having the system serial console [15] in the system connected to a console or terminal server should also help debug reboot issues remotely.

4.3 Change the BIOS (again)

Remember Choose a BIOS password, Section 3.1? Well, then you should now, once you do not need to boot from removable media, to change the default BIOS setup so that it only boots from the hard drive. Make sure you will not lose the BIOS password, otherwise, in the event of a hard disk failure you will not be able to return to the BIOS and change the setup so you can recover it using, for example, a CD-ROM.

Another less secure but more convenient way is to change the setup to have the system boot up from the hard disk and, if it fails, try removable media. By the way, this is often done because most people don't use the BIOS password that often; it's easily forgotten.

4.4 Set a LILO or GRUB password

Anybody can easily get a root-shell and change your passwords by entering <name-of-your-bootimage> init=/bin/sh at the boot prompt. After changing the passwords and rebooting the system, the person has unlimited root-access and can do anything he/she wants to the system. After this procedure you will not have root access to your system, as you do not know the root password.

To make sure that this cannot happen, you should set a password for the boot loader. You can choose between a global password or a password for a certain image.

For LILO you need to edit the config file /etc/lilo.conf and add a password and restricted line as in the example below.

image=/boot/2.2.14-vmlinuz label=Linux read-only password=hackme restricted

Then, make sure that the configuration file is not world readable to prevent local users from reading the password. When done, rerun lilo. Omitting the restricted line causes lilo to always prompt for a password, regardless of whether LILO was passed parameters. The default permissions for /etc/lilo.conf grant read and write permissions to root, and enable read-only access for lilo.conf's group, root.

If you use GRUB instead of LILO, edit /boot/grub/menu.lst and add the following two lines at the top (substituting, of course hackme with the desired password). This prevents users from editing the boot items. timeout 3 specifies a 3 second delay before grub boots the default item.

timeout 3 password hackme

To further harden the integrity of the password, you may store the password in an encrypted form. The utility grub-md5-crypt generates a hashed password which is compatible with GRUB's encrypted password algorithm (MD5). To specify in grub that an MD5 format password will be used, use the following directive:

timeout 3 password --md5 $1$bw0ez$tljnxxKLfMzmnDVaQWgjP0

The --md5 parameter was added to instruct grub to perform the MD5 authentication process. The provided password is the MD5 encrypted version of hackme. Using the MD5 password method is preferable to choosing its clear-text counterpart. More information about grub passwords may be found in the grub-doc package.

4.5 Disable root prompt on the initramfs

Note: This applies to the default kernels provided for releases after Debian 3.1

Linux 2.6 kernels provide a way to access a root shell while booting which will be presented during loading the initramfs on error. This is helpful to permit the administrator to enter a rescue shell with root permissions. This shell can be used to manually load modules when autodetection fails. This behavior is the default for initramfs-tools generated initramfs. The following message will appear:

"ALERT! /dev/sda1 does not exist. Dropping to a shell!

In order to remove this behavior you need to set the following boot argument:panic=0. Either add it to the kopt section of /boot/grub/menu.lst and issue update-grub or to the append section of /etc/lilo.conf.

4.6 Remove root prompt on the kernel

Note: This does not apply to the kernels provided for Debian 3.1 as the timeout for the kernel delay has been changed to 0.

Linux 2.4 kernels provide a way to access a root shell while booting which will be presented just after loading the cramfs file system. A message will appear to permit the administrator to enter an executable shell with root permissions, this shell can be used to manually load modules when autodetection fails. This behavior is the default for initrd's linuxrc. The following message will appear:

Press ENTER to obtain a shell (waits 5 seconds)

In order to remove this behavior you need to change /etc/mkinitrd/mkinitrd.conf and set:

# DELAY The number of seconds the linuxrc script should wait to # allow the user to interrupt it before the system is brought up DELAY=0

Then regenerate your ramdisk image. You can do this for example with:

# cd /boot # mkinitrd -o initrd.img-2.4.18-k7 /lib/modules/2.4.18-k7

or (preferred):

# dpkg-reconfigure -plow kernel-image-2.4.x-yz

4.7 Restricting console login access

Some security policies might force administrators to log in to the system through the console with their user/password and then become superuser (with su or sudo). This policy is implemented in Debian by editing the /etc/login.defs file or /etc/securetty when using PAM. In:

When using PAM, other changes to the login process, which might include restrictions to users and groups at given times, can be configured in /etc/pam.d/login. An interesting feature that can be disabled is the possibility to login with null (blank) passwords. This feature can be limited by removing nullok from the line:

auth required pam_unix.so nullok

4.8 Restricting system reboots through the console

If your system has a keyboard attached to it anyone (yes anyone) can reboot the system through it without login to the system. This might, or might not, adhere to your security policy. If you want to restrict this, you must check the /etc/inittab so that the line that includes ctrlaltdel calls shutdown with the -a switch (remember to run init q after making any changes to this file). The default in Debian includes this switch:

ca:12345:ctrlaltdel:/sbin/shutdown -t1 -a -r now

Now, in order to allow some users to shutdown the system, as the manpage shutdown(8) describes, you must create the file /etc/shutdown.allow and include there the name of users which can boot the system. When the three finger salute (a.k.a. ctrl+alt+del) is given the program will check if any of the users listed in the file are logged in. If none of them is, shutdown will not reboot the system.

4.9 Mounting partitions the right way

When mounting an ext2 or ext3 file system, there are several additional options you can apply to the mount call or to /etc/fstab. For instance, this is my fstab entry for the /tmp partition:

/dev/hda7 /tmp ext2 defaults,nosuid,noexec,nodev 0 2

You see the difference in the options sections. The option nosuid ignores the setuid and setgid bits completely, while noexec forbids execution of any program on that mount point, and nodev ignores device files. This sounds great, but it:

The noexec option prevents binaries from being executed directly, but was easily circumvented in earlier versions of the kernel:

alex@joker:/tmp# mount | grep tmp /dev/hda7 on /tmp type ext2 (rw,noexec,nosuid,nodev) alex@joker:/tmp# ./date bash: ./date: Permission denied alex@joker:/tmp# /lib/ld-linux.so.2 ./date Sun Dec 3 17:49:23 CET 2000

Newer versions of the kernel do however handle the noexec flag properly:

angrist:/tmp# mount | grep /tmp /dev/hda3 on /tmp type ext3 (rw,noexec,nosuid,nodev) angrist:/tmp# ./date bash: ./tmp: Permission denied angrist:/tmp# /lib/ld-linux.so.2 ./date ./date: error while loading shared libraries: ./date: failed to map segment from shared object: Operation not permitted

However, many script kiddies have exploits which try to create and execute files in /tmp. If they do not have a clue, they will fall into this pit. In other words, a user cannot be tricked into executing a trojanized binary in /tmp e.g. when he incidentally adds /tmp into his PATH.

Also be forewarned, some script might depend on /tmp being executable. Most notably, Debconf has (had?) some issues regarding this, for more information see Bug 116448.

The following is a more thorough example. A note, though: /var could be set noexec, but some software [21] keeps its programs under in /var. The same applies to the nosuid option.

/dev/sda6 /usr ext3 defaults,ro,nodev 0 2 /dev/sda12 /usr/share ext3 defaults,ro,nodev,nosuid 0 2 /dev/sda7 /var ext3 defaults,nodev,usrquota,grpquota 0 2 /dev/sda8 /tmp ext3 defaults,nodev,nosuid,noexec,usrquota,grpquota 0 2 /dev/sda9 /var/tmp ext3 defaults,nodev,nosuid,noexec,usrquota,grpquota 0 2 /dev/sda10 /var/log ext3 defaults,nodev,nosuid,noexec 0 2 /dev/sda11 /var/account ext3 defaults,nodev,nosuid,noexec 0 2 /dev/sda13 /home ext3 rw,nosuid,nodev,exec,auto,nouser,async,usrquota,grpquota 0 2 /dev/fd0 /mnt/fd0 ext3 defaults,users,nodev,nosuid,noexec 0 0 /dev/fd0 /mnt/floppy vfat defaults,users,nodev,nosuid,noexec 0 0 /dev/hda /mnt/cdrom iso9660 ro,users,nodev,nosuid,noexec 0 0

4.9.1 Setting /tmp noexec

Be careful if setting /tmp noexec when you want to install new software, since some programs might use it for installation. apt is one such program (see http://bugs.debian.org/116448) if not configured properly APT::ExtractTemplates::TempDir (see apt-extracttemplates(1)). You can set this variable in /etc/apt/apt.conf to another directory with exec privileges other than /tmp.

4.9.2 Setting /usr read-only

If you set /usr read-only you will not be able to install new packages on your Debian GNU/Linux system. You will have to first remount it read-write, install the packages and then remount it read-only. apt can be configured to run commands before and after installing packages, so you might want to configure it properly.

To do this modify /etc/apt/apt.conf and add:

DPkg { Pre-Invoke { "mount /usr -o remount,rw" }; Post-Invoke { "mount /usr -o remount,ro" }; };

Note that the Post-Invoke may fail with a "/usr busy" error message. This happens mainly when you are using files during the update that got updated. You can find these programs by running

# lsof +L1

Stop or restart these programs and run the Post-Invoke manually. Beware! This means you'll likely need to restart your X session (if you're running one) every time you do a major upgrade of your system. You might want to reconsider whether a read-only /usr is suitable for your system. See also this discussion on debian-devel about read-only /usr.

4.10 Providing secure user access

4.10.1 User authentication: PAM

PAM (Pluggable Authentication Modules) allows system administrators to choose how applications authenticate users. Note that PAM can do nothing unless an application is compiled with support for PAM. Most of the applications that are shipped with Debian have this support built in (Debian did not have PAM support before 2.2). The current default configuration for any PAM-enabled service is to emulate UNIX authentication (read /usr/share/doc/libpam0g/Debian-PAM-MiniPolicy.gz for more information on how PAM services should work in Debian).

Each application with PAM support provides a configuration file in /etc/pam.d/ which can be used to modify its behavior:

The following description is far from complete, for more information you might want to read the The Linux-PAM System Administrator's Guide (at the primary PAM distribution site). This document is also provided in the libpam-doc Debian package.

PAM offers you the possibility to go through several authentication steps at once, without the user's knowledge. You could authenticate against a Berkeley database and against the normal passwd file, and the user only logs in if he authenticates correct in both. You can restrict a lot with PAM, just as you can open your system doors very wide. So be careful. A typical configuration line has a control field as its second element. Generally it should be set to requisite, which returns a login failure if one module fails.

The first thing I like to do, is to add MD5 support to PAM applications, since this helps protect against dictionary cracks (passwords can be longer if using MD5). The following two lines should be added to all files in /etc/pam.d/ that grant access to the machine, like login and ssh.

# Be sure to install libpam-cracklib first or you will not be able to log in password required pam_cracklib.so retry=3 minlen=12 difok=3 password required pam_unix.so use_authtok nullok md5

So, what does this incantation do? The first line loads the cracklib PAM module, which provides password strength-checking, prompts for a new password with a minimum length of 12 characters, a difference of at least 3 characters from the old password, and allows 3 retries. Cracklib depends on a wordlist package (such as wenglish, wspanish, wbritish, ...), so make sure you install one that is appropriate for your language or cracklib might not be useful to you at all. [22] The second line introduces the standard authentication module with MD5 passwords and allows a zero length password. The use_authtok directive is necessary to hand over the password from the previous module.

To make sure that the user root can only log into the system from local terminals, the following line should be enabled in /etc/pam.d/login:

auth requisite pam_securetty.so

Then you should modify the list of terminals on which direct root login is allowed in /etc/securetty. Alternatively, you could enable the pam_access module and modify /etc/security/access.conf which allows for a more general and fine-tuned access control, but (unfortunately) lacks decent log messages (logging within PAM is not standardized and is particularly unrewarding problem to deal with). We'll return to access.conf a little later.

Last but not the least, the following line should be enabled in /etc/pam.d/login to set up user resource limits.

session required pam_limits.so

This restricts the system resources that users are allowed (see below in Limiting resource usage: the limits.conf file, Section 4.10.2). For example, you could restrict the number of concurrent logins (of a given group of users, or system-wide), number of processes, memory size etc.

Now edit /etc/pam.d/passwd and change the first line. You should add the option "md5" to use MD5 passwords, change the minimum length of password from 4 to 6 (or more) and set a maximum length, if you desire. The resulting line will look something like:

password required pam_unix.so nullok obscure min=6 max=11 md5

If you want to protect su, so that only some people can use it to become root on your system, you need to add a new group "wheel" to your system (that is the cleanest way, since no file has such a group permission yet). Add root and the other users that should be able to su to the root user to this group. Then add the following line to /etc/pam.d/su:

auth requisite pam_wheel.so group=wheel debug

This makes sure that only people from the group "wheel" can use su to become root. Other users will not be able to become root. In fact they will get a denied message if they try to become root.

If you want only certain users to authenticate at a PAM service, this is quite easy to achieve by using files where the users who are allowed to login (or not) are stored. Imagine you only want to allow user 'ref' to log in via ssh. So you put him into /etc/sshusers-allowed and write the following into /etc/pam.d/ssh:

auth required pam_listfile.so item=user sense=allow file=/etc/sshusers-allowed onerr=fail

Since there have been a number of so called insecure tempfile vulnerabilities, thttpd is one example (see DSA-883-1), the libpam-tmpdir is a good package to install. All you have to do is add the following to /etc/pam.d/common-session:

session optional pam_tmpdir.so

There has also been a discussion about adding this by default in etch. See http://lists.debian.org/debian-devel/2005/11/msg00297.html for more information.

Last, but not least, create /etc/pam.d/other and enter the following lines:

auth required pam_securetty.so auth required pam_unix_auth.so auth required pam_warn.so auth required pam_deny.so account required pam_unix_acct.so account required pam_warn.so account required pam_deny.so password required pam_unix_passwd.so password required pam_warn.so password required pam_deny.so session required pam_unix_session.so session required pam_warn.so session required pam_deny.so

These lines will provide a good default configuration for all applications that support PAM (access is denied by default).

4.10.2 Limiting resource usage: the limits.conf file

You should really take a serious look into this file. Here you can define user resource limits. In old releases this configuration file was /etc/limits.conf, but in newer releases (with PAM) the /etc/security/limits.conf configuration file should be used instead.

If you do not restrict resource usage, any user with a valid shell in your system (or even an intruder who compromised the system through a service or a daemon going awry) can use up as much CPU, memory, stack, etc. as the system can provide. This resource exhaustion problem can be fixed by the use of PAM.

There is a way to add resource limits to some shells (for example, bash has ulimit, see bash(1)), but since not all of them provide the same limits and since the user can change shells (see chsh(1)) it is better to place the limits on the PAM modules as they will apply regardless of the shell used and will also apply to PAM modules that are not shell-oriented.

Resource limits are imposed by the kernel, but they need to be configured through the limits.conf and the PAM configuration of the different services need to load the appropriate PAM. You can check which services are enforcing limits by running:

$ find /etc/pam.d/ \! -name "*.dpkg*" | xargs -- grep limits |grep -v ":#"

Commonly, login, ssh and the graphic session managers (gdm, kdm or xdm) should enforce user limits but you might want to do this in other PAM configuration files, such as cron, to prevent system daemons from taking over all system resources.

The specific limits settings you might want to enforce depend on your system's resources, that's one of the main reasons why no limits are enforced in the default installation.

For example, the configuration example below enforces a 100 process limit for all users (to prevent fork bombs) as well as a limit of 10MB of memory per process and a limit of 10 simultaneous logins. Users in the adm group have higher limits and can produce core files if they want to (there is only a soft limit).

* soft core 0 * hard core 0 * hard rss 1000 * hard memlock 1000 * hard nproc 100 * - maxlogins 1 * hard data 102400 * hard fsize 2048 @adm hard core 100000 @adm hard rss 100000 @adm soft nproc 2000 @adm hard nproc 3000 @adm hard fsize 100000 @adm - maxlogins 10

These would be the limits a default user (including system daemons) would have:

$ ulimit -a core file size (blocks, -c) 0 data seg size (kbytes, -d) 102400 file size (blocks, -f) 2048 max locked memory (kbytes, -l) 10000 max memory size (kbytes, -m) 10000 open files (-n) 1024 pipe size (512 bytes, -p) 8 stack size (kbytes, -s) 8192 cpu time (seconds, -t) unlimited max user processes (-u) 100 virtual memory (kbytes, -v) unlimited

And these are the limits for an administrative user:

$ ulimit -a core file size (blocks, -c) 0 data seg size (kbytes, -d) 102400 file size (blocks, -f) 100000 max locked memory (kbytes, -l) 100000 max memory size (kbytes, -m) 100000 open files (-n) 1024 pipe size (512 bytes, -p) 8 stack size (kbytes, -s) 8192 cpu time (seconds, -t) unlimited max user processes (-u) 2000 virtual memory (kbytes, -v) unlimited

For more information read:

4.10.3 User login actions: edit /etc/login.defs

The next step is to edit the basic configuration and action upon user login. Note that this file is not part of the PAM configuration, it's a configuration file honored by login and su programs, so it doesn't make sense tuning it for cases where neither of the two programs are at least indirectly called (the getty program which sits on the consoles and offers the initial login prompt does invoke login).


This variable should be set to a higher value to make it harder to use the terminal to log in using brute force. If a wrong password is typed in, the possible attacker (or normal user!) has to wait for 10 seconds to get a new login prompt, which is quite time consuming when you test passwords. Pay attention to the fact that this setting is useless if using program other than getty, such as mingetty for example.


If you enable this variable, failed logins will be logged. It is important to keep track of them to catch someone who tries a brute force attack.


If you set this variable to 'yes' it will record unknown usernames if the login failed. It is best if you use 'no' (the default) since, otherwise, user passwords might be inadvertenly logged here (if a user mistypes and they enter their password as the username). If you set it to 'yes', make sure the logs have the proper permissions (640 for example, with an appropriate group setting such as adm).


This one enables logging of su attempts to syslog. Quite important on serious machines but note that this can create privacy issues as well.


The same as SYSLOG_SU_ENAB but applies to the sg program.


As stated above, MD5 sum passwords greatly reduce the problem of dictionary attacks, since you can use longer passwords. If you are using slink, read the docs about MD5 before enabling this option. Otherwise this is set in PAM.


If MD5 passwords are activated in your PAM configuration, then this variable should be set to the same value as used there.

4.10.4 Restricting ftp: editing /etc/ftpusers

The /etc/ftpusers file contains a list of users who are not allowed to log into the host using ftp. Only use this file if you really want to allow ftp (which is not recommended in general, because it uses clear-text passwords). If your daemon supports PAM, you can also use that to allow and deny users for certain services.

FIXME (BUG): Is it a bug that the default ftpusers in Debian does not include all the administrative users (in base-passwd).

A convenient way to add all system accounts to the /etc/ftpusers is to run

$ awk -F : '{if ($3<1000) print $1}' /etc/passwd > /etc/ftpusers

4.10.5 Using su

If you really need users to become the super user on your system, e.g. for installing packages or adding users, you can use the command su to change your identity. You should try to avoid any login as user root and instead use su. Actually, the best solution is to remove su and switch to the sudo mechanism which has a broader logic and more features than su. However, su is more common as it is used on many other Unices.

4.10.6 Using sudo

sudo allows the user to execute defined commands under another user's identity, even as root. If the user is added to /etc/sudoers and authenticates himself correctly, he is able to run commands which have been defined in /etc/sudoers. Violations, such as incorrect passwords or trying to run a program you don't have permission for, are logged and mailed to root.

4.10.7 Disallow remote administrative access

You should also modify /etc/security/access.conf to disallow remote logins to administrative accounts. This way users need to invoke su (or sudo) to use any administrative powers and the appropriate audit trace will always be generated.

You need to add the following line to /etc/security/access.conf, the default Debian configuration file has a sample line commented out:


Remember to enable the pam_access module for every service (or default configuration) in /etc/pam.d/ if you want your changes to /etc/security/access.conf honored.

4.10.8 Restricting users's access

Sometimes you might think you need to have users created in your local system in order to provide a given service (pop3 mail service or ftp). Before doing so, first remember that the PAM implementation in Debian GNU/Linux allows you to validate users with a wide variety of external directory services (radius, ldap, etc.) provided by the libpam packages.

If users need to be created and the system can be accessed remotely take into account that users will be able to log in to the system. You can fix this by giving users a null (/dev/null) shell (it would need to be listed in /etc/shells). If you want to allow users to access the system but limit their movements, you can use the /bin/rbash, equivalent to adding the -r option in bash (RESTRICTED SHELL see bash(1)). Please note that even with restricted shell, a user that access an interactive program (that might allow execution of a subshell) could be able to bypass the limits of the shell.

Debian currently provides in the unstable release (and might be included in the next stable releases) the pam_chroot module (in the libpam-chroot). An alternative to it is to chroot the service that provides remote logging (ssh, telnet). [23]

If you wish to restrict when users can access the system you will have to customize /etc/security/access.conf for your needs.

Information on how to chroot users accessing the system through the ssh service is described in Chroot environment for SSH, Appendix G.

4.10.9 User auditing

If you are really paranoid you might want to add a system-wide configuration to audit what the users are doing in your system. This sections presents some tips using diverse utilities you can use. Input and output audit with script

You can use the script command to audit both what the users run and what are the results of those commands. You cannot setup script as a shell (even if you add it to /etc/shells). But you can have the shell initialization file run the following:

umask 077 exec script -q -a "/var/log/sessions/$USER"

Of course, if you do this system wide it means that the shell would not continue reading personal initialization files (since the shell gets overwritten by script). An alternative is to do this in the user's initialization files (but then the user could remove this, see the comments about this below)

You also need to setup the files in the audit directory (in the example /var/log/sessions/) so that users can write to it but cannot remove the file. This could be done, for example, by creating the user session files in advance and setting them with the append-only flag using chattr.

A useful alternative for sysadmins, which includes date information would be:

umask 077 exec script -q -a "/var/log/sessions/$USER-`date +%Y%m%d`" Using the shell history file

If you want to review what does the user type in the shell (but not what the result of that is) you can setup a system-wide /etc/profile that configures the environment so that all commands are saved into a history file. The system-wide configuration needs to be setup in such a way that users cannot remove audit capabilities from their shell. This is somewhat shell specific so make sure that all users are using a shell that supports this.

For example, for bash, the /etc/profile could be set as follows [24] :

HISTFILE=~/.bash_history HISTSIZE=10000 HISTFILESIZE=999999 # Don't let the users enter commands that are ignored # in the history file HISTIGNORE="" HISTCONTROL="" readonly HISTFILE readonly HISTSIZE readonly HISTFILESIZE readonly HISTIGNORE readonly HISTCONTROL export HISTFILE HISTSIZE HISTFILESIZE HISTIGNORE HISTCONTROL

For this to work, the user can only append information to .bash_history file. You need also to set the append-only option using chattr program for .bash_history for all users. [25].

Note that you could introduce the configuration above in the user's .profile. But then you would need to setup permissions properly in such a way that prevents the user from modifying this file. This includes: having the user's home directories not belong to the user (since he would be able to remove the file otherwise) but at the same time enable them to read the .profile configuration file and write on the .bash_history. It would be good to set the immutable flag (also using chattr) for .profile too if you do it this way. Complete user audit with accounting utilities

The previous example is a simple way to configure user auditing but might be not useful for complex systems or for those in which users do not run shells at all (or exclusively). If this is your case, you need to look at acct, the accounting utilities. These utilities will log all the commands run by users or processes in the system, at the expense of disk space.

When activating accounting, all the information on processes and users is kept under /var/account/, more specifically in the pacct. The accounting package includes some tools (sa, ac and lastcomm) to analyse this data. Other user auditing methods

If you are completely paranoid and want to audit every user's command, you could take bash source code, edit it and have it send all that the user typed into another file. Or have ttysnoop constantly monitor any new ttys [26] and dump the output into a file. Other useful program is snoopy (see also the project page) which is a user-transparent program that hooks in as a library providing a wrapper around execve() calls, any command executed is logged to syslogd using the authpriv facility (usually stored at /var/log/auth.log).

4.10.10 Reviewing user profiles

If you want to see what users are actually doing when they logon to the system you can use the wtmp database that includes all login information. This file can be processed with several utilities, amongst them sac which can output a profile on each user showing in which timeframe they usually log on to the system.

In case you have accounting activated, you can also use the tools provided by it in order to determine when the users access the system and what do they execute.

4.10.11 Setting users umasks

Depending on your user policy you might want to change how information is shared between users, that is, what the default permissions of new files created by users are.

Debian's default umask setting is 022 this means that files (and directories) can be read and accessed by the user's group and by any other users in the system. This definition is set in the standard configuration file /etc/profile which is used by all shells.

If Debian's default value is too permissive for your system you will have to change the umask setting for all the shells. More restrictive umask settings include 027 (no access is allowed to new files for the other group, i.e. to other users in the system) or 077 (no access is allowed to new files to the members the user's group). Debian (by default[27]) creates one group per user so that only the user is included in its group. Consequently 027 and 077 are equivalent as the user's group contains only the user himself.

This change is set by defining a proper umask setting for all users. You can change this by introducing an umask call in the shell configuration files: /etc/profile (source by all Bourne-compatible shells), /etc/csh.cshrc, /etc/csh.login, /etc/zshrc and probably some others (depending on the shells you have installed on your system). You can also change the UMASK setting in /etc/login.defs, Of all of these the last one that gets loaded by the shell takes precedence. The order is: the default system configuration for the user's shell (i.e. /etc/profile and other system-wide configuration files) and then the user's shell (his ~/.profile, ~/.bash_profile, etc...). Some shells, however, can be executed with a nologin value which might skip sourcing some of those files. See your shell's manpage for additional information.

For connections that make use of login the UMASK definition in /etc/login.defs is used before any of the others. However, that value does not apply to user executed programs that do not use login such as those run through su, cron or ssh.

Don't forget to review and maybe modify the dotfiles under /etc/skel/ since these will be new user's defaults when created with the adduser command. Debian default dotfiles do not include any umask call but if there is any in the dotfiles newly created users might a different value.

Note, however that users can modify their own umask setting if they want to, making it more permissive or more restricted, by changing their own dotfiles.

The libpam-umask package adjusts the users' default umask using PAM. Add the following, after installing the package, to /etc/pam.d/common-session:

session optional pam_umask.so umask=077

Finally, you should consider changing root's default 022 umask (as defined in /root/.bashrc) to a more strict umask. That will prevent the system administrator from inadvertenly dropping sensitive files when working as root to world-readable directories (such as /tmp) and having them available for your average user.

4.10.12 Limiting what users can see/access

FIXME: Content needed. Describe the consequences of changing packages permissions when upgrading (an admin this paranoid should chroot his users BTW) if not using dpkg-statoverride.

If you need to grant users access to the system with a shell think about it very carefully. A user can, by default unless in a severely restricted environment (like a chroot jail), retrieve quite a lot of information from your system including:

What can a user see in your system? Probably quite a lot of things, try this (take a deep breath):

find / -type f -a -perm +006 2>/dev/null find / -type d -a -perm +007 2>/dev/null

The output is the list of files that a user can see and the directories to which he has access. Limiting access to other user's information

If you still grant shell access to users you might want to limit what information they can view from other users. Users with shell access have a tendency to create quite a number of files under their $HOMEs: mailboxes, personal documents, configuration of X/GNOME/KDE applications...

In Debian each user is created with one associated group, and no two users belong to the same group. This is the default behavior: when an user account is created, a group of the same name is created too, and the user is assigned to it. This avoids the concept of a common users group which might make it more difficult for users to hide information from other users.

However, users' $HOME directories are created with 0755 permissions (group-readable and world-readable). The group permissions is not an issue since only the user belongs to the group, however the world permissions might (or might not) be an issue depending on your local policy.

You can change this behavior so that user creation provides different $HOME permissions. To change the behavior for new users when they get created, change DIR_MODE in the configuration file /etc/adduser.conf to 0750 (no world-readable access).

Users can still share information, but not directly in their $HOME directories unless they change its permissions.

Note that disabling world-readable home directories will prevent users from creating their personal web pages in the ~/public_html directory, since the web server will not be able to read one component in the path - namely their $HOME directory. If you want to permit users to publish HTML pages in their ~/public_html, then change DIR_MODE to 0751. This will allow the web server to access the final public_html directory (which itself should have a mode of 0755) and provide the content published by users. Of course, we are only talking about a default configuration here; users can generally tune modes of their own files completely to their liking, or you could keep content intended for the web in a separate location which is not a subdirectory of user's $HOME directory.

4.10.13 Generating user passwords

There are many cases when an administrator needs to create many user accounts and provide passwords for all of them. Of course, the administrator could easily just set the password to be the same as the user's account name, but that would not be very sensitive security-wise. A better approach is to use a password generating program. Debian provides makepasswd, apg and pwgen packages which provide programs (the name is the same as the package) that can be used for this purpose. Makepasswd will generate true random passwords with an emphasis on security over pronounceability while pwgen will try to make meaningless but pronounceable passwords (of course this might depend on your mother language). Apg has algorithms to provide for both (there is a client/server version for this program but it is not included in the Debian package).

Passwd does not allow non-interactive assignation of passwords (since it uses direct tty access). If you want to change passwords when creating a large number of users you can create them using adduser with the --disabled-login option and then use usermod or chpasswd [28] (both from the passwd package so you already have them installed). If you want to use a file with all the information to make users as a batch process you might be better off using newusers.

4.10.14 Checking user passwords

User passwords can sometimes become the weakest link in the security of a given system. This is due to some users choosing weak passwords for their accounts (and the more of them that have access to it the greater the chances of this happening). Even if you established checks with the cracklib PAM module and password limits as described in User authentication: PAM, Section 4.10.1 users will still be able to use weak passwords. Since user access might include remote shell access (over ssh, hopefully) it's important to make password guessing as hard as possible for the remote attackers, especially if they were somehow able to collect important information such as usernames or even the passwd and shadow files themselves.

A system administrator must, given a big number of users, check if the passwords they have are consistent with the local security policy. How to check? Try to crack them as an attacker would if he had access to the hashed passwords (the /etc/shadow file).

An administrator can use john or crack (both are brute force password crackers) together with an appropriate wordlist to check users' passwords and take appropriate action when a weak password is detected. You can search for Debian GNU packages that contain word lists using apt-cache search wordlist, or visit the classic Internet wordlist sites such as ftp://ftp.ox.ac.uk/pub/wordlists or ftp://ftp.cerias.purdue.edu/pub/dict.

4.10.15 Logging off idle users

Idle users are usually a security problem, a user might be idle maybe because he's out to lunch or because a remote connection hung and was not re-established. For whatever the reason, idle users might lead to a compromise:

Some remote systems have even been compromised through an idle (and detached) screen.

Automatic disconnection of idle users is usually a part of the local security policy that must be enforced. There are several ways to do this:

The timeoutd or autolog daemons are the preferred method since, after all, users can change their default shell or can, after running their default shell, switch to another (uncontrolled) shell.

4.11 Using tcpwrappers

TCP wrappers were developed when there were no real packet filters available and access control was needed. Nevertheless, they're still very interesting and useful. The TCP wrappers allow you to allow or deny a service for a host or a domain and define a default allow or deny rule (all performed on the application level). If you want more information take a look at hosts_access(5).

Many services installed in Debian are either:

On the one hand, for services configured in /etc/inetd.conf (this includes telnet, ftp, netbios, swat and finger) you will see that the configuration file executes /usr/sbin/tcpd first. On the other hand, even if a service is not launched by the inetd superdaemon, support for the tcp wrappers rules can be compiled into it. Services compiled with tcp wrappers in Debian include ssh, portmap, in.talk, rpc.statd, rpc.mountd, gdm, oaf (the GNOME activator daemon), nessus and many others.

To see which packages use tcpwrappers [29] try:

$ apt-cache rdepends libwrap0

Take this into account when running tcpdchk (a very useful TCP wrappers config file rule and syntax checker). When you add stand-alone services (that are directly linked with the wrapper library) into the hosts.deny and hosts.allow files, tcpdchk will warn you that it is not able to find the mentioned services since it only looks for them in /etc/inetd.conf (the manpage is not totally accurate here).

Now, here comes a small trick, and probably the smallest intrusion detection system available. In general, you should have a decent firewall policy as a first line, and tcp wrappers as the second line of defense. One little trick is to set up a SPAWN [30] command in /etc/hosts.deny that sends mail to root whenever a denied service triggers wrappers:

ALL: ALL: SPAWN ( \ echo -e "\n\ TCP Wrappers\: Connection refused\n\ By\: $(uname -n)\n\ Process\: %d (pid %p)\n\ User\: %u\n\ Host\: %c\n\ Date\: $(date)\n\ " | /usr/bin/mail -s "Connection to %d blocked" root) &

Beware: The above printed example is open to a DoS attack by making many connections in a short period of time. Many emails mean a lot of file I/O by sending only a few packets.

4.12 The importance of logs and alerts

It is easy to see that the treatment of logs and alerts is an important issue in a secure system. Suppose a system is perfectly configured and 99% secure. If the 1% attack occurs, and there are no security measures in place to, first, detect this and, second, raise alarms, the system is not secure at all.

Debian GNU/Linux provides some tools to perform log analysis, most notably swatch, [31] logcheck or log-analysis (all will need some customisation to remove unnecessary things from the report). It might also be useful, if the system is nearby, to have the system logs printed on a virtual console. This is useful since you can (from a distance) see if the system is behaving properly. Debian's /etc/syslog.conf comes with a commented default configuration; to enable it uncomment the lines and restart syslogd (/etc/init.d/syslogd restart):

daemon,mail.*;\ news.=crit;news.=err;news.=notice;\ *.=debug;*.=info;\ *.=notice;*.=warn /dev/tty8

To colorize the logs, you could take a look at colorize, ccze or glark. There is a lot to log analysis that cannot be fully covered here, so a good information resource would be Log Analysis website. In any case, even automated tools are no match for the best analysis tool: your brain.

4.12.1 Using and customizing logcheck

The logcheck package in Debian is divided into the three packages logcheck (the main program), logcheck-database (a database of regular expressions for the program) and logtail (prints loglines that have not yet been read). The Debian default (in /etc/cron.d/logcheck) is that logcheck is run every hour and after reboots.

This tool can be quite useful if properly customized to alert the administrator of unusual system events. Logcheck can be fully customized so that it sends mails based on events found in the logs and worthy of attention. The default installation includes profiles for ignored events and policy violations for three different setups (workstation, server and paranoid). The Debian package includes a configuration file /etc/logcheck/logcheck.conf, sourced by the program, that defines which user the checks are sent to. It also provides a way for packages that provide services to implement new policies in the directories: /etc/logcheck/cracking.d/_packagename_, /etc/logcheck/violations.d/_packagename_, /etc/logcheck/violations.ignore.d/_packagename_, /etc/logcheck/ignore.d.paranoid/_packagename_, /etc/logcheck/ignore.d.server/_packagename_, and /etc/logcheck/ignore.d.workstation/_packagename_. However, not many packages currently do so. If you have a policy that can be useful for other users, please send it as a bug report for the appropriate package (as a wishlist bug). For more information read /usr/share/doc/logcheck/README.Debian.

The best way to configure logcheck is to edit its main configuration file /etc/logcheck/logcheck.conf after installation. Change the default user (root) to whom reports should be mailed. You should set the reportlevel in there, too. logcheck-database has three report levels of increasing verbosity: workstation, server, paranoid. "server" being the default level, paranoid is only recommended for high-security machines running as few services as possible and workstation for relatively sheltered, non-critical machines. If you wish to add new log files just add them to /etc/logcheck/logcheck.logfiles. It is tuned for default syslog install.

Once this is done you might want to check the mails that are sent, for the first few days/weeks/months. If you find you are sent messages you do not wish to receive, just add the regular expressions (see regex(7) and egrep(1)) that correspond to these messages to the /etc/logcheck/ignore.d.reportlevel/local. Try to match the whole logline. Details on howto write rules are explained in /usr/share/doc/logcheck-database/README.logcheck-database.gz. It's an ongoing tuning process; once the messages that are sent are always relevant you can consider the tuning finished. Note that if logcheck does not find anything relevant in your system it will not mail you even if it does run (so you might get a mail only once a week, if you are lucky).

4.12.2 Configuring where alerts are sent

Debian comes with a standard syslog configuration (in /etc/syslog.conf) that logs messages to the appropriate files depending on the system facility. You should be familiar with this; have a look at the syslog.conf file and the documentation if not. If you intend to maintain a secure system you should be aware of where log messages are sent so they do not go unnoticed.

For example, sending messages to the console also is an interesting setup useful for many production-level systems. But for many such systems it is also important to add a new machine that will serve as loghost (i.e. it receives logs from all other systems).

Root's mail should be considered also, many security controls (like snort) send alerts to root's mailbox. This mailbox usually points to the first user created in the system (check /etc/aliases). Take care to send root's mail to some place where it will be read (either locally or remotely).

There are other role accounts and aliases on your system. On a small system, it's probably simplest to make sure that all such aliases point to the root account, and that mail to root is forwarded to the system administrator's personal mailbox.

FIXME: It would be interesting to tell how a Debian system can send/receive SNMP traps related to security problems (jfs). Check: snmptrapfmt, snmp and snmpd.

4.12.3 Using a loghost

A loghost is a host which collects syslog data remotely over the network. If one of your machines is cracked, the intruder is not able to cover his tracks, unless he hacks the loghost as well. So, the loghost should be especially secure. Making a machine a loghost is simple. Just start the syslogd with syslogd -r and a new loghost is born. In order to do this permanently in Debian, edit /etc/default/syslogd and change the line




Next, configure the other machines to send data to the loghost. Add an entry like the following to /etc/syslog.conf:

facility.level @your_loghost

See the documentation for what to use in place of facility and level (they should not be entered verbatim like this). If you want to log everything remotely, just write:

*.* @your_loghost

into your syslog.conf. Logging remotely as well as locally is the best solution (the attacker might presume to have covered his tracks after deleting the local log files). See the syslog(3), syslogd(8) and syslog.conf(5) manpages for additional information.

4.12.4 Log file permissions

It is not only important to decide how alerts are used, but also who has read/modify access to the log files (if not using a remote loghost). Security alerts which the attacker can change or disable are not worth much in the event of an intrusion. Also, you have to take into account that log files might reveal quite a lot of information about your system to an intruder if he has access to them.

Some log file permissions are not perfect after the installation (but of course this really depends on your local security policy). First /var/log/lastlog and /var/log/faillog do not need to be readable by normal users. In the lastlog file you can see who logged in recently, and in the faillog you see a summary of failed logins. The author recommends chmod 660 for both. Take a brief look at your log files and decide very carefully which log files to make readable/writable for a user with a UID other than 0 and a group other than 'adm' or 'root'. You can easily check this in your system with:

# find /var/log -type f -exec ls -l {} \; | cut -c 17-35 |sort -u (see to what users do files in /var/log belong) # find /var/log -type f -exec ls -l {} \; | cut -c 26-34 |sort -u (see to what groups do files in /var/log belong) # find /var/log -perm +004 (files which are readable by any user) # find /var/log \! -group root \! -group adm -exec ls -ld {} \; (files which belong to groups not root or adm)

To customize how log files are created you will probably have to customize the program that generates them. If the log file gets rotated, however, you can customize the behavior of creation and rotation.

4.13 Adding kernel patches

Debian GNU/Linux provides some of the patches for the Linux kernel that enhance its security. These include:

The following security kernel patches are only available for old kernel versions in woody and are deprecated:

However, some patches have not been provided in Debian yet. If you feel that some of these should be included please ask for it at the Work Needing and Prospective Packages.

4.14 Protecting against buffer overflows

Buffer overflow is the name of a common attack to software [33] which makes use of insufficient boundary checking (a programming error, most commonly in the C language) in order to execute machine code through program inputs. These attacks, against server software which listen to connections remotely and against local software which grant higher privileges to users (setuid or setgid) can result in the compromise of any given system.

There are mainly four methods to protect against buffer overflows:

Debian GNU/Linux, as of the 3.0 release, provides software to introduce all of these methods except for the protection on source code compilation (but this has been requested in Bug #213994).

Notice that even if Debian provided a compiler which featured stack/buffer overflow protection all packages would need to be recompiled in order to introduce this feature. This is, in fact, what the Adamantix distribution does (among other features). The effect of this new feature on the stability of software is yet to be determined (some programs or some processor architectures might break due to it).

In any case, be aware that even these workarounds might not prevent buffer overflows since there are ways to circumvent these, as described in phrack's magazine issue 58 or in CORE's Advisory Multiple vulnerabilities in stack smashing protection technologies.

If you want to test out your buffer overflow protection once you have implemented it (regardless of the method) you might want to install the paxtest and run the tests it provides.

4.14.1 Kernel patch protection for buffer overflows

Kernel patches related to buffer overflows include the Openwall patch provides protection against buffer overflows in 2.2 linux kernels. For 2.4 or newer kernels, you need to use the Exec-shield implementation, or the PaX implementation (provided in the grsecurity patch, kernel-patch-2.4-grsecurity, and in the Adamantix patch, kernel-patch-adamantix). For more information on using these patches read the the section Adding kernel patches, Section 4.13.

4.14.2 Testing programs for overflows

The use of tools to detect buffer overflows requires, in any case, of programming experience in order to fix (and recompile) the code. Debian provides, for example: bfbtester (a buffer overflow tester that brute-forces binaries through command line and environment overflows). Other packages of interest would also be rats, pscan, flawfinder and splint.

4.15 Secure file transfers

During normal system administration one usually needs to transfer files in and out from the installed system. Copying files in a secure manner from a host to another can be achieved by using the ssh server package. Another possibility is the use of ftpd-ssl, a ftp server which uses the Secure Socket Layer to encrypt the transmissions.

Any of these methods need special clients. Debian does provide client software, such as scp from the ssh package, which works like rcp but is encrypted completely, so the bad guys cannot even find out WHAT you copy. There is also a ftp-ssl package for the equivalent server. You can find clients for these software even for other operating systems (non-UNIX), putty and winscp provide secure copy implementations for any version of Microsoft's operating system.

Note that using scp provides access to the users to all the file system unless chroot'ed as described in Chrooting ssh, Section 5.1.1. FTP access can be chroot'ed, probably easier depending on you chosen daemon, as described in Securing FTP, Section 5.3. If you are worried about users browsing your local files and want to have encrypted communication you can either use an ftp daemon with SSL support or combine clear-text ftp and a VPN setup (see Virtual Private Networks, Section 8.5).

4.16 File system limits and control

4.16.1 Using quotas

Having a good quota policy is important, as it keeps users from filling up the hard disk(s).

You can use two different quota systems: user quota and group quota. As you probably figured out, user quota limits the amount of space a user can take up, group quota does the equivalent for groups. Keep this in mind when you're working out quota sizes.

There are a few important points to think about in setting up a quota system:

Every partition or directory to which users have full write access should be quota enabled. Calculate and assign a workable quota size for those partitions and directories which combines usability and security.

So, now you want to use quotas. First of all you need to check whether you enabled quota support in your kernel. If not, you will need to recompile it. After this, control whether the package quota is installed. If not you will need this one as well.

Enabling quota for the respective file systems is as easy as modifying the defaults setting to defaults,usrquota in your /etc/fstab file. If you need group quota, substitute usrquota to grpquota. You can also use them both. Then create empty quota.user and quota.group files in the roots of the file systems you want to use quotas on (e.g. touch /home/quota.user /home/quota.group for a /home file system).

Restart quota by doing /etc/init.d/quota stop;/etc/init.d/quota start. Now quota should be running, and quota sizes can be set.

Editing quotas for a specific user can be done by edquota -u <user>. Group quotas can be modified with edquota -g <group>. Then set the soft and hard quota and/or inode quotas as needed.

For more information about quotas, read the quota man page, and the quota mini-howto (/usr/share/doc/HOWTO/en-html/mini/Quota.html). You may also want to look at pam_limits.so.

4.16.2 The ext2 filesystem specific attributes (chattr/lsattr)

In addition to the usual Unix permissions, the ext2 and ext3 filesystems offer a set of specific attributes that give you more control over the files on your system. Unlike the basic permissions, these attributes are not displayed by the usual ls -l command or changed using chmod, and you need two other utilities, lsattr and chattr (in package e2fsprogs) to manage them. Note that this means that these attributes will usually not be saved when you backup your system, so if you change any of them, it may be worth saving the successive chattr commands in a script so that you can set them again later if you have to restore a backup.

Among all available attributes, the two that are most important for increasing security are referenced by the letters 'i' and 'a', and they can only be set (or removed) by the superuser:

These attributes can also be set for directories, in which case everyone is denied the right to modify the contents of a directory list (e.g. rename or remove a file, ...). When applied to a directory, the append attribute only allows file creation.

It is easy to see how the 'a' attribute improves security, by giving to programs that are not running as the superuser the ability to add data to a file without modifying its previous content. On the other hand, the 'i' attribute seems less interesting: after all, the superuser can already use the basic Unix permissions to restrict access to a file, and an intruder that would get access to the superuser account could always use the chattr program to remove the attribute. Such an intruder may first be confused when he sees that he is not able to remove a file, but you should not assume that he is blind - after all, he got into your system! Some manuals (including a previous version of this document) suggest to simply remove the chattr and lsattr programs from the system to increase security, but this kind of strategy, also known as "security by obscurity", is to be absolutely avoided, since it provides a false sense of security.

A secure way to solve this problem is to use the capabilities of the Linux kernel, as described in Proactive defense, Section The capability of interest here is called CAP_LINUX_IMMUTABLE: if you remove it from the capabilities bounding set (using for example the command lcap CAP_LINUX_IMMUTABLE) it won't be possible to change any 'a' or 'i' attribute on your system anymore, even by the superuser ! A complete strategy could be as follows:

  1. Set the attributes 'a' and 'i' on any file you want;

  1. Add the command lcap CAP_LINUX_IMMUTABLE (as well as lcap CAP_SYS_MODULE, as suggested in Proactive defense, Section to one of the startup scripts;

  1. Set the 'i' attribute on this script and other startup files, as well as on the lcap binary itself;

  1. Execute the above command manually (or reboot your system to make sure everything works as planned).

Now that the capability has been removed from the system, an intruder cannot change any attribute on the protected files, and thus cannot change or remove the files. If he forces the machine to reboot (which is the only way to restore the capabilities bounding set), it will easily be detected, and the capability will be removed again as soon as the system restarts anyway. The only way to change a protected file would be to boot the system in single-user mode or using another bootdisk, two operations that require physical access to the machine !

4.16.3 Checking file system integrity

Are you sure /bin/login on your hard drive is still the binary you installed there some months ago? What if it is a hacked version, which stores the entered password in a hidden file or mails it in clear-text version all over the Internet?

The only method to have some kind of protection is to check your files every hour/day/month (I prefer daily) by comparing the actual and the old md5sum of this file. Two files cannot have the same md5sum (the MD5 digest is 128 bits, so the chance that two different files will have the same md5sum is roughly one in 3.4e3803), so you're on the safe site here, unless someone has also hacked the algorithm that creates md5sums on that machine. This is, well, extremely difficult and very unlikely. You really should consider this auditing of your binaries as very important, since it is an easy way to recognize changes at your binaries.

Common tools used for this are sxid, aide (Advanced Intrusion Detection Environment), tripwire, integrit and samhain. Installing debsums will also help you to check the file system integrity, by comparing the md5sums of every file against the md5sums used in the Debian package archive. But beware: those files can easily be changed by an attacker and not all packages provide md5sums listings for the binaries they provided. For more information please read Do periodic integrity checks, Section 10.2 and Taking a snapshot of the system, Section 4.18.

You might want to use locate to index the whole filesystem, if so, consider the implications of that. The Debian findutils package contains locate which runs as user nobody, and so it only indexes files which are visible to everybody. However, if you change it's behaviour you will make all file locations visible to all users. If you want to index all the filesystem (not the bits that the user nobody can see) you can replace locate with the package slocate. slocate is labeled as a security enhanced version of GNU locate, but it actually provides additional file-locating functionality. When using slocate, the user only sees the files he really has access to and you can exclude any files or directories on the system. The slocate package runs its update process with higher privledges than locate, and indexes every file. Users are then able to quickly search for every file which they are able to see. slocate doesn't let them see new files; it filters the output based on your UID.

You might want to use bsign or elfsign. elfsign provides an utility to add a digital signature to an ELF binary and a second utility to verify that signature. The current implementation uses PKI to sign the checksum of the binary. The benefits of doing this are that it enables one to determine if a binary has been modified and who created it. bsign uses GPG, elfsign uses PKI (X.509) certificates (OpenSSL).

4.16.4 Setting up setuid check

The Debian checksecurity package provides a cron job that runs daily in /etc/cron.daily/checksecurity [34]. This cron job will run the /usr/sbin/checksecurity script that will store information of this changes.

The default behavior does not send this information to the superuser but, instead keeps daily copies of the changes in /var/log/setuid.changes. You should set the MAILTO variable (in /etc/checksecurity.conf) to 'root' to have this information mailed to him. See checksecurity(8) for more configuration info.

4.17 Securing network access

FIXME: More (Debian-specific) content needed.

4.17.1 Configuring kernel network features

Many features of the kernel can be modified while running by echoing something into the /proc file system or by using sysctl. By entering /sbin/sysctl -A you can see what you can configure and what the options are, and it can be modified running /sbin/sysctl -w variable=value (see sysctl(8)). Only in rare cases do you need to edit something here, but you can increase security that way as well. For example:

net/ipv4/icmp_echo_ignore_broadcasts = 1

This is a Windows emulator because it acts like Windows on broadcast ping if this option is set to 1. That is, ICMP echo requests sent to the broadcast address will be ignored. Otherwise, it does nothing.

If you want to prevent you system from answering ICMP echo requests, just enable this configuration option:

net/ipv4/icmp_echo_ignore_all = 1

To log packets with impossible addresses (due to wrong routes) on your network use:

/proc/sys/net/ipv4/conf/all/log_martians = 1

For more information on what things can be done with /proc/sys/net/ipv4/* read /usr/src/linux/Documentation/filesystems/proc.txt. All the options are described thoroughly under /usr/src/linux/Documentation/networking/ip-sysctl.txt [35].

4.17.2 Configuring syncookies

This option is a double-edged sword. On the one hand it protects your system against syn packet flooding; on the other hand it violates defined standards (RFCs).

net/ipv4/tcp_syncookies = 1

If you want to change this option each time the kernel is working you need to change it in /etc/network/options by setting syncookies=yes. This will take effect when ever /etc/init.d/networking is run (which is typically done at boot time) while the following will have a one-time effect until the reboot:

echo 1 > /proc/sys/net/ipv4/tcp_syncookies

This option will only be available if the kernel is compiled with the CONFIG_SYNCOOKIES. All Debian kernels are compiled with this option builtin but you can verify it running:

$ sysctl -A |grep syncookies net/ipv4/tcp_syncookies = 1

For more information on TCP syncookies read http://cr.yp.to/syncookies.html.

4.17.3 Securing the network on boot-time

When setting configuration options for the kernel networking you need configure it so that it's loaded every time the system is restarted. The following example enables many of the previous options as well as other useful options.

There are actually two ways to configure your network at boot time. You can configure /etc/sysctl.conf (see: sysctl.conf(5)) or introduce a script that is called when the interface is enabled. The first option will be applied to all interfaces, whileas the second option allows you to configure this on a per-interface basis.

An example of a /etc/sysctl.conf configuration that will secure some network options at the kernel level is shown below. Notice the comment in it, /etc/network/options might override some values if they contradict those in this file when the /etc/init.d/networking is run (which is later than procps on the startup sequence).

# # /etc/sysctl.conf - Configuration file for setting system variables # See sysctl.conf (5) for information. Also see the files under # Documentation/sysctl/, Documentation/filesystems/proc.txt, and # Documentation/networking/ip-sysctl.txt in the kernel sources # (/usr/src/kernel-$version if you have a kernel-package installed) # for more information of the values that can be defined here. # # Be warned that /etc/init.d/procps is executed to set the following # variables. However, after that, /etc/init.d/networking sets some # network options with builtin values. These values may be overridden # using /etc/network/options. # #kernel.domainname = example.com # Additional settings - adapted from the script contributed # by Dariusz Puchala (see below) # Ignore ICMP broadcasts net/ipv4/icmp_echo_ignore_broadcasts = 1 # # Ignore bogus ICMP errors net/ipv4/icmp_ignore_bogus_error_responses = 1 # # Do not accept ICMP redirects (prevent MITM attacks) net/ipv4/conf/all/accept_redirects = 0 # _or_ # Accept ICMP redirects only for gateways listed in our default # gateway list (enabled by default) # net/ipv4/conf/all/secure_redirects = 1 # # Do not send ICMP redirects (we are not a router) net/ipv4/conf/all/send_redirects = 0 # # Do not forward IP packets (we are not a router) # Note: Make sure that /etc/network/options has 'ip_forward=no' net/ipv4/conf/all/forwarding = 0 # # Enable TCP Syn Cookies # Note: Make sure that /etc/network/options has 'syncookies=yes' net/ipv4/tcp_syncookies = 1 # # Log Martian Packets net/ipv4/conf/all/log_martians = 1 # # Turn on Source Address Verification in all interfaces to # prevent some spoofing attacks # Note: Make sure that /etc/network/options has 'spoofprotect=yes' net/ipv4/conf/all/rp_filter = 1 # # Do not accept IP source route packets (we are not a router) net/ipv4/conf/all/accept_source_route = 0

To use the script you need to first create the script, for example, in /etc/network/interface-secure (the name is given as an example) and call it from /etc/network/interfaces like this:

auto eth0 iface eth0 inet static address xxx.xxx.xxx.xxx netmask 255.255.255.xxx broadcast xxx.xxx.xxx.xxx gateway xxx.xxx.xxx.xxx pre-up /etc/network/interface-secure

In this example, before the interface eth0 is enabled the script will be called to secure all network interfaces as shown below.

#!/bin/sh -e # Script-name: /etc/network/interface-secure # # Modifies some default behavior in order to secure against # some TCP/IP spoofing & attacks for all interfaces. # # Contributed by Dariusz Puchalak. # echo 1 > /proc/sys/net/ipv4/icmp_echo_ignore_broadcasts # Broadcast echo protection enabled. echo 0 > /proc/sys/net/ipv4/conf/all/forwarding # IP forwarding disabled. echo 1 > /proc/sys/net/ipv4/tcp_syncookies # TCP syn cookies protection enabled. echo 1 >/proc/sys/net/ipv4/conf/all/log_martians # Log strange packets. # (this includes spoofed packets, source routed packets, redirect packets) # but be careful with this on heavy loaded web servers. echo 1 > /proc/sys/net/ipv4/icmp_ignore_bogus_error_responses # Bad error message protection enabled. # IP spoofing protection. echo 1 > /proc/sys/net/ipv4/conf/all/rp_filter # Disable ICMP redirect acceptance. echo 0 > /proc/sys/net/ipv4/conf/all/accept_redirects echo 0 > /proc/sys/net/ipv4/conf/all/send_redirects # Disable source routed packets. echo 0 > /proc/sys/net/ipv4/conf/all/accept_source_route exit 0

Notice that you can actually have per-interface scripts that will enable different network options for different interfaces (if you have more than one), just change the pre-up line to:

pre-up /etc/network/interface-secure $IFACE

And use a script which will only apply changes to a specific interface, not to all of the interfaces available. Notice that some networking options can only be enabled globally, however. A sample script is this one:

#!/bin/sh -e # Script-name: /etc/network/interface-secure # # Modifies some default behavior in order to secure against # some TCP/IP spoofing & attacks for a given interface. # # Contributed by Dariusz Puchalak. # IFACE=$1 if [ -z "$IFACE" ] ; then echo "$0: Must give an interface name as argument!" echo "Usage: $0 <interface>" exit 1 fi if [ ! -e /proc/sys/net/ipv4/conf/$IFACE/ ]; then echo "$0: Interface $IFACE does not exit (cannot find /proc/sys/net/ipv4/conf/)" exit 1 fi echo 0 > /proc/sys/net/ipv4/conf/$IFACE/forwarding # IP forwarding disabled. echo 1 >/proc/sys/net/ipv4/conf/$IFACE/log_martians # Log strange packets. # (this includes spoofed packets, source routed packets, redirect packets) # but be careful with this on heavy loaded web servers. # IP spoofing protection. echo 1 > /proc/sys/net/ipv4/conf/$IFACE/rp_filter # Disable ICMP redirect acceptance. echo 0 > /proc/sys/net/ipv4/conf/$IFACE/accept_redirects echo 0 > /proc/sys/net/ipv4/conf/$IFACE/send_redirects # Disable source routed packets. echo 0 > /proc/sys/net/ipv4/conf/$IFACE/accept_source_route exit 0

An alternative solution is to create an init.d script and have it run on bootup (using update-rc.d to create the appropriate rc.d links).

4.17.4 Configuring firewall features

In order to have firewall capabilities, either to protect the local system or others behind it, the kernel needs to be compiled with firewall capabilities. The standard Debian 2.2 kernel (Linux 2.2) provides the packet filter ipchains firewall, Debian 3.0 standard kernel (Linux 2.4) provides the stateful packet filter iptables (netfilter) firewall.

In any case, it is pretty easy to use a kernel different from the one provided by Debian. You can find pre-compiled kernels as packages you can easily install in the Debian system. You can also download the kernel sources using the kernel-source-X and build custom kernel packages using make-kpkg from the kernel-package package.

Setting up firewalls in Debian is discussed more thoroughly in Adding firewall capabilities, Section 5.14.

4.17.5 Disabling weak-end hosts issues

Systems with more than one interface on different networks can have services configured so that they will bind only to a given IP address. This usually prevents access to services when requested through any other address. However, this does not mean (although it is a common misconception) that the service is bound to a given hardware address (interface card). [36]

This is not an ARP issue and it's not an RFC violation (it's called weak end host in RFC1122, section Remember, IP addresses have nothing to do with physical interfaces.

On 2.2 (and previous) kernels this can be fixed with:

# echo 1 > /proc/sys/net/ipv4/conf/all/hidden # echo 1 > /proc/sys/net/ipv4/conf/eth0/hidden # echo 1 > /proc/sys/net/ipv4/conf/eth1/hidden .....

On later kernels this can be fixed either with:

Along this text there will be many occasions in which it is shown how to configure some services (sshd server, apache, printer service...) in order to have them listening on any given address, the reader should take into account that, without the fixes given here, the fix would not prevent accesses from within the same (local) network. [39]

FIXME: Comments on Bugtraq indicate there is a Linux specific method to bind to a given interface.

FIXME: Submit a bug against netbase so that the routing fix is standard behavior in Debian?

4.17.6 Protecting against ARP attacks

When you don't trust the other boxes on your LAN (which should always be the case, because it's the safest attitude) you should protect yourself from the various existing ARP attacks.

As you know the ARP protocol is used to link IP addresses to MAC addresses (see RFC826 for all the details). Every time you send a packet to an IP address an ARP resolution is done (first by looking into the local ARP cache then if the IP isn't present in the cache by broadcasting an ARP query) to find the target's hardware address. All the ARP attacks aim to fool your box into thinking that box B's IP address is associated to the intruder's box's MAC address; Then every packet that you want to send to the IP associated to box B will be send to the intruder's box...

Those attacks (ARP cache poisoning, ARP spoofing...) allow the attacker to sniff the traffic even on switched networks, to easily hijack connections, to disconnect any host from the network... ARP attacks are powerful and simple to implement, and several tools exists, such as arpspoof from the dsniff package or arpoison.

However, there is always a solution:

4.18 Taking a snapshot of the system

Before putting the system into production system you could take a snapshot of the whole system. This snapshot could be used in the event of a compromise (see After the compromise (incident response), Chapter 11). You should remake this upgrade whenever the system is upgraded, especially if you upgrade to a new Debian release.

For this you can use a writable removable-media that can be set up read-only, this could be a floppy disk (read protected after use), a CD on a CD-ROM unit (you could use a rewritable CD-ROM so you could even keep backups of md5sums in different dates), or a USB disk or MMC card (if your system can access those and they can be write protected).

The following script creates such a snapshot:

#!/bin/bash /bin/mount /dev/fd0 /mnt/floppy if [ ! -f /usr/bin/md5sum ] ; then echo "Cannot find md5sum. Aborting." exit 1 fi /bin/cp /usr/bin/md5sum /mnt/floppy echo "Calculating md5 database" >/mnt/floppy/md5checksums.txt for dir in /bin/ /sbin/ /usr/bin/ /usr/sbin/ /lib/ /usr/lib/ do find $dir -type f | xargs /usr/bin/md5sum >>/mnt/floppy/md5checksums-lib.txt done echo "post installation md5 database calculated" if [ ! -f /usr/bin/sha1sum ] ; then echo "Cannot find sha1sum" else /bin/cp /usr/bin/sha1sum /mnt/floppy echo "Calculating SHA-1 database" >/mnt/floppy/sha1checksums.txt for dir in /bin/ /sbin/ /usr/bin/ /usr/sbin/ /lib/ /usr/lib/ do find $dir -type f | xargs /usr/bin/sha1sum >>/mnt/floppy/sha1checksums-lib.txt done echo "post installation sha1 database calculated" fi /bin/umount /dev/fd0 exit 0

Note that the md5sum binary (and sha1sum, if available) is placed on the floppy drive so it can be used later on to check the binaries of the system (just in case it gets trojaned). However, if you want to make sure that you are running a legitimate binary, you might want to either compile a static copy of the md5sum binary and use that one (to prevent a trojaned libc library from interfering with the binary) or to use the snapshot of md5sums only from a clean environment such as a rescue CD-ROM or a Live-CD (to prevent a trojaned kernel from interfering). I cannot stress this enough: if you are on a compromised system you cannot trust its output, see After the compromise (incident response), Chapter 11.

The snapshot does not include the files under /var/lib/dpkg/info which includes the MD5 hashes of installed packages (in files ending with .md5sums). You could copy this information along too, however you should notice:

Once the snapshot is done you should make sure to set the medium read-only. You can then store it for backup or place it in the drive and use it to drive a cron check nightly comparing the original md5sums against those on the snapshot.

If you do not want to setup a manual check you can always use any of the integrity systems available that will do this and more, for more information please read Do periodic integrity checks, Section 10.2.

4.19 Other recommendations

4.19.1 Do not use software depending on svgalib

SVGAlib is very nice for console lovers like me, but in the past it has been proven several times that it is very insecure. Exploits against zgv were released, and it was simple to become root. Try to prevent using SVGAlib programs wherever possible.

[ previous ] [ Contents ] [ 1 ] [ 2 ] [ 3 ] [ 4 ] [ 5 ] [ 6 ] [ 7 ] [ 8 ] [ 9 ] [ 10 ] [ 11 ] [ 12 ] [ A ] [ B ] [ C ] [ D ] [ E ] [ F ] [ G ] [ H ] [ next ]

Securing Debian Manual

Version: 3.13, Mon, 10 Nov 2008 23:32:30 +0000

Javier Fernández-Sanguino Peña jfs@debian.org
Authors, Section 1.1