SCAP: Guide To The Secure Configuration of Red Hat Enterprise Linux

NIST

• Guide to the Secure Configuration of Red Hat Enterprise Linux 6 – XCCDF Security Guide

  • Rules with nist Reference in Guide to the Secure Configuration of Red Hat Enterprise Linux 6

• SCAP Guide To The Secure Configuration of Red Hat Enterprise Linux 5

• SCAP Guide To The Secure Configuration of Red Hat Enterprise Linux 4

This guide has been created to assist IT professionals, in effectively securing systems with Red Hat Enterprise Linux 5.

RED HAT ENTERPRISE LINUX 5 SECURITY CHECKLIST

Table of Contents

1 - Introduction

The purpose of this guide is to provide security configuration recommendations for the Red Hat Enterprise Linux (RHEL) 5 operating system. The guidance provided here should be applicable to all variants (Desktop, Server, Advanced Platform) of the product. Recommended settings for the basic operating system are provided, as well as for many commonly-used services that the system can host in a network environment. The guide is intended for system administrators. Readers are assumed to possess basic system administration skills for Unix-like systems, as well as some familiarity with Red Hat’s documentation and administration conventions. Some instructions within this guide are complex. All directions should be followed completely and with understanding of their effects in order to avoid serious adverse effects on the system and its security.

1.1 - General Principles

The following general principles motivate much of the advice in this guide and should also influence any configuration decisions that are not explicitly covered.

1.1.1 - Encrypt Transmitted Data Whenever Possible

Data transmitted over a network, whether wired or wireless, is susceptible to passive monitoring. Whenever practical solutions for encrypting such data exist, they should be applied. Even if data is expected to be transmitted only over a local network, it should still be encrypted. Encrypting authentication data, such as passwords, is particularly important. Networks of RHEL5 machines can and should be configured so that no unencrypted authentication data is ever transmitted between machines.

1.1.2 - Minimize Software to Minimize Vulnerability

The simplest way to avoid vulnerabilities in software is to avoid installing that software. On RHEL, the RPM Package Manager (originally Red Hat Package Manager, abbreviated RPM) allows for careful management of the set of software packages installed on a system. Installed software contributes to system vulnerability in several ways. Packages that include setuid programs may provide local attackers a potential path to privilege escalation. Packages that include network services may give this opportunity to network-based attackers. Packages that include programs which are predictably executed by local users (e.g. after graphical login) may provide opportunities for trojan horses or other attack code to be run undetected. The number of software packages installed on a system can almost always be significantly pruned to include only the software for which there is an environmental or operational need.

1.1.3 - Run Different Network Services on Separate Systems

Whenever possible, a server should be dedicated to serving exactly one network service. This limits the number of other services that can be compromised in the event that an attacker is able to successfully exploit a software flaw in one network service.

1.1.4 - Configure Security Tools to Improve System Robustness

Several tools exist which can be effectively used to improve a system’s resistance to and detection of unknown attacks. These tools can improve robustness against attack at the cost of relatively little configuration effort. In particular, this guide recommends and discusses the use of Iptables for host-based firewalling, SELinux for protection against vulnerable services, and a logging and auditing infrastructure for detection of problems.

1.2 - How to Use This Guide

Readers should heed the following points when using the guide.

1.2.1 - Read Sections Completely and in Order

Each section may build on information and recommendations discussed in prior sections. Each section should be read and understood completely; instructions should never be blindly applied. Relevant discussion will occur after instructions for an action. The system-level configuration guidance in Chapter 2 must be applied to all machines. The guidance for individual services in Chapter 3 must be considered for all machines as well: apply the guidance if the machine is either a server or a client for that service, and ensure that the service is disabled according to the instructions provided if the machine is neither a server nor a client.

1.2.2 - Test in Non-Production Environment

This guidance should always be tested in a non-production environment before deployment. This test environment should simulate the setup in which the system will be deployed as closely as possible.

1.2.3 - Root Shell Environment Assumed

Most of the actions listed in this document are written with the assumption that they will be executed by the root user running the /bin/bash shell. Any commands preceded with a hash mark (#) assume that the administrator will execute the commands as root, i.e. apply the command via sudo whenever possible, or use su to gain root privileges if sudo cannot be used.

1.2.4 - Formatting Conventions

Commands intended for shell execution, as well as configuration file text, are featured in a monospace font. Italics are used to indicate instances where the system administrator must substitute the appropriate information into a command or configuration file.

1.2.5 - Reboot Required

A system reboot is implicitly required after some actions in order to complete the reconfiguration of the system. In many cases, the changes will not take effect until a reboot is performed. In order to ensure that changes are applied properly and to test functionality, always reboot the system after applying a set of recommendations from this guide.

2 - System-wide Configuration

2.1 - Installing and Maintaining Software

The following sections contain information on security-relevant choices during the initial operating system installation process and the setup of software updates.

2.1.1 - Initial Installation Recommendations

The recommendations here apply to a clean installation of the system, where any previous installations are wiped out. The sections presented here are in the same order that the installer presents, but only installation choices with security implications are covered. Many of the configuration choices presented here can also be applied after the system is installed. The choices can also be automatically applied via Kickstart files, as covered in [8].

2.1.1.1 - Disk Partitioning

If using any of the default layouts, check the box to “Review and modify partitioning.” The default layout does not create separate partitions or logical volumes for /var and /tmp. Add logical volumes or partitions for at least /var and /tmp. Adding logical volumes or partitions for /var/log and /var/log/audit may also be necessary, depending on system requirements. (See Section 2.6 for more information about logging and auditing). If user home directories will be stored locally, create a separate partition for /home as well. If creating a custom layout, create the partitions mentioned in the previous paragraph, as well as separate ones for /, /boot and swap space. You may need to make the / logical volume smaller to create space for the additional partitions.

2.1.1.2 - Boot Loader Configuration

Check the box to “Use a boot loader password” and create a password. Once this password is set, anyone who wishes to change the boot loader configuration will need to enter it. More information is available in Section 2.3.5.2. Assigning a boot loader password prevents a local user with physical access from altering the boot loader configuration at system startup.

2.1.1.3 - Network Devices

The default network device configuration uses DHCP, which is not recommended. Unless use of DHCP is absolutely necessary, click the “Edit” button and: * Uncheck “Use Dynamic IP configuration (DHCP).” * Uncheck “Enable IPv4 Support” if the system does not require IPv4. (This is uncommon.) * Uncheck “Enable IPv6 Support” if the system does not require IPv6. * Enter appropriate IPv4 and IPv6 addresses and prefixes as required. With the DHCP setting disabled, the hostname, gateway, and DNS servers should then be assigned on the main screen. Sections 3.9.1 and 3.9.2 contain more information on network configuration and the use of DHCP.

2.1.1.4 - Root Password

The security of the entire system depends on the strength of the root password. The password should be at least 12 characters long, and should include a mix of capitalized and lowercase letters, special characters, and numbers. It should also not be based on any dictionary word.

2.1.1.5 - Software Packages

Uncheck all package groups, including the package groups “Software Development” and “Web Server,” unless there is a specific requirement to install software using the system installer. If the machine will be used as a web server, it is preferable to manually install the necessary RPMs instead of installing the full “Web Server” package group. See Section 3.16 for installation and configuration details. Use the “Customize now” radio box to prune package groups as much as possible. This brings up a two-column view of categories and package groups. If appropriate, uncheck “X Window System” in the “Base System” category to avoid installing X entirely. Any other package groups not necessary for system operation should also be unchecked. Much finer-grained package selection is possible via Kickstart as described in [8].

2.1.1.6 - First-boot Configuration

The system presents more configuration options during the first boot after installation. For the screens listed, implement the security-related recommendations: Screen Recommendation Firewall Leave set to “Enabled.” Only check the “Trusted Services” that this system needs to serve. Uncheck the default selection of SSH if the system does not need to serve SSH. SELinux Leave SELinux set to “Enforcing” mode. Kdump Leave Kdump off unless the feature is required, such as for kernel development and testing. Screen Recommendation Set Up Software Updates If the system is connected to the Internet now, click “Yes, I’d like to register now.” This will require a connection to either the Red Hat Network servers or their proxies or satellites. This can also be configured later as described in Section 2.1.2.1. Create User If the system will require a local user account, it can be created here. Even if the system will be using a network-wide authentication system as described in Section 2.3.6, do not click on the “Use Network Login...” button. Manually applying configuration later is preferable.

2.1.2 - Updating Software

The yum command line tool is used to install and update software packages. Yum replaces the up2date utility used in previous system releases. The system also provides two graphical package managers, pirut and pup. The pirut tool is a graphical front-end for yum that allows users to install and update packages while pup is a simple update tool for packages that are already installed. In the Applications menu, pirut is labeled Add/Remove Software and pup is labeled Software Updater. It is recommended that these tools be used to keep systems up to date with the latest security patches.

2.1.2.1 - Configure Connection to the RHN RPM Repositories

The first step in configuring a system for updates is to register with the Red Hat Network (RHN). For most systems, this is done during the initial installation. Successfully registered systems will appear on the RHN web site. If the system is not listed, run the Red Hat Network Registration tool, which can be found in the Applications menu under System Tools or on the command line: # rhn register Follow the prompts on the screen. If successful, the system will appear on the RHN web site and be subscribed to one or more software update channels. Additionally, a new daemon, rhnsd, will be enabled. If the system will not have access to the Internet, it will not be able to directly subscribe to the RHN update repository. Updates will have to be downloaded from the RHN web site manually. The command line tool yum and the graphical front-ends pirut and pup can be configured to handle this situation.

2.1.2.2 - Disable the rhnsd Daemon

The rhnsd daemon polls the Red Hat Network web site for scheduled actions. Unless it is actually necessary to schedule updates remotely through the RHN website, it is recommended that the service be disabled. # chkconfig rhnsd off The rhnsd daemon is enabled by default, but until the system has been registered with the Red Hat Network, it will not run. However, once the registration process is complete, the rhnsd daemon will run in the background and periodically call the rhn check utility. It is the rhn check utility that communicates with the Red Hat Network web site. This utility is not required for the system to be able to access and install system updates. Once the system has been registered, either use the provided yum-updatesd service or create a cron job to automatically apply updates.

2.1.2.3 - Obtain Software Package Updates with yum

The yum update utility can be run by hand from the command line, called through one of the provided front-end tools, or configured to run automatically at specified intervals.

2.1.2.3.1 - Manually Update Packages Where Appropriate

The following command prints a list of packages that need to be updated: # yum check-update To actually install these updates, run: # yum update

2.1.2.3.2 - Configure Automatic Update Retrieval and Installation with Cron

The yum-updatesd service is not mature enough for an enterprise environment, and the service may introduce unnecessary overhead. When possible, replace this service with a cron job that calls yum directly. Disable the yum-updatesd service: # chkconfig yum-updatesd off Create the file yum.cron, make it executable, and place it in /etc/cron.daily: #!/bin/sh /usr/bin/yum -R 120 -e 0 -d 0 -y update yum /usr/bin/yum -R 10 -e 0 -d 0 -y update This particular script instructs yum to update any packages it finds. Placing the script in /etc/cron.daily ensures its daily execution. To only apply updates once a week, place the script in /etc/cron.weekly instead.

2.1.3 - Software Integrity Checking

The AIDE (Advanced Intrusion Detection Environment) software is included with the system to provide software integrity checking. It is designed to be a replacement for the well-known Tripwire integrity checker. Integrity checking cannot prevent intrusions into your system, but can detect that they have occurred. Any integrity checking software should be configured before the system is deployed and able to provides services to users. Ideally, the integrity checking database would be built before the system is connected to any network, though this may prove impractical due to registration and software updates.

2.1.3.1 - Configure AIDE

Requirements for software integrity checking should be defined by policy, and this is highly dependent on the environment in which the system will be used. As such, a general strategy for implementing integrity checking is provided, but precise recommendations (such as to check a particular file) cannot be. Documentation for AIDE, including the quick-start on which this advice is based, is available in /usr/share/doc/aide-0.12.

2.1.3.1.1 - Install AIDE

AIDE is not installed by default. Install it with the command: # yum install aide

2.1.3.1.2 - Customize Configuration File

Customize /etc/aide.conf to meet your requirements. The default configuration is acceptable for many environments. The man page aide.conf(5) provides detailed information about the configuration file format.

2.1.3.1.3 - Build, Store, and Test Database

Generate a new database: # /usr/sbin/aide --init By default, the database will be written to the file /var/lib/aide/aide.db.new.gz. The database, as well as the configuration file /etc/aide.conf and the binary /usr/sbin/aide (or hashes of these files) should be copied and stored in a secure location. Storing these copies or hashes on read-only media may provide further confidence that they will not be altered. Install the newly-generated database: # cp /var/lib/aide/aide.db.new.gz /var/lib/aide/aide.db.gz Run a manual check: # /usr/sbin/aide --check If this check produces any unexpected output, investigate.

2.1.3.1.4 - Implement Periodic Execution of Integrity Checking

By default, AIDE does not install itself for periodic execution. Implement checking with whatever frequency is required by your security policy. A once-daily check may be suitable for many environments. For example, to implement a daily execution of AIDE at 4:05am, add the following line to /etc/crontab: 05 4 * * * root /usr/sbin/aide --check AIDE output may be an indication of an attack against your system, or it may be the result of something innocuous such as an administrator’s configuration change or a software update. The steps in Section 2.1.3.1.3 should be repeated when configuration changes or software updates necessitate. This will certainly be necessary after applying guidance later in this guide.

2.1.3.1.5 - Manually Verify Integrity of AIDE

Because integrity checking is a means of intrusion detection and not intrusion prevention, it cannot be guaranteed that the AIDE binaries, configuration files, or database have not been tampered with. An attacker could disable or alter these files after a successful intrusion. Because of this, manual and frequent checks on these files is recommended. The safely stored copies (or hashes) of the database, binary, and configuration file were created earlier for this purpose. Manually verify the integrity of the AIDE binaries, configuration file, and database. Possibilities for doing so include: 1. Use sha1sum or md5sum to generate checksums on the files and then visually compare them to those generated from the safely stored versions. This does not, of course, preclude the possibility that such output could also be faked. 2. Mount the stored versions on read-only media and run /bin/diff to verify that there are no differences between the files. 3. Copying the files to another system and performing the hash or file comparisons there may impart additional confidence that the manual verification process is not being interfered with.

2.2 - File Permissions and Masks

Traditional Unix security relies heavily on file and directory permissions to prevent unauthorized users from reading or modifying files to which they should not have access. Adhere to the principle of least privilege — configure each file, directory, and filesystem to allow only the access needed in order for that file to serve its purpose. However, Linux systems contain a large number of files, so it is often prohibitively time-consuming to ensure that every file on a machine has exactly the permissions needed. This section introduces several permission restrictions which are almost always appropriate for system security, and which are easy to test and correct. Note: Several of the commands in this section search filesystems for files or directories with certain characteristics, and are intended to be run on every local ext2 or ext3 partition on a given machine. When the variable PART appears in one of the commands below, it means that the command is intended to be run repeatedly, with the name of each local partition substituted for PART in turn. The following command prints a list of ext2 and ext3 partitions on a given machine: $ mount -t ext2,ext3 | awk '{print $3}' If your site uses a local filesystem type other than ext2 or ext3, you will need to modify this command.

2.2.1 - Restrict Partition Mount Options

System partitions can be mounted with certain options which limit what files on those partitions can do. These options are set in the file /etc/fstab, and can be used to make certain types of malicious behavior more difficult.

2.2.1.1 - Add nodev Option to Non-Root Local Partitions

Edit the file /etc/fstab. The important columns for purposes of this section are column 2 (mount point), column 3 (filesystem type), and column 4 (mount options). For any line which satisfies all of the conditions: * The filesystem type is ext2 or ext3 * The mount point is not / add the text “,nodev” to the list of mount options in column 4. The nodev option prevents users from mounting unauthorized devices on any partition which is known not to contain any authorized devices. The root partition typically contains the /dev partition, which is the primary location for authorized devices, so this option should not be set on /. However, if system programs are being run in chroot jails, this advice may need to be modified further, since it is often necessary to create device files inside the chroot directory for use by the restricted program.

2.2.1.2 - Add nodev, nosuid, and noexec Options to Removable Media Partitions

Edit the file /etc/fstab. Filesystems which represent removable media can be located by finding lines whose mount points contain strings like floppy or cdrom, or whose types are iso9660, vfat, or msdos. For each line representing a removable media mountpoint, add the text ,nodev,nosuid to the list of mount options in column 4. If appropriate, also add the text ,noexec. Users should not be allowed to introduce arbitrary devices or setuid programs to a system. These options are used to prevent that. In addition, while users are usually allowed to add executable programs to a system, the noexec option prevents code from being executed directly from the media itself, and may therefore provide a line of defense against certain types of worms or malicious code.

2.2.2 - Restrict Dynamic Mounting and Unmounting of Filesystems

Linux includes a number of facilities for the automated addition and removal of filesystems on a running system. These facilities may increase convenience, but they all bring some risk, whether direct risk from allowing unprivileged users to introduce arbitrary filesystems to a machine, or risk that software flaws in the automated mount facility itself will allow an attacker to compromise the system. Use caution when enabling any such facility, and find out whether better configuration management or user education might solve the same problem with less risk.

2.2.2.1 - Restrict Console Device Access

The default system configuration grants the console user enhanced privileges normally reserved for the root user, including temporary ownership of most system devices. If not necessary, these privileges should be removed and restricted to root only. Restrict device ownership to root only. Edit /etc/security/console.perms.d/50-default.perms and locate the section prefaced by the following comment: # permission definitions Prepend a # symbol to comment out each line in that section which starts with <console> or <xconsole>: #<console> 0660 <floppy> 0660 root.floppy #<console> 0600 <sound> 0600 root ... #<xconsole> 0600 /dev/console 0600 root.root #<console> 0600 <dri> 0600 root Edit /etc/security/console.perms and make the following changes: <console>=tty[0-9][0-9]* vc/[0-9][0-9]* :0\.[0-9] :0 <xconsole>=:0\.[0-9] :0

2.2.2.2 - Disable USB Device Support

USB flash or hard drives allow an attacker with physical access to a system to quickly copy an enormous amount of data from it.

2.2.2.2.1 - Disable Modprobe Loading of USB Storage Driver

If USB storage devices should not be used, the modprobe program used for automatic kernel module loading should be configured to not load the USB storage driver upon demand. Add the following line to /etc/modprobe.conf to prevent loading of the usb-storage kernel module: install usb-storage : This will prevent the modprobe program from loading the usb-storage module, but will not prevent an administrator (or another program) from using the insmod program to load the module manually.

2.2.2.2.2 - Remove USB Storage Driver

If your system never requires the use of USB storage devices, then the supporting driver can be removed. Though more effective (as USB storage certainly cannot be used if the driver is not available at all), this is less elegant than the method described in Section 2.2.2.2.1. To remove the USB storage driver from the system: rm /lib/modules/kernelversion(s) /kernel/drivers/usb/storage/usb-storage.ko This command will need to be repeated every time the kernel is updated. This command will also cause the command rpm -q --verify kernel to fail, which may be an undesirable side effect. Note that this guidance will not prevent USB storage devices from being mounted if a custom kernel (i.e., not the one supplied with the system) with built-in USB support is used.

2.2.2.2.3 - Disable Kernel Support for USB via Bootloader Configuration

Another means of disabling USB storage is to disable all USB support provided by the operating system. This can be accomplished by adding the “nousb” argument to the kernel’s boot loader configuration. Disabling all kernel support for USB will cause problems for systems with USB-based keyboards, mice, or printers. This guidance is inappropriate for systems which require USB connectivity. To disable kernel support for USB, append “nousb” to the kernel line in /etc/grub.conf as follows: kernel /vmlinuz-version ro vga=ext root=/dev/VolGroup00/LogVol00 rhgb quiet nousb

2.2.2.2.4 - Disable Booting from USB Devices

An attacker with physical access could try to boot the system from a USB flash drive and then access any data on the system’s hard drive, circumventing the normal operating system’s access controls. To prevent this, configure the BIOS to disallow booting from USB drives. Also configure the BIOS or firmware password as described in Section 2.3.5.1 to prevent unauthorized configuration changes.

2.2.2.3 - Disable the Automounter if Possible

If the autofs service is not needed to dynamically mount NFS filesystems or removable media, disable the service: # chkconfig autofs off The autofs daemon mounts and unmounts filesystems, such as user home directories shared via NFS, on demand. In addition, autofs can be used to handle removable media, and the default configuration provides the cdrom device as /misc/cd. However, this method of providing access to removable media is not common, so autofs can almost always be disabled if NFS is not in use. Even if NFS is required, it is almost always possible to configure filesystem mounts statically by editing /etc/ fstab rather than relying on the automounter.

2.2.2.4 - Disable GNOME Automounting if Possible

The system’s default desktop environment, GNOME, runs the program gnome-volume-manager to mount devices and removable media (such as DVDs, CDs and USB flash drives) whenever they are inserted into the system. Execute the following commands to prevent gnome-volume-manager from automatically mounting devices and media: # gconftool-2 --direct \ --config-source xml:readwrite:/etc/gconf/gconf.xml.mandatory \ --type bool \ --set /desktop/gnome/volume_manager/automount_media false # gconftool-2 --direct \ --config-source xml:readwrite:/etc/gconf/gconf.xml.mandatory \ --type bool \ --set /desktop/gnome/volume_manager/automount_drives false Verify the changes by executing the following command, which should return a list of settings: # gconftool-2 -R /desktop/gnome/volume_manager The automount drives and automount media settings should be set to false. Survey the list for any other options that should be adjusted. The system’s capabilities for automatic mounting should be configured to match whatever is defined by security policy. Disabling USB storage as described in Section 2.2.2.2.1 will prevent the use of USB storage devices, but this step can also be taken as an additional layer of prevention and to prevent automatic mounting of CDs and DVDs if required. Particularly for kiosk-style systems, where users should have extremely limited access to the system, more detailed information can be found in Red Hat Desktop: Deployment Guide [5]. The gconf-editor program, available in an RPM of the same name, can be used to explore other settings available in the GNOME environment.

2.2.3 - Verify Permissions on Important Files and Directories

Permissions for many files on a system should be set to conform to system policy. This section discusses important permission restrictions gshadow which should be checked on a regular basis to ensure that no harmful discrepancies have arisen.

2.2.3.1 - Verify Permissions on passwd, shadow, group and gshadow Files

# cd /etc # chown root:root passwd shadow group gshadow # chmod 644 passwd group # chmod 400 shadow gshadow These are the default permissions for these files. Many utilities need read access to the passwd file in order to function properly, but read access to the shadow file allows malicious attacks against system passwords, and should never be enabled.

2.2.3.2 - Verify that All World-Writable Directories Have Sticky Bits Set

Locate any directories in local partitions which are world-writable and do not have their sticky bits set. The following command will discover and print these. Run it once for each local partition PART: # find PART -xdev -type d \( -perm -0002 -a ! -perm -1000 \) -print If this command produces any output, fix each reported directory /dir using the command: # chmod +t /dir When the so-called “sticky bit” is set on a directory, only the owner of a given file may remove that file from the directory. Without the sticky bit, any user with write access to a directory may remove any file in the directory. Setting the sticky bit prevents users from removing each other’s files. In cases where there is no reason for a directory to be world-writable, a better solution is to remove that permission rather than to set the sticky bit. However, if a directory is used by a particular application, consult that application’s documentation instead of blindly changing modes.

2.2.3.3 - Find Unauthorized World-Writable Files

The following command discovers and prints any world-writable files in local partitions. Run it once for each local partition PART: # find PART -xdev -type f -perm -0002 -print If this command produces any output, fix each reported file file using the command: # chmod o-w file Data in world-writable files can be modified by any user on the system. In almost all circumstances, files can be configured using a combination of user and group permissions to support whatever legitimate access is needed without the risk caused by world-writable files. It is generally a good idea to remove global (other) write access to a file when it is discovered. However, check with documentation for specific applications before making changes. Also, monitor for recurring world-writable files, as these may be symptoms of a misconfigured application or user account.

2.2.3.4 - Find Unauthorized SUID/SGID System Executables

The following command discovers and prints any setuid or setgid files on local partitions. Run it once for each local partition PART: # find PART -xdev \( -perm -4000 -o -perm -2000 \) -type f -print If the file does not require a setuid or setgid bit as discussed below, then these bits can be removed with the command: # chmod -s file The following table contains all setuid and setgid files which are expected to be on a stock system. The setuid or setgid bit on these files may be disabled to reduce system risk if only an administrator requires their functionality. The table indicates those files which may not be needed. Note: Several of these files are used for applications which are unlikely to be relevant to most production environments, such as ISDN networking, SSH hostbased authentication, or modification of network interfaces by unprivileged users. It is extremely likely that your site can disable a subset of these files with no loss of functionality. Any files found by the above command which are not in the table should be examined. If the files are not authorized, they should have permissions removed, and further investigation may be warranted. File Set-ID Subsystem/Ref Disable? /bin/mount uid root filesystems no /bin/ping uid root net (3.3.9) no /bin/ping6 uid root net (3.3.9),IPv6 (2.5.3) unless IPv6 is used /bin/su uid root auth (2.3.1.2) no /bin/umount uid root filesystems no /sbin/mount.nfs uid root NFS (3.13) unless NFS is used /sbin/mount.nfs4 uid root NFS (3.13) unless NFSv4 is used /sbin/netreport gid root net (3.3.9) unless users must modify interfaces /sbin/pam timestamp check uid root PAM auth (2.3.3) no /sbin/umount.nfs uid root NFS (3.13) unless NFS is used /sbin/umount.nfs4 uid root NFS (3.13) unless NFSv4 is used /sbin/unix chkpwd uid root PAM auth (2.3.3) no /usr/bin/at uid root cron/at (3.4) no /usr/bin/chage uid root passwd expiry (2.3.1.7) unless users must view expiry info /usr/bin/chfn uid root user info unless users must change finger info /usr/bin/chsh uid root user info unless users must change shells /usr/bin/crontab uid/gid root cron/at (3.4) unless users must use cron /usr/bin/gpasswd uid root group auth no /usr/bin/locate gid slocate locate database no /usr/bin/lockfile gid mail procmail unless procmail is used /usr/bin/newgrp uid root group auth no /usr/bin/passwd uid root passwd auth no /usr/bin/rcp uid root rsh (3.2.3) yes (rsh is obsolete) /usr/bin/rlogin uid root rsh (3.2.3) yes (rsh is obsolete) /usr/bin/rsh uid root rsh (3.2.3) yes (rsh is obsolete) /usr/bin/ssh-agent gid nobody SSH (3.5) no /usr/bin/sudo uid root sudo (2.3.1.3) no /usr/bin/sudoedit uid root sudo (2.3.1.3) no /usr/bin/wall gid tty console messaging unless console messaging is used /usr/bin/write gid tty console messaging unless console messaging is used /usr/bin/Xorg uid root X11 (3.6) unless X11 is used /usr/kerberos/bin/ksu uid root Kerberos auth (2.3.6) unless Kerberos is used /usr/libexec/openssh/ssh-keysign uid root SSH (3.5) unless sshd uses hostbased auth /usr/libexec/utempter/utempter gid utmp terminal support no /usr/lib/squid/pam auth uid root squid (3.19) unless squid is used /usr/lib/squid/ncsa auth uid root squid (3.19) unless squid is used /usr/lib/vte/gnome-pty-helper gid utmp X11, Gnome (3.6) unless X11 is used /usr/sbin/ccreds validate uid root PAM auth (2.3.3) unless PAM auth caching is used /usr/sbin/lockdev gid lock filesystems no /usr/sbin/sendmail.sendmail gid smmsp sendmail client (3.11.2) no /usr/sbin/suexec uid root apache (3.16) unless apache is used /usr/sbin/userhelper uid root PAM auth (2.3.3.4) restrict (see section) /usr/sbin/userisdnctl uid root ISDN unless ISDN is used /usr/sbin/usernetctl uid root user network control unless users must modify interfaces

2.2.3.5 - Find and Repair Unowned Files

The following command will discover and print any files on local partitions which do not belong to a valid user and a valid group. Run it once for each local partition PART: # find PART -xdev \( -nouser -o -nogroup \) -print If this command prints any results, investigate each reported file and either assign it to an appropriate user and group or remove it. Unowned files are not directly exploitable, but they are generally a sign that something is wrong with some system process. They may be caused by an intruder, by incorrect software installation or incomplete software removal, or by failure to remove all files belonging to a deleted account. The files should be repaired so that they will not cause problems when accounts are created in the future, and the problem which led to unowned files should be discovered and addressed.

2.2.4 - Restrict Programs from Dangerous Execution Patterns

The recommendations in this section provide broad protection against information disclosure or other misbehavior. These protections are applied at the system initialization or kernel level, and defend against certain types of badly-configured or compromised programs.

2.2.4.1 - Set Daemon umask

Edit the file /etc/sysconfig/init, and add or correct the following line: umask 027 The settings file /etc/sysconfig/init contains settings which apply to all processes started at boot time. The system umask must be set to at least 022, or daemon processes may create world-writable files. The more restrictive setting 027 protects files, including temporary files and log files, from unauthorized reading by unprivileged users on the system. If a particular daemon needs a less restrictive umask, consider editing the startup script or sysconfig file of that daemon to make a specific exception.

2.2.4.2 - Disable Core Dumps

To disable core dumps for all users, add or correct the following line in /etc/security/limits.conf: * hard core 0 In addition, to ensure that core dumps can never be made by setuid programs, edit /etc/sysctl.conf and add or correct the line: fs.suid_dumpable = 0 A core dump file is the memory image of an executable program when it was terminated by the operating system due to errant behavior. In most cases, only software developers would legitimately need to access these files. The core dump files may also contain sensitive information, or unnecessarily occupy large amounts of disk space. By default, the system sets a soft limit to stop the creation of core dump files for all users. This is accomplished in /etc/profile with the line: ulimit -S -c 0 > /dev/null 2>&1 However, compliance with this limit is voluntary; it is a default intended only to protect users from the annoyance of generating unwanted core files. Users can increase the allowed core file size up to the hard limit, which is unlimited by default. Once a hard limit is set in /etc/security/limits.conf, the user cannot increase that limit within his own session. If access to core dumps is required, consider restricting them to only certain users or groups. See the limits.conf(5) man page for more information. The core dumps of setuid programs are further protected. The sysctl variable fs.suid dumpable controls whether the kernel allows core dumps from these programs at all. The default value of 0 is recommended.

2.2.4.3 - Enable ExecShield

ExecShield comprises a number of kernel features to provide protection against buffer overflows. These features include random placement of the stack and other memory regions, prevention of execution in memory that should only hold data, and special handling of text buffers. This protection is enabled by default, but the sysctl variables kernel.exec-shield and kernel.randomize va space should be checked to ensure that it has not been disabled. To ensure ExecShield (including random placement of virtual memory regions) is activated at boot, add or correct the following settings in /etc/sysctl.conf: kernel.exec-shield = 1 kernel.randomize_va_space = 1 ExecShield uses the segmentation feature on all x86 systems to prevent execution in memory higher than a certain address. It writes an address as a limit in the code segment descriptor, to control where code can be executed, on a per-process basis. When the kernel places a process’s memory regions such as the stack and heap higher than this address, the hardware prevents execution there. However, this cannot always be done for all memory regions in which execution should not occur, so follow guidance in Section 2.2.4.4 to further protect the system.

2.2.4.4 - Enable Execute Disable (XD) or No Execute (NX) Support on x86 Systems

Recent processors in the x86 family support the ability to prevent code execution on a per memory page basis. Generically and on AMD processors, this ability is called No Execute (NX), while on Intel processors it is called Execute Disable (XD). This ability can help prevent exploitation of buffer overflow vulnerabilities and should be activated whenever possible. Extra steps must be taken to ensure that this protection is enabled, particularly on 32-bit x86 systems. Other processors, such as Itanium and POWER, have included such support since inception and the standard kernel for those platforms supports the feature.

2.2.4.4.1 - Check for Processor Support on x86 Systems

Check to see if the processor supports the PAE and NX features: $ cat /proc/cpuinfo If supported, the flags field will contain pae and nx.

2.2.4.4.2 - Install New Kernel on Supported x86 Systems

If supported as determined in the previous section, the kernel-PAE package should be installed to enable XD or NX support: # yum install kernel-PAE The installation process should also have configured the bootloader to load the new kernel at boot. Verify this at reboot and modify /etc/grub.conf if necessary. The kernel-PAE package should not be installed on older systems that do not support the XD or NX bit, as this may prevent them from booting.

2.2.4.4.3 - Enable Support in the BIOS

Computers with the ability to prevent this type of code execution frequently put an option in the BIOS that will allow users to turn the feature on or off at will. Reboot the system and enter the BIOS or “Setup” configuration menu. Navigate the BIOS configuration menu and make sure that the option is enabled. The setting may be located under a “Security” section. Look for Execute Disable (XD) on Intel-based systems and No Execute (NX) on AMD-based systems. See Section 2.3.5.1 for information on protecting this and other BIOS settings.

2.3 - Account and Access Control

In traditional Unix security, if an attacker gains shell access to a certain login account, he can perform any action or access any file to which that account has access. Therefore, making it more difficult for unauthorized people to gain shell access to accounts, particularly to privileged accounts, is a necessary part of securing a system. This section introduces mechanisms for restricting access to accounts under RHEL5.

2.3.1 - Protect Accounts by Restricting Password-Based Login

Conventionally, Unix shell accounts are accessed by providing a username and password to a login program, which tests these values for correctness using the /etc/passwd and /etc/shadow files. Password-based login is vulnerable to guessing of weak passwords, and to sniffing and man-in-the-middle attacks against passwords entered over a network or at an insecure console. Therefore, mechanisms for accessing accounts by entering usernames and passwords should be restricted to those which are operationally necessary.

2.3.1.1 - Restrict Root Logins to System Console

Edit the file /etc/securetty. Ensure that the file contains only the following lines: * The primary system console device: console * The virtual console devices: tty1 tty2 tty3 tty4 tty5 tty6 ... * If required by your organization, the deprecated virtual console interface may be retained for backwards compatibility: vc/1 vc/2 vc/3 vc/4 vc/5 vc/6 ... * If required by your organization, the serial consoles may be added: ttyS0 ttyS1 Direct root logins should be allowed only for emergency use. In normal situations, the administrator should access the system via a unique unprivileged account, and use su or sudo to execute privileged commands. Discouraging administrators from accessing the root account directly ensures an audit trail in organizations with multiple administrators. Locking down the channels through which root can connect directly reduces opportunities for password-guessing against the root account. The login program uses the file /etc/securetty to determine which interfaces should allow root logins. The virtual devices /dev/console and /dev/tty* represent the system consoles (accessible via the Ctrl-Alt-F1 through Ctrl-Alt-F6 keyboard sequences on a default installation). The default securetty file also contains /dev/vc/*. These are likely to be deprecated in most environments, but may be retained for compatibility. Root should also be prohibited from connecting via network protocols. See Section 3.5 for instructions on preventing root from logging in via SSH.

2.3.1.2 - Limit su Access to the Root Account

1. Ensure that the group wheel exists, and that the usernames of all administrators who should be allowed to execute commands as root are members of that group. # grep ^wheel /etc/group 2. Edit the file /etc/pam.d/su. Add, uncomment, or correct the line: auth required pam_wheel.so use_uid The su command allows a user to gain the privileges of another user by entering the password for that user’s account. It is desirable to restrict the root user so that only known administrators are ever allowed to access the root account. This restricts password-guessing against the root account by unauthorized users or by accounts which have been compromised. By convention, the group wheel contains all users who are allowed to run privileged commands. The PAM module pam wheel.so is used to restrict root access to this set of users.

2.3.1.3 - Configure sudo to Improve Auditing of Root Access

1. Ensure that the group wheel exists, and that the usernames of all administrators who should be allowed to execute commands as root are members of that group. # grep ^wheel /etc/group 2. Edit the file /etc/sudoers. Add, uncomment, or correct the line: %wheel ALL=(ALL) ALL The sudo command allows fine-grained control over which users can execute commands using other accounts. The primary benefit of sudo when configured as above is that it provides an audit trail of every command run by a privileged user. It is possible for a malicious administrator to circumvent this restriction, but, if there is an established procedure that all root commands are run using sudo, then it is easy for an auditor to detect unusual behavior when this procedure is not followed. Editing /etc/sudoers by hand can be dangerous, since a configuration error may make it impossible to access the root account remotely. The recommended means of editing this file is using the visudo command, which checks the file’s syntax for correctness before allowing it to be saved. Note that sudo allows any attacker who gains access to the password of an administrator account to run commands as root. This is a downside which must be weighed against the benefits of increased audit capability and of being able to heavily restrict the use of the high-value root password (which can be logistically difficult to change often). As a basic precaution, never use the NOPASSWD directive, which would allow anyone with access to an administrator account to execute commands as root without knowing the administrator’s password. The sudo command has many options which can be used to further customize its behavior. See the sudoers(5) man page for details.

2.3.1.4 - Block Shell and Login Access for Non-Root System Accounts

Do not perform the steps in this section on the root account. Doing so might cause the system to become inaccessible. Using /etc/passwd, obtain a listing of all users, their UIDs, and their shells, for instance by running: # awk -F: '{print $1 ":" $3 ":" $7}' /etc/passwd Identify the system accounts from this listing. These will primarily be the accounts with UID numbers less than 500, other than root. For each identified system account SYSACCT , lock the account: # usermod -L SYSACCT and disable its shell: # usermod -s /sbin/nologin SYSACCT These are the accounts which are not associated with a human user of the system, but which exist to perform some administrative function. Make it more difficult for an attacker to use these accounts by locking their passwords and by setting their shells to some non-valid shell. The RHEL5 default non-valid shell is /sbin/nologin, but any command which will exit with a failure status and disallow execution of any further commands, such as /bin/false or /dev/null, will work.

2.3.1.5 - Verify that No Accounts Have Empty Password Fields

Run the command: # awk -F: '($2 == "") {print}' /etc/shadow If this produces any output, fix the problem by locking each account (see Section 2.3.1.4 above) or by setting a password. If an account has an empty password, anybody may log in and run commands with the privileges of that account. Accounts with empty passwords should never be used in operational environments.

2.3.1.6 - Verify that No Non-Root Accounts Have UID 0

This command will print all password file entries for accounts with UID 0: # awk -F: '($3 == "0") {print}' /etc/passwd This should print only one line, for the user root. If any other lines appear, ensure that these additional UID-0 accounts are authorized, and that there is a good reason for them to exist. In general, the best practice solution for auditing use of the root account is to restrict the set of cases in which root must be accessed anonymously by requiring use of su or sudo in almost all cases. Some sites choose to have more than one account with UID 0 in order to differentiate between administrators, but this practice may have unexpected side effects, and is therefore not recommended.

2.3.1.7 - Set Password Expiration Parameters

Edit the file /etc/login.defs to specify password expiration settings for new accounts. Add or correct the following lines: PASS_MAX_DAYS=180 PASS_MIN_DAYS=7 PASS_MIN_LEN=8 PASS_WARN_AGE=7 For each existing human user USER , modify the current expiration settings to match these: # chage -M 180 -m 7 -W 7 USER Users should be forced to change their passwords, in order to decrease the utility of compromised passwords. However, the need to change passwords often should be balanced against the risk that users will reuse or write down passwords if forced to change them too often. Forcing password changes every 90-360 days, depending on the environment, is recommended. Set the appropriate value as PASS MAX DAYS and apply it to existing accounts with the -M flag. The PASS MIN DAYS (-m) setting prevents password changes for 7 days after the first change, to discourage password cycling. If you use this setting, train users to contact an administrator for an emergency password change in case a new password becomes compromised. The PASS WARN AGE (-W) setting gives users 7 days of warnings at login time that their passwords are about to expire.

2.3.1.8 - Remove Legacy + Entries from Password Files

The command: # grep "^+:" /etc/passwd /etc/shadow /etc/group should produce no output. The + symbol was used by systems to include data from NIS maps into existing files. However, a certain configuration error in which a NIS inclusion line appears in /etc/passwd, but NIS is not running, could lead to anyone being able to access the system with the username + and no password. Therefore, it is important to verify that no such line appears in any of the relevant system files. The correct way to tell the local system to consult network databases such as LDAP or NIS for user information is to make appropriate modifications to /etc/nsswitch.conf.

2.3.2 - Use Unix Groups to Enhance Security

The access control policies which can be enforced by standard Unix permissions are limited, and configuring SELinux (Section 2.4) is frequently a better choice. However, this guide recommends that security be enhanced to the extent possible by enforcing the Unix group policies outlined in this section.

2.3.2.1 - Create a Unique Default Group for Each User

When running useradd, do not use the -g flag or otherwise override the default group. The Red Hat default is that each new user account should have a unique primary group whose name is the same as that of the account. This default is recommended, in order to provide additional protection against files which are created with group write permission enabled.

2.3.2.2 - Create and Maintain a Group Containing All Human Users

Identify all user accounts on the system which correspond to human users. Depending on your system configuration, this may be all entries in /etc/passwd with UID values of at least 500. Once, you have identified such a set of users, create a group named usergroup (substitute some name appropriate to your environment) and populate it with each human user: # groupadd usergroup # usermod -G usergroup human1 # usermod -G usergroup human2 ... # usermod -G usergroup humanN Then modify your procedure for creating new user accounts by adding -G usergroup to the set of flags with which useradd is invoked, so that new human users will be placed in the correct group by default. Creating a group of human users does not, by itself, enhance system security. However, as you work on securing your system, you will often find commands which never need to be run by system accounts, or which are only ever needed by users logged into the graphical console (which should only ever be available to human users, even on workstations). Once a group of users has been created, it is easy to restrict access to a given command, for instance /path/to/graphical/command , to authorized users: # chgrp usergroup /path/to/graphical/command # chmod 750 /path/graphical/command Without a group of human users, it is necessary to restrict access by somehow preventing each system account from running the command, which is an error-prone process even when it is possible at all.

2.3.3 - Protect Accounts by Configuring PAM

PAM, or Pluggable Authentication Modules, is a system which implements modular authentication for Linux programs. PAM is well-integrated into Linux’s authentication architecture, making it difficult to remove, but it can be configured to minimize your system’s exposure to unnecessary risk. This section contains guidance on how to accomplish that, and how to ensure that the modules used by your PAM configuration do what they are supposed to do. PAM is implemented as a set of shared objects which are loaded and invoked whenever an application wishes to authenticate a user. Typically, the application must be running as root in order to take advantage of PAM. Traditional privileged network listeners (e.g. sshd) or SUID programs (e.g. sudo) already meet this requirement. An SUID root application, userhelper, is provided so that programs which are not SUID or privileged themselves can still take advantage of PAM. PAM looks in the directory /etc/pam.d for application-specific configuration information. For instance, if the program login attempts to authenticate a user, then PAM’s libraries follow the instructions in the file /etc/ pam.d/login to determine what actions should be taken. One very important file in /etc/pam.d is /etc/pam.d/system-auth. This file, which is included by many other PAM configuration files, defines “default” system authentication measures. Modifying this file is a good way to make far-reaching authentication changes, for instance when implementing a centralized authentication service. Be careful when making changes to PAM’s configuration files. The syntax for these files is complex, and modifications can have unexpected consequences.1 The default configurations shipped with applications should be sufficient for most users. Running authconfig or system-config-authentication will re-write the PAM configuration files, destroying any manually made changes and replacing them with a series of system defaults. 1One reference to the configuration file syntax can be found at http://www.kernel.org/pub/linux/libs/pam/Linux-PAM-html/ sag-configuration-file.html.

2.3.3.1 - Set Password Quality Requirements

The default pam cracklib PAM module provides strength checking for passwords. It performs a number of checks, such as making sure passwords are not similar to dictionary words, are of at least a certain length, are not the previous password reversed, and are not simply a change of case from the previous password. The pam passwdqc PAM module provides the ability to enforce even more stringent password strength requirements. It is provided in an RPM of the same name. The man pages pam cracklib(8) and pam passwdqc(8) provide information on the capabilities and configuration of each. If password strength stronger than that guaranteed by pam cracklib is required, configure PAM to use pam passwdqc. To activate pam passwdqc, locate the following line in /etc/pam.d/system-auth: password requisite pam_cracklib.so try_first_pass retry=3 and then replace it with the line: password requisite pam_passwdqc.so min=disabled,disabled,16,12,8 If necessary, modify the arguments (min=disabled,disabled,16,12,8) to ensure compliance with your organization’s security policy. Configuration options are described in the man page pam passwdqc(8) and also in /usr/share/doc/pam passwdqc-version. The minimum lengths provided here supercede that specified by the argument PASS MIN LEN as described in Section 2.3.1.7. The options given in the example above set a minimum length for each of the password “classes” that pam passwdqc recognizes. Setting a particular minimum value to disabled will stop users from choosing a password that falls into that category alone.

2.3.3.2 - Set Lockouts for Failed Password Attempts

The pam tally2 PAM module provides the capability to lock out user accounts after a number of failed login attempts. Its documentation is available in /usr/share/doc/pam-version/txts/README.pam tally2. If locking out accounts after a number of incorrect login attempts is required by your security policy, implement use of pam tally2.so for the relevant PAM-aware programs such as login, sshd, and vsftpd. Find the following line in /etc/pam.d/system-auth: auth sufficient pam_unix.so nullok try_first_pass and then change it so that it reads as follows: auth required pam_unix.so nullok try_first_pass In the same file, comment out or delete the lines: auth requisite pam_succeed_if.so uid >= 500 quiet auth required pam_deny.so To enforce password lockout, add the following to the individual programs’ configuration files in /etc/pam.d. First, add to end of the auth lines: auth required pam_tally2.so deny=5 onerr=fail Second, add to the end of the account lines: account required pam_tally2.so Adjust the deny argument to conform to your system security policy. The pam tally2 utility can be used to unlock user accounts as follows: # /sbin/pam_tally2 --user username --reset Locking out user accounts presents the risk of a denial-of-service attack. The security policy regarding system lockout must weigh whether the risk of such a denial-of-service attack outweighs the benefits of thwarting password guessing attacks. The pam tally2 utility can be run from a cron job on a hourly or daily basis to try and offset this risk.

2.3.3.3 - Use pam deny.so to Quickly Deny Access to a Service

In order to deny access to a service SVCNAME via PAM, edit the file /etc/pam.d/SVCNAME . Prepend this line to the beginning of the file: auth requisite pam_deny.so Under most circumstances, there are better ways to disable a service than to deny access via PAM. However, this should suffice as a way to quickly make a service unavailable to future users (existing sessions which have already been authenticated, are not affected). The requisite tag tells PAM that, if the named module returns failure, authentication should fail, and PAM should immediately stop processing the configuration file. The pam deny.so module always returns failure regardless of its input.

2.3.3.4 - Restrict Execution of userhelper to Console Users

If your environment has defined a group, usergroup containing all the human users of your system, restrict execution of the userhelper program to only that group: # chgrp usergroup /usr/sbin/userhelper # chmod 4710 /usr/sbin/userhelper The userhelper program provides authentication for graphical services which must run with root privileges, such as the system-config- family of graphical configuration utilities. Only human users logged into the system console are likely to ever have a legitimate need to run these utilities. This step provides some protection against possible flaws in userhelper’s implementation, and against further privilege escalation when system accounts are compromised. See Section 2.3.2.2 for more information on creating a group of human users. The userhelper program is configured by the files in /etc/security/console.apps/. Each file specifies, for some program, what user the program should run as, and what program should be executed after successful authentication. Note: The configuration in /etc/security/console.apps/ is applied in combination with the PAM configuration of the service defined in /etc/pam.d/. First, userhelper determines what user the service should run as. (Typically, this will be root.) Next, userhelper uses the PAM API to allow the user who ran the program to attempt to authenticate as the desired user. The PAM API exchange is wrapped in a GUI if the application’s configuration requests one.

2.3.4 - Secure Session Configuration Files for Login Accounts

When a user logs into a Unix account, the system configures the user’s session by reading a number of files. Many of these files are located in the user’s home directory, and may have weak permissions as a result of user error or misconfiguration. If an attacker can modify or even read certain types of account configuration information, he can often gain full access to the affected user’s account. Therefore, it is important to test and correct configuration file permissions for interactive accounts, particularly those of privileged users such as root or system administrators.

2.3.4.1 - Ensure that No Dangerous Directories Exist in Roots Path '

The active path of the root account can be obtained by starting a new root shell and running: # echo $PATH This will produce a colon-separated list of directories in the path. For each directory DIR in the path, ensure that DIR is not equal to a single . character. Also ensure that there are no “empty” elements in the path, such as in these examples: PATH=:/bin PATH=/bin: PATH=/bin::/sbin These empty elements have the same effect as a single . character. For each element in the path, run: # ls -ld DIR and ensure that write permissions are disabled for group and other. It is important to prevent root from executing unknown or untrusted programs, since such programs could contain malicious code. Therefore, root should not run programs installed by unprivileged users. Since root may often be working inside untrusted directories, the . character, which represents the current directory, should never be in the root path, nor should any directory which can be written to by an unprivileged or semi-privileged (system) user. It is a good practice for administrators to always execute privileged commands by typing the full path to the command.

2.3.4.2 - Ensure that User Home Directories are not Group-Writable or World-Readable

Sections 2.3.4.2–2.3.4.5 recommend modifying user home directories. Notify your user community, and solicit input if appropriate, before making this type of change. For each human user USER of the system, view the permissions of the user’s home directory: # ls -ld /home/USER Ensure that the directory is not group-writable and that it is not world-readable. If necessary, repair the permissions: # chmod g-w /home/USER # chmod o-rwx /home/USER User home directories contain many configuration files which affect the behavior of a user’s account. No user should ever have write permission to another user’s home directory. Group shared directories can be configured in subdirectories or elsewhere in the filesystem if they are needed. Typically, user home directories should not be world-readable. If a subset of users need read access to one another’s home directories, this can be provided using groups.

2.3.4.3 - Ensure that User Dot-Files are not World-writable

For each human user USER of the system, view the permissions of all dot-files in the user’s home directory: # ls -ld /home/USER /.[A-Za-z0-9]* Ensure that none of these files are group- or world-writable. Correct each misconfigured file FILE by executing: # chmod go-w /home/USER /FILE A user who can modify another user’s configuration files can likely execute commands with the other user’s privileges, including stealing data, destroying files, or launching further attacks on the system.

2.3.4.4 - Ensure that Users Have Sensible Umask Values

1. Edit the global configuration files /etc/profile, /etc/bashrc, and /etc/csh.cshrc. Add or correct the line: umask 077 2. Edit the user definitions file /etc/login.defs. Add or correct the line: UMASK 077 3. View the additional configuration files /etc/csh.login and /etc/profile.d/*, and ensure that none of these files redefine the umask to a more permissive value unless there is a good reason for it. 4. Edit the root shell configuration files /root/.bashrc, /root/.bash profile, /root/.cshrc, and /root/.tcshrc. Add or correct the line: umask 077 With a default umask setting of 077, files and directories created by users will not be readable by any other user on the system. Users who wish to make specific files group- or world-readable can accomplish this using the chmod command. Additionally, users can make all their files readable to their group by default by setting a umask of 027 in their shell configuration files. If default per-user groups exist (that is, if every user has a default group whose name is the same as that user’s username and whose only member is the user), then it may even be safe for users to select a umask of 007, making it very easy to intentionally share files with group s of which the user is a member. In addition, it may be necessary to change root’s umask temporarily in order to install software or files which must be readable by other users, or to change the default umasks of certain service accounts such as the FTP user. However, setting a restrictive default protects the files of users who have not taken steps to make their files more available, and preventing files from being inadvertently shared.

2.3.4.5 - Ensure that Users do not Have .netrc Files

For each human user USER of the system, ensure that the user has no .netrc file. The command: # ls -l /home/USER /.netrc should return the error “No such file or directory”. If any user has such a file, approach that user to discuss removing this file. The .netrc file is a configuration file used to make unattended logins to other systems via FTP. When this file exists, it frequently contains unencrypted passwords which may be used to attack other systems.

2.3.5 - Protect Physical Console Access

It is impossible to fully protect a system from an attacker with physical access, so securing the space in which the system is located should be considered a necessary step. However, there are some steps which, if taken, make it more difficult for an attacker to quickly or undetectably modify a system from its console.

2.3.5.1 - Set BIOS Password

The BIOS (on x86 systems) is the first code to execute during system startup and controls many important system parameters, including which devices the system will try to boot from, and in which order. Assign a password to prevent any unauthorized changes to the BIOS configuration. The exact steps will vary depending on your machine, but are likely to include: 1. Reboot the machine. 2. Press the appropriate key during the initial boot screen (F2 is typical). 3. Navigate the BIOS configuration menu to add a password. The exact process will be system-specific and the system’s hardware manual may provide detailed instructions. This password should prevent attackers with physical access from attempting to change important parameters, such as those described in Sections 2.5.2.2.1 and 2.2.2.2.4. However, an attacker with physical access can usually clear the BIOS password. The password should be written down and stored in a physically-secure location, such as a safe, in the event that it is forgotten and must be retrieved.

2.3.5.2 - Set Boot Loader Password

During the boot process, the boot loader is responsible for starting the execution of the kernel and passing options to it. The boot loader allows for the selection of different kernels – possibly on different partitions or media. Options it can pass to the kernel include “single-user mode,” which provides root access without any authentication, and the ability to disable SELinux. To prevent local users from modifying the boot parameters and endangering security, the boot loader configuration should be protected with a password. The default RHEL boot loader for x86 systems is called GRUB. To protect its configuration: 1. Select a password and then generate a hash from it by running: # grub-md5-crypt 2. Insert the following line into /etc/grub.conf immediately after the header comments. (Use the output from grub-md5-crypt as the value of password-hash ): password --md5 password-hash 3. Verify the permissions on /etc/grub.conf (which is a symlink to ../boot/grub/grub.conf): # chown root:root /etc/grub.conf # chmod 600 /etc/grub.conf Boot loaders for other platforms should offer a similar password protection feature.

2.3.5.3 - Require Authentication for Single-User Mode

Single-user mode is intended as a system recovery method, providing a single user root access to the system by providing a boot option at startup. By default, no authentication is performed if single-user mode is selected. This provides a trivial mechanism of bypassing security on the machine and gaining root access. To require entry of the root password even if the system is started in single-user mode, add the following line to the /etc/inittab file: ~~:S:wait:/sbin/sulogin

2.3.5.4 - Disable Interactive Boot

Edit the file /etc/sysconfig/init. Add or correct the setting: PROMPT=no The PROMPT option allows the console user to perform an interactive system startup, in which it is possible to select the set of services which are started on boot. Using interactive boot, the console user could disable auditing, firewalls, or other services, weakening system security.

2.3.5.5 - Implement Inactivity Time-out for Login Shells

If the system does not run X Windows, then the login shells can be configured to automatically log users out after a period of inactivity. The following instructions are not practical for systems which run X Windows, as they will close terminal windows in the X environment. For information on how to automatically lock those systems, see Section 2.3.5.6. To implement a 10-minute idle time-out for the default /bin/bash shell, create a new file tmout.sh in the directory /etc/profile.d with the following lines: TMOUT=600 readonly TMOUT export TMOUT To implement a 10-minute idle time-out for the tcsh shell, create a new file autologout.csh in the directory /etc/profile.d with the following line: set -r autologout 10 Similar actions should be taken for any other login shells used. The example time-out here of 10 minutes should be adjusted to whatever your security policy requires. The readonly line for bash and the -r option for tcsh can be omitted if policy allows users to override the value. The automatic shell logout only occurs when the shell is the foreground process. If, for example, a vi session is left idle, then automatic logout would not occur. When logging in through a remote connection, as with SSH, it may be more effective to set the timeout value directly through that service. To learn how to set automatic timeout intervals for SSH, see Section 3.5.2.3.

2.3.5.6 - Configure Screen Locking

When a user must temporarily leave an account logged-in, screen locking should be employed to prevent passersby from abusing the account. User education and training is particularly important for screen locking to be effective. A policy should be implemented that trains all users to lock the screen when they plan to temporarily step away from a logged-in account. Automatic screen locking is only meant as a safeguard for those cases where a user forgot to lock the screen.

2.3.5.6.1 - Configure GUI Screen Locking

In the default GNOME desktop, the screen can be locked by choosing Lock Screen from the System menu. The gconftool-2 program can be used to enforce mandatory screen locking settings for the default GNOME environment. Run the following commands to enforce idle activation of the screen saver, screen locking, a blank-screen screensaver, and 10-minute idle activation time: # gconftool-2 --direct \ --config-source xml:readwrite:/etc/gconf/gconf.xml.mandatory \ --type bool \ --set /apps/gnome-screensaver/idle_activation_enabled true # gconftool-2 --direct \ --config-source xml:readwrite:/etc/gconf/gconf.xml.mandatory \ --type bool \ --set /apps/gnome-screensaver/lock_enabled true # gconftool-2 --direct \ --config-source xml:readwrite:/etc/gconf/gconf.xml.mandatory \ --type string \ --set /apps/gnome-screensaver/mode blank-only # gconftool-2 --direct \ --config-source xml:readwrite:/etc/gconf/gconf.xml.mandatory \ --type int \ --set /apps/gnome-screensaver/idle_delay 10 The default setting of 10 minutes for idle activation is reasonable for many office environments, but the setting should conform to whatever policy is defined. The screensaver mode blank-only is selected to conceal the contents of the display from passersby. Because users should be trained to lock the screen when they step away from the computer, the automatic locking feature is only meant as a backup. The Lock Screen icon from the System menu can also be dragged to the taskbar in order to facilitate even more convenient screen-locking. The root account cannot be screen-locked, but this should have no practical effect as the root account should never be used to log into an X Windows environment, and should only be used to for direct login via console in emergency circumstances. For more information about configuring GNOME screensaver, see http://live.gnome.org/GnomeScreensaver. For more information about enforcing preferences in the GNOME environment using the GConf configuration system, see http://www.gnome.org/projects/gconf and the man page gconftool-2(1).

2.3.5.6.2 - Configure Console Screen Locking

A console screen locking mechanism is provided in the vlock package, which is not installed by default. If the ability to lock console screens is necessary, install the vlock package: # yum install vlock Instruct users to invoke the program when necessary, in order to prevent passersby from abusing their login: $ vlock The -a option can be used to prevent switching to other virtual consoles.

2.3.6 - Use a Centralized Authentication Service

A centralized authentication service is any method of maintaining central control over account and authentication data and of keeping this data synchronized between machines. Such services can range in complexity from a script which pushes centrally-generated password files out to all machines, to a managed scheme such as LDAP or Kerberos. If authentication information is not centrally managed, it quickly becomes inconsistent, leading to out-of-date credentials and forgotten accounts which should have been deleted. In addition, many older protocols (such as NFS) make use of the UID to identify users over a network. This is not a good practice, and these protocols should be avoided if possible. However, since most sites must still make use of some older protocols, having consistent UIDs and GIDs site-wide is a significant benefit. Centralized authentication services do have the disadvantage that authentication information must be transmitted over a network, leading to a risk that credentials may be intercepted or manipulated. Therefore, these services must be deployed carefully. The following precautions should be taken when configuring any authentication service: * Ensure that authentication information and any sensitive account information are never sent over the network unencrypted. * Ensure that the root account has a local password, to allow recovery in case of network outage or authentication server failure. This guide recommends the use of LDAP. Secure configuration of OpenLDAP for clients and servers is described in Section 3.12. Kerberos is also a good choice for a centralized authentication service, but a description of its configuration is beyond the scope of this guide. The NIS service is not recommended, and should be considered obsolete. (See Section 3.2.4.)

2.3.7 - Warning Banners for System Accesses

Each system should expose as little information about itself as possible. System banners, which are typically displayed just before a login prompt, give out information about the service or the host’s operating system. This might include the distribution name and the system kernel version, and the particular version of a network service. This information can assist intruders in gaining access to the system as it can reveal whether the system is running vulnerable software. Most network services can be configured to limit what information is displayed. Many organizations implement security policies that require a system banner provide notice of the system’s ownership, provide warning to unauthorized users, and remind authorized users of their consent to monitoring.

2.3.7.1 - Modify the System Login Banner

The contents of the file /etc/issue are displayed on the screen just above the login prompt for users logging directly into a terminal. Remote login programs such as SSH or FTP can be configured to display /etc/issue as well. Instructions for configuring each server daemon to show this file can be found in the relevant sections of Chapter 3. By default, the system will display the version of the OS, the kernel version, and the host name. Edit /etc/issue. Replace the default text with a message compliant with the local site policy or a legal disclaimer.

2.3.7.2 - Implement a GUI Warning Banner

In the default graphical environment, users logging directly into the system are greeted with a login screen provided by the GNOME display manager. The warning banner should be displayed in this graphical environment for these users. The files for the default RHEL theme can be found in /usr/share/gdm/themes/RHEL. Add the following sample block of XML to /usr/share/gdm/themes/RHEL/RHEL.xml after the first two ”pixmap” entries: <item type="rect"/> <pos anchor="n" x="50%" y="10" width="box" height="box"/> <box> <item type="label"/> <normal font="Sans 14" color="#ffffff"/> <text> Insert the text of your warning banner here. </text> </item> </box> </item> The full syntax that GDM theme files expect is documented elsewhere, but the above XML will create a text box centered at the top of the screen. The font, text color, and exact positioning can all be easily modified by editing the appropriate values. The latest current GDM theme manual can be found at http://www.gnome.org/ projects/gdm/docs/thememanual.html.

2.4 - SELinux

SELinux is a feature of the Linux kernel which can be used to guard against misconfigured or compromised programs. SELinux enforces the idea that programs should be limited in what files they can access and what actions they can take. The default SELinux policy, as configured on RHEL5, has been sufficiently developed and debugged that it should be usable on almost any Red Hat machine with minimal configuration and a small amount of system administrator training. This policy prevents system services — including most of the common network-visible services such as mail servers, ftp servers, and DNS servers — from accessing files which those services have no valid reason to access. This action alone prevents a huge amount of possible damage from network attacks against services, from trojaned software, and so forth. This guide recommends that SELinux be enabled using the default (targeted) policy on every Red Hat system, unless that system has requirements which make a stronger policy appropriate.

2.4.1 - How SELinux Works

In the traditional Linux/Unix security model, known as Discretionary Access Control (DAC), processes run under a user and group identity, and enjoy that user and group’s access rights to all files and other objects on the system. This system brings with it a number of security problems, most notably: that processes frequently do not need and should not have the full rights of the user who ran them; that user and group access rights are not very granular, and may require administrators to allow too much access in order to allow the access that is needed; that the Unix filesystem contains many resources (such as temporary directories and world-readable files) which are accessible to users who have no legitimate reason to access them; and that legitimate users can easily provide open access to their own resources through confusion or carelessness. SELinux provides a Mandatory Access Control (MAC) system that greatly augments the DAC model. Under SELinux, every process and every object (e.g. file, socket, pipe) on the system is given a security context, a label which include detailed type information about the object. The kernel allows processes to access objects only if that access is explicitly allowed by the policy in effect. The policy defines transitions, so that a user can be allowed to run software, but the software can run under a different context than the user’s default. This automatically limits the damage that the software can do to files accessible by the calling user — the user does not need to take any action to gain this benefit. For an action to occur, both the traditional DAC permissions must be satisifed as well as SELinux’s MAC rules. If either do not permit the action, then it will not be allowed. In this way, SELinux rules can only make a system’s permissions more restrictive and secure. SELinux requires a complex policy in order to allow all the actions required of a system under normal operation. Three such policies have been designed for use with RHEL5, and are included with the system. In increasing order of power and complexity, they are: targeted, strict, and mls. The targeted SELinux policy consists mostly of Type Enforcement (TE) rules, and a small number of Role-Based Access Control (RBAC) rules. It restricts the actions of many types of programs, but leaves interactive users largely unaffected. The strict policy also uses TE and RBAC rules, but on more programs and more aggressively. The mls policy implements Multi-Level Security (MLS), which introduces even more kinds of labels — sensitivity and category — and rules that govern access based on these. The remainder of this section provides guidance for the configuration of the targeted policy and the administration of systems under this policy. Some pointers will be provided for readers who are interested in further strengthening their systems by using one of the stricter policies provided with RHEL5 or in writing their own policy.

2.4.2 - Enable SELinux

Edit the file /etc/selinux/config. Add or correct the following lines: SELINUX=enforcing SELINUXTYPE=targeted Edit the file /etc/grub.conf. Ensure that the following arguments DO NOT appear on any kernel command line in the file: selinux=0 enforcing=0 The directive SELINUX=enforcing enables SELinux at boot time. If SELinux is causing a lot of problems or preventing the system from booting, it is possible to boot into the warning-only mode SELINUX=permissive for debugging purposes. Make certain to change the mode back to enforcing after debugging, set the filesystems to be relabelled for consistency using the command touch /.autorelabel, and reboot. However, the RHEL5 default SELinux configuration should be sufficiently reasonable that most systems will boot without serious problems. Some applications that require deep or unusual system privileges, such as virtual machine software, may not be compatible with SELinux in its default configuration. However, this should be uncommon, and SELinux’s application support continues to improve. In other cases, SELinux may reveal unusual or insecure program behavior by design. The directive SELINUXTYPE=targeted configures SELinux to use the default targeted policy. See Section 2.4.6 if a stricter policy is appropriate for your site. The SELinux boot mode specified in /etc/selinux/config can be overridden by command-line arguments passed to the kernel. It is necessary to check grub.conf to ensure that this has not been done and to protect the bootloader as described in Section 2.3.5.2.

2.4.3 - Disable Unnecessary SELinux Daemons

Several daemons are installed by default as part of the RHEL5 SELinux support mechanism. These daemons may improve the system’s ability to enforce SELinux policy in a useful fashion, but may also represent unnecessary code running on the machine, increasing system risk. If these daemons are not needed on your system, they should be disabled.

2.4.3.1 - Disable and Remove SETroubleshoot if Possible

Is there a mission-critical reason to allow users to view SELinux denial information using the sealert GUI? If not, disable the service and remove the RPM: # chkconfig setroubleshoot off # yum erase setroubleshoot The setroubleshoot service is a facility for notifying the desktop user of SELinux denials in a user-friendly fashion. SELinux errors may provide important information about intrusion attempts in progress, or may give information about SELinux configuration problems which are preventing correct system operation. In order to maintain a secure and usable SELinux installation, error logging and notification is necessary. However, setroubleshoot is a service which has complex functionality, which runs a daemon and uses IPC to distribute information which may be sensitive, or even to allow users to modify SELinux settings, and which does not yet implement real authentication mechanisms. This guide recommends disabling setroubleshoot and using the kernel audit functionality to monitor SELinux’s behavior. In addition, since setroubleshoot automatically runs client-side code whenever a denial occurs, regardless of whether the setroubleshootd daemon is running, it is recommended that the program be removed entirely unless it is needed.

2.4.3.2 - Disable MCS Translation Service (mcstrans) if Possible

Unless there is some overriding need for the convenience of category label translation, disable the MCS translation service: # chkconfig mcstrans off The mcstransd daemon provides the category label translation information defined in /etc/selinux/targeted/ setrans.conf to client processes which request this information. Category labelling is unlikely to be used except in sites with special requirements. Therefore, it should be disabled in order to reduce the amount of potentially vulnerable code running on the system. See Section 2.4.6 for more information about systems which use category labelling.

2.4.3.3 - Restorecon Service (restorecond)

The restorecond daemon monitors a list of files which are frequently created or modified on running systems, and whose SELinux contexts are not set correctly. It looks for creation events related to files listed in /etc/ selinux/restorecond.conf, and sets the contexts of those files when they are discovered. The restorecond program is fairly simple, so it brings low risk, but, in its default configuration, does not add much value to a system. An automated program such as restorecond may be used to monitor problematic files for context problems, or system administrators may be trained to check file contexts of newly-created files using the command ls -lZ, and to repair contexts manually using the restorecon command. This guide makes no recommendation either for or against the use of restorecond.

2.4.4 - Check for Unconfined Daemons

Daemons that SELinux policy does not know about will inherit the context of the parent process. Because daemons are launched during startup and descend from the init process, they inherit the initrc t context. This is a problem because it may cause AVC denials, or it could allow privileges that the daemon does not require. To check for unconfined daemons, run the following command: # ps -eZ | egrep "initrc" | egrep -vw "tr|ps|egrep|bash|awk" | tr ':' ' ' | awk It should produce no output in a well-configured system.

2.4.5 - Debugging SELinux Policy Errors

SELinux’s default policies have improved significantly over time, and most systems should have few problems using the targeted SELinux policy. However, policy problems may still occasionally prevent accesses which should be allowed. This is especially true if your site runs any custom or heavily modified applications. This section gives some brief guidance on discovering and repairing SELinux-related access problems. Guidance given here is necessarily incomplete, but should provide a starting point for debugging. If you suspect that a permission error or other failure may be caused by SELinux (and are certain that misconfiguration of the traditional Unix permissions are not the cause of the problem), search the audit logs for AVC events: # ausearch -m AVC,USER_AVC -sv no The output of this command will be a set of events. The timestamp, along with the comm and pid fields, should indicate which line describes the problem. Look up the context under which the process is running. Assuming the process ID is PID , find the context by running: # ps -p PID -Z The AVC denial message should identify the offending file or directory. The name field should contain the filename (not the full pathname by default), and the ino field can be used to search by inode, if necessary. Assuming the file is FILE , find its SELinux context: # ls -Z FILE An administrator should suspect an SELinux misconfiguration whenever a program gets a “permission denied” error but the standard Unix permissions appear to be correct, or a program fails mysteriously on a task which seems to involve file access or network communication. As described in Section 2.4.1, SELinux augments each process with a context providing detailed type information about that process. The contexts under which processes run may be referred to as subject contexts. Similarly, each filesystem object is given a context. The targeted policy consists of a set of rules, each of which allows a subject type to perform some operation on a given object type. The kernel stores information about these access decisions in an structure known as an Access Vector Cache (AVC), so authorization decisions made by the system are audited with the type AVC. It is also possible for userspace modules to implement their own policies based on SELinux, and these decisions are audited with the type USER AVC. AVC denials are logged by the kernel audit facility (see Section 2.6.2 for configuration guidance on this subsystem) and may also be visible via setroubleshoot. This guide recommends the use of the audit userspace utilities to find AVC errors. It is possible to manually locate these errors by looking in the file /var/log/audit/audit.log or in /var/log/messages (depending on the syslog configuration in effect), but the ausearch tool allows finegrained searching on audit event types, which may be necessary if system call auditing is enabled as well. The command line above tells ausearch to look for kernel or userspace AVC messages (-m AVC,USER AVC) where the access attempt did not succeed (-sv no). If an AVC denial occurs when it should not have, the problem is generally one of the following: * The program is running with the wrong subject context. This could happen as a result of an incorrect context on the program’s executable file, which could happen if 3rd party software is installed and not given appropriate SELinux file contexts. * The file has the wrong object context because the current file’s context does not match the specification. This can occur when files are created or modified in certain ways. It is not atypical for configuration files to get the wrong contexts after a system configuration change performed by an administrator. To repair the file, use the command: # restorecon -v FILE This should produce output indicating that the file’s context has been changed. The /usr/bin/chcon program can be used to manually change a file’s context, but this is problematic because the change will not persist if it does not agree with the policy-defined contexts applied by restorecon. * The file has the wrong object context because the specification is either incorrect or does not match the way the file is being used on this system. In this case, it will be necessary to change the system file contexts. Run the system-config-selinux tool, and go to the “File Labeling” menu. This will give a list of files and wildcards corresponding to file labelling rules on the system. Add a rule which maps the file in question to the desired context. As an alternative, file contexts can be modified from the command line using the semanage(8) tool. * The program and file have the correct contexts, but the policy should allow some operation between those two contexts which is currently not allowed. In this case, it will be necessary to modify the SELinux policy. Run the system-config-selinux tool, and go to the “Boolean” menu. If your configuration is supported, but is not the Red Hat default, then there will be a boolean allowing real-time modification of the SELinux policy to fix the problem. Browse through the items in this menu, looking for one which is related to the service which is not working. As an alternative, SELinux booleans can be modified from the command line using the getsebool(8) and setsebool(8) tools. If there is no boolean, it will be necessary to create and load a policy module. A simple way to build a policy module is to use the audit2allow tool. This tool can take input in the format of AVC denial messages, and generate syntactically correct Type Enforcement rules which would be sufficient to prevent those denials. For example, to generate and display rules which would allow all kernel denials seen in the past five minutes, run: # ausearch -m AVC -sv no -ts recent | audit2allow It is possible to use audit2allow to directly create a module package suitable for loading into the kernel policy. To do this, invoke audit2allow with the -M flag: # ausearch -m AVC -sv no -ts recent | audit2allow -M localmodule If this is successful, several lines of output should appear. Review the generated TE rules in the file localmodule .te and ensure that they express what you wish to allow. The file localmodule .pp should also have been created. This file is a policy module package that can be loaded into the kernel. To do so, use system-config-selinux, go to the “Policy Module” menu and use the “Add” button to enable your module package in SELinux, or load it from the command line using semodule(8): # semodule -i localmodule .pp Section 45.2 of [9] covers this procedure in detail.

2.4.6 - Further Strengthening

The recommendations up to this point have discussed how to configure and maintain a system under the default configuration of the targeted policy, which constrains only the actions of daemons and system software. This guide strongly recommends that any site which is not currently using SELinux at all transition to the targeted policy, to gain the substantial security benefits provided by that policy. However, the default policy provides only a subset of the full security gains available from using SELinux. In particular, the SELinux policy is also capable of constraining the actions of interactive users, of providing compartmented access by sensitivity level (MLS) and/or category (MCS), and of restricting certain types of system actions using booleans beyond the RHEL5 defaults. This section introduces other uses of SELinux which may be possible, and provides links to some outside resources about their use. Detailed description of how to implement these steps is beyond the scope of this guide.

2.4.6.1 - Strengthen the Default SELinux Boolean Configuration

SELinux booleans are used to enable or disable segments of policy to comply with site policy. Booleans may apply to the entire system or to an individual daemon. For instance, the boolean allow execstack, if enabled, allows programs to make part of their stack memory region executable. This would apply to all programs on the system. The boolean ftp home dir allows ftpd processes to access user home directories, and applies only to daemons which implement FTP. The command $ getsebool -a lists the values of all SELinux booleans on the system. Section 2.4.5 discussed loosening boolean values in order to debug functionality problems which occur under more restrictive defaults. It is also useful to examine and strengthen the boolean settings, to disable functionality which is not required by legitimate programs on your system, but which might be symptomatic of an attack. See the manpages booleans(8), getsebool(8), and setsebool(8) for general information about booleans. There are also manual pages for several subsystems which discuss the use of SELinux with those systems. Examples include ftpd selinux(8), httpd selinux(8), and nfs selinux(8). Another good reference is the html documentation distributed with the selinux-policy RPM. This documentation is stored under /usr/share/doc/selinux-policy-version /html/ The pages global tunables.html and global booleans.html may be useful when examining booleans.

2.4.6.2 - Use a Stronger Policy

Using a stronger policy can greatly enhance security, but will generally require customization to be compatible with the particular system’s purpose, and this may be costly or time consuming. Under the targeted policy, interactive processes are given the type unconfined t, so interactive users are not constrained by SELinux even if they attempt to take strange or malicious actions. The first alternative policy available with RHEL5’s SELinux distribution, called strict, extends the protections offered by the default policy from daemons and system processes to all processes. To use the strict policy, first ensure that the policy module is installed: # yum install selinux-policy-strict Then edit /etc/selinux/config and correct the line: SELINUXTYPE=strict The mls policy type can be used to enforce sensitivity or category labelling, and requires site-specific configuration of these labels in order to be useful. To use this policy, install the appropriate policy module: # yum install selinux-policy-mls Then edit /etc/selinux/config and correct the line: SELINUXTYPE=mls Note: Switching between policies typically requires the entire disk to be relabelled, so that files get the appropriate SELinux contexts under the new policy. Boot with the additional grub command-line options enforcing=0 single autorelabel to relabel the disk in single-user mode, then reboot normally.

2.4.7 - SELinux References

* NSA SELinux resources: – Web page: http://www.nsa.gov/selinux/ – Mailing list: [email protected] List information at: http://www.nsa.gov/selinux/info/list.cfm * Fedora SELinux resources: – FAQ: http://docs.fedoraproject.org/selinux-faq/ – Wiki: http://fedoraproject.org/wiki/SELinux/ – Mailing list: [email protected] List information at: https://www.redhat.com/mailman/listinfo/fedora-selinux-list * Chapters 43–45 of Red Hat Enterprise Linux 5: Deployment Guide [9] * The book SELinux by Example: Using Security Enhanced Linux [13]

2.5 - Network Configuration and Firewalls

Most machines must be connected to a network of some sort, and this brings with it the substantial risk of network attack. This section discusses the security impact of decisions about networking which must be made when configuring a system. This section also discusses firewalls, network access controls, and other network security frameworks, which allow system-level rules to be written that can limit attackers’ ability to connect to your system. These rules can specify that network traffic should be allowed or denied from certain IP addresses, hosts, and networks. The rules can also specify which of the system’s network services are available to particular hosts or networks.

2.5.1 - Kernel Parameters which Affect Networking

The sysctl utility is used to set a number of parameters which affect the operation of the Linux kernel. Several of these parameters are specific to networking, and the configuration options in this section are recommended.

2.5.1.1 - Network Parameters for Hosts Only

Is this system going to be used as a firewall or gateway to pass IP traffic between different networks? If not, edit the file /etc/sysctl.conf and add or correct the following lines: net.ipv4.ip forward = 0 net.ipv4.conf.all.send redirects = 0 net.ipv4.conf.default.send redirects = 0 These settings disable hosts from performing network functionality which is only appropriate for routers.

2.5.1.2 - Network Parameters for Hosts and Routers

Edit the file /etc/sysctl.conf and add or correct the following lines: net.ipv4.conf.all.accept_source_route = 0 net.ipv4.conf.all.accept_redirects = 0 net.ipv4.conf.all.secure_redirects = 0 net.ipv4.conf.all.log_martians = 1 net.ipv4.conf.default.accept_source_route = 0 net.ipv4.conf.default.accept_redirects = 0 net.ipv4.conf.default.secure_redirects = 0 net.ipv4.icmp_echo_ignore_broadcasts = 1 net.ipv4.icmp_ignore_bogus_error_messages = 1 net.ipv4.tcp_syncookies = 1 net.ipv4.conf.all.rp_filter = 1 net.ipv4.conf.default.rp_filter = 1 These options improve Linux’s ability to defend against certain types of IPv4 protocol attacks. The accept source route, accept redirects, and secure redirects options are turned off to disable IPv4 protocol features which are considered to have few legitimate uses and to be easy to abuse. The net.ipv4.conf.all.log martians option logs several types of suspicious packets, such as spoofed packets, source-routed packets, and redirects. The icmp echo ignore broadcasts icmp ignore bogus error messages options protect against ICMP attacks. The tcp syncookies option uses a cryptographic feature called SYN cookies to allow machines to continue to accept legitimate connections when faced with a SYN flood attack. See [12] for further information on this option. The rp filter option enables RFC-recommended source validation. It should not be used on machines which are routers for very complicated networks, but is helpful for end hosts and routers serving small networks. For more information on any of these, see the kernel source documentation file /Documentation/networking/ip-sysctl.txt.2

2.5.2 - Wireless Networking

Wireless networking (sometimes referred to as 802.11 or Wi-Fi) presents a serious security risk to sensitive or classified systems and networks. Wireless networking hardware is much more likely to be included in laptop or portable systems than desktops or servers. See Section 3.3.14 for information on Bluetooth wireless support. Bluetooth serves a different purpose and possesses a much shorter range, but it still presents serious security risks. Removal of hardware is the only way to absolutely ensure that the wireless capability remains disabled. If it is completely impractical to remove the wireless hardware, and site policy still allows the device to enter sensitive spaces, every effort to disable the capability via software should be made. In general, acquisition policy should include provisions to prevent the purchase of equipment that will be used in sensitive spaces and includes wireless capabilities.

2.5.2.1 - Remove Wireless Hardware if Possible

Identifying the wireless hardware is the first step in removing it. The system’s hardware manual should contain information on its wireless capabilities. Wireless hardware included with a laptop typically takes the form of a mini-PCI card or PC card. Other forms include devices which plug into USB or Ethernet ports, but these should be readily apparent and easy to remove from the base system. A PC Card (originally called a PCMCIA card) is designed to be easy to remove, though it may be hidden when inserted into the system. Frequently, there will be one or more buttons near the card slot that, when pressed, eject the card from the system. If no card is ejected, the slot is empty. A mini-PCI card is approximately credit-card sized and typically accessible via a removable panel on the underside of the laptop. Removing the panel may require simple tools. In addition to manually inspecting the hardware, it is also possible to query the system for its installed hardware devices. The commands /sbin/lspci and /sbin/lsusb will show a list of all recognized devices on their respective buses, and this may indicate the presence of a wireless device.

2.5.2.2 - Disable Wireless Through Software Configuration

If it is impossible to remove the wireless hardware from the device in question, disable as much of it as possible through software. The following methods can disable software support for wireless networking, but note that these methods do not prevent malicious software or careless users from re-activating the devices.

2.5.2.2.1 - Disable Wireless in BIOS

Some laptops that include built-in wireless support offer the ability to disable the device through the BIOS. This is system-specific; consult your hardware manual or explore the BIOS setup during boot. 2A recent version of this file can be found online at http://lxr.linux.no/source/Documentation/networking/ip-sysctl.txt.

2.5.2.2.2 - Deactivate Wireless Interfaces

Deactivating the wireless interfaces should prevent normal usage of the wireless capability. First, identify the interfaces available with the command: # ifconfig -a Additionally,the following command may also be used to determine whether wireless support (“extensions”) is included for a particular interface, though this may not always be a clear indicator: # iwconfig After identifying any wireless interfaces (which may have names like wlan0, ath0, wifi0, or eth0), deactivate the interface with the command: # ifdown interface These changes will only last until the next reboot. To disable the interface for future boots, remove the appropriate interface file from /etc/sysconfig/network-scripts: # rm /etc/sysconfig/network-scripts/ifcfg-interface

2.5.2.2.3 - Disable Wireless Drivers

Removing the kernel drivers that provide support for wireless Ethernet devices will prevent users from easily activating the devices. To remove the wireless drivers from the system: # rm -r /lib/modules/kernelversion(s) /kernel/drivers/net/wireless This command must also be repeated every time the kernel is upgraded.

2.5.3 - IPv6

The system includes support for Internet Protocol version 6. A major and often-mentioned improvement over IPv4 is its enormous increase in the number of available addresses. Another important feature is its support for automatic configuration of many network settings.

2.5.3.1 - Disable Support for IPv6 unless Needed

Because the IPv6 networking code is relatively new and complex, it is particularly important that it be disabled unless needed. Despite configuration that suggests support for IPv6 has been disabled, link-local IPv6 address autoconfiguration occurs even when only an IPv4 address is assigned. The only way to effectively prevent execution of the IPv6 networking stack is to prevent the kernel from loading the IPv6 kernel module.

2.5.3.1.1 - Disable Automatic Loading of IPv6 Kernel Module

To prevent the IPv6 kernel module (ipv6) from being loaded, add the following line to /etc/modprobe.conf: alias net-pf-10 off The unexpected name is a result of how the kernel requests modules for network protocol families; net-pf-10 is an alias for the ipv6 module.

2.5.3.1.2 - Disable Interface Usage of IPv6

To prevent configuration of IPv6 for all interfaces, add or correct the following lines in /etc/sysconfig/ network: NETWORKING_IPV6=no IPV6INIT=no For each network interface IFACE , add or correct the following lines in /etc/sysconfig/network-scripts/ ifcfg-IFACE as an additional prevention mechanism: IPV6INIT=no If it becomes necessary later to configure IPv6, only the interfaces requiring it should be enabled.

2.5.3.2 - Configure IPv6 Settings if Necessary

A major feature of IPv6 is the extent to which systems implementing it can automatically configure their networking devices using information from the network. From a security perspective, manually configuring important configuration information is always preferable to accepting it from the network in an unauthenticated fashion.

2.5.3.2.1 - Disable Automatic Configuration

Disable the system’s acceptance of router advertisements and redirects by adding or correcting the following line in /etc/sysconfig/network (note that this does not disable sending router solicitations): IPV6_AUTOCONF=no

2.5.3.2.2 - Manually Assign Global IPv6 Address

To manually assign an IP address for an interface IFACE, edit the file /etc/sysconfig/network-scripts/ ifcfg-IFACE. Add or correct the following line (substituting the correct IPv6 address): IPV6ADDR=2001:0DB8::ABCD/64 Manually assigning an IP address is preferable to accepting one from routers or from the network otherwise. The example address here is an IPv6 address reserved for documentation purposes, as defined by RFC3849.

2.5.3.2.3 - Use Privacy Extensions for Address if Necessary

To introduce randomness into the automatic generation of IPv6 addresses, add or correct the following line in /etc/sysconfig/network-scripts/ifcfg-IFACE: IPV6_PRIVACY=rfc3041 Automatically-generated IPv6 addresses are based on the underlying hardware (e.g. Ethernet) address, and so it becomes possible to track a piece of hardware over its lifetime using its traffic. If it is important for a system’s IP address to not trivially reveal its hardware address, this setting should be applied.

2.5.3.2.4 - Manually Assign IPv6 Router Address

Edit the file /etc/sysconfig/network-scripts/ifcfg-IFACE , and add or correct the following line (substituting your gateway IP as appropriate): IPV6_DEFAULTGW=2001:0DB8::0001 Router addresses should be manually set and not accepted via any autoconfiguration or router advertisement.

2.5.3.2.5 - Limit Network-Transmitted Configuration

Add the following lines to /etc/sysctl.conf to limit the configuration information requested from other systems, and accepted from the network: net.ipv6.conf.default.router_solicitations = 0 net.ipv6.conf.default.accept_ra_rtr_pref = 0 net.ipv6.conf.default.accept_ra_pinfo = 0 net.ipv6.conf.default.accept_ra_defrtr = 0 net.ipv6.conf.default.autoconf = 0 net.ipv6.conf.default.dad_transmits = 0 net.ipv6.conf.default.max_addresses = 1 The router solicitations setting determines how many router solicitations are sent when bringing up the interface. If addresses are statically assigned, there is no need to send any solicitations. The accept ra pinfo setting controls whether the system will accept prefix info from the router. The accept ra defrtr setting controls whether the system will accept Hop Limit settings from a router advertisement. Setting it to 0 prevents a router from changing your default IPv6 Hop Limit for outgoing packets. The autoconf setting controls whether router advertisements can cause the system to assign a global unicast address to an interface. The dad transmits setting determines how many neighbor solicitations to send out per address (global and link-local) when bringing up an interface to ensure the desired address is unique on the network. The max addresses setting determines how many global unicast IPv6 addresses can be assigned to each interface. The default is 16, but it should be set to exactly the number of statically configured global addresses required.

2.5.4 - TCP Wrapper

TCP Wrapper is a library which provides simple access control and standardized logging for supported applications which accept connections over a network. Historically, TCP Wrapper was used to support inetd services. Now that inetd is deprecated (see Section 3.2.1), TCP Wrapper supports only services which were built to make use of the libwrap library. To determine whether a given executable daemon /path/to/daemon supports TCP Wrapper, check the documentation, or run: $ ldd /path/to/daemon | grep libwrap.so If this command returns any output, then the daemon probably supports TCP Wrapper. An alternative to TCP Wrapper support is packet filtering using iptables. Note that iptables works at the network level, while TCP Wrapper works at the application level. This means that iptables filtering is more efficient and more resistant to flaws in the software being protected, but TCP Wrapper provides support for logging, banners, and other application-level tricks which iptables cannot provide.

2.5.4.1 - How TCP Wrapper Protects Services

TCP Wrapper provides access control for the system’s network services using two configuration files. When a connection is attempted: 1. The file /etc/hosts.allow is searched for a rule matching the connection. If one is found, the connection is allowed. 2. Otherwise, the file /etc/hosts.deny is searched for a rule matching the connection. If one is found, the connection is rejected. 3. If no matching rules are found in either file, then the connection is allowed. By default, TCP Wrapper does not block access to any services. In the simplest case, each rule in /etc/hosts.allow and /etc/hosts.deny takes the form: daemon : client where daemon is the name of the server process for which the connection is destined, and client is the partial or full hostname or IP address of the client. It is valid for daemon and client to contain one item, a comma-separated list of items, or a special keyword like ALL, which matches any service or client. (See the hosts access(5) manpage for a list of other keywords.) Note: Partial hostnames start at the root domain and are delimited by the . character. So the client machine host03.dev.example.com, with IP address 10.7.2.3, could be matched by any of the specifications: .example.com .dev.example.com 10.7.2.

2.5.4.2 - Reject All Connections From Other Hosts if Appropriate

Restrict all connections to non-public services to localhost only. Suppose pubsrv1 and pubsrv2 are the names of daemons which must be accessed remotely. Configure TCP Wrapper as follows. Edit /etc/hosts.allow. Add the following lines: pubsrv1 ,pubsrv2 : ALL ALL: localhost Edit /etc/hosts.deny. Add the following line: ALL: ALL These rules deny connections to all TCP Wrapper enabled services from any host other than localhost, but allow connections from anywhere to the services which must be publicly accessible. (If no public services exist, the first line in /etc/hosts.allow may be omitted.)

2.5.4.3 - Allow Connections Only From Hosts in This Domain if Appropriate

For each daemon, domainsrv , which only needs to be contacted from inside the local domain, example.com , configure TCP Wrapper to deny remote connections. Edit /etc/hosts.allow. Add the following line: domainsrv : .example.com Edit /etc/hosts.deny. Add the following line: domainsrv : ALL There are many possible examples of services which need to communicate only within the local domain. If a machine is a local compute server, it may be necessary for users to connect via SSH from their desktop workstations, but not from outside the domain. In that case, you should protect the daemon sshd using this method. As another example, RPC-based services such as NFS might be enabled within the domain only, in which case the daemon portmap should be protected. Note: This example protects only the service domainsrv . No filtering is done on other services unless a line is entered into /etc/hosts.deny which refers to those services by name, or which restricts the special service ALL.

2.5.4.4 - Monitor Syslog for Relevant Connections and Failures

Ensure that the following line exists in /etc/syslog.conf. (This is the default, so it is likely to be correct if the configuration has not been modified): authpriv.* /var/log/secure Configure logwatch or other log monitoring tools to periodically summarize failed connections reported by TCP Wrapper at the facility authpriv.info. By default, TCP Wrapper audits all rejected connections at the facility authpriv, level info. In the log file, TCP Wrapper rejections will contain the substring: daemon [pid ]: refused connect from ipaddr These lines can be used to detect malicious scans, and to debug failures resulting from an incorrect TCP Wrapper configuration. If appropriate, it is possible to change the syslog facility and level used by a given TCP Wrapper rule by adding the severity option to each desired configuration line in /etc/hosts.deny: daemon : client : severity facility .level By default, successful connections are not logged by TCP Wrapper. See Section 2.6 for more information about system auditing.

2.5.4.5 - Further Resources

For more information about TCP Wrapper, see the tcpd(8) and hosts access(5) manpages and the documentation directory /usr/share/doc/tcp wrappers-version . Some information may be available from the Tools section of the author’s website, http://www.porcupine.org, and from the RHEL4 Reference Guide [6].

2.5.5 - Iptables and Ip6tables

A host-based firewall called Netfilter is included as part of the Linux kernel distributed with the system. It is activated by default. This firewall is controlled by the program iptables, and the entire capability is frequently referred to by this name. An analogous program called ip6tables handles filtering for IPv6. Unlike TCP Wrappers, which depends on the network server program to support and respect the rules written, Netfilter filtering occurs at the kernel level, before a program can even process the data from the network packet. As such, any program on the system is affected by the rules written. This section provides basic information about strengthening the iptables and ip6tables configurations included with the system. For more complete information that may allow the construction of a sophisticated ruleset tailored to your environment, please consult the references at the end of this section.

2.5.5.1 - Inspect and Activate Default Rules

View the currently-enforced iptables rules by running the command: # iptables -nL --line-numbers The command is analogous for the ip6tables program. If the firewall does not appear to be active (i.e., no rules appear), activate it and ensure that it starts at boot by issuing the following commands (and analogously for ip6tables): # service iptables restart # chkconfig iptables on The default iptables rules are: Chain INPUT (policy ACCEPT) num target prot opt source destination 1 RH-Firewall-1-INPUT all -- 0.0.0.0/0 0.0.0.0/0 Chain FORWARD (policy ACCEPT) num target prot opt source destination 1 RH-Firewall-1-INPUT all -- 0.0.0.0/0 0.0.0.0/0 Chain OUTPUT (policy ACCEPT) num target prot opt source destination Chain RH-Firewall-1-INPUT (2 references) num target prot opt source destination 1 ACCEPT all -- 0.0.0.0/0 0.0.0.0/0 2 ACCEPT icmp -- 0.0.0.0/0 0.0.0.0/0 icmp type 255 3 ACCEPT esp -- 0.0.0.0/0 0.0.0.0/0 4 ACCEPT ah -- 0.0.0.0/0 0.0.0.0/0 5 ACCEPT udp -- 0.0.0.0/0 224.0.0.251 udp dpt:5353 6 ACCEPT udp -- 0.0.0.0/0 0.0.0.0/0 udp dpt:631 7 ACCEPT tcp -- 0.0.0.0/0 0.0.0.0/0 tcp dpt:631 8 ACCEPT all -- 0.0.0.0/0 0.0.0.0/0 state RELATED,ESTABLISHED 9 ACCEPT tcp -- 0.0.0.0/0 0.0.0.0/0 state NEW tcp dpt:22 10 REJECT all -- 0.0.0.0/0 0.0.0.0/0 reject-with icmp-host-prohibited The ip6tables default rules are similar, with its rules 2 and 10 reflecting protocol naming and addressing differences. Instead of rule 8, however, ip6tables includes two rules that accept all incoming udp and tcp packets with a particular destination port range. This is because the current Netfilter implementation for IPv6 lacks reliable connection-tracking functionality.

2.5.5.2 - Understand the Default Ruleset

Understanding and creating firewall rules can be a challenging activity, filled with corner cases and difficult-todebug problems. Because of this, administrators should develop a thorough understanding of the default ruleset before carefully modifying it. The default ruleset is divided into four sections, each of which is called a chain: INPUT, FORWARD, OUTPUT, and RH-Firewall-1-INPUT. INPUT, OUTPUT, and FORWARD are built-in chains. * The INPUT chain is activated on packets destined for (i.e., addressed to) the system. * The OUTPUT chain is activated on packets which are originating from the system. * The FORWARD chain is activated for packets that the system will process and send through another interface, if so configured. * The RH-Firewall-1-INPUT chain is a custom (or user-defined) chain, which is used by the INPUT and FORWARD chains. A packet starts at the first rule in the appropriate chain and proceeds until it matches a rule. If a match occurs, then control will jump to the specified target. The default ruleset uses the built-in targets ACCEPT and REJECT, and also the user-defined target/chain RH-Firewall-1-INPUT. Jumping to the target ACCEPT means to allow the packet through, while REJECT means to drop the packet and send an error message to the sending host. A related target called DROP means to drop the packet on the floor without even sending an error message. The default policy for all of the built-in chains (shown after their names in the rule output above) is set to ACCEPT. This means that if no rules in the chain match the packets, they are allowed through. Because no rules at all are written for the OUTPUT chain, this means that iptables does not stop any packets originating from the system. The INPUT and FORWARD chains jump to the user-defined target RH-Firewall-1-INPUT for all packets. RH-Firewall-1-INPUT tries to match, in order, the following rules for both iptables and ip6tables: * Rule 1 appears to accept all packets. However, this appears true only because the rules are not presented in verbose mode. Executing the command # iptables -vnL --line-numbers reveals that this rule applies only to the loopback (lo) interface (see column in), while all other rules apply to all interfaces. Thus, packets not coming from the loopback interface do not match and proceed to the next rule. * Rule 2 explicitly allows all icmp packet types; iptables uses the code 255 to mean all icmp types. * Rule 3 explicitly allows all esp packets; these are packets which contain IPsec ESP headers. * Rule 4 explicitly allows all ah packets; these are packets which contain an IPsec authentication header SPI. * Rule 5 allows inbound communication on udp port 5353 (mDNS), which the avahi daemon uses. * Rules 6 and 7 allows inbound communication on both tcp and udp port 631, which the cups daemon uses. * Rule 8, in the iptables rules, allows inbound packets that are part of a session initiated by the system. In ip6tables, rules 8 and 9 allow any inbound packets with a destination port address between 32768 and 61000. * Rule 9 (10, for ip6tables) allows inbound connections in tcp port 22, which is the SSH protocol. * Rule 10 (11, for ip6tables) rejects all other packets and sends an error message to the sender. Because this is the last rule and matches any packet, it effectively prevents any packet from reaching the chain’s default ACCEPT target. Preventing the acceptance of any packet that is not explicitly allowed is proper design for a firewall.

2.5.5.3 - Strengthen the Default Ruleset

The default rules can be strengthened. The system scripts that activate the firewall rules expect them to be defined in the configuration files iptables and ip6tables in the directory /etc/sysconfig. Many of the lines in these files are similar to the command line arguments that would be provided to the programs /sbin/iptables or /sbin/ip6tables – but some are quite different. The program system-config-securitylevel allows additional services to penetrate the default firewall rules and automatically adjusts /etc/ sysconfig/ iptables . This program is only useful if the default ruleset meets your security requirements. Otherwise, this program should not be used to make changes to the firewall configuration because it re-writes the saved configuration file. The following recommendations describe how to strengthen the default ruleset configuration file. An alternative to editing this configuration file is to create a shell script that makes calls to the iptables program to load in rules, and then invokes service iptables save to write those loaded rules to /etc/sysconfig/iptables. The following alterations can be made directly to /etc/sysconfig/iptables and /etc/sysconfig/ip6tables. Instructions apply to both unless otherwise noted. Language and address conventions for regular iptables are used throughout this section; configuration for ip6tables will be either analogous or explicitly covered.

2.5.5.3.1 - Change the Default Policies

Change the default policy to DROP (from ACCEPT) for the INPUT and FORWARD built-in chains: *filter :INPUT DROP [0:0] :FORWARD DROP [0:0] Changing the default policy in this way implements proper design for a firewall, i.e. any packets which are not explicitly permitted should not be accepted.

2.5.5.3.2 - Restrict ICMP Message Types

In /etc/sysconfig/iptables, the accepted ICMP messages types can be restricted. To accept only ICMP echo reply, destination unreachable, and time exceeded messages, remove the line: -A RH-Firewall-1-INPUT -p icmp --icmp-type any -j ACCEPT and insert the lines: -A RH-Firewall-1-INPUT -p icmp --icmp-type echo-reply -j ACCEPT -A RH-Firewall-1-INPUT -p icmp --icmp-type destination-unreachable -j ACCEPT -A RH-Firewall-1-INPUT -p icmp --icmp-type time-exceeded -j ACCEPT To allow the system to respond to pings, also insert the following line: -A RH-Firewall-1-INPUT -p icmp --icmp-type echo-request -j ACCEPT Ping responses can also be limited to certain networks or hosts by using the -s option in the previous rule. Because IPv6 depends so heavily on ICMPv6, it is preferable to deny the ICMPv6 packets you know you don’t need (e.g. ping requests) in /etc/sysconfig/ip6tables, while letting everything else through: -A RH-Firewall-1-INPUT -p icmpv6 --icmpv6-type echo-request -j DROP If you are going to statically configure the machine’s address, it should ignore Router Advertisements which could add another IPv6 address to the interface or alter important network settings: -A RH-Firewall-1-INPUT -p icmpv6 --icmpv6-type router-advertisement -j DROP Restricting other ICMPv6 message types in /etc/sysconfig/ip6tables is not recommended because the operation of IPv6 depends heavily on ICMPv6. Thus, more care must be taken when blocking ICMPv6 types.

2.5.5.3.3 - Remove IPsec Rules

If the system will not process IPsec traffic, then remove the following rules: -A RH-Firewall-1-INPUT -p 50 -j ACCEPT -A RH-Firewall-1-INPUT -p 51 -j ACCEPT

2.5.5.3.4 - Log and Drop Packets with Suspicious Source Addresses

Packets with non-routable source addresses should be rejected, as they may indicate spoofing. Because the modified policy will reject non-matching packets, you only need to add these rules if you are interested in also logging these spoofing or suspicious attempts before they are dropped. If you do choose to log various suspicious traffic, add identical rules with a target of DROP after each LOG. To log and then drop these IPv4 packets, insert the following rules in /etc/sysconfig/iptables (excepting any that are intentionally used): -A INPUT -i eth0 -s 10.0.0.0/8 -j LOG --log-prefix "IP DROP SPOOF A: " -A INPUT -i eth0 -s 172.16.0.0/12 -j LOG --log-prefix "IP DROP SPOOF B: " -A INPUT -i eth0 -s 192.168.0.0/16 -j LOG --log-prefix "IP DROP SPOOF C: " -A INPUT -i eth0 -s 224.0.0.0/4 -j LOG --log-prefix "IP DROP MULTICAST D: " -A INPUT -i eth0 -s 240.0.0.0/5 -j LOG --log-prefix "IP DROP SPOOF E: " -A INPUT -i eth0 -d 127.0.0.0/8 -j LOG --log-prefix "IP DROP LOOPBACK: " Similarly, you might wish to log packets containing some IPv6 reserved addresses if they are not expected on your network: -A INPUT -i eth0 -s ::1 -j LOG --log-prefix "IPv6 DROP LOOPBACK: " -A INPUT -s 2002:E000::/20 -j LOG --log-prefix "IPv6 6to4 TRAFFIC: " -A INPUT -s 2002:7F00::/24 -j LOG --log-prefix "IPv6 6to4 TRAFFIC: " -A INPUT -s 2002:0000::/24 -j LOG --log-prefix "IPv6 6to4 TRAFFIC: " -A INPUT -s 2002:FF00::/24 -j LOG --log-prefix "IPv6 6to4 TRAFFIC: " -A INPUT -s 2002:0A00::/24 -j LOG --log-prefix "IPv6 6to4 TRAFFIC: " -A INPUT -s 2002:AC10::/28 -j LOG --log-prefix "IPv6 6to4 TRAFFIC: " -A INPUT -s 2002:C0A8::/32 -j LOG --log-prefix "IPv6 6to4 TRAFFIC: " If you are not expecting to see site-local multicast or auto-tunneled traffic, you can log those: -A INPUT -s FF05::/16 -j LOG --log-prefix "IPv6 SITE-LOCAL MULTICAST: " -A INPUT -s ::0.0.0.0/96 -j LOG --log-prefix "IPv4 COMPATIBLE IPv6 ADDR: " If you wish to block multicasts to all link-local nodes (e.g. if you are not using router autoconfiguration and do not plan to have any services that multicast to the entire local network), you can block the link-local all-nodes multicast address (before accepting incoming ICMPv6): -A INPUT -d FF02::1 -j LOG --log-prefix "Link-local All-Nodes Multicast: " However, if you’re going to allow IPv4 compatible IPv6 addresses (of the form ::0.0.0.0/96), you should then consider logging the non-routable IPv4-compatible addresses: -A INPUT -s ::0.0.0.0/104 -j LOG --log-prefix "IP NON-ROUTABLE ADDR: " -A INPUT -s ::127.0.0.0/104 -j LOG --log-prefix "IP DROP LOOPBACK: " -A INPUT -s ::224.0.0.0.0/100 -j LOG --log-prefix "IP DROP MULTICAST D: " -A INPUT -s ::255.0.0.0/104 -j LOG --log-prefix "IP BROADCAST: " If you are not expecting to see any IPv4 (or IPv4-compatible) traffic on your network, consider logging it before it gets dropped: -A INPUT -s ::FFFF:0.0.0.0/96 -j LOG --log-prefix "IPv4 MAPPED IPv6 ADDR: " -A INPUT -s 2002::/16 -j LOG --log-prefix "IPv6 6to4 ADDR: " The following rule will log all traffic originating from a site-local address, which is deprecated address space: -A INPUT -s FEC0::/10 -j LOG --log-prefix "SITE-LOCAL ADDRESS TRAFFIC: "

2.5.5.3.5 - Log and Drop All Other Packets

To log before dropping all packets that are not explicitly accepted by previous rules, change the final lines from -A RH-Firewall-1-INPUT -j REJECT --reject-with icmp-host-prohibited COMMIT to -A RH-Firewall-1-INPUT -j LOG -A RH-Firewall-1-INPUT -j DROP COMMIT The rule to log all dropped packets must be used with care. Chatty but otherwise non-malicious network protocols (e.g. NetBIOS) may result in voluminous logs; insertion of earlier rules to explicitly drop their packets without logging may be appropriate.

2.5.5.4 - Further Strengthening

Further strengthening, particularly as a result of customization to a particular environment, is possible for the iptables rules. Consider the following options, though their practicality depends on the network environment and usage scenario: * Restrict outgoing traffic. As shown above, the OUTPUT chain’s default policy can be changed to DROP, and rules can be written to specifically allow only certain types of outbound traffic. Such a policy could prevent casual usage of insecure protocols such as ftp and telnet, or even disrupt spyware. However, it would still not prevent a sophisticated user or program from using a proxy to circumvent the intended effects, and many client programs even try to automatically tunnel through port 80 to avoid such restrictions. * SYN flood protection. SYN flood protection can be provided by iptables, but might run into limiting issues for servers. For example, the iplimit match can be used to limit simultaneous connections from a given host or class. Similarly, the recent match allows the firewall to deny additional connections from any host within a given period of time (e.g. more than 3 –state NEW connections on port 22 within a minute to prevent dictionary login attacks). A more precise option for DoS protection is using TCP SYN cookies. (See Section 2.5.1.2 for more information.)

2.5.5.5 - Further Resources

More complex, restrictive, and powerful rulesets can be created, but this requires careful customization that relies on knowledge of the particular environment. The following resources provide more detailed information: * The iptables(8) man page * The Netfilter Project’s documentation at http://www.netfilter.org * The Red Hat Enterprise Linux Reference Guide

2.5.6 - Secure Sockets Layer Support

The Secure Sockets Layer (SSL) protocol provides encrypted and authenticated network communications, and many network services include support for it. Using SSL is recommended, especially to avoid any plaintext transmission of sensitive data, even over a local network. The SSL implementation included with the system is called OpenSSL. Recent implementations of SSL may also be referred to as Transport Layer Security (TLS). SSL uses public key cryptography to provide authentication and encryption. Public key cryptography involves two keys, one called the public key and the other called the private key. These keys are mathematically related such that data encrypted with one key can only be decrypted by the other, and vice versa. As their names suggest, public keys can be distributed to anyone while a private key must remain known only to its owner. SSL uses certificates, which are files that hold cryptographic data: a public key, and a signature of that public key. In SSL authentication, a server presents a client with its certificate as a means of demonstrating that it is who it claims it is. If everything goes correctly, the client can verify the server’s certificate by determining that the signature inside the certificate could only have been generated by a third party whom the client trusts. This third party is called a Certificate Authority (CA). Each client system should also have certificates from trusted CAs, and the client uses these CA certificates to verify the authenticity of the server’s certificate. After authenticating a server using its certificate and a CA certificate, SSL provides encryption by using the server certificate to securely negotiate a shared secret key. If your server must communicate using SSL with systems that might not be able to securely accept a new CA certificate prior to any SSL communication, then paying an established CA (whose certificates your clients already have) to sign your server certificates is recommended. The steps for doing this vary by vendor. Once the signed certificates have been obtained, configuration of the services is the same whether they were purchased from a vendor or signed by your own CA. For setting up an internal network and encrypting local traffic, creating your own CA to sign SSL certificates can be appropriate. The major steps in this process are: 1. Create a CA to sign certificates 2. Create SSL certificates for servers using that CA 3. Enable client support by distributing the CA’s certificate

2.5.6.1 - Create a CA to Sign Certificates

The following instructions apply to OpenSSL since it is included with the system, but creating a CA is possible with any standards-compliant SSL toolkit. The security of certificates depends on the security of the CA that signed them, so performing these steps on a secure machine is critical. The system used as a CA should be physically secure and not connected to any network. It should receive any certificate signing requests (CSRs) via removable media and output certificates onto removable media. The script /etc/pki/tls/misc/CA is included to assist in the process of setting up a CA. This script uses many settings in /etc/pki/tls/openssl.cnf. The settings in this file can be changed to suit your needs and allow easier selection of default settings, particularly in the [req distinguished name] section. To create the CA: # cd /etc/pki/tls/misc # ./CA -newca * When prompted, press enter to create a new CA key with the default name cakey.pem. * When prompted, enter a password that will protect the private key, then enter the same password again to verify it. * At the prompts, fill out as much of the CA information as is relevant for your site. You must specify a common name, or generation of the CA certificate will fail. * Next, you will be prompted for the password, so that the script can re-open the private key in order to write the certificate. This step performs the following actions: * creates the directory /etc/pki/CA (by default), which contains files necessary for the operation of a certificate authority. These are: – serial, which contains the current serial number for certificates signed by the CA – index.txt, which is a text database file that contains information about certificates signed – crl, which is a directory for holding revoked certificates – private, a directory which stores the CA’s private key * creates a public-private key pair for the CA in the file /etc/pki/CA/private/cakey.pem. The private key must be kept private in order to ensure the security of the certificates the CA will later sign. * signs the public key (using the corresponding private key, in a process called self-signing) to create the CA certificate, which is then stored in /etc/pki/CA/cacert.pem. When the CA later signs a server certificate using its private key, it means that it is vouching for the authenticity of that server. A client can then use the CA’s certificate (which contains its public key) to verify the authenticity of the server certificate. To accomplish this, it is necessary to distribute the CA certificate to any clients as covered in Section 2.5.6.3.

2.5.6.2 - Create SSL Certificates for Servers

Creating an SSL certificate for a server involves the following steps: 1. A public-private key pair for the server must be generated. 2. A certificate signing request (CSR) must be created from the key pair. 3. The CSR must be signed by a certificate authority (CA) to create the server certificate. If a CA has been set up as described in Section 2.5.6.1, it can sign the CSR. 4. The server certificate and keys must be installed on the server. Instructions on how to generate and sign SSL certificates are provided for the following common services: * Mail server, in Section 3.11.4.6. * Dovecot, in Section 3.17.2.2. * Apache, in Section 3.16.4.1.

2.5.6.3 - Enable Client Support

The system ships with certificates from well-known commercial CAs. If your server certificates were signed by one of these established CAs, then this step is not necessary since the clients should include the CA certificate already. If your servers use certificates signed by your own CA, some user applications will warn that the server’s certificate cannot be verified because the CA is not recognized. Other applications may simply fail to accept the certificate and refuse to operate, or continue operating without ever having properly verified the server certificate. To avoid this warning, and properly authenticate the servers, your CA certificate must be exported to every application on every client system that will be connecting to an SSL-enabled server.

2.5.6.3.1 - Adding a Trusted CA for Firefox

Firefox needs to have a certificate from the CA that signed the web server’s certificate, so that it can authenticate the web server. To import a new CA certificate into Firefox 1.5: 1. Launch Firefox and choose Preferences from the Edit menu. 2. Click the Advanced button. 3. Select the Security pane. 4. Click the View Certificates button. 5. Click the Authorities tab. 6. Click the Import button at the bottom of the screen. 7. Navigate to the CA certificate and import it.

2.5.6.3.2 - Adding a Trusted CA for Thunderbird

Thunderbird needs to have a certificate from the CA that signed the mail server’s certificates, so that it can authenticate the mail server(s). To import a new CA certificate into Thunderbird 1.5: 1. Launch Thunderbird and choose Account Settings from the Edit menu. 2. Select Security under the account name. 3. Click the View Certificates button. 4. Click the Authorities tab. 5. Click the Import button at the bottom of the screen. 6. Navigate to the CA certificate and import it.

2.5.6.3.3 - Adding a Trusted CA for Evolution

The Evolution e-mail client needs to have a certificate from the CA that signed the mail server’s certificates, so that it can authenticate the mail server(s). To import a new CA certificate into Evolution: 1. Launch Evolution and choose Preferences from the Edit menu. 2. Select Certificates from the icon list on the left. 3. Click the Authorities tab. 4. Click the Import button. 5. Navigate to the CA certificate and import it.

2.5.6.4 - Further Resources

* The OpenSSL Project home page at http://www.openssl.org * The openssl(1) man page * Jeremy Mates’s how-to: http://sial.org/howto/openssl

2.6 - Logging and Auditing

Successful local or network attacks on systems do not necessarily leave clear evidence of what happened. It is necessary to build a configuration in advance that collects this evidence, both in order to determine that something anomalous has occurred, and in order to respond appropriately. In addition, a well-configured logging and audit infrastructure will show evidence of any misconfiguration which might leave the system vulnerable to attack. Logging and auditing take different approaches to collecting data. A logging infrastructure provides a framework for individual programs running on the system to report whatever events are considered interesting: the sshd program may report each successful or failed login attempt, while the sendmail program may report each time it sends an e-mail on behalf of a local or remote user. An auditing infrastructure, on the other hand, reports each instance of certain low-level events, such as entry to the setuid system call, regardless of which program caused the event to occur. Auditing has the advantage of being more comprehensive, but the disadvantage of reporting a large amount of information, most of which is uninteresting. Logging (particularly using a standard framework like syslog) has the advantage of being compatible with a wide variety of client applications, and of reporting only information considered important by each application, but the disadvantage that the information reported is not consistent between applications. A robust infrastructure will perform both logging and auditing, and will use configurable automated methods of summarizing the reported data, so that system administrators can remove or compress reports of events known to be uninteresting in favor of alert monitoring for events known to be interesting. This section discusses how to configure logging, log monitoring, and auditing, using tools included with RHEL5. It is recommended that syslog be used for logging, with logwatch providing summarization, and that auditd be used for auditing, with aureport providing summarization.

2.6.1 - Configure Syslog

Syslog has been the default Unix logging mechanism for many years. It has a number of downsides, including inconsistent log format, lack of authentication for received messages, and lack of authentication, encryption, or reliable transport for messages sent over a network. However, due to its long history, syslog is an accepted standard which is supported by almost all Unix applications. This section discusses how to configure syslog for best effect, and how to use tools provided with the system to maintain and monitor your logs.

2.6.1.1 - Ensure All Important Messages are Captured

Edit the file /etc/syslog.conf. Add or correct whichever of the following lines are appropriate for your environment: auth,info.* /var/log/messages kern.* /var/log/kern.log daemon.* /var/log/daemon.log syslog.* /var/log/syslog lpr,news,uucp,local0,local1,local2,local3,local4,local5,local6.* /var/log/unused.log When a message is sent to syslog for logging, it is sent with a facility name (such as mail, auth, or local2), and a priority (such as debug, notice, or emerg). Each line of syslog’s configuration file is a directive which specifies a set of facility/priority pairs, and then gives a filename or host to which log messages of matching types should be sent. In order for a message to match a type, the facility must match, and the priority must be the priority named in the rule or any higher priority. (See syslog.conf(5) for an ordered list of priorities.) Older versions of syslog mandated a very restrictive format for the syslog.conf file. However, the version of syslog shipped with RHEL5 allows any sort of whitespace (spaces or tabs, not just tabs) to separate the selection criteria from the message disposition, and allows the use of facility.* as a wildcard matching a given facility at any priority. The default RHEL5 syslog configuration stores the facilities authpriv, cron, and mail in named logs. This guide describes the implementation of the following configuration, but any configuration which stores the important facilities and is usable by the administrators will suffice: * Store each of the facilities kern, daemon, and syslog in its own log, so that it will be easy to access information about messages from those facilities. * Restrict the information stored in /var/log/messages to only the facilities auth and user, and store all messages from those facilities. Messages can easily become cluttered otherwise. * Store information about all facilities which should not be in use at this site in a file called /var/log/ unused.log. If any messages are logged to this file at some future point, this may be an indication that an unknown service is running, and should be investigated. In addition, if news and uucp are not in use at this site, remove the directive from the default syslog.conf which stores those facilities. Making use of the local facilities is also recommended. Specific configuration is beyond the scope of this guide, but applications such as SSH can easily be configured to log to a local facility which is not being used for anything else. If this is done, reconfigure /etc/syslog.conf to store this facility in an appropriate named log or in /var/log/messages, rather than in /var/log/unused.log.

2.6.1.2 - Confirm Existence and Permissions of System Log Files

For each log file LOGFILE referenced in /etc/syslog.conf, run the commands: # touch LOGFILE # chown root:root LOGFILE # chmod 0600 LOGFILE Syslog will refuse to log to a file which does not exist. All messages intended for that file will be silently discarded, so it is important to verify that all log files exist. Some logs may contain sensitive information, so it is better to restrict permissions so that only administrative users can read or write logfiles.

2.6.1.3 - Send Logs to a Remote Loghost

Edit /etc/syslog.conf. Add or correct the line: *.* @loghost.example.com where loghost.example.com is the name of your central log server. If system logs are to be useful in detecting malicious activities, it is necessary to send logs to a remote server. An intruder who has compromised the root account on a machine may delete the log entries which indicate that the system was attacked before they are seen by an administrator. However, it is recommended that logs be stored on the local host in addition to being sent to the loghost, because syslog uses the UDP protocol to send messages over a network. UDP does not guarantee reliable delivery, and moderately busy sites will lose log messages occasionally, especially in periods of high traffic which may be the result of an attack. In addition, remote syslog messages are not authenticated in any way, so it is easy for an attacker to introduce spurious messages to the central log server. Also, some problems cause loss of network connectivity, which will prevent the sending of messages to the central server. For all of these reasons, it is better to store log messages both centrally and on each host, so that they can be correlated if necessary.

2.6.1.4 - Enable syslogd to Accept Remote Messages on Loghosts Only

Is this machine the central log server for your organization? If so, edit the file /etc/sysconfig/syslog. Add or correct the following line: SYSLOGD_OPTIONS="-m 0 -r -s example.com " where example.com is the name of your domain. If the machine is not a log server, edit /etc/sysconfig/syslog, and instead add or correct the line: SYSLOGD_OPTIONS="-m 0" By default, RHEL5’s syslog does not listen over the network for log messages. The -r flag enables syslogd to listen over a network, and should be used only if necessary. The -s example.com flag strips the domain name example.com from each sending machine’s hostname before logging messages from that host, to reduce the amount of redundant information placed in log files. See the syslogd(8) man page for further information.

2.6.1.5 - Ensure All Logs are Rotated by logrotate

Edit the file /etc/logrotate.d/syslog. Find the first line, which should look like this (wrapped for clarity): /var/log/messages /var/log/secure /var/log/maillog /var/log/spooler \ /var/log/boot.log /var/log/cron { Edit this line so that it contains a one-space-separated listing of each log file referenced in /etc/syslog.conf. All logs in use on a system must be rotated regularly, or the log files will consume disk space over time, eventually interfering with system operation. The file /etc/logrotate.d/syslog is the configuration file used by the logrotate program to maintain all log files written by syslog. By default, it rotates logs weekly and stores four archival copies of each log. These settings can be modified by editing /etc/logrotate.conf, but the defaults are sufficient for purposes of this guide. Note that logrotate is run nightly by the cron job /etc/cron.daily/logrotate. If particularly active logs need to be rotated more often than once a day, some other mechanism must be used.

2.6.1.6 - Monitor Suspicious Log Messages using Logwatch

The system includes an extensible program called Logwatch for reporting on unusual items in syslog. Logwatch is valuable because it provides a parser for the syslog entry format and a number of signatures for types of lines which are considered to be mundane or noteworthy. Logwatch has a number of downsides: the signatures can be inaccurate and are not always categorized consistently, and you must be able to program in Perl in order to customize the signature database. However, it is recommended that all Linux sites which do not have time to deploy a third-party log monitoring application run Logwatch in its default configuration. This provides some useful information about system activity in exchange for very little administrator effort. This guide recommends that Logwatch be run only on the central logserver, if your site has one, in order to focus administrator attention by sending all daily logs in a single e-mail.

2.6.1.6.1 - Configure Logwatch on the Central Log Server

Is this machine the central log server? If so, edit the file /etc/logwatch/conf/logwatch.conf. Add or correct the following lines: HostLimit = no SplitHosts = yes MultiEmail = no Service = -zz-disk_space Ensure that logwatch.pl is run nightly from cron. (This is the default): # cd /etc/cron.daily # ln -s /usr/share/logwatch/scripts/logwatch.pl 0logwatch On a central logserver, you want Logwatch to summarize all syslog entries, including those which did not originate on the logserver itself. The HostLimit setting tells Logwatch to report on all hosts, not just the one on which it is running. If SplitHosts is set, Logwatch will separate entries by hostname. This makes the report longer but significantly more usable. If it is not set, then Logwatch will not report which host generated a given log entry, and that information is almost always necessary. If MultiEmail is set, then each host’s information will be sent in a separate e-mail message. This is a matter of preference. The Service directive -zz-disk space tells Logwatch not to run the zz-disk space report, which reports on free disk space. Since all log monitoring is being done on the central logserver, the disk space listing will always be that of the logserver, regardless of which host is being monitored. This is confusing, so disable that service. Note that this does mean that Logwatch will not monitor disk usage information. Many workarounds are possible, such as running df on each host daily via cron and sending the output to syslog so that it will be reported to the logserver.

2.6.1.6.2 - Disable Logwatch on Clients if a Logserver Exists

Does your site have a central logserver which has been configured to report on logs received from all systems? If so: # rm /etc/cron.daily/0logwatch If no logserver exists, it will be necessary for each machine to run Logwatch individually. Using a central logserver provides the security and reliability benefits discussed earlier, and also makes monitoring logs easier and less time-intensive for administrators.

2.6.2 - System Accounting with auditd

The audit service is the current Linux recommendation for kernel-level auditing. By default, the service audits about SELinux AVC denials and certain types of security-relevant events such as system logins, account modifications, and authentication events performed by programs such as sudo. Under its default configuration, auditd has modest disk space requirements, and should not noticeably impact system performance. The audit service, in its default configuration, is strongly recommended for all sites, regardless of whether they are running SELinux. Sites with substantial auditing requirements may be able to configure auditd’s system call auditing to meet these requirements.

2.6.2.1 - Enable the auditd Service

Ensure that the auditd service is enabled (this is the default): # chkconfig auditd on By default, auditd logs only SELinux denials, which are helpful for debugging SELinux and discovering intrusion attempts, and certain types of security events, such as modifications to user accounts (useradd, passwd, etc), login events, and calls to sudo. Data is stored in /var/log/audit/audit.log. By default, auditd rotates 4 logs by size (5MB), retaining a maximum of 20MB of data in total, and refuses to write entries when the disk is too full. This minimizes the risk of audit data filling its partition and impacting other services. However, it is possible to lose audit data if the system is busy.

2.6.2.2 - Use aureport to Summarize Audit Logs

Familiarize yourself with the aureport(8) man page, then design a short series of audit reporting commands suitable for a daily summary of information important to your site. Install these commands into the system cron infrastructure by placing an appropriately named file in /etc/cron. daily. Retrieve information from the audit log using the ausearch and aureport commands. The former gives sufficiently detailed information to unambiguously find the source of SELinux problems. The latter provides summary output suitable for a daily report. For instance, if your site requires a daily report of every user to login to the machine, the following command could be run from cron: # aureport -l -i -ts yesterday -te today More details about using ausearch to find an SELinux problem can be found in Section 2.4.5.

2.6.2.3 - Configure auditd for Sites with Further Auditing Requirements

The auditd program can perform comprehensive monitoring of system calls, so it is suitable for most kernel-level auditing requirements. This section describes a few configuration settings which are relevant when performing higher-volume auditing, but a full description of how to perform such auditing is beyond the scope of this guide. The mailing list [email protected] may be a good source of further information.

2.6.2.3.1 - Increase auditd Data Retention

* Determine STOREMB , the amount of audit data (in megabytes) which should be retained in each log file. Edit the file /etc/audit/auditd.conf. Add or modify the following line: max_log_file = STOREMB * Using your site’s standard procedure, create a dedicated partition for log files. The partition should be larger than the maximum space which auditd will ever use, which is the maximum size of each log file (max log file) multiplied by the number of log files (num logs). Mount the new partition on /var/log/audit. * If your site requires that the machine be disabled when auditing cannot be performed, configure auditd to halt the system when disk space for auditing runs low. Edit /etc/audit/auditd.conf, and add or correct the following lines: space_left_action = email action_mail_acct = root admin_space_left_action = halt The default action to take when the logs reach their maximum size is to rotate the log files, discarding the oldest one. If it is more important to retain all possible auditing information, even if that opens the possibility of running out of space and taking the action defined by admin space left action, add or correct the line: max_log_file_action = keep_logs By default, auditd retains 4 log files of size 5Mb apiece. For a busy system or a system which is auditing any additional system calls, this is likely to be insufficient. The log file size needed will depend heavily on what types of events are being audited. First configure auditing to log all the events of interest. Then monitor the log size manually for awhile to determine what file size will allow you to keep the required data for the correct time period. Using a dedicated partition for /var/log/audit prevents the auditd logs from disrupting system functionality if they fill, and, more importantly, prevents other activity in /var from filling the partition and stopping the audit trail. (The audit logs are size-limited and therefore unlikely to grow without bound unless configured to do so.) Some machines may have requirements that no actions occur which cannot be audited. If this is the case, then auditd can be configured to halt the machine if it runs out of space. Note: Since older logs are rotated, configuring auditd this way does not prevent older logs from being rotated away before they can be viewed. If your system is configured to halt when logging cannot be performed, make sure this can never happen under normal circumstances! Ensure that /var/ log/ audit is on its own partition, and that this partition is larger than the maximum amount of data auditd will retain normally. 3List information can be found at http://www.redhat.com/mailman/listinfo/linux-audit

2.6.2.3.2 - Configure Rules for Comprehensive Auditing

Edit the file /etc/audit/audit.rules. Add one line for each desired audit rule. See the auditctl(8) manpage for details about the syntax of audit rules. The syntax of audit rules will not be discussed here. The audit subsystem supports extensive auditing, including: * Tracing of arbitrary system calls (identified by name or number) on entry or exit. * Filtering by PID, UID, call success, system call argument (with some limitations), etc. * Monitoring of specific files for modifications to the file’s contents or metadata. The audit documentation directory can be found at: /usr/share/doc/audit-* and contains several example rulesets for various purposes.

3 - Services

3.1 - DisableAllUnneededServicesatBootTime

The best protection against vulnerable software is running less software. This section describes how to review the software which Red Hat Enterprise Linux installs on a system and disable software which is not needed. It then enumerates the software packages installed on a default RHEL5 system and provides guidance about which ones can be safely disabled.

3.1.1 - Determine which Services are Enabled at Boot

Run the command: # chkconfig --list | grep :on The first column of this output is the name of a service which is currently enabled at boot. Review each listed service to determine whether it can be disabled. If it is appropriate to disable some service srvname , do so using the command: # chkconfig srvname off Use the guidance below for information about unfamiliar services.

3.1.2 - Guidance on Default Services

The table in this section contains a list of all services which are enabled at boot by a default RHEL5 installation. For each service, one of the following recommendations is made: * Enable: The service provides a significant capability with limited risk exposure. Leave the service enabled. * Configure: The service either is required for most systems to function properly or provides an important security function. It should be left enabled by most environments. However, it must be configured securely on all machines, and different options may be needed for workstations than for servers. See the referenced section for recommended configuration of this service. * Disable if possible: The service opens the system to some risk, but may be required by some environments. See the appropriate section of the guide, and disable the service if at all possible. * Servers only: The service provides some function to other machines over the network. If that function is needed in the target environment, the service should remain enabled only on a small number of dedicated servers, and should be disabled on all other machines on the network. Service name Action Reference acpid Enable 3.3.15.2 anacron Disable if possible 3.4 apmd Disable if possible 3.3.15.1 atd Configure 3.4 auditd Configure 2.6.2 Service name Action Reference autofs Disable if possible 2.2.2.3 avahi-daemon Disable if possible 3.7 bluetooth Disable if possible 3.3.14 cpuspeed Enable 3.3.15.3 crond Configure 3.4 cups Disable if possible 3.8 firstboot Disable if possible 3.3.1 gpm Disable if possible 3.3.2 haldaemon Disable if possible 3.3.13.2 hidd Disable if possible 3.3.14.2 hplip Disable if possible 3.8.4.1 ip6tables Configure 2.5.5 iptables Configure 2.5.5 irqbalance Enable 3.3.3 isdn Disable if possible 3.3.4 kdump Disable if possible 3.3.5 kudzu Disable if possible 3.3.6 mcstrans Disable if possible 2.4.3.2 (SELinux) mdmonitor Disable if possible 3.3.7 messagebus Disable if possible 3.3.13.1 microcode ctl Disable if possible 3.3.8 netfs Disable if possible 3.13 (NFS) network Enable 3.3.9 nfslock Disable if possible 3.13 (NFS) pcscd Disable if possible 3.3.10 portmap Disable if possible 3.13 (NFS) readahead early Disable if possible 3.3.12 readahead later Disable if possible 3.3.12 restorecond Enable 2.4.3.3 (SELinux) rhnsd Disable if possible 2.1.2.2 rpcgssd Disable if possible 3.13 (NFS) rpcidmapd Disable if possible 3.13 (NFS) sendmail Configure 3.11 setroubleshoot Disable if possible 2.4.3.1 (SELinux) smartd Enable 3.3.11 sshd Servers only 3.5 syslog Configure 2.6.1 xfs Disable if possible 3.6 (X11) yum-updatesd Disable if possible 2.1.2.3.2

3.1.3 - Guidance for Unfamiliar Services

If the system is running any services which have not been covered, determine what these services do, and disable them if they are not needed or if they pose a high risk. If a service srvname is unknown, try running: $ rpm -qf /etc/init.d/srvname to discover which RPM package installed the service. Then, run: $ rpm -qi rpmname for a brief description of what that RPM does.

3.2 - Obsolete Services

This section discusses a number of network-visible services which have historically caused problems for system security, and for which disabling or severely limiting the service has been the best available guidance for some time. As a result of this consensus, these services are not installed as part of RHEL5 by default. Organizations which are running these services should prioritize switching to more secure services which provide the needed functionality. If it is absolutely necessary to run one of these services for legacy reasons, care should be taken to restrict the service as much as possible, for instance by configuring host firewall software (see Section 2.5.5) to restrict access to the vulnerable service to only those remote hosts which have a known need to use it.

3.2.1 - Inetd and Xinetd

Is there an operational need to run the deprecated inetd or xinetd software packages? If not, ensure that they are removed from the system: # yum erase inetd xinetd Beginning with Red Hat Enterprise Linux 5, the xinetd service is no longer installed by default. This change represents increased awareness that the dedicated network listener model does not improve security or reliability of services, and that restriction of network listeners is better handled using a granular model such as SELinux than using xinetd’s limited security options.

3.2.2 - Telnet

Is there a mission-critical reason for users to access the system via the insecure telnet protocol, rather than the more secure SSH protocol? If not, ensure that the telnet server is removed from the system: # yum erase telnet-server The telnet protocol uses unencrypted network communication, which means that data from the login session, including passwords and all other information transmitted during the session, can be stolen by eavesdroppers on the network, and also that outsiders can easily hijack the session to gain authenticated access to the telnet server. Organizations which use telnet should be actively working to migrate to a more secure protocol. See Section 3.5 for information about the SSH service.

3.2.3 - Rlogin, Rsh, and Rcp

The Berkeley r-commands are legacy services which allow cleartext remote access and have an insecure trust model.

3.2.3.1 - Remove the Rsh Server Commands from the System

Is there a mission-critical reason for users to access the system via the insecure rlogin, rsh, or rcp commands rather than the more secure ssh and scp? If not, ensure that the rsh server is removed from the system: # yum erase rsh-server SSH was designed to be a drop-in replacement for the r-commands, which suffer from the same hijacking and eavesdropping problems as telnet. There is unlikely to be a case in which these commands cannot be replaced with SSH.

3.2.3.2 - Remove .rhosts Support from PAM Configuration Files

Check that pam rhosts authentication is not used by any PAM services. Run the command: # grep -l pam rhosts /etc/pam.d/* This command should return no output. The RHEL5 default is not to rely on .rhosts or /etc/hosts.equiv for any PAM-based services, so, on an uncustomized system, this command should return no output. If any files do use pam rhosts, modify them to make use of a more secure authentication method instead. For more information about PAM, see Section 2.3.3.

3.2.4 - NIS

The NIS client service ypbind is not activated by default. In the event that it was activated at some point, disable it by executing the command: # chkconfig ypbind off The NIS server package is not installed by default. In the event that it was installed at some point, remove it from the system by executing the command: # yum erase ypserv The Network Information Service (NIS), also known as “Yellow Pages” (YP), and its successor NIS+ have been made obsolete by Kerberos, LDAP, and other modern centralized authentication services. NIS should not be used because it suffers from security problems inherent in its design, such as inadequate protection of important authentication information.

3.2.5 - TFTP Server

Is there an operational need to run the deprecated TFTP server software? If not, ensure that it is removed from the system: # yum erase tftp-server TFTP is a lightweight version of the FTP protocol which has traditionally been used to configure networking equipment. However, TFTP provides little security, and modern versions of networking operating systems fre77 quently support configuration via SSH or other more secure protocols. A TFTP server should be run only if no more secure method of supporting existing equipment can be found.

3.3 - BaseServices

This section addresses the base services that are configured to start up on boot in a RHEL5 default installation. Some of these services listen on the network and should be treated with particular discretion. The other services are local system utilities that may or may not be extraneous. Each of these services should be disabled if not required.

3.3.1 - Installation Helper Service (firstboot)

Firstboot is a daemon specific to the Red Hat installation process. It handles “one-time” configuration following successful installation of the operating system. As such, there is no reason for this service to remain enabled. Disable firstboot by issuing the command: # chkconfig firstboot off

3.3.2 - Console Mouse Service (gpm)

GPM is the service that controls the text console mouse pointer. (The X Windows mouse pointer is unaffected by this service.) If mouse functionality in the console is not required, disable this service: # chkconfig gpm off Although it is preferable to run as few services as possible, the console mouse pointer can be useful for preventing administrator mistakes in runlevel 3 by enabling copy-and-paste operations.

3.3.3 - Interrupt Distribution on Multiprocessor Systems (irqbalance)

The goal of the irqbalance service is to optimize the balance between power savings and performance through distribution of hardware interrupts across multiple processors. In a server environment with multiple processors, this provides a useful service and should be left enabled. If a machine has only one processor, the service may be disabled: # chkconfig irqbalance off

3.3.4 - ISDN Support (isdn)

The ISDN service facilitates Internet connectivity in the presence of an ISDN modem. If an ISDN modem is not being used, disable this service: # chkconfig isdn off

3.3.5 - Kdump Kernel Crash Analyzer (kdump)

Kdump is a new kernel crash dump analyzer. It uses kexec to boot a secondary kernel (“capture” kernel) following a system crash. The kernel dump from the system crash is loaded into the capture kernel for analysis. Unless the system is used for kernel development or testing, disable the service: # chkconfig kdump off

3.3.6 - Kudzu Hardware Probing Utility (kudzu)

Is there a mission-critical reason for console users to add new hardware to the system? If not: # chkconfig kudzu off Kudzu, Red Hat’s hardware detection program, represents an unnecessary security risk as it allows unprivileged users to perform hardware configuration without authorization. Unless this specific functionality is required, Kudzu should be disabled.

3.3.7 - Software RAID Monitor (mdmonitor)

The mdmonitor service is used for monitoring a software RAID (hardware RAID setups do not use this service). This service is extraneous unless software RAID is in use (which is not common). If software RAID monitoring is not required, disable this service: # chkconfig mdmonitor off

3.3.8 - IA32 Microcode Utility(microcodectl)

microcode ctl is a microcode utility for use with Intel IA32 processors (Pentium Pro, PII, Celeron, PIII, Xeon, Pentium 4, etc) If the system is not running an Intel IA32 processor, disable this service: # chkconfig microcode ctl off

3.3.9 - Network Service (network)

The network service allows associated network interfaces to access the network. This section contains general guidance for controlling the operation of the service. For kernel parameters which affect networking, see Section

3.3.9.1 - Disable All Networking if Not Needed

If the system is a standalone machine with no need for network access or even communication over the loopback device, then disable this service: # chkconfig network off

3.3.9.2 - Disable All External Network Interfaces if Not Needed

If the system does not require network communications but still needs to use the loopback interface, remove all files of the form ifcfg-interface except for ifcfg-lo from /etc/sysconfig/network-scripts: # rm /etc/sysconfig/network-scripts/ifcfg-interface

3.3.9.3 - Disable Zeroconf Networking

Zeroconf networking allows the system to assign itself an IP address and engage in IP communication without a statically-assigned address or even a DHCP server. Automatic address assignment via Zeroconf (or DHCP) is not recommended. To disable Zeroconf automatic route assignment in the 169.245.0.0 subnet, add or correct the following line in /etc/sysconfig/network: NOZEROCONF=yes Zeroconf addresses are in the network 169.254.0.0. The networking scripts add entries to the system’s routing table for these addresses. Zeroconf address assignment commonly occurs when the system is configured to use DHCP but fails to receive an address assignment from the DHCP server.

3.3.10 - Smart Card Support (pcscd)

The pcscd service provides support for Smart Cards and Smart Card Readers. If Smart Cards are not in use on the system, disable this service: # chkconfig pcscd off

3.3.11 - SMART Disk Monitoring Support (smartd)

SMART (Self-Monitoring, Analysis, and Reporting Technology) is a feature of hard drives that allows them to detect symptoms of disk failure and relay an appropriate warning. This technology is considered to bring relatively low security risk, and can be useful. Leave this service running if the system’s hard drives are SMART-capable. Otherwise, disable it: # chkconfig smartd off

3.3.12 - Boot Caching (readahead early/readahead later)

The following services provide one-time caching of files belonging to some boot services, with the goal of allowing the system to boot faster. It is recommended that this service be disabled on most machines: # chkconfig readahead early off # chkconfig readahead later off The readahead services do not substantially increase a system’s risk exposure, but they also do not provide great benefit. Unless the system is running a specialized application for which the file caching substantially improves system boot time, this guide recommends disabling the services.

3.3.13 - Application Support Services

The following services are software projects of freedesktop.org that are meant to provide system integration through a series of common APIs for applications. They are heavily integrated into the X Windows environment. If the system is not using X Windows, these services can typically be disabled.

3.3.13.1 - D-Bus IPC Service (messagebus)

D-Bus is an IPC mechanism that provides a common channel for inter-process communication. If no services which require D-Bus are in use, disable this service: # chkconfig messagebus off A number of default services make use of D-Bus, including X Windows (Section 3.6), Bluetooth (Section 3.3.14) and Avahi (Section 3.7). This guide recommends that D-Bus and all its dependencies be disabled unless there is a mission-critical need for them. Stricter configuration of D-Bus is possible and documented in the man page dbus-daemon(1). D-Bus maintains two separate configuration files, located in /etc/dbus-1/, one for system-specific configuration and the other for session-specific configuration.

3.3.13.2 - HAL Daemon (haldaemon)

The haldaemon service provides a dynamic way of managing device interfaces. It automates device configuration and provides an API for making devices accessible to applications through the D-Bus interface.

3.3.13.2.1 - Disable HAL Daemon if Possible

HAL provides valuable attack surfaces to attackers as an intermediary to privileged operations and should be disabled unless necessary: # chkconfig haldaemon off

3.3.13.2.2 - Configure HAL Daemon if Necessary

HAL provides a limited user the ability to mount system devices. This is primarily used by X utilities such as gnome-volume-manager to perform automounting of removable media. HAL configuration is currently only possible through a series of fdi files located in /usr/share/hal/fdi/ Note: The HAL future road map includes a mandatory framework for managing administrative privileges called PolicyKit. To prevent users from accessing devices through HAL, create the file /etc/hal/fdi/policy/99-policy-all-drives.fdi with the contents: <?xml version="1.0" encoding="UTF-8"?> <deviceinfo version="0.2"> <device> <match key="info.capabilities" contains="volume"> <merge key="volume.ignore" type="bool">true</merge> </match> </device> </deviceinfo> The above code matches any device labeled with the volume capability (any device capable of being mounted will be labeled this way) and sets the corresponding volume.ignore key to true, indicating that the volume should be ignored. This both makes the volume invisible to the UI, and denies mount attempts by unprivileged users.

3.3.14 - Bluetooth Support

Bluetooth provides a way to transfer information between devices such as mobile phones, laptops, PCs, printers, digital cameras, and video game consoles over a short-range wireless link. Any wireless communication presents a serious security risk to sensitive or classified systems. Section 2.5.2 contains information on the related topic of wireless networking. Removal of hardware is the only way to ensure that the Bluetooth wireless capability remains disabled. If it is completely impractical to remove the Bluetooth hardware module, and site policy still allows the device to enter sensitive spaces, every effort to disable the capability via software should be made. In general, acquisition policy should include provisions to prevent the purchase of equipment that will be used in sensitive spaces and includes Bluetooth capabilities.

3.3.14.1 - Bluetooth Host Controller Interface Daemon (bluetooth)

The bluetooth service enables the system to use Bluetooth devices. If the system requires no Bluetooth devices, disable this service: # chkconfig bluetooth off

3.3.14.2 - Bluetooth Input Devices (hidd)

The hidd service provides support for Bluetooth input devices. If the system has no Bluetooth input devices (e.g. keyboard or mouse), disable this service: # chkconfig hidd off

3.3.14.3 - Disable Bluetooth Kernel Modules

The kernel’s module loading system can be configured to prevent loading of the Bluetooth module. Add the following to /etc/modprobe.conf to prevent the loading of the Bluetooth module: alias net-pf-31 off The unexpected name, net-pf-31, is a result of how the kernel requests modules for network protocol families; it is an alias for the bluetooth module.

3.3.15 - Power Management Support

The following services provide an interface to power management functions. These functions include monitoring battery power, system hibernate/suspend, CPU throttling, and various power-save utilities.

3.3.15.1 - Advanced Power Management Subsystem (apmd)

The apmd service provides last generation power management support. If the system is capable of ACPI support, or if power management is not necessary, disable this service: # chkconfig apmd off APM is being replaced by ACPI and should be considered deprecated. As such, it can be disabled if ACPI is supported by your hardware and kernel. If the file /proc/acpi/info exists and contains ACPI version information, then APM can safely be disabled without loss of functionality.

3.3.15.2 - Advanced Configuration and Power Interface (acpid)

The acpid service provides next generation power management support. Unless power management features are not necessary, leave this service enabled.

3.3.15.3 - CPU Throttling (cpuspeed)

The cpuspeed service uses hardware support to throttle the CPU when the system is idle. Unless CPU power optimization is unnecessary, leave this service enabled.

3.4 - Cron and At Daemons

The cron and at services are used to allow commands to be executed at a later time. The cron service is required by almost all systems to perform necessary maintenance tasks, while at may or may not be required on a given system. Both daemons should be configured defensively.

3.4.1 - Disable anacron if Possible

Is this a machine which is designed to run all the time, such as a server or a workstation which is left on at night? If so: # yum erase anacron The anacron subsystem is designed to provide cron functionality for machines which may be shut down during the normal times that system cron jobs run, frequently in the middle of the night. Laptops and workstations which are shut down at night should keep anacron enabled, so that standard system cron jobs will run when the machine boots. However, on machines which do not need this additional functionality, anacron represents another piece of privileged software which could contain vulnerabilities. Therefore, it should be removed when possible to reduce system risk.

3.4.2 - Restrict Permissions on Files Used by cron

1. Restrict the permissions on the primary system crontab file: # chown root:root /etc/crontab # chmod 600 /etc/crontab 2. If anacron has not been removed, restrict the permissions on its primary configuration file: # chown root:root /etc/anacrontab # chmod 600 /etc/anacrontab 3. Restrict the permission on all system crontab directories: # cd /etc # chown -R root:root cron.hourly cron.daily cron.weekly cron.monthly cron.d # chmod -R go-rwx cron.hourly cron.daily cron.weekly cron.monthly cron.d 4. Restrict the permissions on the spool directory for user crontab files: # chown root:root /var/spool/cron # chmod -R go-rwx /var/spool/cron Cron and anacron make use of a number of configuration files and directories. The system crontabs need only be edited by root, and user crontabs are edited using the setuid root crontab command. If unprivileged users can modify system cron configuration files, they may be able to gain elevated privileges, so all unnecessary access to these files should be disabled.

3.4.3 - Restrict at and cron to Authorized Users

1. Remove the cron.deny file: # rm /etc/cron.deny 2. Edit /etc/cron.allow, adding one line for each user allowed to use the crontab command to create cron jobs. 3. Remove the at.deny file: # rm /etc/at.deny 4. Edit /etc/at.allow, adding one line for each user allowed to use the at command to create at jobs. The /etc/cron.allow and /etc/at.allow files contain lists of users who are allowed to use cron and at to delay execution of processes. If these files exist and if the corresponding files /etc/cron.deny and /etc/at.deny do not exist, then only users listed in the relevant allow files can run the crontab and at commands to submit jobs to be run at scheduled intervals. On many systems, only the system administrator needs the ability to schedule jobs. Note that even if a given user is not listed in cron.allow, cron jobs can still be run as that user. The cron.allow file controls only administrative access to the crontab command for scheduling and modifying cron jobs.

3.5 - SSH Server

The SSH protocol is recommended for remote login and remote file transfer. SSH provides confidentiality and integrity for data exchanged between two systems, as well as server authentication, through the use of public key cryptography. The implementation included with the system is called OpenSSH, and more detailed documentation is available from its website, http://www.openssh.org. Its server program is called sshd and provided by the RPM package openssh-server.

3.5.1 - Disable OpenSSH Server if Possible

Unless the system needs to provide the remote login and file transfer capabilities of SSH, disable and remove the OpenSSH server and its configuration.

3.5.1.1 - Disable and Remove OpenSSH Software

Disable and remove openssh-server with the commands: # chkconfig sshd off # yum erase openssh-server Users of the system will still be able to use the SSH client program /usr/bin/ssh to access SSH servers on other systems.

3.5.1.2 - Remove SSH Server iptables Firewall Exception

Edit the files /etc/sysconfig/iptables and /etc/sysconfig/ip6tables (if IPv6 is in use). In each file, locate and delete the line: -A RH-Firewall-1-INPUT -m state --state NEW -m tcp -p tcp --dport 22 -j ACCEPT By default, inbound connections to SSH’s port are allowed. If the SSH server is not being used, this exception should be removed from the firewall configuration. See Section 2.5.5 for more information about Iptables.

3.5.2 - Configure OpenSSH Server if Necessary

If the system needs to act as an SSH server, then certain changes should be made to the OpenSSH daemon configuration file /etc/ssh/sshd config. The following recommendations can be applied to this file. See the sshd config(5) man page for more detailed information.

3.5.2.1 - Ensure Only Protocol 2 Connections Allowed

Only SSH protocol version 2 connections should be permitted. Version 1 of the protocol contains security vulnerabilities. The default setting shipped in the configuration file is correct, but it is important enough to check. Verify that the following line appears: Protocol 2

3.5.2.2 - Limit Users SSH Access'

By default, the SSH configuration allows any user to access the system. In order to allow all users to login via SSH but deny only a few users, add or correct the following line: DenyUsers USER1 USER2 Alternatively, if it is appropriate to allow only a few users access to the system via SSH, add or correct the following line: AllowUsers USER1 USER2

3.5.2.3 - Set Idle Timeout Interval for User Logins

SSH allows administrators to set an idle timeout interval. After this interval has passed, the idle user will be automatically logged out. Find and edit the following lines in /etc/ssh/sshd config as follows: ClientAliveInterval interval ClientAliveCountMax 0 The timeout interval is given in seconds. To have a timeout of 5 minutes, set interval to 300. If a shorter timeout has already been set for the login shell, as in Section 2.3.5.5, that value will prempt any SSH setting made here. Keep in mind that some processes may stop SSH from correctly detecting that the user is idle.

3.5.2.4 - Disable .rhosts Files

SSH can emulate the behavior of the obsolete rsh command in allowing users to enable insecure access to their accounts via .rhosts files. To ensure that this behavior is disabled, add or correct the following line: IgnoreRhosts yes

3.5.2.5 - Disable Host-Based Authentication

SSH’s cryptographic host-based authentication is slightly more secure than .rhosts authentication, since hosts are cryptographically authenticated. However, it is not recommended that hosts unilaterally trust one another, even within an organization. To disable host-based authentication, add or correct the following line: HostbasedAuthentication no

3.5.2.6 - Disable root Login via SSH

The root user should never be allowed to login directly over a network, as this both reduces auditable information about who ran privileged commands on the system and allows direct attack attempts on root’s password. To disable root login via SSH, add or correct the following line: PermitRootLogin no

3.5.2.7 - Disable Empty Passwords

To explicitly disallow remote login from accounts with empty passwords, add or correct the following line: PermitEmptyPasswords no Measures should also be taken to disable accounts with empty passwords system-wide, as described in Section 2.3.1.5.

3.5.2.8 - Enable a Warning Banner

Section 2.3.7 contains information on how to create an appropriate warning banner. To enable a warning banner, add or correct the following line: Banner /etc/issue

3.5.2.9 - Strengthen Firewall Configuration if Possible

If the SSH server must only accept connections from the local network, then strengthen the default firewall rule for the SSH service. Determine an appropriate network block, netwk, and network mask, mask, representing the machines on your network which must be allowed to access this SSH server. Edit the files /etc/sysconfig/iptables and /etc/sysconfig/ip6tables (if IPv6 is in use). In each file, locate the line: -A RH-Firewall-1-INPUT -m state --state NEW -m tcp -p tcp --dport 22 -j ACCEPT and replace it with: -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p tcp --dport 22 -j ACCEPT If your site uses IPv6, and you are editing ip6tables, use the line: -A RH-Firewall-1-INPUT -s ipv6netwk::/ipv6mask -m tcp -p tcp --dport 22 -j ACCEPT instead because Netfilter does not yet reliably support stateful filtering for IPv6. See Section 2.5.5 for more information about Iptables configuration.

3.6 - X Window System

The X Window System implementation included with the system is called X.org.

3.6.1 - Disable X Windows if Possible

Unless there is a mission-critical reason for the machine to run a GUI login screen, prevent X from starting automatically at boot. There is usually no reason to run X Windows on a dedicated server machine, since administrators can login via SSH or on the text console.

3.6.1.1 - Disable X Windows at System Boot

Edit the file /etc/inittab, and correct the line id:5:initdefault: to: id:3:initdefault: This action changes the default boot runlevel of the system from 5 to 3. These two runlevels should be identical except that runlevel 5 starts X on boot, while runlevel 3 does not.

3.6.1.2 - Remove X Windows from the System if Possible

Remove the X11 RPMs from the system: # yum groupremove "X Window System" As long as X.org remains installed on the system, users can still run X Windows by typing startx at the shell prompt. This may run X Windows using configuration settings which are less secure than the system defaults. Therefore, if the machine is a dedicated server which does not need to provide graphical logins at all, it is safest to remove the X.org software entirely. The command given here will remove over 100 packages. It should safely and effectively remove X from machines which do not need it.

3.6.1.3 - Lock Down X Windows startx Configuration if Necessary

If X is not to be started at boot time but the software must remain installed, users will be able to run X manually using the startx command. In some cases, this runs X with a configuration which is less safe than the default. Follow these instructions to mitigate risk from this configuration.

3.6.1.3.1 - Disable X Font Server

Disable the xfs helper service: # chkconfig xfs off The system’s X.org requires the X Font Server service (xfs) to function. The xfs service will be started automatically if X.org is activated via startx. Therefore, it is safe to prevent xfs from starting at boot when X is disabled, even if users are allowed to run X manually.

3.6.1.3.2 - Disable X Window System Listening

To prevent X.org from listening for remote connections, create the file /etc/X11/xinit/xserverrc and fill it with the following line: exec X :0 -nolisten tcp $@ One of X.org’s features is the ability to provide remote graphical display. This feature should be disabled unless it is required. If the system uses runlevel 5, which is the default, the GDM display manager starts X safely, with remote listening disabled. However, if X is started from the command line with the startx command, then the server will listen for new connections on X’s default port, 6000. See the xinit(1), startx(1), and Xserver(1) man pages for more information.

3.6.2 - Configure X Windows if Necessary

If there is a mission-critical reason for this machine to run a GUI, improve the security of the default X configuration by following the guidance in this section.

3.6.2.1 - Create Warning Banners for GUI Login Users

Edit the file /etc/gdm/custom.conf. Locate the [greeter] section, and correct that section to contain the lines: [greeter] InfoMsgFile=/etc/issue See Section 2.3.7 for an explanation of banner file use. This setting will cause the system greeting banner to be displayed in a box prior to GUI login. If the default banner font is inappropriate, it can be changed by specifying the InfoMsgFont directive as well, for instance: InfoMsgFont=Sans 12

3.7 - Avahi Server

The Avahi daemon implements the DNS Service Discovery and Multicast DNS protocols, which provide service and host discovery on a network. It allows a system to automatically identify resources on the network, such as printers or web servers. This capability is also known as mDNSresponder and is a major part of Zeroconf networking. By default, it is enabled.

3.7.1 - Disable Avahi Server if Possible

Because the Avahi daemon service keeps an open network port, it is subject to network attacks. Disabling it is particularly important to reduce the system’s vulnerability to such attacks.

3.7.1.1 - Disable Avahi Server Software

Issue the command: # chkconfig avahi-daemon off

3.7.1.2 - Remove Avahi Server iptables Firewall Exception

Edit the files /etc/sysconfig/iptables and /etc/sysconfig/ip6tables (if IPv6 is in use). In each file, locate and delete the line: -A RH-Firewall-1-INPUT -p udp --dport 5353 -d 224.0.0.251 -j ACCEPT By default, inbound connections to Avahi’s port are allowed. If the Avahi server is not being used, this exception should be removed from the firewall configuration. See Section 2.5.5 for more information about the Iptables firewall.

3.7.2 - Configure Avahi if Necessary

If your system requires the Avahi daemon, its configuration can be restricted to improve security. The Avahi daemon configuration file is /etc/avahi/avahi-daemon.conf. The following security recommendations should be applied to this file. See the avahi-daemon.conf(5) man page or documentation at http://www.avahi.org for more detailed information about the configuration options.

3.7.2.1 - Serve Only via Required Protocol

The default setting in the configuration file allows Avahi to use both IPv4 and IPv6 sockets. If you are using only IPv4, edit /etc/avahi/avahi-daemon.conf and ensure the following line exists in the [server] section: use-ipv6=no Similarly, if you are using only IPv6, disable IPv4 sockets with the line: use-ipv4=no

3.7.2.2 - Check Responses TTL Field '

Avahi can be set to ignore IP packets unless their TTL field is 255. To make Avahi ignore packets unless the TTL field is 255, edit /etc/avahi/avahi-daemon.conf and ensure the following line appears in the [server] section: check-response-ttl=yes This helps to ensure that only mDNS responses from the local network are processed, because the TTL field in a packet is decremented from its initial value of 255 whenever it is routed from one network to another. Although a properly-configured router or firewall should not allow mDNS packets into the local network at all, this option provides another check to ensure they are not trusted.

3.7.2.3 - Prevent Other Programs from Using Avahis Port '

Avahi can stop other mDNS stacks from running on the host by preventing other processes from binding to port 5353. To prevent other mDNS stacks from running, edit /etc/avahi/avahi-daemon.conf and ensure the following line appears in the [server] section: disallow-other-stacks=yes This is designed to help ensure that only Avahi is responsible for mDNS traffic coming from that port on the system.

3.7.2.4 - Disable Publishing if Possible

The default setting in the configuration file allows the avahi-daemon to send information about the local host, such as its address records and the services it offers, to the local network. To stop sending this information but still allow Avahi to query the network for services, ensure the configuration file includes the following line in the [publish] section: disable-publishing=yes This line may be particularly useful if Avahi is needed for printer discovery, but not to advertise services. This configuration is highly recommended for client systems that should not advertise their services (or existence).

3.7.2.5 - Restrict Published Information

If it is necessary to publish some information to the network, it should not be joined by any extraneous information, or by information supplied by a non-trusted source on the system. Prevent user applications from using Avahi to publish services by adding or correcting the following line in the [publish] section: disable-user-service-publishing=yes Implement as many of the following lines as possible, to restrict the information published by Avahi: publish-addresses=no publish-hinfo=no publish-workstation=no publish-domain=no Inspect the files in the directory /etc/avahi/services/. Unless there is an operational need to publish information about each of these services, delete the corresponding file. These options should be used even if publishing is disabled entirely via disable-publishing, since that option prevents publishing attempts from succeeding, while these options prevent the attempts from being made in the first place. Using both approaches is recommended for completeness.

3.8 - Print Support

The Common Unix Printing System (CUPS) service provides both local and network printing support. A system running the CUPS service can accept print jobs from other systems, process them, and send them to the appropriate printer. It also provides an interface for remote administration through a web browser. The CUPS service is installed and activated by default. The project homepage and more detailed documentation are available at http://www.cups.org. The HP Linux Imaging and Printing service (HPLIP) is a separate package that provides support for some of the additional features that HP printers provide that CUPS may not necessarily support. It relies upon the CUPS service.

3.8.1 - Disable the CUPS Service if Possible

Do you need the ability to print from this machine or to allow others to print to it? If not: # chkconfig cups off

3.8.2 - Disable Firewall Access to Printing Service if Possible

Does this system need to operate as a network print server? If not, edit the files /etc/sysconfig/iptables and /etc/sysconfig/ip6tables (if IPv6 is in use). In each file, locate and delete the lines: -A RH-Firewall-1-INPUT -p udp -m udp --dport 631 -j ACCEPT -A RH-Firewall-1-INPUT -p tcp -m tcp --dport 631 -j ACCEPT By default, inbound connections to the Internet Printing Protocol port are allowed. If the print server does not need to be accessed, either because the machine is not running the print service at all or because the machine is not providing a remote network printer to other machines, this exception should be removed from the firewall configuration. See Section 2.5.5 for more information about the Iptables firewall.

3.8.3 - Configure the CUPS Service if Necessary

CUPS provides the ability to easily share local printers with other machines over the network. It does this by allowing machines to share lists of available printers. Additionally, each machine that runs the CUPS service can potentially act as a print server. Whenever possible, the printer sharing and print server capabilities of CUPS should be limited or disabled. The following recommendations should demonstrate how to do just that.

3.8.3.1 - Limit Printer Browsing

By default, CUPS listens on the network for printer list broadcasts on UDP port 631. This functionality is called printer browsing.

3.8.3.1.1 - Disable Printer Browsing Entirely if Possible

To disable printer browsing entirely, edit the CUPS configuration file, located at /etc/cups/cupsd.conf: Browsing Off BrowseAllow none The CUPS print service can be configured to broadcast a list of available printers to the network. Other machines on the network, also running the CUPS print service, can be configured to listen to these broadcasts and add and configure these printers for immediate use. By disabling this browsing capability, the machine will no longer generate or receive such broadcasts.

3.8.3.1.2 - Limit Printer Browsing to a Particular Subnet if Necessary

It is possible to disable outgoing printer list broadcasts without affecting incoming broadcasts from other machines. To do so, open the CUPS configuration file, located at /etc/cups/cupsd.conf. Look for the line that begins with BrowseAddress and remove it. The line will look like the following: BrowseAddress @LOCAL If the intent is not to block printer sharing, but to limit it to a particular set of machines, you can limit the UDP printer broadcasts to trusted network addresses. BrowseAddress ip-address :631 Likewise, to ignore incoming UDP printer list broadcasts, or to limit the set of machines to listen to, use the BrowseAllow and BrowseDeny directives. BrowseDeny all BrowseAllow ip-address This combination will deny incoming broadcasts from any machine except those that are explicitly allowed with BrowseAllow. By default, when printer sharing is enabled, CUPS will broadcast to every network that its host machine is connected to through all available network interfaces on port 631. It will also listen to incoming broadcasts from other machines on the network. Either list one BrowseAddress line for each client machine and one BrowseAllow line for each print server or use one of the supported shorthand notations that the CUPS service recognizes. Please see the cupsd.conf(5) man page or the documentation provided at http://www.cups.org for more information on other ways to format these directives.

3.8.3.2 - Disable Print Server Capabilities if Possible

Disabling the print server capabilities in this manner will also disable the Web Administration interface. To prevent remote users from potentially connecting to and using locally configured printers, disable the CUPS print server sharing capabilities. To do so, limit how the server will listen for print jobs by removing the more generic port directive from /etc/cups/cupsd.conf: Port 631 and replacing it with the Listen directive: Listen localhost:631 This will prevent remote users from printing to locally configured printers while still allowing local users on the machine to print normally. By default, locally configured printers will not be shared over the network, but if this functionality has somehow been enabled, these recommendations will disable it again. Be sure to disable outgoing printer list broadcasts, or remote users will still be able to see the locally configured printers, even if they cannot actually print to them. To limit print serving to a particular set of users, use the Policy directive.

3.8.3.3 - Limit Access to the Web Administration Interface

By default, access to the CUPS web administration interface is limited to the local machine. It is recommended that this not be changed, especially since the authentication mechanisms that CUPS provides are limited in their effectiveness. If it is absolutely necessary to allow remote users to administer locally installed printers, be sure to limit that access as much as possible by taking advantage of the Location and Policy directive blocks. For example, to enable remote access for ip-address for user username, modify each of the Location and Policy directive blocks as follows: <Location /> AuthType Basic Require user username Order allow,deny Allow localhost Allow ip-address </Location> As with the BrowseAllow directive, use one Allow directive for each machine that needs access or use one of the available CUPS directive definition shortcuts to enable access from a class of machines at once. The Require user directive can take a list of individual users, a group of users (prefixed with @), or the shorthand valid-user. Host-based authentication has known limitations, especially since IP addresses are easy to spoof. Requiring users to authenticate themselves can alleviate this problem, but it cannot eliminate it. Do not use the root account to manage and administer printers. Create a separate account for this purpose and limit access to valid users with Require valid-user or Require user printeradmin .

3.8.3.4 - Take Further Security Measures When Appropriate

Whenever possible, limit outside networks’ access to port 631. Consider using CUPS directives that limit the number of incoming clients, such as MaxClients or MaxClientsPerHost. Additionally, there are a series of Policy and Location directives intended to limit which users can perform different printing tasks. When used together, these may help to mitigate the possibility of a denial of service attack. See cupsd.conf(5) for a full list of possible directives.

3.8.4 - The HP Linux Imaging and Printing (HPLIP) Toolkit

The HPLIP package is an HP printing support utility that is installed and enabled in a default installation. The HPLIP package is comprised of two separate components. The first is the main HPLIP service and the second is a smaller subcomponent called HPIJS. HPLIP is a feature-oriented network service that provides higher level printing support (such as bi-directional I/O, scanning, photo card, and toolbox functionality). HPIJS is a lower level basic printing driver that provides basic support for non-PostScript HP printers.

3.8.4.1 - Disable HPLIP Service if Possible

Since the HPIJS driver will still function without the added HPLIP service, HPLIP should be disabled unless the specific higher level functions that HPLIP provides are needed by a non-PostScript HP printer on the system. # chkconfig hplip off Note: If installing the HPLIP package from scratch, it should be noted that HPIJS can be installed directly without HPLIP. Please see the FAQ at the HPLIP web site at http://hplip.sourceforge.net/faqs.html for more information on how to do this.

3.9 - DHCP

The Dynamic Host Configuration Protocol (DHCP) allows systems to request and obtain an IP address and many other parameters from a server. In general, sites use DHCP either to allow a large pool of mobile or unknown machines to share a limited number of IP addresses, or to standardize installations by avoiding static, individual IP address configuration on hosts. It is recommended that sites avoid DHCP as much as possible. Since DHCP authentication is not well-supported, DHCP clients are open to attacks from rogue DHCP servers. Such servers can give clients incorrect information (e.g. malicious DNS server addresses) which could lead to their compromise. If a machine must act as a DHCP client or server, configure it defensively using the guidance in this section. This guide recommends configuring networking on clients by manually editing the appropriate files under /etc/sysconfig. It is also possible to use the graphical front-end programs system-config-network and system-config-network-tui, but these programs rewrite configuration files from scratch based on their defaults – destroying any manual changes – and should therefore be used with caution.

3.9.1 - Disable DHCP Client if Possible

For each interface IFACE on the system (e.g. eth0), edit /etc/sysconfig/network-scripts/ifcfg-IFACE and make the following changes: 1. Correct the BOOTPROTO line to read: BOOTPROTO=static 2. Add or correct the following lines, substituting the appropriate values based on your site’s addressing scheme: NETMASK=255.255.255.0 IPADDR=192.168.1.2 GATEWAY=192.168.1.1 DHCP is the default network configuration method provided by the system installer, so it may be enabled on many systems.

3.9.2 - Configure DHCP Client if necessary

If DHCP must be used, then certain configuration changes can minimize the amount of information it receives and applies from the network, and thus the amount of incorrect information a rogue DHCP server could successfully distribute. For more information on configuring dhclient, see the dhclient(8) and dhclient.conf(5) man pages.

3.9.2.1 - Minimize the DHCP-Configured Options

Create the file /etc/dhclient.conf, and add an appropriate setting for each of the ten configuration settings which can be obtained via DHCP. For each setting, setting , do one of the following: * If the setting should not be configured remotely by the DHCP server, select an appropriate static value, and add the line: supersede setting value ; * If the setting should be configured remotely by the DHCP server, add the lines: request setting ; require setting ; For example, suppose the DHCP server should provide only the IP address itself and the subnet mask. Then the entire file should look like: supersede domain-name "example.com "; supersede domain-name-servers 192.168.1.2 ; supersede nis-domain ""; supersede nis-servers ""; supersede ntp-servers "ntp.example.com "; supersede routers 192.168.1.1 ; supersede time-offset -18000 ; request subnet-mask; require subnet-mask; By default, the DHCP client program, dhclient, requests and applies ten configuration options (in addition to the IP address) from the DHCP server: subnet-mask, broadcast-address, time-offset, routers, domain-name, domain-name-servers, host-name, nis-domain, nis-servers, and ntp-servers. Many of the options requested and applied by dhclient may be the same for every system on a network. It is recommended that almost all configuration options be assigned statically, and only options which must vary on a host-by-host basis be assigned via DHCP. This limits the damage which can be done by a rogue DHCP server. If appropriate for your site, it is also possible to supersede the host-name directive in /etc/dhclient.conf, establishing a static hostname for the machine. However, dhclient does not use the host name option provided by the DHCP server (instead using the value provided by a reverse DNS lookup). Note: In this example, the options nis-servers and nis-domain are set to empty strings, on the assumption that the deprecated NIS protocol is not in use. (See Section 3.2.4.) It is necessary to supersede settings for unused services so that they cannot be set by a hostile DHCP server. If an option is set to an empty string, dhclient will typically not attempt to configure the service.

3.9.3 - Disable DHCP Server if possible

If the dhcp package has been installed on a machine which does not need to operate as a DHCP server, disable the daemon: # chkconfig dhcpd off If possible, remove the software as well: # yum erase dhcp The DHCP server dhcpd is not installed or activated by default. If the software was installed and activated, but the system does not need to act as a DHCP server, it should be disabled and removed. Unmanaged DHCP servers will provide faulty information to clients, interfering with the operation of a legitimate site DHCP server if there is one, or causing misconfigured machines to exhibit unpredictable behavior if there is not.

3.9.4 - Configure the DHCP Server if necessary

If the system must act as a DHCP server, the configuration information it serves should be minimized. Also, support for other protocols and DNS-updating schemes should be explicitly disabled unless needed. The configuration file for dhcpd is called /etc/dhcpd.conf. The file begins with a number of global configuration options. The remainder of the file is divided into sections, one for each block of addresses offered by dhcpd, each of which contains configuration options specific to that address block.

3.9.4.1 - Do Not Use Dynamic DNS

To prevent the DHCP server from receiving DNS information from clients, edit /etc/dhcpd.conf, and add or correct the following global option: ddns-update-style none; The Dynamic DNS protocol is used to remotely update the data served by a DNS server. DHCP servers can use Dynamic DNS to publish information about their clients. This setup carries security risks, and its use is not recommended. If Dynamic DNS must be used despite the risks it poses, it is critical that Dynamic DNS transactions be protected using TSIG or some other cryptographic authentication mechanism. See Section 3.14 for more information about DNS servers, including further information about TSIG and Dynamic DNS. Also see dhcpd.conf(5) for more information about protecting the DHCP server from passing along malicious DNS data from its clients. Note: The ddns-update-style option controls only whether the DHCP server will attempt to act as a Dynamic DNS client. As long as the DNS server itself is correctly configured to reject DDNS attempts, an incorrect ddns-update-style setting on the client is harmless (but should be fixed as a best practice).

3.9.4.2 - Deny Decline Messages

Edit /etc/dhcpd.conf and add or correct the following global option to prevent the DHCP server from responding the DHCPDECLINE messages, if possible: deny declines; The DHCPDECLINE message can be sent by a DHCP client to indicate that it does not consider the lease offered by the server to be valid. By issuing many DHCPDECLINE messages, a malicious client can exhaust the DHCP server’s pool of IP addresses, causing the DHCP server to forget old address allocations.

3.9.4.3 - Deny BOOTP Queries

Unless your network needs to support older BOOTP clients, disable support for the bootp protocol by adding or correcting the global option: deny bootp; The bootp option tells dhcpd to respond to BOOTP queries. If support for this simpler protocol is not needed, it should be disabled to remove attack vectors against the DHCP server.

3.9.4.4 - Minimize Served Information

Edit /etc/dhcpd.conf. Examine each address range section within the file, and ensure that the following options are not defined unless there is an operational need to provide this information via DHCP: option domain-name option domain-name-servers option nis-domain option nis-servers option ntp-servers option routers option time-offset Because the configuration information provided by the DHCP server could be maliciously provided to clients by a rogue DHCP server, the amount of information provided via DHCP should be minimized. Remove these definitions from the DHCP server configuration to ensure that legitimate clients do not unnecessarily rely on DHCP for this information. Note: By default, the RHEL5 client installation uses DHCP to request much of the above information from the DHCP server. In particular, domain-name, domain-name-servers, and routers are configured via DHCP. These settings are typically necessary for proper network functionality, but are also usually static across machines at a given site. See Section 3.9.2.1 for a description of how to configure static site information within the DHCP client configuration.

3.9.4.5 - Configure Logging

Ensure that the following line exists in /etc/syslog.conf: daemon.* /var/log/daemon.log Configure logwatch or other log monitoring tools to summarize error conditions reported by the dhcpd process. By default, dhcpd logs notices to the daemon facility. Sending all daemon messages to a dedicated log file is part of the syslog configuration outlined in Section 2.6.1.1.

3.9.4.6 - Further Resources

* The man pages dhcpd.conf(5) and dhcpd(8) * ISC web page http://isc.org/products/DHCP

3.10 - Network Time Protocol

The Network Time Protocol is used to manage the system clock over a network. Computer clocks are not very accurate, so time will drift unpredictably on unmanaged systems. Central time protocols can be used both to ensure that time is consistent among a network of machines, and that their time is consistent with the outside world. Local time synchronization is recommended for all networks. If every machine on your network reliably reports the same time as every other machine, then it is much easier to correlate log messages in case of an attack. In addition, a number of cryptographic protocols (such as Kerberos) use timestamps to prevent certain types of attacks. If your network does not have synchronized time, these protocols may be unreliable or even unusable. Depending on the specifics of the network, global time accuracy may be just as important as local synchronization, or not very important at all. If your network is connected to the Internet, it is recommended that you make use of a public timeserver, since globally accurate timestamps may be necessary if you need to investigate or respond to an attack which originated outside of your network. Whether or not you use an outside timeserver, configure the network to have a small number of machines operating as NTP servers, and the remainder obtaining time information from those internal servers.

3.10.1 - Select NTP Software

The Network Time Protocol (RFC 1305) is designed to synchronize time with a very high degree of accuracy even on an unreliable network. NTP is therefore a complex protocol. The Simple Network Time Protocol (RFC 4330) implements a subset of NTP which is intended to be good enough to meet the time requirements of most networks. The primary implementation of NTP comes from ntp.org, and is shipped with RHEL5 as the ntp RPM. An alternative is OpenNTPD, which is an implementation of SNTP, and which can be obtained as source code from http://www.openntpd.org. OpenNTPD may be simpler to configure than the reference NTP implementation, at the cost of the need to install and maintain third-party software. This guide does not recommend the use of a particular NTP/SNTP software package, but does recommend that some NTP software be selected and installed on all machines. The remainder of this section describes how to securely configure NTP clients and servers, and discusses both the reference NTP implementation and OpenNTPD.

3.10.2 - Configure Reference NTP if Appropriate

The ntp RPM implements the reference NTP server.

3.10.2.1 - Configure an NTP Client

There are a number of options for configuring clients to work with the reference NTP server. It is possible to run the ntp daemon on each host, configuring clients so that the ntp protocol ignores all network access. This introduces an additional network listener on client machines, and is therefore not recommended. This guide recommends running a network type synchronization program periodically using cron. This guide describes the syntax for the ntpdate program, but it is also possible to run ntpd itself with the -q flag, if additional time accuracy is desired at the expense of running more complex code on each machine. Alternately, even if the server is running the reference NTP implementation, it is possible for clients to access it using SNTP. See Section 3.10.3.2 for information about configuring SNTP clients.

3.10.2.1.1 - Run ntpdate using Cron

Create a file /etc/cron.d/ntpdate containing the following crontab: 15 * * * * root /usr/sbin/ntpdate ntp-server where ntp-server is the hostname or IP address of the site NTP server. This crontab will use ntpdate to synchronize the time to the NTP server at 15 minutes past every hour. (It is possible to choose a different minute, or to vary the minute between machines in order to avoid heavy traffic to the NTP server.) Hourly synchronization should be sufficiently frequent that clock drift will not be noticeable.

3.10.2.2 - Configure an NTP Server

The NTP server contacts a central NTP server, probably either one provided by your ISP or a public time server, to obtain accurate time data. The server then allows other machines on your network to request the time data. The NTP server configuration file is located at /etc/ntp.conf.

3.10.2.2.1 - Enable the NTP Daemon

If this machine is an NTP server, ensure that ntpd is enabled at boot time: # chkconfig ntpd on

3.10.2.2.2 - Deny All Access to ntpd by Default

Edit the file /etc/ntp.conf. Prepend or correct the following line: restrict default ignore Since ntpd is a complex software package which listens for network connections and runs as root, it must be protected from network access by unauthorized machines. This setting uses ntpd’s internal authorization to deny all access to any machine, server or client, which is not specifically authorized by other policy settings.

3.10.2.2.3 - Specify a Remote NTP Server for Time Data

Find the IP address, server-ip , of an appropriate remote NTP server. Edit the file /etc/ntp.conf, and add or correct the following lines: restrict server-ip mask 255.255.255.255 nomodify notrap noquery server server-ip If your site does not require time data to be accurate, but merely to be synchronized among local machines, this step can be omitted, and the NTP server will default to providing time data from the local clock. However, it is a good idea to periodically synchronize the clock to some source of accurate time, even if it is not appropriate to do so automatically. The previous step disabled all remote access to this NTP server’s state data. This NTP server must contact a remote server to obtain accurate data, so NTP’s configuration must allow that remote data to be used to modify the system clock. The restrict line changes the default access permissions for that remote server. The server line specifies the remote server as the preferred NTP server for time data. If you intend to synchronize to more than one server, specify restrict and server lines for each server. Note: It would be possible to specify a hostname, rather than an IP address, for the server field. However, the restrict setting applies only to network blocks of IP addresses, so it is considered more maintainable to use the IP address in both fields.

3.10.2.2.4 - Allow Legitimate NTP Clients to Access the Server

Determine an appropriate network block, netwk , and network mask, mask , representing the machines on your network which will synchronize to this server. Edit /etc/ntp.conf and add the line: restrict netwk mask mask nomodify notrap Edit /etc/sysconfig/iptables. Add the following line, ensuring that it appears before the final LOG and DROP lines for the RH-Firewall-1-INPUT chain: -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p udp --dport 123 -j ACCEPT If the clients are spread across more than one netblock, separate restrict and ACCEPT lines should be added for each netblock. The iptables configuration is needed because the default iptables configuration does not allow inbound access to any services. See Section 2.5.5 for more information about iptables. Note: The reference NTP implementation will refuse to serve time data to clients until enough time has elapsed that the server host’s time can be assumed to have settled to an accurate value. While testing, wait ten minutes after starting ntpd before attempting to synchronize clients.

3.10.3 - Configure OpenNTPD if Appropriate

OpenNTPD is an implementation of the SNTP protocol which is provided as a simple alternative to the reference NTP server. Advantages of OpenNTPD include simplicity of configuration, built-in privilege separation and chroot jailing of the NTP protocol code, and a small codebase which lacks many of the management and other protocol features used by the reference NTP server. This simplicity comes at the cost of degraded time accuracy, but SNTP is probably accurate enough for most sites with typical monitoring requirements.

3.10.3.1 - Obtain NTP Software

If your site intends to use the OpenNTPD implementation, it is necessary to compile and install the software. (If your site intends to use the reference NTP implementation, no installation is necessary.) 1. Obtain the software by downloading an appropriate source version, openntpd-version .tar.gz, from http://www.openntpd.org/portable.html. 2. Unpack the source code: $ tar xzf openntpd-version .tar.gz 3. Configure and compile the source. (By default, the code will be compiled for installation into /usr/ local): $ cd openntpd-version $ ./configure --with-privsep-user=ntp $ make 4. As root, install the resulting program into /usr/local: # make install The configuration option --with-privsep-user=ntp tells OpenNTPD to use the existing system account ntp for the non-root portion of its operation.

3.10.3.2 - Configure an SNTP Client

OpenNTPD runs only in daemon mode — there is no command line suitable to be run from cron. However, this is considered reasonably safe for client use because the daemon does not listen on any network ports by default, and because OpenNTPD is a small codebase with no remote management interface or other complex features. However, it is possible to run a time-stepping program, such as rdate(1), from cron instead of configuring the daemon as outlined in this section.

3.10.3.2.1 - Enable the NTP Daemon

Edit the file /etc/rc.local. Add or correct the following line: /usr/local/sbin/ntpd -s

3.10.3.2.2 - Configure the Client NTP Daemon to Use the Local Server

Edit the file /usr/local/etc/ntpd.conf. Add or correct the following line: server local-server.example.com where local-server.example.com is the hostname of the site’s local NTP or SNTP server.

3.10.3.3 - Configure an SNTP Server

The SNTP server obtains time data from a remote server, and then listens on a network interface for time queries from local machines.

3.10.3.3.1 - Enable the NTP Daemon

Edit the file /etc/rc.local. Add or correct the following line: /usr/local/sbin/ntpd -s Since OpenNTPD is third-party software, it does not have a standard startup script, so the daemon is started at boot using the local facility.

3.10.3.3.2 - Listen for Client Connections

Edit the file /usr/local/etc/ntpd.conf. Add or correct the following line: listen on ipaddr where ipaddr is the primary IP address of this server. By default, ntpd does not listen for any connections over a network. Listening must be actively enabled on NTP servers so that clients may obtain time data.

3.10.3.3.3 - Allow Legitimate NTP Clients to Access the Server

Determine an appropriate network block, netwk , and network mask, mask , representing the machines on your network which will synchronize to this server. Edit /etc/sysconfig/iptables. Add the following line, ensuring that it appears before the final LOG and DROP lines for the RH-Firewall-1-INPUT chain: -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p udp --dport 123 -j ACCEPT The iptables configuration is needed because the default iptables configuration does not allow inbound access to any services. See Section 2.5.5 for more information about iptables.

3.10.3.3.4 - Specify a Remote NTP Server for Time Data

Find the hostname, server-host , of an appropriate remote NTP server. Edit the file /usr/local/etc/ ntpd.conf, and add or correct the following line: server server-host This setting configures ntpd to obtain time data from the remote host. To use multiple time servers, add one line for each server.

3.11 - Mail Transfer Agent

Mail servers are used to send and receive mail over a network on behalf of site users. Mail is a very common service, and MTAs are frequent targets of network attack. Ensure that machines are not running MTAs unnecessarily, and configure needed MTAs as defensively as possible.

3.11.1 - Select Mail Server Software and Configuration

Select one of the following options for configuring e-mail on the machine: * If this machine does not need to operate as a mail server, follow the instructions in Section 3.11.2 to run sendmail in submission-only mode. * If the machine must operate as a mail server, read the strategies for MTA configuration in Section 3.11.3 for information about configuration options. Then apply both the MTA-independent operating system configuration guidance in Section 3.11.4, and the specific guidance for your MTA: – If the Sendmail MTA is preferred, see Section 3.11.5. – If the Postfix MTA is preferred, see Section 3.11.6. – If another MTA is preferred, use that MTA’s documentation to implement the ideas in Section 3.11.3. It is recommended that very few machines at any site be configured to receive mail over a network. However, it may be necessary for most machines at a given site to send e-mail, for instance so that cron jobs can report output to an administrator. Sendmail supports a submission-only mode in which mail can be sent from the machine to a central site MTA, but the machine cannot receive mail over a network. If a Mail Transfer Agent (MTA) is needed, the system default is Sendmail. Postfix, a popular alternative written with security in mind, is also available. Postfix can be more effectively contained by SELinux as its modular design has resulted in separate processes performing specific actions. More information on these MTAs is available from their respective websites, http://www.sendmail.org and http://www.postfix.org.

3.11.2 - Configure SMTP For Mail Client

This guide discusses the use of Sendmail for submission-only e-mail configuration. It is also possible to use Postfix.

3.11.2.1 - Disable the Listening Sendmail Daemon

Edit the file /etc/sysconfig/sendmail. Add or modify the line: DAEMON=no The MTA performs two functions: listening over a network for incoming SMTP e-mail requests, and sending mail from the local machine. Since outbound mail may be delayed due to network outages or other problems, the outbound MTA runs in a queue-only mode, in which it periodically attempts to resend any delayed mail. Setting DAEMON=no tells sendmail to execute only the queue runner on this machine, and never to receive SMTP mail requests.

3.11.2.2 - Configure Mail Submission if Appropriate

If it is appropriate to configure mail submission with a central MTA, edit /etc/mail/submit.cf. Locate the line beginning with D{MTAHost}, and modify it to read: D{MTAHost}mailserver where mailserver is the hostname of the server to which this machine should forward its outgoing mail. This suggestion is provided as a simple way to migrate away from a configuration in which each machine at a site runs its own MTA, to a configuration in which client machines do not run listening daemons. If this modification is made to /etc/mail/submit.cf, then, when a local process on a machine attempts to send mail, the message will be forwarded to the machine mailserver for processing. Modifying /etc/mail/submit.cf directly is only appropriate if your site does not perform any other mailserver customization on clients. If other customization is done, use your usual Sendmail change procedure to define the MTA host. Note: In addition to making this change on the client, it may also be necessary to reconfigure the MTA on mailserver so that it will relay mail on behalf of this host.

3.11.3 - Strategies for MTA Security

This section discusses several types of MTA configuration which should be performed in order to protect against attacks involving the mail system. Though configuration syntax will differ depending on which MTA is in use (see Section 3.11.5 for Sendmail configuration syntax and Section 3.11.6 for Postfix), these strategies are generally advisable for any MTA, including ones not covered by this guide.

3.11.3.1 - Use Resource Limits to Mitigate Denial of Service

It is often desirable to constrain an attacker’s ability to consume a mail server’s resources simply by sending otherwise valid mail at a high rate, whether maliciously or accidentally. Relevant resource limits include con106 CHAPTER 3. SERVICES straints on: the number of MTA daemons which may run at one time, the rate at which incoming messages may be received, the size and complexity of each message, or the amount of mail queue space which must remain free in order for mail to be delivered. That last parameter deserves additional explanation. Most MTAs require queue space for temporary files in order to process existing messages in their queues. Therefore, if the queue filesystem is allowed to fill completely in a denial of service, the MTA will not be able to clear its own queue even when the malicious traffic has stopped. This will delay recovery from an attack.

3.11.3.2 - Configure SMTP Greeting Banner

When remote mail senders connect to the MTA on port 25, they are greeted by an initial banner as part of the SMTP dialogue. This banner is necessary, but it frequently gives away too much information, including the MTA software which is in use, and sometimes also its version number. Remote mail senders do not need this information in order to send mail, so the banner should be changed to reveal only the hostname (which is already known and may be useful) and the word ESMTP, to indicate that the modern SMTP protocol variant is supported.

3.11.3.3 - Control Mail Relaying

The sending of Unsolicited Bulk E-mail, referred to variously as UBE, UCE, or spam, is a major problem on the Internet today. The security implications of spam are that it operates as a Denial of Service attack on legitimate e-mail use. Strategies for fighting spam receipt at your site are complex and quickly evolving, and thus far beyond the scope of this guide. The problem of relaying unauthorized e-mail, however, can and should be addressed by any network-connected site. Most MTAs perform two functions: to accept mail from remote sites on behalf of local users, and to allow local users to send mail to remote sites. The former function is relatively easy — mail whose recipient address is local can be assumed to be destined for a local user. The latter function is more complex. Since it is typically considered neither secure nor desirable for users to log in to the MTA host itself to send mail, the MTA must be able to remotely accept mail addressed to anyone from the user’s workstation. If the MTA is running very old software or is configured poorly, it can be possible for attackers to take advantage of this feature, using your MTA to relay their spam from one remote site to another. This is undesirable for many reasons, not least that your site will quickly be blacklisted as a spam source, leaving you unable to send legitimate e-mail to your correspondents. The simplest solution described in this guide is to configure the MTA to relay mail only from the local site’s address range, and some variant on this is the default for most modern MTAs. That solution may be insufficient for sites whose users need to send mail from remote machines, for instance while travelling, as well as for sites where mail submission must be accepted from network ranges which are not considered secure, either because authorized machines are unmanaged or because it is possible to connect unauthorized machines to the network. If remote or mobile hosts are authorized to relay, or if local clients exist in insecure netblocks, the SMTP AUTH protocol should be used to require mail senders to authenticate before submitting messages. For better protection and to allow support for a wide range of authentication mechanisms without sending passwords over a network in clear text, SMTP AUTH transactions should be encrypted using SSL. Another approach is to require mail to be submitted on port 587, the designated Message Submission Port. Use of a separate port allows the mail relay function to be entirely separated from the mail delivery function. This may become a best practice in the future, but description of how to configure the Message Submission Port is currently beyond the scope of this guide. See RFC 2476 for information about this configuration.

3.11.4 - Configure Operating System to Protect Mail Server

The guidance in this section is appropriate for any host which is operating as a site MTA, whether the mail server runs using Sendmail, Postfix, or some other software.

3.11.4.1 - Use Separate Hosts for External and Internal Mail if Possible

The mail server is a frequent target of network attack from the outside. However, since all site users receive mail, the mail server must be open to some connection from each inside users. It is strongly recommended that these functions be separated, by having an externally visible mail server which processes all incoming and outgoing mail, then forwards internal mail to a separate machine from which users can access it.

3.11.4.2 - Protect the MTA Host from User Access

The mail server contains privileged data belonging to all users and performs a vital network function. Preventing users from logging into this server is a precaution against privilege escalation or denial of service attacks which might compromise the mail service. Take steps to ensure that only system administrators are allowed shell access to the MTA host.

3.11.4.3 - Restrict Remote Access to the Mail Spool

If users directly connect to this machine to receive mail, ensure that there is a single, well-secured mechanism for access to the directory /var/spool/mail (the directory /var/mail is a symlink to this). Allowing unrestricted access to /var/spool/mail can be dangerous, since this directory contains sensitive information belonging to all users. Protocols such as NFS, which have an insecure authorization mechanism by default, should be considered insufficient for these purposes. See Section 3.17 for details on secure configuration of POP3 or IMAP, which are the preferred ways to provide user access to mail.

3.11.4.4 - Configure iptables to Allow Access to the Mail Server

Edit /etc/sysconfig/iptables. Add the following line, ensuring that it appears before the final LOG and DROP lines for the RH-Firewall-1-INPUT chain: -A RH-Firewall-1-INPUT -m state --state NEW -p tcp --dport 25 -j ACCEPT The default Iptables configuration does not allow inbound access to the SMTP service. This modification allows that access, while keeping other ports on the server in their default protected state. See Section 2.5.5 for more information about Iptables.

3.11.4.5 - Verify System Logging and Log Permissions for Mail

Edit the file /etc/syslog.conf. Add or correct the following line if necessary (this is the default): mail.* -/var/log/maillog Run the following commands to ensure correct permissions on the mail log: # chown root:root /var/log/maillog # chmod 600 /var/log/maillog The mail server logs contain a record of every e-mail which is sent or received on the system, which is considered sensitive information by most sites. It is necessary that these logs be collected for purposes of debugging and statistics, but their contents should be protected from unauthorized access.

3.11.4.6 - Configure SSL Certificates for Use with SMTP AUTH

If SMTP AUTH is to be used (see Section 3.11.3.3 for a description of possible anti-relaying mechanisms), the use of SSL to protect credentials in transit is strongly recommended. There are also configurations for which it may be desirable to encrypt all mail in transit from one MTA to another, though such configurations are beyond the scope of this guide. In either event, the steps for creating and installing an SSL certificate are independent of the MTA in use, and are described here.

3.11.4.6.1 - Create an SSL Certificate

Note: This step must be performed on your CA system, not on the MTA host itself. If you will have a commercial CA sign certificates, then this step should be performed on a separate, physically secure system devoted to that purpose. Change into the CA certificate directory: # cd /etc/pki/tls/certs Generate a key pair for the mail server: # openssl genrsa -out mailserverkey.pem 2048 Next, generate a certificate signing request (CSR) for the CA to sign, making sure to supply your mail server’s fully qualified domain name as the Common Name: # openssl req -new -key mailserverkey.pem -out mailserver.csr Next, the mail server CSR must be signed to create the mail server certificate. You can either send the CSR to an established CA or sign it with your CA. To sign mailserver.csr using your CA: # openssl ca -in mailserver.csr -out mailservercert.pem This step creates a private key, mailserverkey.pem, and a public certificate, mailservercert.pem. The mail server will use these to prove its identity by demonstrating that it has a certificate which has been signed by a CA. Mail clients at your site should be willing to send their mail only to a server they can authenticate.

3.11.4.6.2 - Install the SSL Certificate

Create the PKI directory for mail certificates, if it does not already exist: # mkdir /etc/pki/tls/mail # chown root:root /etc/pki/tls/mail # chmod 755 /etc/pki/tls/mail Using removable media or some other secure transmission format, install the files generated in the previous step onto the mail server: * /etc/pki/tls/mail/serverkey.pem: the private key mailserverkey.pem * /etc/pki/tls/mail/servercert.pem: the certificate file mailservercert.pem Verify the ownership and permissions of these files: # chown root:root /etc/pki/tls/mail/serverkey.pem # chown root:root /etc/pki/tls/mail/servercert.pem # chmod 600 /etc/pki/tls/mail/serverkey.pem # chmod 644 /etc/pki/tls/mail/servercert.pem Verify that the CA’s public certificate file has been installed as /etc/pki/tls/CA/cacert.pem, and has the correct permissions: # chown root:root /etc/pki/tls/CA/cacert.pem # chmod 644 /etc/pki/tls/CA/cacert.pem

3.11.5 - Configure Sendmail Server if Necessary

When sendmail is configured to act as a server for incoming mail, it listens on port 25 for connections, and responds to those connections using the configuration in /etc/mail/sendmail.cf. This file has a somewhat opaque format, and modifying it directly is generally not recommended. Instead, the following procedure should be used to modify the sendmail configuration: 1. Install the sendmail-cf RPM, which is required in order to compile a new configuration file: # yum install sendmail-cf 2. Edit the M4 source file /etc/mail/sendmail.mc as directed by the configuration step you are applying. 3. Inside the directory /etc/mail/, use make to build the configuration according to the Makefile provided by Sendmail: # cd /etc/mail # make sendmail.cf

3.11.5.1 - Limit Denial of Service Attacks

Edit /etc/mail/sendmail.mc, and add or correct the following options: define(`confMAX_DAEMON_CHILDREN',`40 ')dnl define(`confCONNECTION_RATE_THROTTLE', `3 ')dnl define(`confMIN_FREE_BLOCKS',`20971520 ')dnl define(`confMAX_HEADERS_LENGTH',`51200 ')dnl define(`confMAX_MESSAGE_SIZE',`10485760 ')dnl define(`confMAX_RCPTS_PER_MESSAGE',`100 ')dnl Note: The values given here are examples, and may need to be modified for any particular site, especially one with high e-mail volume. These configuration options serve to make it more difficult for attackers to consume resources on the MTA host. (See Section 3.11.3.1 for details on why this is done.) The MAX DAEMON CHILDREN option limits the number of sendmail processes which may be deployed to handle incoming connections at any one time, while CONNECTION RATE THROTTLE limits the number of connections per second which each listener may receive. The MIN FREE BLOCKS option stops e-mail receipt when the queue filesystem is close to full. The MAX HEADERS LENGTH (bytes), MAX MESSAGE SIZE (bytes), and MAX RCPTS PER MESSAGE (distinct recipients) options place bounds on the legal sizes of messages received via SMTP.

3.11.5.2 - Configure SMTP Greeting Banner

Edit /etc/mail/sendmail.mc, and add or correct the following line, substituting an appropriate greeting string for $j : define(`confSMTP_LOGIN_MSG', `$j ')dnl and recompile sendmail's configuration. The default greeting banner discloses that the listening mail process is Sendmail rather than some other MTA, and also provides the version number. See Section 2.3.7 for more about warning banners, and Section 3.11.3.2 for strategies regarding SMTP greeting banners in particular. The Sendmail variable $j contains the hostname of the mail server, which may be an appropriate greeting string for most environments.

3.11.5.3 - Control Mail Relaying

This guide will discuss two mechanisms for controlling mail relaying in Sendmail. The /etc/mail/relay-domains file contains a list of hostnames that are allowed to relay mail. Follow the guidance in Section 3.11.5.3.1 to configure relaying for trusted machines. If there are machines which must be allowed to relay mail, but which cannot be trusted to relay unconditionally, configure SMTP AUTH with TLS support using the guidance in Sections 3.11.5.3.2 and following.

3.11.5.3.1 - Configure Trusted Networks and Hosts

* If all machines which share a common domain or subdomain name may relay, then edit /etc/mail/ relay-domains, adding a line for each domain or subdomain, e.g.: example.com trusted-subnet.school.edu ... * If the machines which are allowed to relay must be specified on a per-host basis, then edit /etc/mail/ relay-domains, adding a line for each such host: host1.example.com host5.subnet.example.com smtp.trusted-subnet.school.edu Then edit /etc/mail/sendmail.mc, add or correct the line: FEATURE(`relay_hosts_only')dnl and recompile sendmail’s configuration. The file /etc/mail/relay-domains must contain only the set of machines for which this MTA should unconditionally relay mail. This configures both inbound and outbound relaying, that is, hosts mentioned in relay-domains may send mail through the MTA, and the MTA will also accept inbound mail addressed to such hosts. This is a trust relationship — if spammers gain access to these machines, your site will effectively become an open relay. It is recommended that only machines which are managed by you or by another trusted organization be placed in relay-domains, and that users of all other machines be required to use SMTP AUTH to send mail. Note: The relay-domains file must be configured to contain either a list of domains (in which case every host in each of those domains will be allowed to relay) or a list of hosts (in which case each individual relaying host must be listed and the sendmail.cf must be reconfigured to interpret the relay-domains file in the desired way).

3.11.5.3.2 - Require SMTP AUTH Before Relaying from Untrusted Clients

By default, Sendmail uses the Cyrus-SASL library to provide authentication. To enable the use of SASL authentication for relaying, edit /etc/mail/sendmail.mc and add or correct the following settings: TRUST_AUTH_MECH(`LOGIN PLAIN') define(`confAUTH_MECHANISMS', `LOGIN PLAIN') and recompile sendmail.cf. Then edit /usr/lib/sasl2/Sendmail.conf and add or correct the following lines: pwcheck_method: saslauthd Enable the saslauthd daemon: # chkconfig saslauthd on The AUTH MECHANISMS configuration option tells sendmail to allow the specified authentication mechanisms to be used during the SMTP dialogue. The two listed mechanisms use SASL to test a password provided by the user. Since these mechanisms transmit plaintext passwords, they should be protected using TLS as described in the next section. The TRUST AUTH MECH command tells sendmail that senders who successfully authenticate using the specified mechanism may relay mail through this MTA even if their addresses are not in relay-domains. The file /usr/lib/sasl/Sendmail.conf is the Cyrus-SASL configuration file for Sendmail. The pwcheck method directive tells SASL how to find passwords. The simplest method, described here, is to run a separate authentication daemon, saslauthd, which is able to communicate with the system authentication service. On Red Hat, saslauthd uses PAM by default, which should work in most cases. If you have a centralized authentication system which does not work via PAM, look at the saslauthd(8) manpage to determine how to configure saslauthd for your environment.

3.11.5.3.3 - Require TLS for SMTP AUTH

Edit /etc/mail/sendmail.mc, add or correct the following lines: define(`confAUTH_OPTIONS', `A p')dnl define(`confCACERT_PATH', `/etc/pki/tls/CA')dnl define(`confCACERT', `/etc/pki/tls/CA/cacert.pem')dnl define(`confSERVER_CERT', `/etc/pki/tls/mail/servercert.pem')dnl define(`confSERVER_KEY', `/etc/pki/tls/mail/serverkey.pem')dnl and recompile sendmail.cf. These options, combined with the previous settings, tell Sendmail to protect all SMTP AUTH transactions using TLS. The first four options describe the location of the necessary TLS certificate and key files. The AUTH OPTIONS parameter configures the SMTP AUTH dialogue. The A option is enabled by default, and simply says that authentication is allowed if an appropriate mechanism can be found. The p option tells Sendmail to protect against passive attacks. The PLAIN and LOGIN authentication mechanisms, recommended by this guide for compatibility with PAM, send passwords in the clear. (Cleartext password transmissions are vulnerable to passive attack.) Therefore, if p is set, the SMTP daemon will not make the AUTH command available until after the client has used the STARTTLS command to encrypt the session. If other authentication mechanisms were enabled which did not send passwords in the clear, then TLS would not necessarily be required.

3.11.6 - Configure Postfix if Necessary

Postfix stores its configuration files in the directory /etc/postfix by default. The primary configuration file is /etc/postfix/main.cf. Other files will be introduced as needed.

3.11.6.1 - Limit Denial of Service Attacks

Edit /etc/postfix/main.cf. Add or correct the following lines: default_process_limit = 100 smtpd_client_connection_count_limit = 10 smtpd_client_connection_rate_limit = 30 queue_minfree = 20971520 header_size_limit = 51200 message_size_limit = 10485760 smtpd_recipient_limit = 100 Note: The values given here are examples, and may need to be modified for any particular site. By default, the Postfix anvil process gathers mail receipt statistics. To get information about about what connection rates are typical at your site, look in /var/log/maillog for lines with the daemon name postfix/anvil. These configuration options serve to make it more difficult for attackers to consume resources on the MTA host. (See Section 3.11.3.1 for details on why this is done.) The default process limit parameter controls how many smtpd processes can exist at a time, while smtpd client connection count limit controls the number of those which can be occupied by any one remote sender, and smtpd client connection rate limit controls the number of connections any one client can make per minute. By default, local hosts (those in mynetworks) are exempted from per-client rate limiting. The queue minfree parameter establishes a free space threshold, in order to stop e-mail receipt before the queue filesystem is entirely full. The header size limit, message size limit, and smtpd recipient limit parameters place bounds on the legal sizes of messages received via SMTP.

3.11.6.2 - Configure SMTP Greeting Banner

Edit /etc/postfix/main.cf, and add or correct the following line, substituting some other wording for the banner information if you prefer: smtpd_banner = $myhostname ESMTP The default greeting banner discloses that the listening mail process is Postfix. See Section 2.3.7 for more about warning banners, and Section 3.11.3.2 for strategies regarding SMTP greeting banners in particular.

3.11.6.3 - Control Mail Relaying

Postfix’s mail relay controls are implemented with the help of the smtpd recipient restrictions option, which controls the restrictions placed on the SMTP dialogue once the sender and recipient envelope addresses are known. The guidance in Sections 3.11.6.3.1–3.11.6.3.2 should be applied to all machines. If there are machines which must be allowed to relay mail, but which cannot be trusted to relay unconditionally, configure SMTP AUTH with SSL support using the guidance in Sections 3.11.6.3.3 and following.

3.11.6.3.1 - Configure Trusted Networks and Hosts

Edit /etc/postfix/main.cf, and configure the contents of the mynetworks variable in one of the following ways: * If any machine in the subnet containing the MTA may be trusted to relay messages, add or correct the line: mynetworks_style = subnet * If only the MTA host itself is trusted to relay messages, add or correct: mynetworks_style = host * If the set of machines which can relay is more complicated, manually specify an entry for each netblock or IP address which is trusted to relay by setting the mynetworks variable directly: mynetworks = 10.0.0.0/16 , 192.168.1.0/24 , 127.0.0.1 The mynetworks variable must contain only the set of machines for which this MTA should unconditionally relay mail. This is a trust relationship — if spammers gain access to these machines, your site will effectively become an open relay. It is recommended that only machines which are managed by you or by another trusted organization be placed in mynetworks, and users of all other machines be required to use SMTP AUTH to send mail.

3.11.6.3.2 - Allow Unlimited Relaying for Trusted Networks Only

Edit /etc/postfix/main.cf, and add or correct the smtpd recipient restrictions definition so that it contains at least: smtpd_recipient_restrictions = ... permit_mynetworks, reject_unauth_destination, ... The full contents of smtpd recipient restrictions will vary by site, since this is a common place to put spam restrictions and other site-specific options. The permit mynetworks option allows all mail to be relayed from the machines in mynetworks. Then, the reject unauth destination option denies all mail whose destination address is not local, preventing any other machines from relaying. These two options should always appear in this order, and should usually follow one another immediately unless SMTP AUTH is used.

3.11.6.3.3 - Require SMTP AUTH Before Relaying from Untrusted Clients

SMTP authentication allows remote clients to relay mail safely by requiring them to authenticate before submitting mail. Postfix’s SMTP AUTH uses an authentication library called SASL, which is not part of Postfix itself. This section describes how to configure authentication using the Cyrus-SASL implementation. See below for a discussion of other options. To enable the use of SASL authentication, edit /etc/postfix/main.cf and add or correct the following settings: smtpd_sasl_auth_enable = yes smtpd_recipient_restrictions = ... permit_mynetworks, permit_sasl_authenticated, reject_unauth_destination, ... Then edit /usr/lib/sasl/smtpd.conf and add or correct the following line with the correct authentication mechanism for SASL to use: pwcheck_method: saslauthd Enable the saslauthd daemon: # chkconfig saslauthd on Postfix can use either the Cyrus library or Dovecot as a source for SASL authentication. If this host is running Dovecot for some other reason, it is recommended that Dovecot’s SASL support be used instead of running the Cyrus code as well. See http://www.postfix.org/SASL README.html for instructions on implementing that configuration, which is not described in this guide. In Postfix’s configuration, the directive smtpd sasl auth enable tells smtpd to allow the use of the SMTP AUTH command during the SMTP dialogue, and to support that command by getting authentication information from SASL. The smtpd recipient restrictions directive is changed so that, if the client is not connecting from a trusted address, it is allowed to attempt authentication (permit sasl authenticated) in order to relay mail. The file /usr/lib/sasl/smtpd.conf is the Cyrus-SASL configuration file. The pwcheck method directive tells SASL how to find passwords. The simplest method, described above, is to run a separate authentication daemon, saslauthd, which is able to communicate with the system authentication system. On RHEL5, saslauthd uses PAM by default, which should work in most cases. If you have a centralized authentication system which does not work via PAM, look at the saslauthd(8) manpage to find out how to configure saslauthd for your environment.

3.11.6.4 - Require TLS for SMTP AUTH

Edit /etc/postfix/main.cf, and add or correct the following lines: smtpd_tls_CApath = /etc/pki/tls/CA smtpd_tls_CAfile = /etc/pki/tls/CA/cacert.pem smtpd_tls_cert_file = /etc/pki/tls/mail/servercert.pem smtpd_tls_key_file = /etc/pki/tls/mail/serverkey.pem smtpd_tls_security_level = may smtpd_tls_auth_only = yes These options tell Postfix to protect all SMTP AUTH transactions using TLS. The first four options describe the locations of the necessary TLS key files. The smtpd tls security level directive tells smtpd to allow the STARTTLS command during the SMTP protocol exchange, but not to require it for mail senders. (Unless your site receives mail only from other trusted sites whose sysadmins can be asked to maintain a copy of your site certificate, you do not want to require TLS for all SMTP exchanges.) The smtpd tls auth only directive tells smtpd to require the STARTTLS command before allowing the client to attempt to authenticate for relaying using SMTP AUTH. It may not be possible to use this directive if you must allow relaying from non-TLS-capable client software. If this is the case, simply omit that line.

3.12 - LDAP

LDAP is a popular directory service, that is, a standardized way of looking up information from a central database. It is relatively simple to configure a RHEL5 machine to obtain authentication information from an LDAP server. If your network uses LDAP for authentication, be sure to configure both clients and servers securely.

3.12.1 - Use OpenLDAP to Provide LDAP Service if Possible

The system’s default LDAP client/server program is called OpenLDAP. Its documentation is available at the project web page: http://www.openldap.org.

3.12.2 - Configure OpenLDAP Clients

Before configuring any machine to be an LDAP client, ensure that a working LDAP server is present on the network. See Section 3.12.3 for instructions on configuring an LDAP server. This guide recommends configuring OpenLDAP clients by manually editing the appropriate configuration files. RHEL5 provides an automated configuration tool called authconfig and a graphical wrapper for authconfig called system-config-authentication. However, these tools do not give sufficient flexibility over configuration. The authconfig tools do not allow you to specify locations of SSL certificate files, which is useful when trying to use SSL cleanly across several protocols. They are also overly aggressive in placing services such as netgroups and automounter maps under LDAP control, where it is safer to use LDAP only for services to which it is relevant in your environment.

3.12.2.1 - Configure the Appropriate LDAP Parameters for the Domain

Assume the fully qualified host name of your LDAP server is ldap.example.com and the base DN of your domain is dc=example,dc=com (it is conventional to use the domain name as a base DN). Edit /etc/ldap. conf, and add or correct the following lines: base dc=example,dc=com uri ldap://ldap.example.com / Then edit /etc/openldap/ldap.conf, and add or correct the following lines: BASE dc=example,dc=com URI ldap://ldap.example.com / The machine whose hostname is given here must be configured as an LDAP server, serving data identified by the base DN used here. See Section 3.12.3 for details on configuring an LDAP server.

3.12.2.2 - Configure LDAP to Use TLS for All Transactions

1. Ensure a copy of the site’s CA certificate has been placed in the file /etc/pki/tls/CA/cacert.pem. 2. Configure LDAP to enforce TLS use and to trust certificates signed by the site’s CA. First, edit the file /etc/ldap.conf, and add or correct the following lines: ssl start_tls tls_checkpeer yes tls_cacertdir /etc/pki/tls/CA tls_cacertfile /etc/pki/tls/CA/cacert.pem Then edit /etc/openldap/ldap.conf, and add or correct the following lines: TLS_CACERTDIR /etc/pki/tls/CA TLS_CACERT /etc/pki/tls/CA/cacert.pem Section 2.5.6 describes the system-wide configuration of SSL for your enterprise. It is possible to place your certificate information under some directory other than /etc/pki/tls, but using a consistent directory structure across all SSL services at your site is recommended. The LDAP server must be configured with a certificate signed by the CA certificate named here.

3.12.2.3 - Configure Authentication Services to Use OpenLDAP

Edit the file /etc/ldap.conf, and add or correct the following lines: pam_password md5 Edit the file /etc/nsswitch.conf, and add or correct the following lines: passwd: files ldap shadow: files ldap group: files ldap Edit the file /etc/pam.d/system-auth-ac. Make the following changes, which will add references to LDAP in each of the four sections of the file: * Immediately before the last line in the auth section (the one containing pam deny.so), insert the line: auth sufficient pam_ldap.so use_first_pass * Modify the first line in the account section by adding the option broken shadow. The line should then read: account required pam_unix.so broken_shadow * Immediately before the last line in the account section (the one containing pam permit.so), insert the line: account [default=bad success=ok user_unknown=ignore] pam_ldap.so * Immediately before the last line in the password section (the one containing pam deny.so), insert the line: password sufficient pam_ldap.so use_authtok * At the end of the file (after the last line in the session section), append the line: session optional pam_ldap.so The first modification tells LDAP to expect passwords in MD5 hash format, rather than clear text. Red Hat systems use the file /etc/nsswitch.conf to determine the appropriate sources to search for certain kinds of data, such as usernames, groups, hostnames, netgroups, or protocols. It is possible to manage many other types of data using LDAP, but this guide recommends that only usernames (passwd data), passwords (shadow data), and groups (group data) be managed using LDAP. If your site uses netgroups, it may be appropriate to manage these via LDAP as well. However, data which almost never changes, such as the contents of the /etc/services file, is a poor choice for central administration, since it introduces risk with little benefit. It is recommended that the automounter not be used at all, so LDAP control of automounter maps is unlikely to be appropriate. The file /etc/pam.d/system-auth-ac is used by PAM to control access to most authenticated services. The syntax of the PAM configuration file is somewhat cryptic. The lines recommended here have the combined effect of using LDAP to find authentication data for users who cannot be found in the local /etc/passwd file. This means that, for instance, it is still possible to use a local root password. The details of options such as broken shadow, use authtok, and use first pass may be looked up in the man pages for the various PAM modules. Their basic effect is to attempt to authenticate given a password against both the local /etc/shadow and the central LDAP server, without forcing the user to type the password more than once. PAM configuration is discussed further in Section 2.3.3.

3.12.3 - Configure OpenLDAP Server

This section contains guidance on how to configure an OpenLDAP server to securely provide information for use in a centralized authentication service. This is not a comprehensive guide to maintaining an OpenLDAP server, but may be helpful in transitioning to an OpenLDAP infrastructure nonetheless.

3.12.3.1 - Install OpenLDAP Server RPM

Is this machine the OpenLDAP server? If so: # yum install openldap-servers # chkconfig ldap on The openldap-servers RPM is not installed by default on RHEL5 machines. It is needed only by the OpenLDAP server, not by the clients which use LDAP for authentication.

3.12.3.2 - Configure Domain-Specific Parameters

Edit the file /etc/openldap/slapd.conf. Add or correct the following lines: suffix "dc=example,dc=com " rootdn "cn=Manager,dc=example,dc=com " where dc=example,dc=com is the same root you will use on the LDAP clients. These are basic LDAP configuration directives. The suffix parameter gives the root name of all information served by this LDAP server, and should be some name related to your domain. The rootdn parameter names LDAP’s privileged user, who is allowed to read or write all data managed by this LDAP server.

3.12.3.3 - Configure an LDAP Root Password

Ensure that the configuration file has reasonable permissions before putting the hashed root password in that file: # chown root:ldap /etc/openldap/slapd.conf # chmod 640 /etc/openldap/slapd.conf Generate a hashed password using the slappasswd utility: # slappasswd New password: Re-enter new password: This will output a hashed password string. Edit the file /etc/openldap/slapd.conf, and add or correct the line: rootpw {SSHA}hashed-password-string Be sure to select a secure password for the LDAP root user, since this user has permission to read and write all LDAP data, so a compromise of the LDAP root password will probably enable a full compromise of your site. Protect configuration files containing the hashed password the same way you would protect other files, such as /etc/shadow, which contain hashed authentication data. In addition, be sure to use a reasonably strong hash function, such as SHA-1,1 rather than an insecure scheme such as crypt.

3.12.3.4 - Configure the LDAP Server to Require TLS for All Transactions

Because LDAP queries and responses, particularly those containing authentication information or other sensitive data, must be protected from disclosure or modification while in transit over the network, this guide recommends using SSL to protect all transactions. In order to do this, it is necessary to have a site-wide SSL infrastructure in which a CA certificate is used to verify that other certificates, such as that presented by the LDAP server to its clients, are authentic. Therefore, this procedure involves using the CA system to create a certificate for the LDAP server, then installing that certificate on the LDAP server and configuring slapd to require its use. See Section 2.5.6 for details about the process of creating SSL certificates for use by servers at your site.

3.12.3.4.1 - Create the Certificate for the LDAP Server

Note: This step must be performed on the CA system, not on the LDAP server itself. Change into the CA certificate directory: # cd /etc/pki/tls/certs Generate a key pair for the LDAP server: # openssl genrsa -out ldapserverkey.pem 2048 Next, generate a certificate signing request (CSR) for the CA to sign: # openssl req -new -key ldapserverkey.pem -out ldapserver.csr Sign the ldapserver.csr request: # openssl ca -in ldapserver.csr -out ldapservercert.pem This step creates a private key, ldapserverkey.pem, and a public certificate, ldapservercert.pem. The LDAP server will use these to prove its identity by demonstrating that it has a certificate which has been signed by the site CA. LDAP clients at your site should only be willing to accept authentication data from a verified LDAP server.

3.12.3.4.2 - Install the Certificate on the LDAP Server

Create the PKI directory for LDAP certificates if it does not already exist: # mkdir /etc/pki/tls/ldap # chown root:root /etc/pki/tls/ldap # chmod 755 /etc/pki/tls/ldap Using removable media or some other secure transmission format, install the files generated in the previous step onto the LDAP server: 1If you are using SHA-1, the hashed password string will begin with “{SHA}” or “{SSHA}”. * /etc/pki/tls/ldap/serverkey.pem: the private key ldapserverkey.pem * /etc/pki/tls/ldap/servercert.pem: the certificate file ldapservercert.pem Verify the ownership and permissions of these files: # chown root:ldap /etc/pki/tls/ldap/serverkey.pem # chown root:ldap /etc/pki/tls/ldap/servercert.pem # chmod 640 /etc/pki/tls/ldap/serverkey.pem # chmod 640 /etc/pki/tls/ldap/servercert.pem Verify that the CA’s public certificate file has been installed as /etc/pki/tls/CA/cacert.pem, and has the correct permissions: # mkdir /etc/pki/tls/CA # chown root:root /etc/pki/tls/CA/cacert.pem # chmod 644 /etc/pki/tls/CA/cacert.pem As a result of these steps, the LDAP server will have access to its own private certificate and the key with which that certificate is encrypted, and to the public certificate file belonging to the CA. Note that it would be possible for the key to be protected further, so that processes running as ldap could not read it. If this were done, the LDAP server process would need to be restarted manually whenever the server rebooted.

3.12.3.4.3 - Configure slapd to Use the Certificates

Edit the file /etc/openldap/slapd.conf. Add or correct the following lines: TLSCACertificateFile /etc/pki/tls/CA/cacert.pem TLSCertificateFile /etc/pki/tls/ldap/servercert.pem TLSCertificateKeyFile /etc/pki/tls/ldap/serverkey.pem security simple_bind=128 The first set of lines tell slapd where to find the appropriate SSL certificates to present to clients when they request an encrypted transaction. The last setting tells slapd never to allow clients to present credentials (i.e. passwords) in an unencrypted session. It is a good security principle never to allow unencrypted passwords to traverse a network, so ensure that LDAP mandates this.

3.12.3.5 - Install Account Information into the LDAP Database

There are many ways to maintain an OpenLDAP database. Methods include: * Input entries in ldif(5) format into a file /path/to/new entries , and use slapadd to import those entries while slapd is not running: # slapadd -l /path/to/new entries * Write a script to create and modify LDAP entries by connecting to the LDAP server normally. The Perl Net::LDAP module is appropriate for this, there is a Python API called python-ldap, and functionality is likely available for other scripting languages as well. * Use an LDAP front-end program which provides an interface for editing the database. If the front-end program is web-based or otherwise accessible over a network, ensure that authentication information is protected via SSL between the administrator’s client and the program, as well as between the program and the LDAP database. Any of these methods or others may be appropriate for your site. This guide does not provide a recommendation, and there will be no further discussion of the syntax of entering LDAP data into the database.

3.12.3.5.1 - Create Top-level LDAP Structure for Domain

Create a structure for the domain itself with at least the following attributes: dn: dc=example,dc=com objectClass: dcObject objectClass: organization dc: example o: Organization Description This is a placeholder for the root of the domain’s LDAP tree. Without this entry, LDAP will not be able to find any other entries for the domain.

3.12.3.5.2 - Create LDAP Structures for Users and Groups

Create LDAP structures for people (users) and for groups with at least the following attributes: dn: ou=people,dc=example,dc=com ou: people structuralObjectClass: organizationalUnit objectClass: organizationalUnit dn: ou=groups,dc=example,dc=com ou: groups structuralObjectClass: organizationalUnit objectClass: organizationalUnit Posix users and groups are the two top-level items which will be needed in order to use LDAP for authentication. These organizational units are used to identify the two categories within LDAP.

3.12.3.5.3 - Create Unix Accounts

For each Unix user, create an LDAP entry with at least the following attributes (others may be appropriate for your site as well), using variable values appropriate to that user. dn: uid=username ,ou=people,dc=example,dc=com structuralObjectClass: inetOrgPerson objectClass: inetOrgPerson objectClass: posixAccount objectClass: shadowAccount cn: fullname sn: surname gecos: fullname gidNumber: primary-group-id homeDirectory: /home/username loginShell: /path/to/shell uid: username uidNumber: uid userPassword: {MD5}md5-hashed-password If your site implements password expiration in which passwords must be changed every N days (see Section 2.3.1.7), then each entry should also have the attribute: shadowMax: N In general, the LDAP schemas for users use uid to refer to the text username, and uidNumber for the numeric UID. This usage may be slightly confusing when compared to the standard Unix usage. You should not create entries for the root account or for system accounts which are unique to individual systems, but only for user accounts which are to be shared across machines, and which have authentication information (such as a password) associated with them.

3.12.3.5.4 - Create Unix Groups

For each Unix group, create an LDAP entry with at least the following attributes: dn: cn=groupname ,ou=groups,dc=example,dc=com cn: groupname structuralObjectClass: posixGroup objectClass: posixGroup gidNumber: gid memberUid: username1 memberUid: username2 ... memberUid: usernameN Note that each user has a primary group, identified by the gidNumber field in the user’s account entry. That group must be created, but it is not necessary to list the user as a memberUid of the group. This behavior should be familiar to administrators, since it is identical to the handling of the /etc/passwd and /etc/group files. Do not create entries for the root group or for system groups, but only for groups which contain human users or which are shared across systems.

3.12.3.5.5 - Create Groups to Administer LDAP

If a group of LDAP administrators, admins , is desired, that group must be created somewhat differently. The specification should have these attributes: dn: cn=admins ,ou=groups,dc=example,dc=com cn: admins structuralObjectClass: groupOfUniqueNames objectClass: groupOfUniqueNames uniqueMember: cn=Manager,dc=example,dc=com uniqueMember: uid=admin1-username ,ou=people,dc=example,dc=com uniqueMember: uid=admin2-username ,ou=people,dc=example,dc=com ... uniqueMember: uid=adminN-username ,ou=people,dc=example,dc=com LDAP cannot use Posix groups for its own internal authentication — it needs to compare the username specified in an authenticated bind to some internal groupOfUniqueNames. If you do not specify an LDAP administrators’ group, then all LDAP management will need to be done using the LDAP root user (Manager). For reasons of auditing and error detection, it is recommended that LDAP administrators have unique identities. (See Section 2.3.1.3 for similar reasoning applied to the use of sudo for privileged system commands.)

3.12.3.6 - Configure slapd to Protect Authentication Information

Edit the file /etc/openldap/slapd.conf. Add or correct the following access specifications: 1. Protect the user’s password by allowing the user himself or the LDAP administrators to change it, allowing the anonymous user to authenticate against it, and allowing no other access: access to attrs=userPassword by self write by group/groupOfUniqueNames/uniqueMember="cn=admins ,ou=groups,dc=example,dc=com " write by anonymous auth by * none access to attrs=shadowLastChange by self write by group/groupOfUniqueNames/uniqueMember="cn=admins ,ou=groups,dc=example,dc=com " write by * read 2. Allow anyone to read other information, and allow the administrators to change it: access to * by group/groupOfUniqueNames/uniqueMember="cn=admins ,ou=groups,dc=example,dc=com " write by * read Access rules are applied in the order encountered, so more specific rules should appear first. In particular, the rule restricting access to userPassword must appear before the rule allowing access to all data. The shadowLastChange attribute is a timestamp, and is only critical if your site implements password expiration. If your site does not have an LDAP administrators group, the LDAP root user (called Manager in this guide) will be able to change data without an explicit access statement.

3.12.3.7 - Correct Permissions on LDAP Server Files

Correct the permissions on the ldap server’s files: # chown ldap:root /var/lib/ldap/* Some manual methods of inserting information into the LDAP database may leave these files with incorrect permissions. This will prevent slapd from starting correctly.

3.12.3.8 - Configure iptables to Allow Access to the LDAP Server

Determine an appropriate network block, netwk , and network mask, mask , representing the machines on your network which will synchronize to this server. Edit /etc/sysconfig/iptables. Add the following lines, ensuring that they appear before the final LOG and DROP lines for the RH-Firewall-1-INPUT chain: -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p tcp --dport 389 -j ACCEPT -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p tcp --dport 636 -j ACCEPT The default Iptables configuration does not allow inbound access to any services. These modifications allow access to the LDAP primary (389) and encrypted-only (636) ports, while keeping all other ports on the server in their default protected state. See Section 2.5.5 for more information about Iptables. Note: Even if the LDAP server restricts connections so that only encrypted queries are allowed, it will probably be necessary to allow traffic to the default port 389. This is true because many LDAP clients implement encryption by connecting to the primary port and issuing the STARTTLS command.

3.12.3.9 - Configure Logging for LDAP

1. Edit the file /etc/syslog.conf. Add or correct the following line: local4.* /var/log/ldap.log 2. Create the log file with safe permissions: # touch /var/log/ldap.log # chown root:root /var/log/ldap.log # chmod 0600 /var/log/ldap.log 3. Edit the file /etc/logrotate.d/syslog and add the pathname /var/log/ldap.log to the space-separated list in the first line. 4. Edit the LDAP configuration file /etc/openldap/slapd.conf and set a reasonable set of default log parameters, such as: loglevel stats2 OpenLDAP sends its log data to the syslog facility local4 at priority debug. By default, RHEL5 does not store this facility at all. The syslog configuration suggested here will store any output logged by slapd in the file /var/log/ldap.log, and will include that file in the standard log rotation for syslog files. By default, LDAP’s logging is quite verbose. The loglevel parameter is a space-separated list of items to be logged. Specifying stats2 will reduce the log output somewhat, but this level will still produce some logging every time an LDAP query is made. (This may be appropriate, depending on your site’s auditing requirements.) In order to capture only slapd startup messages, specify loglevel none. See slapd.conf(5) for detailed information about the loglevel parameter. See Section 2.6.1 for more information about syslog.

3.13 - NFS and RPC

The Network File System is the most popular distributed filesystem for the Unix environment, and is very widely deployed. Unfortunately, NFS was not designed with security in mind, and has a number of weaknesses, both in terms of the protocol itself and because any NFS installation must expose several daemons, running on both servers and clients, to network attack. This section discusses the circumstances under which it is possible to disable NFS and its dependencies, and then details steps which should be taken to secure, as much as possible, NFS’s configuration. This section is relevant to machines operating as NFS clients, as well as to those operating as NFS servers.

3.13.1 - Disable All NFS Services if Possible

The steps in Section 3.13.1 will prevent a machine from operating as either an NFS client or an NFS server. Only perform these steps on machines which do not need NFS at all. Is there a mission-critical reason for this machine to operate as either an NFS client or an NFS server? If not, follow all instructions in the remainder of Section 3.13.1 to disable subsystems required by NFS. NFS is a commonly used mechanism for sharing data between machines in an organization. However, its use opens many potential security holes. If NFS is not universally needed in your organization, improve the security posture of any machine which does not require NFS by disabling it entirely.

3.13.1.1 - Disable Services Used Only by NFS

If NFS is not needed, perform the following steps to disable NFS client daemons: # chkconfig nfslock off # chkconfig rpcgssd off # chkconfig rpcidmapd off The nfslock, rpcgssd, and rpcidmapd daemons all perform NFS client functions. All of these daemons run with elevated privileges, and many listen for network connections. If they are not needed, they should be disabled to improve system security posture.

3.13.1.2 - Disable netfs if Possible

Determine whether any network filesystems handled by netfs are mounted on this system: # mount -t nfs,nfs4,smbfs,cifs,ncpfs If this command returns no output, disable netfs to improve system security: # chkconfig netfs off The netfs script manages the boot-time mounting of several types of networked filesystems, of which NFS and Samba (see Section 3.18) are the most common. If these filesystem types are not in use, the script can be disabled, protecting the system somewhat against accidental or malicious changes to /etc/fstab and against flaws in the netfs script itself.

3.13.1.3 - Disable RPC Portmapper if Possible

If: * NFS is not needed * The site does not rely on NIS for authentication information, and * The machine does not run any other RPC-based service then disable the RPC portmapper service: # chkconfig portmap off By design, the RPC model does not require particular services to listen on fixed ports, but instead uses a daemon, portmap, to tell prospective clients which ports to use to contact the services they are trying to reach. This model weakens system security by introducing another privileged daemon which may be directly attacked, and is unnecessary because RPC was never adopted by enough services to risk using up all the ports on a system. Unfortunately, the portmapper is central to RPC design, so it cannot be disabled if your site is using any RPCbased services, including NFS, NIS (see Section 3.2.4 for information about NIS, which is not recommended), or any third-party or custom RPC-based program. If none of these programs are in use, however, portmap should be disabled to improve system security. In order to get more information about whether portmap may be disabled on a given host, query the local portmapper using the command: # rpcinfo -p If the only services listed are portmapper and status, it is safe to disable the portmapper. If other services are listed and your site is not running NFS or NIS, investigate these services and disable them if possible.

3.13.2 - Configure All Machines which Use NFS

The steps in this section are appropriate for all machines which run NFS, whether they operate as clients or as servers.

3.13.2.1 - Make Each Machine a Client or a Server, not Both

If NFS must be used, it should be deployed in the simplest configuration possible to avoid maintainability problems which may lead to unnecessary security exposure. Due to the reliability and security problems caused by NFS, it is not a good idea for machines which act as NFS servers to also mount filesystems via NFS. At the least, crossed mounts (the situation in which each of two servers mounts a filesystem from the other) should never be used.

3.13.2.2 - Restrict Access to the Portmapper

Edit the file /etc/hosts.deny. Add or correct the line: portmap: ALL Edit the file /etc/hosts.allow. Add or correct the line: portmap: IPADDR1 , IPADDR2 , ... where each IPADDR is the IP address of a server or client with which this machine shares NFS filesystems. If the machine is an NFS server, it may be simpler to use an IP netblock specification, such as 10.3.2. (this is the TCP Wrappers syntax representing the netblock 10.3.2.0/24), or a hostname specification, such as .subdomain.example.com. The use of hostnames is not recommended. The /etc/hosts.allow and /etc/hosts.deny files are used by TCP Wrappers to determine whether specified remote hosts are allowed to access certain services. The default portmapper shipped with RHEL5 has TCP Wrappers support built in, so this specification can be used to provide some protection against network attacks on the portmapper. (See Section 2.5.4 for more information about TCP Wrappers.) Note: This step protects only the portmap service itself. It is still possible for attackers to guess the port numbers of NFS services and attack those services directly, even if they are denied access to the portmapper.

3.13.2.3 - Configure NFS Services to Use Fixed Ports

Edit the file /etc/sysconfig/nfs. Add or correct the following lines: LOCKD_TCPPORT=lockd-port LOCKD_UDPPORT=lockd-port MOUNTD_PORT=mountd-port RQUOTAD_PORT=rquotad-port STATD_PORT=statd-port STATD_OUTGOING_PORT=statd-outgoing-port where each X-port is a port which is not used by any other service on your network. Firewalling should be done at each host and at the border firewalls to protect the NFS daemons from remote access, since NFS servers should never be accessible from outside the organization. However, by default, the portmapper assigns each NFS service to a port dynamically at service startup time. Dynamic ports cannot be protected by port filtering firewalls such as iptables (Section 2.5.5). Therefore, restrict each service to always use a given port, so that firewalling can be done effectively. Note that, because of the way RPC is implemented, it is not possible to disable the portmapper even if ports are assigned statically to all RPC services.

3.13.3 - Configure NFS Clients

The steps in this section are appropriate for machines which operate as NFS clients.

3.13.3.1 - Disable NFS Server Daemons

# chkconfig nfs off # chkconfig rpcsvcgssd off There is no need to run the NFS server daemons except on a small number of properly secured machines designated as NFS servers. Ensure that these daemons are turned off on clients.

3.13.3.2 - Mount Remote Filesystems with Restrictive Options

Edit the file /etc/fstab. For each filesystem whose type (column 3) is nfs or nfs4, add the text ,nodev,nosuid to the list of mount options in column 4. If appropriate, also add ,noexec. See Section 2.2.1.2 for a description of the effects of these options. In general, execution of files mounted via NFS should be considered risky because of the possibility that an adversary could intercept the request and substitute a malicious file. Allowing setuid files to be executed from remote servers is particularly risky, both for this reason and because it requires the clients to extend root-level trust to the NFS server.

3.13.4 - Configure NFS Servers

The steps in this section are appropriate for machines which operate as NFS servers.

3.13.4.1 - Configure the Exports File Restrictively

Linux’s NFS implementation uses the file /etc/exports to control what filesystems and directories may be accessed via NFS. (See the exports(5) manpage for more information about the format of this file.) The syntax of the exports file is not necessarily checked fully on reload, and syntax errors can leave your NFS configuration more open than intended. Therefore, exercise caution when modifying the file. The syntax of each line in /etc/exports is /DIR ipaddr1 (opt1 ,opt2 ) ipaddr2 (opt3 ) where /DIR is a directory or filesystem to export, ipaddrN is an IP address, netblock, hostname, domain, or netgroup to which to export, and optN is an option.

3.13.4.1.1 - Use Access Lists to Enforce Authorization Restrictions on Mounts

Edit /etc/exports. Ensure that each export line contains a set of IP addresses or hosts which are allowed to access that export. If no IP addresses or hostnames are specified on an export line, then that export is available to any remote host which requests it. All lines of the exports file should specify the hosts (or subnets, if needed) which are allowed to access the exported directory, so that unknown or remote hosts will be denied.

3.13.4.1.2 - Use Root-Squashing on All Exports

Edit /etc/exports. Ensure that no line contains the option no root squash. If a filesystem is exported using root squashing, requests from root on the client are considered to be unprivileged (mapped to a user such as nobody). This provides some mild protection against remote abuse of an NFS server. Root squashing is enabled by default, and should not be disabled.

3.13.4.1.3 - Restrict NFS Clients to Privileged Ports

Edit /etc/exports. Ensure that no line contains the option insecure. By default, Linux’s NFS implementation requires that all client requests be made from ports less than 1024. If your organization has control over machines connected to its network, and if NFS requests are prohibited at the border firewall, this offers some protection against malicious requests from unprivileged users. Therefore, the default should not be changed.

3.13.4.1.4 - Export Filesystems Read-Only if Possible

Edit /etc/exports. Ensure that every line contains the option ro and does not contain the option rw, unless there is an operational need for remote clients to modify that filesystem. If a filesystem is being exported so that users can view the files in a convenient fashion, but there is no need for users to edit those files, exporting the filesystem read-only removes an attack vector against the server. The default filesystem export mode is ro, so do not specify rw without a good reason.

3.13.4.2 - Allow Legitimate NFS Clients to Access the Server

Determine an appropriate network block, netwk , and network mask, mask , representing the machines on your network which must mount NFS filesystems from this server. Edit /etc/sysconfig/iptables. Add the following lines, ensuring that they appear before the final LOG and DROP lines for the RH-Firewall-1-INPUT chain: -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p udp --dport 111 -j ACCEPT -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p tcp --dport 111 -j ACCEPT -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p tcp --dport 2049 -j ACCEPT -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p tcp --dport lockd-port -j ACCEPT -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p udp --dport lockd-port -j ACCEPT -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p tcp --dport mountd-port -j ACCEPT -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p udp --dport mountd-port -j ACCEPT -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p tcp --dport rquotad-port -j ACCEPT -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p udp --dport rquotad-port -j ACCEPT -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p tcp --dport statd-port -j ACCEPT -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p udp --dport statd-port -j ACCEPT where the variable port numbers match those selected in Section 3.13.2.3 The default iptables configuration does not allow inbound access to any services. This modification will allow the specified block of remote hosts to initiate connections to the set of NFS daemons, while keeping all other ports on the server in their default protected state. See Section 2.5.5 for more information about iptables.

3.14 - DNS Server

Most organizations have an operational need to run at least one nameserver. However, there are many common attacks involving DNS, be configured defensively.

3.14.1 - Disable DNS Server if Possible

Is there an operational need for this machine to act as a DNS server for this site? If not, disable the software and remove it from the system: # chkconfig named off # yum erase bind DNS software should be disabled on any machine which does not need to be a nameserver. Note that the BIND DNS server software is not installed on RHEL5 by default. The remainder of this section discusses secure configuration of machines which must be nameservers.

3.14.2 - Run the BIND9 Software if DNS Service is Needed

It is highly recommended that the BIND9 software be used to provide DNS service. BIND is the Internet standard Unix nameserver, and, while it has had security problems in the past, it is also well-maintained and Red Hat is likely to quickly issue updates in response to any problems discovered in the future. In addition, BIND version 9 has new security features and more secure default settings than earlier versions. In particular, BIND version 4 is no longer recommended for production use, and BIND4 servers should be upgraded to a newer version as soon as possible.

3.14.3 - Isolate DNS from Other Services

This section discusses mechanisms for preventing the DNS server from interfering with other services. This is done both to protect the remainder of the network should a nameserver be compromised, and to make direct attacks on nameservers more difficult.

3.14.3.1 - Run DNS Software on Dedicated Servers if Possible

Since DNS is a high-risk service which must frequently be made available to the entire Internet, it is strongly recommended that no other services be offered by machines which act as organizational DNS servers.

3.14.3.2 - Run DNS Software in a chroot Jail

Install the bind-chroot package: # yum install bind-chroot Place a valid named.conf file inside the chroot jail: # cp /etc/named.conf /var/named/chroot/etc/named.conf # chown root:root /var/named/chroot/etc/named.conf # chmod 644 /var/named/chroot/etc/named.conf Create and populate an appropriate zone directory within the jail, based on the options directive. If your named.conf includes: options { directory "/path/to/DIRNAME "; ... } then copy that directory and its contents from the original zone directory: # cp -r /path/to/DIRNAME /var/named/chroot/DIRNAME Edit the file /etc/sysconfig/named. Add or correct the line: ROOTDIR=/var/named/chroot Chroot jails are not foolproof. However, they serve to make it more difficult for a compromised program to be used to attack the entire host. They do this by restricting a program’s ability to traverse the directory upward, so that files outside the jail are not visible to the chrooted process. Since RHEL5 supports a standard mechanism for placing BIND in a chroot jail, you should take advantage of this feature. Note: If you are running BIND in a chroot jail, then you should use the jailed named.conf as the primary nameserver configuration file. That is, when this guide recommends editing /etc/named.conf, you should instead edit /var/named/chroot/etc/named.conf.

3.14.3.3 - Configure Firewalls to Protect the DNS Server

Edit the file /etc/sysconfig/iptables. Add the following lines, ensuring that they appear before the final LOG and DROP lines for the RH-Firewall-1-INPUT chain: -A RH-Firewall-1-INPUT -m state --state NEW -p udp --dport 53 -j ACCEPT -A RH-Firewall-1-INPUT -m state --state NEW -p tcp --dport 53 -j ACCEPT These lines are necessary in order to allow remote machines to contact the DNS server. If this server is only available to the local network, it may be appropriate to insert a -s flag into this rule to allow traffic only from packets on the local network. See Section 3.5.1.2 for an example of such a modification. See Section 2.5.5 for general information about iptables.

3.14.4 - Protect DNS Data from Tampering or Attack

This section discusses DNS configuration options which make it more difficult for attackers to gain access to private DNS data or to modify DNS data.

3.14.4.1 - Run Separate DNS Servers for External and Internal Queries if Possible

Is it possible to run external and internal nameservers on separate machines? If so, follow the configuration guidance in this section. If not, see Section 3.14.4.2 for an alternate approach using BIND9. On the external nameserver, edit /etc/named.conf. Add or correct the following directives: options { allow-query { any; }; recursion no; ... }; zone "example.com " IN { ... }; On the internal nameserver, edit /etc/named.conf. Add or correct the following directives, where SUBNET is the numerical IP representation of your organization in the form xxx.xxx.xxx.xxx/xx: acl internal { SUBNET ; localhost; }; options { allow-query { internal; }; ... }; zone "internal.example.com " IN { ... }; Enterprise nameservers generally serve two functions. One is to provide public information about the machines in a domain for the benefit of outside users who wish to contact those machines, for instance in order to send mail to users in the enterprise, or to visit the enterprise’s external web page. The other is to provide nameservice to client machines within the enterprise. Client machines require both private information about enterprise machines (which may be different from the public information served to the rest of the world) and public information about machines outside the enterprise, which is used to send mail or visit websites outside of the organization. In order to provide the public nameservice function, it is necessary to share data with untrusted machines which request it — otherwise, the enterprise cannot be conveniently contacted by outside users. However, internal data should be protected from disclosure, and serving irrelevant public name queries for outside domains leaves the DNS server open to cache poisoning and other attacks. Therefore, local network nameservice functions should not be provided to untrusted machines. Separate machines should be used to fill these two functions whenever possible.

3.14.4.2 - Use Views to Partition External and Internal Information if Necessary

If it is not possible to run external and internal nameservers on separate physical machines, run BIND9 and simulate this feature using views. Edit /etc/named.conf. Add or correct the following directives (where SUBNET is the numerical IP representation of your organization in the form xxx.xxx.xxx.xxx/xx): acl internal { SUBNET ; localhost; }; view "internal-view" { match-clients { internal; }; zone "." IN { type hint; file "db.cache"; }; zone "internal.example.com " IN { ... }; }; view "external-view" { match-clients { any; }; recursion no; zone "example.com " IN { ... }; }; The view feature is provided by BIND9 as a way to allow a single nameserver to make different sets of data available to different sets of clients. If possible, it is always better to run external and internal nameservers on separate machines, so that even complete compromise of the external server cannot be used to obtain internal data or confuse internal DNS clients. However, this is not always feasible, and use of a feature like views is preferable to leaving internal DNS data entirely unprotected. Note: As shown in the example, database files which are required for recursion, such as the root hints file, must be available to any clients which are allowed to make recursive queries. Under typical circumstances, this includes only the internal clients which are allowed to use this server as a general-purpose nameserver.

3.14.4.3 - Disable Zone Transfers from the Nameserver if Possible

Is it necessary for a secondary nameserver to receive zone data via zone transfer from the primary server? If not, follow the instructions in this section. If so, see the next section for instructions on protecting zone transfers. Edit /etc/named.conf. Add or correct the following directive: options { allow-transfer { none; }; ... } If both the primary and secondary nameserver are under your control, or if you have only one nameserver, it may be possible to use an external configuration management mechanism to distribute zone updates. In that case, it is not necessary to allow zone transfers within BIND itself, so they should be disabled to avoid the potential for abuse.

3.14.4.4 - Authenticate Zone Transfers if Necessary

If it is necessary for a secondary nameserver to receive zone data via zone transfer from the primary server, follow the instructions here. Use dnssec-keygen to create a symmetric key file in the current directory: # cd /tmp # dnssec-keygen -a HMAC-MD5 -b 128 -n HOST dns.example.com Kdns.example.com .+aaa +iiiii This output is the name of a file containing the new key. Read the file to find the base64-encoded key string: # cat Kdns.example.com .+NNN +MMMMM .key dns.example.com IN KEY 512 3 157 base64-key-string Edit /etc/named.conf on the primary nameserver. Add the directives: key zone-transfer-key { algorithm hmac-md5; secret "base64-key-string "; }; zone "example.com " IN { type master; allow-transfer { key zone-transfer-key; }; ... } Edit /etc/named.conf on the secondary nameserver. Add the directives: key zone-transfer-key { algorithm hmac-md5; secret "base64-key-string "; }; server IP-OF-MASTER { keys { zone-transfer-key; }; }; zone "example.com " IN { type slave; masters { IP-OF-MASTER ; }; ... }; The BIND transaction signature (TSIG) functionality allows primary and secondary nameservers to use a shared secret to verify authorization to perform zone transfers. This method is more secure than using IP-based limiting to restrict nameserver access, since IP addresses can be easily spoofed. However, if you cannot configure TSIG between your servers because, for instance, the secondary nameserver is not under your control and its administrators are unwilling to configure TSIG, you can configure an allow-transfer directive with numerical IP addresses or ACLs as a last resort. Note: The purpose of the dnssec-keygen command is to create the shared secret string base64-key-string . Once this secret has been obtained and inserted into named.conf on the primary and secondary servers, the key files Kdns.example.com .+NNN +MMMMM .key and Kdns.example.com .+NNN +MMMMM .private are no longer needed, and may safely be deleted.

3.14.4.5 - Disable Dynamic Updates if Possible

Is there a mission-critical reason to enable the risky dynamic update functionality? If not: Edit /etc/named.conf. For each zone specification, correct the following directive if necessary: zone "example.com " IN { allow-update { none; }; ... } Dynamic updates allow remote servers to add, delete, or modify any entries in your zone file. Therefore, they should be considered highly risky, and disabled unless there is a very good reason for their use. If dynamic updates must be allowed, IP-based ACLs are insufficient protection, since they are easily spoofed. Instead, use TSIG keys (see the previous section for an example), and consider using the update-policy directive to restrict changes to only the precise type of change needed.

3.15 - FTPServer

FTP is a common method for allowing remote access to files. Like telnet, the FTP protocol is unencrypted, which means that passwords and other data transmitted during the session can be captured and that the session is vulnerable to hijacking. Therefore, running the FTP server software is not recommended. However, there are some FTP server configurations which may be appropriate for some environments, particularly those which allow only read-only anonymous access as a means of downloading data available to the public.

3.15.1 - Disable vsftpd if Possible

Is there a mission-critical reason for the machine to act as an FTP server? If not, disable vsftpd if it has been installed: # chkconfig vsftpd off

3.15.2 - Use vsftpd to Provide FTP Service if Necessary

If this machine must operate as an FTP server, install the vsftpd package via the standard channels: # yum install vsftpd After RHEL 2.1, Red Hat switched from distributing wu-ftpd with RHEL to distributing vsftpd. For security and for consistency with future Red Hat releases, the use of vsftpd is recommended.

3.15.3 - Configure vsftpd Securely

The primary vsftpd configuration file is /etc/vsftpd.conf, if that file exists, or /etc/vsftpd/vsftpd.conf if it does not. For the remainder of this section, the phrase “the configuration file” will refer to whichever of those files is appropriate for your environment.

3.15.3.1 - Enable Logging of All FTP Transactions

Edit the vsftpd configuration file. Add or correct the following configuration options: xferlog_std_format=NO log_ftp_protocol=YES The modifications above ensure that all commands sent to the ftp server are logged using the verbose vsftpd log format. The default vsftpd log file is /var/log/vsftpd.log. Note: If verbose logging to vsftpd.log is done, sparse logging of downloads to /var/log/xferlog will not also occur. However, the information about what files were downloaded is included in the information logged to vsftpd.log.

3.15.3.2 - Create Warning Banners for All FTP Users

Edit the vsftpd configuration file. Add or correct the following configuration options: banner_file=/etc/issue See Section 2.3.7 for an explanation of banner file use. This setting will cause the system greeting banner to be used for FTP connections as well.

3.15.3.3 - Restrict the Set of Users Allowed to Access FTP

This section describes how to disable non-anonymous (password-based) FTP logins, or, if it is not possible to do this entirely due to legacy applications, how to restrict insecure FTP login to only those users who have an identified need for this access.

3.15.3.3.1 - Restrict Access to Anonymous Users if Possible

Is there a mission-critical reason for users to transfer files to/from their own accounts using FTP, rather than using a secure protocol like SCP/SFTP? If not: Edit the vsftpd configuration file. Add or correct the following configuration option: local_enable=NO If non-anonymous FTP logins are necessary, follow the guidance in the remainder of this section to secure these logins as much as possible. The use of non-anonymous FTP logins is strongly discouraged. Since SSH clients and servers are widely available, and since SSH provides support for a transfer mode which resembles FTP in user interface, there is no good reason to allow password-based FTP access. See Section 3.5 for more information about SSH.

3.15.3.3.2 - Limit Users Allowed FTP Access if Necessary

If there is a mission-critical reason for users to access their accounts via the insecure FTP protocol, limit the set of users who are allowed this access. Edit the vsftpd configuration file. Add or correct the following configuration options: userlist_enable=YES userlist_file=/etc/vsftp.ftpusers userlist_deny=NO Edit the file /etc/vsftp.ftpusers. For each user USERNAME who should be allowed to access the system via ftp, add a line containing that user’s name. USERNAME If anonymous access is also required, add the anonymous usernames to /etc/vsftp.ftpusers as well: anonymous ftp Historically, the file /etc/ftpusers contained a list of users who were not allowed to access the system via ftp. It was used to prevent system users such as the root user from logging in via the insecure ftp protocol. However, when the configuration option userlist deny=NO is set, vsftpd interprets ftpusers as the set of users who are allowed to login via ftp. Since it should be possible for most users to access their accounts via secure protocols, it is recommended that this setting be used, so that non-anonymous ftp access can be limited to legacy users who have been explicitly identified.

3.15.3.4 - Disable FTP Uploads if Possible

Is there a mission-critical reason for users to upload files via FTP? If not: Edit the vsftpd configuration file. Add or correct the following configuration options: write_enable=NO If FTP uploads are necessary, follow the guidance in the remainder of this section to secure these transactions as much as possible. Anonymous FTP can be a convenient way to make files available for universal download. However, it is less common to have a need to allow unauthenticated users to place files on the FTP server. If this must be done, it is necessary to ensure that files cannot be uploaded and downloaded from the same directory.

3.15.3.5 - Place the FTP Home Directory on its Own Partition

By default, the anonymous FTP root is the home directory of the ftp user account. The df command can be used to verify that this directory is on its own partition. If there is a mission-critical reason for anonymous users to upload files, precautions must be taken to prevent these users from filling a disk used by other services.

3.15.3.6 - Configure Firewalls to Protect the FTP Server

Edit the file /etc/sysconfig/iptables. Add the following lines, ensuring that they appear before the final LOG and DROP lines for the RH-Firewall-1-INPUT chain: -A RH-Firewall-1-INPUT -m state --state NEW -p tcp --dport 21 -j ACCEPT Edit the file /etc/sysconfig/iptables-config. Ensure that the space-separated list of modules contains the FTP connection tracking module: IPTABLES_MODULES="ip_conntrack_ftp" These settings configure iptables to allow connections to an FTP server. The first line allows initial connections to the FTP server port. FTP is an older protocol which is not very compatible with firewalls. During the initial FTP dialogue, the client and server negotiate an arbitrary port to be used for data transfer. The ip conntrack ftp module is used by iptables to listen to that dialogue and allow connections to the data ports which FTP negotiates. This allows an FTP server to operate on a machine which is running a firewall.

3.16 - Web Server

The web server is responsible for providing access to content via the HTTP protocol. Web servers represent a significant security risk because: * The HTTP port is commonly probed by malicious sources * Web server software is very complex, and includes a long history of vulnerabilities * The HTTP protocol is unencrypted and vulnerable to passive monitoring The system’s default web server software is Apache 2 and is provided in the RPM package httpd.

3.16.1 - Disable Apache if Possible

If Apache was installed and activated, but the system does not need to act as a web server, then it should be disabled and removed from the system: # chkconfig httpd off # yum erase httpd

3.16.2 - Install Apache if Necessary

If the Apache web server must be run, follow these guidelines to install it defensively. Then follow the guidelines in the remainder of Section 3.16 to configure the web server machine and software as securely as possible.

3.16.2.1 - Install Apache Software Safely

Install the Apache 2 package from the standard Red Hat distribution channel: # yum install httpd Note: This method of installation is recommended over installing the “Web Server” package group during the system installation process. The Web Server package group includes many packages which are likely extraneous, while the command-line method installs only the required httpd package itself.

3.16.2.2 - Confirm Minimal Built-in Modules

The default Apache installation minimizes the number of modules that are compiled directly into the binary (core prefork http core mod so). This minimizes risk by limiting the capabilities allowed by the webserver. Query the set of compiled-in modules using the following command: $ httpd -l If the number of compiled-in modules is significantly larger than the aforementioned set, this guide recommends reinstallating Apache with a reduced configuration.

3.16.3 - Secure the Apache Configuration

The Apache configuration file is /etc/httpd/conf/httpd.conf. Apply the recommendations in the remainder of this section to this file.

3.16.3.1 - Restrict Information Leakage

The ServerTokens and ServerSignature directives determine how much information the web server discloses about the configuration of the system. ServerTokens Prod restricts information in page headers, returning only the word “Apache.” ServerSignature Off keeps Apache from displaying the server version on error pages. It is a good security practice to limit the information provided to clients. Add or correct the following directives in /etc/httpd/conf/httpd.conf so that as little information as possible is released: ServerTokens Prod ServerSignature Off

3.16.3.2 - Minimize Loadable Modules

A default installation of Apache includes a plethora of “dynamically shared objects” (DSO) that are loaded at run-time. Unlike the aforementioned “compiled-in” modules, a DSO can be disabled in the configuration file by removing the corresponding LoadModule directive. Note: A DSO only provides additional functionality if associated directives are included in the Apache configuration file. It should also be noted that removing a DSO will produce errors on Apache startup if the configuration file contains directives that apply to that module. Refer to http://httpd.apache.org/docs/ for details on which directives are associated with each DSO. Follow each DSO removal, the configuration can be tested with the following command to check if everything still works: # services httpd configtest The purpose of each of the modules loaded by default will now be addressed one at a time. If none of a module’s directives are being used, remove it.

3.16.3.2.1 - Apache Core Modules

These modules comprise a basic subset of modules that are likely needed for base Apache functionality; ensure they are not commented out in /etc/httpd/conf/httpd.conf: LoadModule auth_basic_module modules/mod_auth_basic.so LoadModule authn_default_module modules/mod_authn_default.so LoadModule authz_host_module modules/mod_authz_host.so LoadModule authz_user_module modules/mod_authz_user.so LoadModule authz_groupfile_module modules/mod_authz_groupfile.so LoadModule authz_default_module modules/mod_authz_default.so LoadModule log_config_module modules/mod_log_config.so LoadModule logio_module modules/mod_logio.so LoadModule setenvif_module modules/mod_setenvif.so LoadModule mime_module modules/mod_mome.so LoadModule autoindex_module modules/mod_autoindex.so LoadModule negotiation_module modules/mod_negotiation.so LoadModule dir_module modules/mod_dir.so LoadModule alias_module modules/mod_alias.so

3.16.3.2.2 - HTTP Basic Authentication

The following modules are necessary if this web server will provide content that will be restricted by a password. Authentication can be performed using local plain text password files (authn file), local DBM password files (authn dbm) or an LDAP directory (see Section 3.16.3.2.5). The only module required by the web server depends on your choice of authentication. Comment out the modules you don’t need from the following: LoadModule authn_file_module modules/mod_authn_file.so LoadModule authn_dbm_module modules/mod_authn_dbm.so authn alias allows for authentication based on aliases. authn anon allows anonymous authentication similar to that of anonymous ftp sites. authz owner allows authorization based on file ownership. authz dbm allows for authorization based on group membership if the web server is using DBM authentication. If the above functionality is unnecessary, comment out the related module: #LoadModule authn_alias_module modules/mod_authn_alias.so #LoadModule authn_anon_module modules/mod_authn_anon.so #LoadModule authz_owner_module modules/mod_authz_owner.so #LoadModule authz_dbm_module modules/mod_authz_dbm.so

3.16.3.2.3 - HTTP Digest Authentication

This module provides encrypted authentication sessions. However, this module is rarely used and considered experimental. Alternate methods of encrypted authentication are recommended, such as SSL (Section 3.16.4.1) If the above functionality is unnecessary, comment out the related module: #LoadModule auth_digest_module modules/mod_auth_digest.so

3.16.3.2.4 - mod rewrite

The mod rewrite module is very powerful and can protect against certain classes of web attacks. However, it is also very complex and has a significant history of vulnerabilities itself. If the above functionality is unnecessary, comment out the related module: #LoadModule rewrite_module modules/mod_rewrite.so

3.16.3.2.5 - LDAP Support

This module provides HTTP authentication via an LDAP directory. If the above functionality is unnecessary, comment out the related modules: #LoadModule ldap_module modules/mod_ldap.so #LoadModule authnz_ldap_module modules/mod_authnz_ldap.so If LDAP is to be used, SSL encryption (Section 3.16.4.1) should be used as well.

3.16.3.2.6 - Server Side Includes

Server Side Includes provide a method of dynamically generating web pages through the insertion of server-side code. However, the technology is also deprecated and introduces significant security concerns. If the above functionality is unnecessary, comment out the related module: #LoadModule include_module modules/mod_include.so If there is a critical need for Server Side Includes, they should be enabled with the option IncludesNoExec to prevent arbitrary code execution. Additionally, user supplied data should be encoded to prevent cross-site scripting vulnerabilities.

3.16.3.2.7 - MIME Magic

This module provides a second layer of MIME support that in most configurations is likely extraneous. If the above functionality is unnecessary, comment out the related module: #LoadModule mime_magic_module modules/mod_mime_magic.so

3.16.3.2.8 - WebDAV (Distributed Authoring and Versioning)

WebDAV is an extension of the HTTP protocol that provides distributed and collaborative access to web content. Due to a number of security concerns with WebDAV, its use is not recommended. If the above functionality is unnecessary, comment out the related modules: #LoadModule dav_module modules/mod_dav.so #LoadModule dav_fs_module modules/mod_dav_fs.so If there is a critical need for WebDAV, extra care should be taken in its configuration. Since DAV access allows remote clients to manipulate server files, any location on the server that is DAV enabled should be protected by encrypted authentication.

3.16.3.2.9 - Server Activity Status

This module provides real-time access to statistics on the internal operation of the web server. This is an unnecessary information leak and should be disabled. If the above functionality is unnecessary, comment out the related module: #LoadModule status_module modules/mod_status.so If there is a critical need for this module, ensure that access to the status page is properly restricted to a limited set of hosts in the status handler configuration.

3.16.3.2.10 - Web Server Configuration Display

This module creates a web page illustrating the configuration of the web server. This is an unnecessary security leak and should be disabled. If the above functionality is unnecessary, comment out the related module: #LoadModule info_module modules/mod_info.so If there is a critical need for this module, use the Location directive to provide an access control list to restrict access to the information.

3.16.3.2.11 - URL Correction on Misspelled Entries

This module attempts to find a document match by allowing one misspelling in an otherwise failed request. If the above functionality is unnecessary, comment out the related module: #LoadModule speling_module modules/mod_speling.so This functionality weakens server security by making site enumeration easier.

3.16.3.2.12 - User-specific directories

The UserDir directive provides user-specific directory translation, allowing URLs based on associated usernames. If the above functionality is unnecessary, comment out the related module: #LoadModule userdir_module modules/mod_userdir.so If there is a critical need for this module, include the line UserDir disabled root (at a minimum) in the configuration file. Ideally, UserDir should be disabled, and then enabled on a case-by-case basis for specific users that require this functionality. Note: A web server’s users can be trivially enumerated using this module.

3.16.3.2.13 - Proxy Support

This module provides proxying support, allowing Apache to forward requests and serve as a gateway for other servers. If the above functionality is unnecessary, comment out the related modules: #LoadModule proxy_module modules/mod_proxy.so #LoadModule proxy_balancer_module modules/mod_proxy_balancer.so #LoadModule proxy_ftp_module modules/mod_proxy_ftp.so #LoadModule proxy_http_module modules/mod_proxy_http.so #LoadModule proxy_connect_module modules/mod_proxy_connect.so If proxy support is needed, load proxy and the appropriate proxy protocol handler module (one of proxy http, proxy ftp, or proxy connect). Additionally, make certain that a server is secure before enabling proxying, as open proxy servers are a security risk. proxy balancer enables load balancing, but requires that mod status be enabled. Since mod status is not recommended, proxy balancer should be avoided as well.

3.16.3.2.14 - Cache Support

This module allows Apache to cache data, optimizing access to frequently accessed content. However, not only is it an experimental module, but it also introduces potential security flaws into the web server such as the possibility of circumventing Allow and Deny directives. If the above functionality is unnecessary, comment out the related modules: #LoadModule cache_module modules/mod_cache.so #LoadModule disk_cache_module modules/mod_disk_cache.so #LoadModule file_cache_module modules/mod_file_cache.so #LoadModule mem_cache_module modules/mod_mem_cache.so If caching is required, it should not be enabled for any limited-access content.

3.16.3.2.15 - CGI Support (and Related Modules)

This module allows HTML to interact with the CGI web programming language. If the above functionality is unnecessary, comment out the related modules: #LoadModule cgi_module modules/mod_cgi.so #LoadModule env_module modules/mod_env.so #LoadModule actions_module modules/mod_actions.so #LoadModule suexec_module modules/mod_suexec.so If the web server requires the use of CGI, enable the cgi module. If extended CGI functionality is required, include the appropriate modules. env allows for control of the environment passed to CGI scripts. actions allows CGI events to be triggered when files of a certain type are requested. su exec allows CGI scripts to run as a specified user/group instead of as the server’s user/group.

3.16.3.2.16 - Various Optional Components

The following modules perform very specific tasks, sometimes providing access to just a few additional directives. If this functionality is not required (or if you are not using these directives), comment out the associated module: * External filtering (response passed through external program prior to client delivery) #LoadModule ext_filter_module modules/mod_ext_filter.so * User-specified Cache Control and Expiration #LoadModule expires_module modules/mod_expires.so * Compression Output Filter (provides content compression prior to client delivery) #LoadModule deflate_module modules/mod_deflate.so * HTTP Response/Request Header Customization #LoadModule headers_module modules/mod_headers.so * User activity monitoring via cookies #LoadModule usertrack_module modules/mod_usertrack.so * Dynamically configured mass virtual hosting #LoadModule vhost_alias_module modules/mod_vhost_alias.so

3.16.3.3 - Minimize Configuration Files Included

The Include directive directs Apache to load supplementary configuration files from a provided path. The default configuration loads all files that end in .conf from the /etc/httpd/conf.d directory. To restrict excess configuration, the following line should be commented out and replaced with Include directives that only reference required configuration files: #Include conf.d/*.conf If the above change was made, ensure that the SSL encryption remains loaded by explicitly including the corresponding configuration file: (see Section 3.16.4.1 for further details on SSL configuration) Include conf.d/ssl.conf If PHP is necessary, a similar alteration must be made: (see Section 3.16.4.4.1 for further details on PHP configuration) Include conf.d/php.conf

3.16.3.4 - Directory Restrictions

The Directory tags in the web server configuration file allow finer grained access control for a specified directory. All web directories should be configured on a case-by-case basis, allowing access only where needed.

3.16.3.4.1 - Restrict Root Directory

The Apache root directory should always have the most restrictive configuration enabled. <Directory / > Options None AllowOverride None Order allow,deny </Directory>

3.16.3.4.2 - Restrict Web Directory

The default configuration for the web (/var/www/html) Directory allows directory indexing (Indexes)and the following of symbolic links (FollowSymLinks). Neither of these is recommended. The /var/www/html directory hierarchy should not be viewable via the web, and symlinks should only be followed if the owner of the symlink also owns the linked file. Ensure that this policy is adhered to by altering the related section of the configuration: <Directory "/var/www/html"> # ... Options SymLinksIfOwnerMatch # ... </Directory>

3.16.3.4.3 - Restrict Other Critical Directories

All accessible web directories should be configured with similar restrictive settings. The Options directive should be limited to necessary functionality and the AllowOverride directive should be used only if needed. The Order and Deny access control tags should be used to deny access by default, allowing access only where necessary.

3.16.3.5 - Configure Authentication if Applicable

Basic authentication is handled in plaintext over the network. Therefore, all login attempts are vulnerable to password sniffing. For increased protection against passive monitoring, encrypted authentication over a secure channel such as SSL (Section 3.16.4.1) is recommended. 1. Set up a password file. If a password file doesn’t yet exist, one must be generated with the following command: # htpasswd -cs passwdfile user WARNING: This command will overwrite an existing file at this location. Once a password file has been generated, subsequent users can be added with the following command: # htpasswd -s passwdfile user 2. Optionally, set up a group file (if using group authentication). The group file is a plain text file of the following format (each group is on its own line, followed by a colon and a list of users that belong to that group, separated by spaces): group : user1 user2 group2 : user3 3. Modify file permissions so that Apache can read the group and passwd files: # chgrp apache passwdfile groupfile # chmod 640 passwdfile groupfile 4. Turn on authentication for desired directories Add the following options inside the appropriate Directory tag: * For single-user authentication: <Directory "directory "> # ... AuthName "Private Data" AuthType Basic AuthUserFile passwdfile require user user # ... </Directory> * For multiple-user authentication restricted by groups: <Directory "directory "> # ... AuthName "Private Data" AuthType Basic AuthUserFile passwdfile AuthGroupFile groupfile require group group # ... </Directory> * For multiple-user authentication restricted by valid user accounts: <Directory "directory "> # ... AuthName "Private Data" AuthType Basic AuthUserFile passwdfile require valid-user # ... </Directory> The AuthName directive specifies a label for the protected content. The AuthType directive specifies the kind of authentication (if using Digest authentication, this line would instead read AuthType Digest) The AuthUserFile and AuthGroupFile directives point to the password and group files (if using Digest authentication, these directives would instead be AuthDigestFile and AuthDigestGroupFile.) The require user directive restricts access to a single user. The require group directive restricts access to multiple users in a designated group. The short-hand require valid-user directive restricts access to any user in the passwdfile Note: Make sure the AuthUserFile and AuthGroupFile locations are outside the web server document tree to prevent remote clients from having access to restricted usernames and passwords. This guide recommends /etc/httpd/conf as a location for these files.

3.16.3.6 - Limit Available Methods

Web server methods are defined in section 9 of RFC 2616 (http://www.ietf.org/rfc/rfc2616.txt). If a web server does not require the implementation of all available methods, they should be disabled. Note: GET and POST are the most common methods. A majority of the others are limited to the WebDAV protocol. <Directory /var/www/html> # ... # Only allow specific methods (this command is case-sensitive!) <LimitExcept GET POST> Order allow,deny </LimitExcept> # ... </Directory>

3.16.4 - Use Appropriate Modules to Improve Apaches Security'

Among the modules available for Apache are several whose use may improve the security of the web server installation. This section recommends and discusses the deployment of security-relevant modules.

3.16.4.1 - Deploy mod ssl

Because HTTP is a plain text protocol, all traffic is susceptible to passive monitoring. If there is a need for confidentiality, SSL should be configured and enabled to encrypt content. Note: mod nss is a FIPS 140-2 certified alternative to mod ssl. The modules share a considerable amount of code and should be nearly identical in functionality. If FIPS 140-2 validation is required, then mod nss should be used. If it provides some feature or its greater compatibility is required, thenmod ssl should be used.

3.16.4.1.1 - Install mod ssl

Install mod ssl: # yum install mod ssl

3.16.4.1.2 - Create an SSL Certificate

On your CA (if you are using your own) or on another physically secure system, generate a key pair for the web server: # cd /etc/pki/tls/certs # openssl genrsa -des3 -out httpserverkey.pem 2048 When prompted, enter a strong, unique passphrase to protect the web server key pair. Next, generate a Certificate Signing Request (CSR) from the key for the CA: # openssl req -new -key httpserverkey.pem -out httpserver.csr Enter the passphrase for the web server key pair and then fill out the fields as completely as possible (or hit return to accept defaults); the Common Name field is especially important. It must match the fullyqualified domain name of your server exactly (e.g. www.example.com) or the certificate will not work. The /etc/pki/tls/openssl.conf file will determine which other fields (e.g. Country Name, Organization Name, etc) must match between the server request and the CA. Leave the challenge password and an optional company name blank. Next, the web server CSR must be signed to create the web server certificate. You can either send the CSR to an established CA or sign it with your CA. To sign httpserver.csr using your CA: # openssl ca -in httpserver.csr -out httpservercert.pem When prompted, enter the CA passphrase to continue and then complete the process. The httpservercert. pem certificate needed to enable SSL on the web server is now in the directory. Finally, the web server key and certificate file need to be moved to the web server. Use removable media if possible. Place the server key and certificate file in /etc/pki/tls/http/, naming them serverkey.pem and servercert.pem, respectively.

3.16.4.1.3 - Install SSL Certificate

Add or modify the configuration file /etc/httpd/conf.d/ssl.conf to match the following: # establish new listening port Listen 443 # seed appropriately SSLRandomSeed startup file:/dev/urandom 1024 SSLRandomSeed connect file:/dev/urandom 1024 <VirtualHost site-on-certificate.com:443> # Enable SSL SSLEngine On # Path to server certificate + private key SSLCertificateFile /etc/pki/tls/http/servercert.pem SSLCertificateKeyFile /etc/pki/tls/http/serverkey.pem SSLProtocol All -SSLv2 # Weak ciphers and null authentication should be denied unless absolutely necessary # (and even then, such cipher weakening should occur within a Location enclosure) SSLCipherSuite HIGH:MEDIUM:!aNULL:+MD5 </VirtualHost> Ensure that all directories that house SSL content are restricted to SSL access only in /etc/httpd/conf/ httpd.conf: <Directory /var/www/html/secure> # require SSL for access SSLRequireSSL SSLOptions +StrictRequire # require domain to match certificate domain SSLRequire %{HTTP HOST} eq "site-on-certificate.com" # rather than reply with 403 error, redirect user to appropriate site # this is OPTIONAL - uncomment to apply # ErrorDocument 403 https://site-on-certificate.com </Directory>

3.16.4.2 - Deploy mod security

mod security provides an application level firewall for Apache. Following the installation of mod security with the base ruleset, specific configuration advice can be found at http://www.modsecurity.org/ to design a policy that best matches the security needs of the web applications.

3.16.4.2.1 - Install mod security

Install mod security: # yum install mod_security

3.16.4.2.2 - Configure mod security Filtering

mod security supports a significant number of options, far too many to be fully covered in this guide. However, the following list comprises a smaller subset of suggested filters to be added to /etc/httpd/conf/ httpd.conf: # enable mod security SecFilterEngine On # enable POST filtering SecFilterScanPost On # Make sure that URL encoding is valid SecFilterCheckURLEncoding On # Accept almost all byte values SecFilterForceByteRange 1 255 # Prevent directory traversal SecFilter "\.\./" # Filter on specific system specific paths SecFilter /etc/passwd SecFilter /bin/ # Prevent cross-site scripting SecFilter "<[[:space:]]* script" # Prevent SQL injection SecFilter "delete[[:space:]]+from" SecFilter "insert[[:space:]]+into" SecFilter "select.+from"

3.16.4.3 - Use Denial-of-Service Protection Modules

Denial-of-service attacks are difficult to detect and prevent while maintaining acceptable access to authorized users. However, there are a number of traffic-shaping modules that attempt to address the problem. Well-known DoS protection modules include: mod_throttle mod_bwshare mod_limitipconn mod_dosevasive It is recommended that denial-of-service prevention be implemented for the web server. However, this guide leaves specific configuration details to the discretion of the reader.

3.16.4.4 - Configure Supplemental Modules Appropriately

Any required functionality added to the web server via additional modules should be configured appropriately.

3.16.4.4.1 - Configure PHP Securely

PHP is a widely used and often misconfigured server-side scripting language. It should be used with caution, but configured appropriately when needed. Make the following changes to /etc/php.ini: # Do not expose PHP error messages to external users display_errors = Off # Enable safe mode safe_mode = On # Only allow access to executables in isolated directory safe_mode_exec_dir = php-required-executables-path # Limit external access to PHP environment safe_mode_allowed_env_vars = PHP_ # Restrict PHP information leakage expose_php = Off # Log all errors log_errors = On # Do not register globals for input data register_globals = Off # Minimize allowable PHP post size post_max_size = 1K # Ensure PHP redirects appropriately cgi.force_redirect = 0 # Disallow uploading unless necessary file_uploads = Off # Disallow treatment of file requests as fopen calls allow_url_fopen = Off # Enable SQL safe mode sql.safe_mode = On

3.16.5 - Configure Operating System to Protect Web Server

The following configuration steps should be taken on the machine which hosts the web server, in order to provide as safe an environment as possible for the web server.

3.16.5.1 - Restrict File and Directory Access

Minimize access to critical Apache files and directories: # chmod 511 /usr/sbin/httpd # chmod 750 /var/log/httpd/ # chmod 750 /etc/httpd/conf/ # chmod 640 /etc/httpd/conf/* # chgrp -R apache /etc/httpd/conf

3.16.5.2 - Configure iptables to Allow Access to the Web Server

Edit /etc/sysconfig/iptables. Add the following lines, ensuring that they appear before the final LOG and DROP lines for the RH-Firewall-1-INPUT chain: -A RH-Firewall-1-INPUT -m state --state NEW -p tcp --dport 80 -j ACCEPT -A RH-Firewall-1-INPUT -m state --state NEW -p tcp --dport 443 -j ACCEPT The default Iptables configuration does not allow inbound access to the HTTP (80) and HTTPS (443) ports used by the web server. This modification allows that access, while keeping other ports on the server in their default protected state. See Section 2.5.5 for more information about Iptables.

3.16.5.3 - Run Apache in a chroot Jail if Possible

Putting Apache in a chroot jail minimizes the damage done by a potential break-in by isolating the web server to a small section of the filesystem. In order to configure Apache to run from a chroot directory, edit the Apache configuration file, /etc/httpd/ conf/httpd.conf, and add the directive: SecChrootDir /chroot/apache It is also necessary to place all files required by Apache inside the filesystem rooted at /chroot/apache , including Apache’s binaries, modules, configuration files, and served web pages. The details of this configuration are beyond the scope of this guide.

3.16.6 - Additional Resources

Further resources should be consulted if your web server requires more extensive configuration guidance, especially if particular applications need to be secured. In particular, [26] is recommended as a more comprehensive guide to securing Apache.

3.17 - IMAP and POP3 Server

Dovecot provides IMAP and POP3 services. It is not installed by default. The project page at http://www. dovecot.org contains more detailed information about Dovecot configuration.

3.17.1 - Disable Dovecot if Possible

If the system does not need to operate as an IMAP or POP3 server, disable and remove Dovecot if it was installed: # chkconfig dovecot off # yum erase dovecot

3.17.2 - Configure Dovecot if Necessary

Dovecot’s main configuration file is /etc/dovecot.conf. The settings which appear, commented out, in the file are the defaults.

3.17.2.1 - Support Only the Necessary Protocols

Edit /etc/dovecot.conf. Add or correct the following lines, replacing PROTOCOL with only the subset of protocols (imap, imaps, pop3, pop3s) required: protocols = PROTOCOL Dovecot supports the IMAP and POP3 protocols, as well as SSL-protected versions of those protocols. Configure the Dovecot server to support only the protocols needed by your site. If possible, require SSL protection for all transactions. The SSL protocol variants listen on alternate ports (995 instead of 110 for pop3s, and 993 instead of 143 for imaps), and require SSL-aware clients. An alternate approach is to listen on the standard port and require the client to use the STARTTLS command before authenticating.

3.17.2.2 - Enable SSL Support

SSL should be used to encrypt network traffic between the Dovecot server and its clients. Users must authenticate to the Dovecot server in order to read their mail, and passwords should never be transmitted in clear text. In addition, protecting mail as it is downloaded is a privacy measure, and clients may use SSL certificates to authenticate the server, preventing another system from impersonating the server. See Section 2.5.6 for general SSL information, including the setup of a Certificate Authority (CA).

3.17.2.2.1 - Create an SSL Certificate

Note: The following steps should be performed on your CA system, and not on the Dovecot server itself. If you will have a commercial CA sign certificates, then these steps should be performed on a separate, physically secure system devoted to that purpose. On your CA (if you are using your own) or on another physically secure system, generate a key pair for the Dovecot server: # cd /etc/pki/tls/certs # openssl genrsa -out imapserverkey.pem 2048 Next, generate a certificate signing request (CSR) for the CA to sign, making sure to enter the server’s fully-qualified domain name when prompted for the Common Name: # openssl req -new -key imapserverkey.pem -out imapserver.csr Next, the mail server CSR must be signed to create the Dovecot server certificate. You can either send the CSR to an established CA or sign it with your CA. To sign imapserver.csr using your CA: # openssl ca -in imapserver.csr -out imapservercert.pem This step creates a private key, imapserverkey.pem, and a public certificate, imapservercert.pem. The Dovecot server will use these to prove its identity by demonstrating that it has a certificate which has been signed by a CA. POP3 or IMAP clients at your site should only be willing to provide users’ credentials to a server they can authenticate.

3.17.2.2.2 - Install the SSL Certificate

Create the PKI directory for POP and IMAP certificates if it does not already exist: # mkdir /etc/pki/tls/imap # chown root:root /etc/pki/tls/imap # chmod 755 /etc/pki/tls/imap Using removable media or some other secure transmission format, install the files generated in the previous step onto the Dovecot server: * /etc/pki/tls/imap/serverkey.pem: the private key imapserverkey.pem * /etc/pki/tls/imap/servercert.pem: the certificate file imapservercert.pem Verify the permissions on these files: # chown root:root /etc/pki/tls/imap/serverkey.pem # chown root:root /etc/pki/tls/imap/servercert.pem # chmod 600 /etc/pki/tls/imap/serverkey.pem # chmod 600 /etc/pki/tls/imap/servercert.pem Verify that the CA’s public certificate file has been installed as /etc/pki/tls/CA/cacert.pem, and has the correct permissions: # chown root:root /etc/pki/tls/CA/cacert.pem # chmod 644 /etc/pki/tls/CA/cacert.pem

3.17.2.2.3 - Configure Dovecot to Use the SSL Certificate

Edit /etc/dovecot.conf and add or correct the following lines (ensuring they reference the appropriate files): ssl_cert_file = /etc/pki/tls/imap/servercert.pem ssl_key_file = /etc/pki/tls/imap/serverkey.pem ssl_ca_file = /etc/pki/tls/CA/cacert.pem These options tell Dovecot where to find the TLS configuration, allowing clients to make encrypted connections.

3.17.2.2.4 - Disable Plaintext Authentication

To prevent Dovecot from attempting plaintext authentication of clients, edit /etc/dovecot.conf and add or correct the following line: disable_plaintext_auth = yes The disable plaintext auth command disallows login-related commands until an encrypted session has been negotiated using SSL. If client compatibility requires you to allow connections to the pop3 or imap ports, rather than the alternate SSL ports, you should use this command to require STARTTLS before authentication.

3.17.2.3 - Enable Dovecot Options to Protect Against Code Flaws

Edit /etc/dovecot.conf and add or correct the following line: login_process_per_connection = yes mail_drop_priv_before_exec = yes IMAP and POP3 are remote authenticated protocols, meaning that the server must accept remote connections from anyone, but provide substantial services only to clients who have successfully authenticated. To protect against security problems, Dovecot splits these functions into separate server processes. The imap-login and/or pop3-login processes accept connections from unauthenticated users, and only spawn imap or pop3 processes on successful authentication. However, the imap-login and pop3-login processes themselves may contain vulnerabilities. Since each of these processes operates as a daemon, handling multiple sequential client connections from different users, bugs in the code could allow unauthenticated users to steal credential data. If the login process per connection option is enabled, then a separate imap-login or pop3-login process is created for each new connection, protecting against this class of problems. This option has an efficiency cost, but is strongly recommended. If the mail drop priv before exec option is on, the imap-login or pop3-login process will drop privileges to the user’s ID after authentication and before executing the imap or pop3 process itself. Under some very limited circumstances, this could protect against privilege escalation by authenticated users. However, if the mail executable option is used to run code before starting each user’s session, it is important to drop privileges to prevent the custom code from running as root.

3.17.2.4 - Allow IMAP Clients to Access the Server

Edit /etc/sysconfig/iptables. Add the following line, ensuring that it appears before the final LOG and DROP lines for the RH-Firewall-1-INPUT chain: -A RH-Firewall-1-INPUT -m state --state NEW -p tcp --dport 143 -j ACCEPT The default iptables configuration does not allow inbound access to any services. This modification will allow remote hosts to initiate connections to the IMAP daemon, while keeping all other ports on the server in their default protected state. See Section 2.5.5 for more information about iptables.

3.18 - Samba(SMB) Microsoft Windows File Sharing Server

When properly configured, the Samba service allows Linux machines to provide file and print sharing to Microsoft Windows machines. There are two software packages that provide Samba support. The first, samba-client, provides a series of command line tools that enable a client machine to access Samba shares. The second, simply labeled samba, provides the Samba service. It is this second package that allows a Linux machine to act as an Active Directory server, a domain controller, or as a domain member. Only the samba-client package is installed by default.

3.18.1 - Disable Samba if Possible

If the Samba service has been enabled and will not be used, disable it: # chkconfig smb off Even after the Samba server package has been installed, it will remain disabled. Do not enable this service unless it is absolutely necessary to provide Microsoft Windows file and print sharing functionality.

3.18.2 - Configure Samba if Necessary

All settings for the Samba daemon can be found in /etc/samba/smb.conf. Settings are divided between a [global] configuration section and a series of user created share definition sections meant to describe file or print shares on the system. By default, Samba will operate in user mode and allow client machines to access local home directories and printers. It is recommended that these settings be changed or that additional limitations be set in place.

3.18.2.1 - Testing the Samba Configuration File

To test the configuration file for syntax errors, use the testparm command. It will also list all settings currently in place, including defaults that may not appear in the configuration file. # testparm -v

3.18.2.2 - Choosing the Appropriate security Parameter

There are two kinds of security in Samba, share-level (share) and user-level. User-level security is further subdivided into four separate implementations: user, domain, ads, and server. It is recommended that the share and server security modes not be used. In share security, everyone is given the same password for each share, preventing individual user accountability. server security mode has been superseded by the domain and ads security modes. It may now be considered obsolete. The security parameter is set in the [global] section of the Samba configuration file. It determines how the server will handle user names and passwords. Some security modes require additional parameters, such as workgroup, realm, or password server names. All security modes will require that each remote user have a matching local account. One workaround to this problem is to use the winbindd daemon. Please consult the official Samba documentation to learn more.

3.18.2.2.1 - Use user Security for Servers Not in a Domain Context

This is the default setting with a new Samba installation and the best choice when operating outside of a domain security context. The relevant parameters in /etc/samba/smb.conf will read as follows: security = user workgroup = MYGROUP Set the value of workgroup so that it matches the value of other machines on the network. In user mode, authentication requests are handled locally and not passed on to a separate authentication server. This is the desired behavior for standalone servers and domain controllers.

3.18.2.2.2 - Use domain Security for Servers in a Domain Context

When using Samba as a Primary or Backup Domain Controller, use security = user, not security = domain. This tells Samba that the local machine is hosting the authentication backend. First, change the security parameter to domain. Next, set the workgroup and netbios name parameters (if necessary): security = domain workgroup = WORKGROUP netbios name = NETBIOSNAME domain mode is used for any machine that will act as a domain member server. It lets Samba know that the authentication information it needs can be found on another machine. Primary and Backup Domain Controllers host copies of this information. Samba will try to automatically determine which machine it should authenticate against on a domain network. If this detection fails, it may be necessary to specify the location manually. Unlike the Microsoft Windows implementation of the SMB standard, a Samba machine can freely change roles within a domain without requiring that the machine be reinstalled (such roles include primary and backup domain controllers, domain member servers, and ordinary domain workstations). However, there are some limitations on how each machine can fulfill each role in a mixed network.

3.18.2.2.3 - Use ads (Active Directory Service) Security For Servers in an ADS Domain

Context The security mode ads enables a Samba machine to act as an ADS domain member server. Since ADS requires Kerberos, be sure to set the realm parameter appropriately and configure the local copy of Kerberos. If necessary, it is also possible to manually set the password server parameter. security = ads realm = MY REALM password server = your.kerberos.server Currently, it is possible to act as an Active Directory domain member server, but not as a domain controller. Be sure to operate in mixed mode. Native mode may not work yet in current versions of Samba. Future support for ADS should be forthcoming in Samba 4. See the Samba project web site at http://www.samba.org for more details.

3.18.2.3 - Disable Guest Access and Local Login Support

Do not allow guest users to access local file or printer shares. In global or in each share, set the parameter guest ok to no: [share] guest ok = no It is safe to disable local login support for remote Samba users. Consider changing the add user account script to set remote user shells to /sbin/nologin.

3.18.2.4 - Disable Root Access

Administrators should not use administrator accounts to access Samba file and printer shares. If possible, disable the root user and the wheel administrator group: [share] invalid users = root @wheel If administrator accounts cannot be disabled, ensure that local machine passwords and Samba service passwords do not match. Typically, administrator access is required when Samba must create user and machine accounts and shares. Domain member servers and standalone servers may not need administrator access at all. If that is the case, add the invalid users parameter to [global] instead.

3.18.2.5 - Set the Allowed Authentication Negotiation Levels

By default, Samba will attempt to negotiate with Microsoft Windows machines to set a common communication protocol. Whenever possible, be sure to disable LANMAN authentication, as it is far weaker than the other supported protocols. [global] client lanman auth = no Newer versions of Microsoft Windows may require the use of NTLMv2. NTLMv2 is the preferred protocol for authentication, but since older machines do not support it, Samba has disabled it by default. If possible, reenable it. [global] client ntlmv2 auth = yes For the sake of backwards compatibility, most modern Windows machines will still allow other machines to communicate with them over weak protocols such as LANMAN. On Samba, by enabling NTLMv2, you are also disabling LANMAN and NTLMv1. If NTLMv1 is required, it is still possible to individually disable LANMAN.

3.18.2.6 - Let Domain Controllers Create Machine Trust Accounts On-the-Fly

Add or correct an add machine script entry to the [global] section of /etc/samba/smb.conf to allow Samba to dynamically create Machine Trust Accounts: [global] add machine script = /usr/sbin/useradd -n -g machines -d /dev/null -s /sbin/nologin %u Make sure that the group machines exists. If not, add it with the following command: /usr/sbin/groupadd machines When acting as a PDC, it becomes necessary to create and store Machine Trust Accounts for each machine that joins the domain. On a Microsoft Windows PDC, this account is created with the Server Manager tool, but on a Samba PDC, two accounts must be created. The first is the local machine account, and the second is the Samba account. For security purposes, it is recommended to let Samba create these accounts on-the-fly. When Machine Trust Accounts are created manually, there is a small window of opportunity in which a rogue machine could join the domain in place of the new server.

3.18.2.7 - Restrict Access to the [IPC$] Share

Limit access to the [IPC$] share so that only machines in your network will be able to connect to it: [IPC$] hosts allow = 192.168.1. 127.0.0.1 hosts deny = 0.0.0.0/0 The [IPC$] share allows users to anonymously fetch a list of shared resources from a server. It is intended to allow users to browse the list of available shares. It also can be used as a point of attack into a system. Disabling it completely may break some functionality, so it is recommended that you merely limit access to it instead.

3.18.2.8 - Restrict File Sharing

Only users with local user accounts will be able to log in to Samba shares by default. Shares can be limited to particular users or network addresses. Use the hosts allow and hosts deny directives accordingly, and consider setting the valid users directive to a limited subset of users or to a group of users. Separate each address, user, or user group with a space as follows: [share] hosts allow = 192.168.1. 127.0.0.1 valid users = userone usertwo @usergroup It is also possible to limit read and write access to particular users with the read list and write list options, though the permissions set by the system itself will override these settings. Set the read only attribute for each share to ensure that global settings will not accidentally override the individual share settings. Then, as with the valid users directive, separate each user or group of users with a space: [share] read only = yes write list = userone usertwo @usergroup The Samba service is only required for sharing files and printers with Microsoft Windows workstations, and even then, other options may exist. Do not use the Samba service to share files between Unix or Linux machines.

3.18.2.9 - Restrict Printer Sharing

By default, Samba utilizes the CUPS printing service to enable printer sharing with Microsoft Windows workstations. If there are no printers on the local machine, or if printer sharing with Microsoft Windows is not required, disable the printer sharing capability by commenting out the following lines, found in /etc/ samba/smb.conf: [global] ; load printers = yes ; cups options = raw [printers] comment = All Printers path = /usr/spool/samba browseable = no guest ok = no writable = no printable = yes There may be other options present, but these are the only options enabled and uncommented by default. Removing the [printers] share should be enough for most users. If the Samba printer sharing capability is needed, consider disabling the Samba network browsing capability or restricting access to a particular set of users or network addresses. Set the valid users parameter to a small subset of users or restrict it to a particular group of users with the shorthand @. Separate each user or group of users with a space. For example, under the [printers] share: [printers] valid users = user @printerusers The CUPS service is capable of sharing printers with other Unix and Linux machines on the local network without the Samba service. The Samba service is only required when a Microsoft Windows machine needs printer access on a Unix or Linux host.

3.18.2.10 - Configure iptables to Allow Access to the Samba Server

Determine an appropriate network block, netwk , and network mask, mask , representing the machines on your network which should operate as clients of the Samba server. Edit /etc/sysconfig/iptables. Add the following lines, ensuring that they appear before the final LOG and DROP lines for the RH-Firewall-1-INPUT chain: -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p tcp --dport 137 -j ACCEPT -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p tcp --dport 138 -j ACCEPT -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p tcp --dport 139 -j ACCEPT -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p tcp --dport 445 -j ACCEPT The default Iptables configuration does not allow inbound access to the ports used by the Samba service. This modification allows that access, while keeping other ports on the server in their default protected state. Since these ports are frequent targets of network scanning attacks, restricting access to only the network segments which need to access the Samba server is strongly recommended. See Section 2.5.5 for more information about Iptables.

3.18.3 - Avoid the Samba Web Administration Tool (SWAT)

SWAT is a web based configuration tool provided by the Samba team that enables both local and remote configuration management. It is not installed by default. It is recommended that SWAT not be used, as it requires the use of a Samba administrator account and sends that password in the clear over a network connection. If SWAT is absolutely required, limit access to the local machine or tunnel SWAT connections through SSL with stunnel.

3.19 - Proxy Server

A proxy server is a very desirable target for a potential adversary because much (or all) sensitive data for a given infrastructure may flow through it. Therefore, if one is required, the machine acting as a proxy server should be dedicated to that purpose alone and be stored in a physically secure location. The system’s default proxy server software is Squid, and provided in an RPM package of the same name.

3.19.1 - Disable Squid if Possible

If Squid was installed and activated, but the system does not need to act as a proxy server, then it should be disabled and removed: # chkconfig squid off # yum erase squid

3.19.2 - Configure Squid if Necessary

The Squid configuration file is /etc/squid/squid.conf. The following recommendations can be applied to this file. Note: If a particular tag is not present in the configuration file, Squid falls back to the default setting (which is often illustrated by a comment).

3.19.2.1 - Listen on Uncommon Port

The default listening port for the Squid service is 3128. As such, it is frequently scanned by adversaries looking for proxy servers. Select an arbitrary (but uncommon) high port to use as the Squid listening port and make the corresponding change to the configuration file: http port port Run the following command to add a new SELinux port mapping for the service: # semanage port -a -t http_cache_port_t -p tcp port

3.19.2.2 - Verify Default Secure Settings

Several security-enhancing settings in the Squid configuration file are enabled by default, but appear as comments in the configuration file (as mentioned in Section 3.19.2). In these instances, the explicit directive is not present, which means it is implicitly enabled. If you are operating with a default configuration file, this section can be ignored. Ensure that the following security settings are NOT explicitly changed from their default values: ftp_passive on ftp_sanitycheck on check_hostnames on request_header_max_size 20 KB reply_header_max_size 20 KB cache_effective_user squid cache_effective_group squid ignore_unknown_nameservers on ftp passive forces FTP passive connections. ftp sanitycheck performs additional sanity checks on FTP data connections. check hostnames ensures that hostnames meet RFC compliance. request header max size and reply header max size place an upper limit on HTTP header length, precautions against denial-of-service and buffer overflow vulnerabilities. cache effective user and cache effective group designate the EUID and EGID of Squid following initialization (it is essential that the EUID/EGID be set to an unprivileged sandbox account). ignore unknown nameservers checks to make sure that DNS responses come from the same IP the request was sent to.

3.19.2.3 - Change Default Insecure Settings

The default configuration settings for the following tags are considered to be weak security and NOT recommended. Add or modify the configuration file to include the following lines: allow_underscore off httpd_suppress_version_string on forwarded_for off log_mime_hdrs on allow underscore enforces RFC 1034 compliance on hostnames by disallowing the use of underscores. httpd suppress version string prevents Squid from revealing version information in web headers and error pages. forwarded for reveals proxy client IP addresses in HTTP headers and should be disabled to prevent the leakage of internal network configuration details. log mime hdrs enables logging of HTTP response/request headers.

3.19.2.4 - Configure Authentication if Applicable

Note: Authentication cannot be used in the case of transparent proxies due to limitations of the TCP/IP protocol. Similar to web servers, two of the available options are Basic and Digest authentication. The other options are NTLM and Negotiate authentication. As noted in Section 3.16.3.5, Basic authentication transmits passwords in plain-text and is susceptible to passive monitoring. If network sniffing is a concern, basic authentication should not be used. Negotiate is the newest and most secure protocol. It attempts to use Kerberos authentication and falls back to NTLM if it cannot. It should be noted that Kerberos requires a third-party Key Distribution Center (KDC) to function properly, whereas the other methods of authentication are two-party schemes. Squid also offers the ability to choose a custom external authenticator. Designating an external authenticator (also known as a “helper” module) allows Squid to offer pluggable third-party authentication schemes. LDAP is one example of a helper module that exists and is in use today. There are comments under the auth param tag inside /etc/squid/squid.conf that provide extensive detail on how to configure each of these methods. If authentication is necessary, choose a method of authentication and configure appropriately. The recommended minimum configurations illustrated for each method are acceptable. To force an ACL (as discussed in Section 3.19.2.5) to require authentication, use the following directive: acl name-of-ACL proxy_auth REQUIRED Note: The keyword REQUIRED can be replaced with a user or list of users to further restrict access to a smaller subset of users.

3.19.2.5 - Access Control Lists (ACL)

Be very careful with the order of access control tags. Access control is handled top-down. The first rule that matches is the only rule adhered to. The last rule on the list defines the default behavior in the case of no rule match. The acl and http access tags are used in combination to allow filtering based on a series of access control lists. Squid has a list of default ACLs for localhost, SSL ports, and “safe” ports. Following the definition of these ACLs, a series of http access directives establish the following default filtering policy: * Allow cachemgr access only from localhost * Allow access to only ports in the “safe” access control list * Limit CONNECT method to SSL ports only * Allow access from localhost * Deny all other requests The default ACL policies are reasonable from a security standpoint. However, the number of ports listed as “safe” could be significantly trimmed depending on the needs of your network. Out of the box, ports 21, 70, 80, 210, 280, 443, 488, 591, 777, and 1025 through 65535 are all considered safe. Some of these ports are associated with deprecated or rarely used protocols. As such, this list could be trimmed to further tighten filtering. The following actions should be taken to tighten the ACL policies: 1. There is a filter line in the configuration file that is recommended but commented out. This line should be uncommented or added to prevent access to localhost from the proxy: http access deny to_localhost 2. An access list should be setup for the specific network or networks that the proxy is intended to serve. Only this subset of IP addresses should be allowed access. Add these lines where the following comment appears: # INSERT YOUR OWN RULE(S) HERE TO ALLOW ACCESS FROM YOUR CLIENTS acl your-network-acl-name src ip-range http_access allow your-network-acl-name Note: ip-range is of the format xxx.xxx.xxx.xxx/xx 3. Ensure that the final http access line to appear in the document is the following: http access deny all This guarantees that all traffic not meeting an explicit filtering rule is denied. Further filters should be established to meet the specific needs of a network, explicitly allowing access only where necessary. 4. Consult the chart below. Corresponding acl entries for unused protocols should be commented out and thus denied. Port Service Summary Recommendation 21 ftp File Transfer Protocol(FTP) is a widely used file transfer protocol. ALLOW 70 gopher The gopher protocol is a deprecated search and retrieval protocol that is almost extinct, with as few as 100 gopher servers present worldwide. Support for gopher is disabled in most modern browsers. DENY 80 http A web proxy needs to allow access to HTTP traffic. ALLOW 210 wais The Wide Area Information Server port is similar to gopher, serving as a text searching system to scour indexes on remote machines. Today, it is deprecated and nearly non-existent on the Internet. DENY 280 http-mgmt No documentation of any kind could be found on the obscure service that resides on this port. DENY 443 https SSL traffic is likely (and recommended) for any proxy and should be allowed. ALLOW 488 gss-http No documentation of any kind could be found on the obscure service that resides on this port. DENY 591 filemaker Filemaker is a database application originally offered by Apple in the 1980s. Although development continues and it remains in use today, it should be disabled if your network does not require such traffic. DENY the obscure service that resides on this port. DENY Port Service Summary Recommendation 1025-65535 unregistered ports Random high ports are used by a variety of applications and should be allowed. ALLOW

3.19.2.6 - Configure Internet Cache Protocol (ICP) if Necessary

The ICP protocol is a cache communication protocol that allows multiple Squid servers to communicate. The ICP protocol was designed with no security in mind, relying on user-defined access control lists alone to determine which ICP messages to allow. If a Squid server is standalone, the ICP port should be disabled by adding or correcting the following line in the configuration file: icp_port 0 If the Squid server is meant to speak with peers, strict ACLs should be established to only allow ICP traffic from trusted neighbors. To accomplish this, add or correct the following lines: icp_access allow acl-defining-trusted-neighbors icp_access deny all

3.19.2.7 - Configure iptables to Allow Access to the Proxy Server

Determine an appropriate network block, netwk , and network mask, mask , representing the machines on your network which should operate as clients of the proxy server. Edit /etc/sysconfig/iptables. Add the following line, ensuring that it appears before the final LOG and DROP lines for the RH-Firewall-1-INPUT chain: -A RH-Firewall-1-INPUT -s netwk /mask -m state --state NEW -p tcp --dport port -j ACCEPT For port , use either the default 3128 or the alternate port was selected in Section 3.19.2.1. The default Iptables configuration does not allow inbound access to the Squid proxy service. This modification allows that access, while keeping other ports on the server in their default protected state. See Section 2.5.5 for more information about Iptables.

3.19.2.8 - Forward Log Messages to Syslog Daemon

The default behavior of Squid is to record its log messages in /var/log/squid.log. This behavior can be supplemented so that Squid also sends messages to syslog as well. This is useful for centralizing log data, particularly in instances where multiple Squid servers are present. Squid provides a command line argument to enable syslog forwarding. Modify the SQUID OPTS line in /etc/init.d/squid to include the -s option: SQUID_OPTS="${SQUID_OPTS:-"-D"} -s"

3.19.2.9 - Do Not Run as Root

Since Squid is loaded by the system’s service utility, it starts as root and then changes its effective UID to the UID specified by the cache effective user directive. However, since it was still executed by root, the program maintains a saved UID of root even after changing its effective UID. To prevent this undesired behavior, Squid must either be configured to run in a chroot environment or it must be executed by a non-privileged user in non-daemon mode (the service utility must not be used).

3.19.2.9.1 - Run Squid in a chroot Jail

Chrooting Squid can be a very complicated task. Documentation for the process is vague and a great deal of trial and error may be required to determine all the files that need to be transitioned over to the chroot environment. Therefore, this guide recommends instead the method detailed in Section 3.19.2.9.2 to lower privileges. If chrooting Squid is still desired, it can be enabled with the following directive in the configuration file: chroot chroot-path Then, all the necessary files used by Squid must be copied into the chroot-path directory. The specifics of this step cannot be covered in this guide because they are highly dependent on the external programs used in the Squid configuration. Note: The strace utility is a valuable resource for discovering the files needed for the chroot environment.

3.19.2.9.2 - Modify Service Entry to Lower Privileges

The following modification to /etc/init.d/squid forces the service utility to execute Squid as the squid user instead of the root user: # determine the name of the squid binary [ -f /usr/sbin/squid ] && SQUID="sudo -u squid squid" Making this change prevents Squid from writing its pid to /var/run. This pid file is used by service to check to see if the program started successfully. Therefore, a new location must be chosen for this pid file that the squid user has access to, and the corresponding references in /etc/init.d/squid must be altered to point to it. Make the following modification to the Squid configuration file: pid_filename /var/spool/squid/squid.pid Edit the file /etc/init.d/squid by changing all occurrences of /var/run/squid.pid to /var/spool/squid/ squid.pid Also modify the following line in /etc/init.d/squid: [ $RETVAL -eq 0 ] && touch /var/lock/subsys/squid and add the following lines immediately after it: rm -f /var/lock/subsys/squid status squid

3.20 - SNMP Server

The Simple Network Management Protocol allows administrators to monitor the state of network devices, including computers. Older versions of SNMP were well-known for weak security, such as plaintext transmission of the community string (used for authentication) and also usage of easily-guessable choices for community string.

3.20.1 - Disable SNMP Server if Possible

The system includes an SNMP daemon that allows for its remote monitoring, though it not installed by default. If it was installed and activated, it is important that the software be disabled and removed. If there is not a mission-critical need for hosts at this site to be remotely monitored by a SNMP tool, then disable and remove SNMP as follows: # chkconfig snmpd off # yum erase net-snmpd

3.20.2 - Configure SNMP Server if Necessary

If it is necessary to run the snmpd agent on the system, some best practices should be followed to minimize the security risk from the installation. The multiple security models implemented by SNMP cannot be fully covered here so only the following general configuration advice can be offered: * use only SNMP version 3 security models and enable the use of authentication and encryption for those * write access to the MIB (Management Information Base) should be allowed only if necessary * all access to the MIB should be restricted following a principle of least privilege * network access should be limited to the maximum extent possible including restricting to expected network addresses both in the configuration files and in the system firewall rules * ensure SNMP agents send traps only to, and accept SNMP queries only from, authorized management stations * ensure that permissions on the snmpd.conf configuration file (by default, in /etc/snmp) are 640 or more restrictive * ensure that any MIB files’ permissions are also 640 or more restrictive

3.20.2.1 - Further Resources

The following resources provide more detailed information about the SNMP software: * The CERT SNMP Vulnerabilities FAQ at http://www.cert.org/tech tips/snmp faq.html * The Net-SNMP project web page at http://net-snmp.sourceforge.net * The snmp config(5) man page * the snmpd.conf(5) man page

Trademark Information

Red Hat and Red Hat Enterprise Linux are either registered trademarks or trademarks of Red Hat, Inc in the United States and other countries.

All other names are registered trademarks or trademarks of their respective companies.