Resources

The proper startup and shutdown of services required for a system to function and fulfill its purpose is central to its management. While Unix init scripts (also known as System V or Sys V scripts, due to their origin in Unix System V) have a long history and are in one form or another, common to all Unix and Unux systems, the way in which they are managed, sequenced, enabled, disabled, and the preferred script format often differs somewhat between distributions. The primary areas to be aware of for Debian startup and shutdown scripts include the purpose of run levels, dependency-based sequencing, and utilities available for administering the boot sequence.

"What is the best distribution for my needs? What do I need to know to administer a Debian system? What's different about Debian? What is the best way to handle something specific in Debian? I ran an Internet search on these questions and got millions of results. Now, what do I do? Can someone help me?"

The answer to the last question is yes. Answering the others requires a bit of background. This discussion is oriented towards those who are new to Debian. In it, we'll cover Debian's place among the various Linux distributions, project organization (and how that impacts administration), and licensing issues. Those who are already familiar with Debian may wish to skip ahead to the next chapter.

So, what are some of the utilities provided for Debian administration? We've already mentioned a couple of specialized ones for GRUB and sendmail. There are many others that handle specific software or services, either included in the Debian package itself, or in a separate package. Here, however, we're concerned with more general utilities. The two main ones are dpkg-reeonfigure and gadmin.

In addition to Webmin, there is a server called Usermin which may be installed from the same repository as Webmin. It allows individual users to perform a number of functions more easily, such as changing their password, accessing their files, read and manage their email.andmanagingsomeaspectsoftheiruserprofile.Itis also modular and has the same security features as Webmin.

dselect, APT, aptitude, and Synaptic, all provide some form of automatic dependency resolution. In rare cases, a dependency can't be resolved automatically, and user intervention is required. aptitude will calculate alternatives and ask the user to select from them. APT and Synaptic generally require the user to add packages to the command line or selection list manually.

Such problems generally only occur in the testing and unstable releases, where software dependencies are constantly updated and some may not completely resolve until all of the software involved has been updated and placed in the release. However, one common source of this problem occurs in the stable release as well, and is due to a dependency on a virtual package.

A virtual package is not the name of an actual package, but the name of a library or function that anyone of a number of packages can provide. Since there are usually multiple packages that can satisfy the dependency, the user must choose one to manually install, after which the remainder of the original dependencies can be satisfied automatically. TIlis rarely occurs during a standard upgrade, and almost never during a distribution upgrade, where such virtual packages are selected automatically.

In general, you will only see this probland then rarely, when installing single packages manually.

For many years, the University of Maryland (UoM) Computer Science department was the source of quality, free software that rivaled or even surpassed proprietary solutions. The Advanced Maryland Automatic Network Disk Archiver (AMANDA), is one such solution. Although no longer supported by UoM, it is now hosted on SourceForge, where it remains in active development. In addition to the free Community edition, there is a paid Enterprise edition that includes additional features, such as a graphical configuration utility.

Originally oriented heavily towards centralized tape backup of many networked systems, it now supports disk and even cloud-based storage of backup data as well. Amanda requires a software client running on the systems to be backed up. Oients are available for most Unix type systems, as well as Mac as/X and various Windows releases. The server side will run on pretty much any Unix-or Linux-based system. This makes Amanda especially useful for large, heterogeneous sites.

Amanda uses standard tape and disk file formats, which allows standard tools, such as mt and tar, to be used to browse or even recover data if desired. Amanda provides for parallel backups of many systems at once, backup file management; restore utilities that are easy to use, and several layers of security (including encryption of the backup data over the network, and encryption of the backup files). Amanda is implemented as a single central server that communicates with multiple clients.

A discussion of Amanda configuration could take up a whole book and is very dependent on the type of backup media you use and the systems you are backing up. However, briefly, the community (free) version of Amanda must be configured manually, by adding subdirectories and configuration files to the /ete/amanda/ configuration directory. Refer to the documentation that comes with Amanda, or visit the www •amanda. org website for further information. Also, several commercially published books on backup and recovery have chapters on Amanda configuration.

The actual installation is quite straightforward, and is considered one of the simplest distributions for base installs. Boot up the installation disk, and answer the necessary questions. A help button is frequently available to provide additional information during the installation. Also, standard and advanced installation subjects (including much of what is discussed in this chapter) are covered in detail in the Debian installation guide for the current release, available at http://www.debian.org/

releases/stable/installmanual.

software (such as the desktop environment which installs KDE, GNOME, XFCE, or LXDE, "Laptop" which installs software commonly used on laptops, and so on, all described io the installation guide). Doing this installs a standard set of additional software that suits most users' needs.

Root access to a system is a serious issue. Anyone who knows the root password can do great damage, and the more people who know it, the greater the chance of the password getting into the wrong hands. To avoid this problem, the Budo package provides the ability to run commands as if one were the root user (including, if desired, a shell with full root permissions) without requiring the root password, only the user's own password and proper authorization. The advantage of this is that authorized commands are logged with the user's ID, providing an audit of who actually performed the command.

When Budo is used on a system, the only remaining use for the administrator password is when the system is booted into maintenance (single user) mode. Most Linux distributions, including Debian, reqtrire the root password before providing the root shell prompt in single user mode. However, this requirement can be bypassed in several ways by someone with access to the physical console. Therefore, many administrators completely disable the root password. When this is done, booting into maintenance or single user mode no longer prompts for the root password.

By disabling the root password completely, root-level access to the system via any method other than Budo, other programs that provide root privileges to normal users (also known as setuid programs), or via boot into single user mode is not possible. Note that, since booting into maintenance mode no longer requires a password, physical access to the system console must be controlled in some other way. This may be done by physical means, by requiring a password for the system boot via BIOS (Basic Input/Output System) or UEFI (Unified Extensible Firmware Interface), or by password protecting the GRUB or other system boot loader.

Best practice is to disable the root account login completely. Ubuntu , now does this by default, and Debian is moving in that direction Make sure you either restrict physical access to the console or password protect the boot process.

The /ete/sudoers file and files in the /ete/sudoers.d directory are where sudo access is configured. The details are provided in the sudo package documentation. Briefly, access to commands can be allowed or restricted by individual user, group members, or even lists and can allow access to commands depending on the remote host from which the user is accessing the system.

There are various fonnats supported for clustered systems, including AFS and GFS2. In general, they are not used for the basic system files required for booting, but are better suited for shared data. It is possible to set up such systems for booting, but this is beyond the scope of this discussion. Ifyou are interested, there are a number of publications available on Linux clustering. A good starting point might be the Wikipedia article on clustered filesystems at http:/len. wikipedia.org/wikil Clustered_file_system.

Low-level disk image or so-called Bare Metal backups, are byte-level copies of a full partition or even a complete logical or physical disk. Some utilities will only include used sectors in the copy, thus reducing the size and speed of the backup and restore operations, but this is not always the case. When restoring such a backup, some sort of media, such as a live CD, is required in order to boot up the software to restore the image.

Full disk byte-Ievel backups are useful ifyou need to duplicate or restore a system quickly, as long as the disk being restored is the same or larger. Partition backups of this type may also be done, but in order to restore them you will need some way to restore the partition information to the disk partition table.

Low-level backups are useful in two situations:

• When a system must be replaced and restored quickly with identical hardware
• When a system needs to be duplicated many times on identical hardware

A good example of the latter is when a company provides workers with preconfigured systems all with identical or nearly identical initial configurations. In this case, itis easy just to restore the disk from a standardized copy whenever a new system is required or an old one needs to be returned to its original condition for a new employee.

Now that we have the background on how Debian is organized and developed, the question to be answered is, "how does this affect the administrator."

There are three primary areas which an administrator needs to be aware of. They are as follows:

• The availability of support
• The availability of proprietary features
• licensing issues

Aside from configuring individual software packages, an administrator is responsible for managing how the various services on his systems are started and stopped, managing network connections, maintaining the filesystem, managing system logs, and configuring the face the system shows to the users.

One of the stability requirements for Debian is that when a package is installed, it should not adversely affect the system. Often, this is as simple as providing a standard configuration, such as Apache's simple It works! page. In other cases, however, at least some information must be provided during installation in order to create a basic, working configuration. This is handled by having the installation procedure ask the necessary questions during the installation process in order to create a working configuration. Infact, Debian developers have created the extensive debconf subsystem for this purpose.

The debconf subsystem provides the ability to configure packages before, during, and after installation, and it is integrated into all Debian packages that require configuration. Itsupports a number of frontend user interfaces, prewritten answers that can be input to the installation process, a simple interface for asking questions and providing the answers to the installation scripts, and even ways to reconfigure packages any time after installation. In general, this method of configuration provides only the most commonly used configuration options. More complex configurations must be handled by using any utilities provided by the packages, or by editing the configuration files directly.

Even if you edit configuration files directly, Debian admirristrators have passed on various techniques to the developers to ease this process. In some cases, they have created alternative and more understandable configuration files that are scanned and used to create the actual, complex configuration by scripts provided in the package. A good example of this is the GRUB2 boot loader. An administrator makes whatever modifications to the configuration files in Jete/default and /ete/grub. d, and then runs update-grub to create the actual file that GRUB uses to load the operating system. The script then executes the necessary commands for creating and installing the configuration, as well as the necessary updates of other files affected by GRUB reconfiguration. This is just one example of how various utilities created by the Debian developers and other third parties have simplified Debian administration.

Another good example is sendmail. Administrators often modify the files and various database files required by sendmail to tailor it for large, corporate installations. However, instead of compiling the configuration and databases, then forcing sendmail to load the new configuration, a single command, sendmaileonfig, will perform all of the necessary steps at once, with minimal interruption in server operation.

Security is always a concern when remote access to a system is involved. Webmin handles this by requiring authentication and providing for detailed access restrictions that provide a layer of control beyond the firewall. Webmin users can be defined separately, or certain local users can be designated. Access to the various modules in Webmin can be restricted to certain users or groups of users, and detailed logs of Webmin actions are kept.

One of the most useful package types is called meta packages. These are packages that contain no software themselves, but that require other packages to be installed, and thus provide a unified set of software for a particular purpose. This works because, although no actual software is in the meta package, the package manager will select and install all of the required dependent packages, providing a complete set of features in a single, easy installation step.

Frequently, there are multiple packages with slightly different names that install a slightly different set of dependencies. A good example is GNOME. You can install the GNOME meta package, which will provide a complete GNOME installation, including many extras. Or, you can elect to install gnome -core, which provides only the basic desktop environment, and select from any of the additional packages that provide additional features and functions, such as:

• evolution (e-mail software, similar to Outlook)
• gnome-documents (document management features)
• gnome-games
• gnome-media (multimedia applications)
• libreoffice-gnome (office suite) and others

One minor problem with meta packages is that there is no easy way to search for them. Many do have 'meta' in their descriptions, which can facilitate some searches, but this is not universally true. The GNOME packages mentioned previously do not follow this convention as of Debian 7. Nevertheless, they are often fairly obvious, and not too difficult to find.

If you're in charge of one or more systems that are installed at a distant location, backing them up individually can be a large chore. Fortunately, there are a number of packages that can help. Most backup packages, even those intended to create a backup of a single machine, have options to send the data to a remote location for actual storage. Ofcourse, any of the packages with this capability can be used on multiple machines. However, there are two popular packages that not only provide backup services to multiple hosts from a central location, but provide for management of backup cycles and automated runs as well. They are Amanda and Bacula. Of course, other such packages exist, and there are some excellent third-party backup solutions as well. However, both Amanda and Bacula are provided as part of Debian, so they are free as well as able to handle many systems and a variety of backup media.

There are two primary disadvantages to disk or directory encryption. The first is probably the most serious; if the password is forgotten, the data is permanently lost and completely unrecoverable unless unencrypted backups (or encrypted backups where the password has not been lost) are available. The second disadvantage is performance. Most software encryption modules perform well, but there is no avoiding some overhead, even if it is minimal. Hardware encryption, such as that provided by the motherboard or the disk controller or drive itself, generally does not have significant overhead.

Best practice is to evaluate whether the disadvantages of encryption outweigh the potential damage from compromised or stolen systems.

Most corporate policies require partial or full (if available) disk encryption on laptops, and directory encryption as a minimum on public servers.

While the traditional user/group/world permissions structure is adequate for many systems, there are times when more fine-grained control is required. This may be handled by using Access Control Lists (ACL). ACLs are implemented using the extended attributes available in various filesystems and are available by default in the EXT filesystems normally used in Debian Linux. The details may be found in the Debian ael package. In particular, check out the manual pages for the getael, setael, and ehael commands.

While ACLs are useful, there are some issues to be aware of. For example, most graphical file managers do not support them, and some utilities (EMACS comes to mind) clear the file ACL list when updating a file. This can be addressed somewhat by setting a default ACL for the directory, but non-default file ACLs will still need to be reconstructed ifthey are lost. Another issue is performance. On some filesystems, the existence of an ACL delays the initial file access due to the additional access verification required. Since this occurs only on the initial access, its impact is often minimal in practice. Finally, there is the issue of backups when you use ACLs. Most backup utilities now recognize ACLs. In Debian Linux, both rsyne and tar now have options to allow them to recognize and back up ACLs, but they are only handled ifyou use the proper command options. The EXT backup utilities dump and restore also handle ACLs. However, some utilities may not recognize or handle ACLs properly, so always check the package documentation and manual pages if you use Access Control Lists.

Note that, if you do use ACLs, you are not reqtrired to use them for all mes and directories. It is quite possible to set them on only those files and directories that need them. Judicious use of file ACLs and default ACLs set for certain directories will help streamline ACL management.

B-Tree File System (Btrfs) is the next Linux filesystem format. It focuses on faulttolerance, repair, and easy administration, with the ability to scale up to larger storage configurations. ext-based systems can be easily converted to Btrfs. For the moment, Btrfs is still under heavy development, although only forward-compatible fonnat changes are anticipated. Debian 7 does allow it to be used, but it is not yet recommended for production systems.

In general, there are two types of backups: low-level image backups, and file-level backups.