Tutorial For Lpi Exam 201: Part 6

Topic 211: System Maintenance

David Mertz, Ph.D.
Professional Neophyte
August, 2005

Welcome to "System Maintenance", the sixth of eight tutorials designed to prepare you for LPI exam 201. In this tutorial you will learn some basic concept of system logging, software packaging, and backup strategies.

Before You Start

About this series

The Linux Professional Institute (LPI) certifies Linux system administrators at junior and intermediate levels. There are two exams at each certification level. This series of eight tutorials helps you prepare for the first of the two LPI intermediate level system administrator exams--LPI exam 201. A companion series of tutorials is available for the other intermediate level exam--LPI exam 202. Both exam 201 and exam 202 are required for intermediate level certification. Intermediate level certification is also known as certification level 2.

Each exam covers several or topics and each topic has a weight. The weight indicate the relative importance of each topic. Very roughly, expect more questions on the exam for topics with higher weight. The topics and their weights for LPI exam 201 are:

Topic 201: Linux Kernel (5) Topic 202: System Startup (5) Topic 203: Filesystems (10) Topic 204: Hardware (8) Topic 209: File Sharing Servers (8) Topic 211: System Maintenance (4) Topic 213: System Customization and Automation (3) Topic 214: Troubleshooting (6)

About this tutorial

Welcome to "System Maintenance", the sixth of eight tutorials designed to prepare you for LPI exam 201. In this tutorial you will learn some basic concept of system logging, software packaging, and backup strategies.


To get the most from this tutorial, you should already have a basic knowledge of Linux and a working Linux system on which you can practice the commands covered in this tutorial.

About system maintenance

This tutorial--and the corresponding LPI exam, is a little bit of a grab-bag of several topics that do not cleanly fall in other categories. System logging and analyzing log files is an important task for a system administrator to be familiar with. Likewise, a maintained system should carry out a sensible backup strategy using standard Linux tools.

Not every system administrator will need to create custom software packages, but for administrators of multiple (similar) installations, installing site- or company-specific software packages will be part of an administrators duties. This tutorial looks at the Debian and RPM package formats, as well as touching on basic "tarballs".

System Logging

About logging

Many processes and servers under a Linux system append changing status information to "log files." These log files often live in the /var/log/ directory, and often begin with a time stamp indicating when the described event occurred. But for better or worse, there is no real consistency in the precise format of log files. The one feature you can pretty much count on is that Linux log files are plain ASCII files, and contain one "event" per line of the file. Often--but not always--log files contain a (relatively) consistent set of space- or tab-delimited data field.

Some processes, especially internet services, handle log file writes within their own process. At heart, writing to a log file is just an append to an open file handle. But many programs (especially daemons and cron'd jobs) use the standard syslog API to let the syslogd or klogd daemons handle the specific logging process.

Parsing log files

Exactly how you go about parsing a log file depends greatly on the specific format it takes. For log files with regular table format, tools like cut, split, head and tail are likely to be useful. For all log files, grep is a great tool for finding and filtering contents of interest. For more complex processing tasks, you are likely to think of sed, awk, perl or python as tools of choice.

For a good introduction to many of the text processing tools you are most likely to use in processing and analyzing log files, see David's IBM developerWorks tutorial on the GNU text processing utilities. A number of good higher-level tools also exist for working with log files, but these tools are usually distribution specific and/or non-standard (but often Free Software) utilities.

Logging with syslogd and klogd

The daemon klogd intercepts and logs Linux kernel messages. As a rule, klogd will utilize the more general syslogd capabilities, but in special cases it may log messages directly to a file.

The general daemon syslogd provides logging for many programs. Every logged message contains at least a time and a hostname field, and usually a program name field. The specific behavior of 'syslogd is controlled by the configuration file /etc/syslog.conf. Application (including kernel) messages may be logged to files, usually living under /var/log/, or remotely over a network socket.

Configuring /etc/syslog.conf

The file /etc/syslog.conf contains a set of rules, one per line. Empty lines and lines starting with a # are ignored. Each rule consists of two whitespace-separated fields, a selector field and an action field. The selector, in turn, contains one of more dot-separated facility/priority pairs. A facility is a subsystem that wishes to log messages, and can have the (case-insensitive) values: "auth", "authpriv", "cron", "daemon", "ftp", "kern", "lpr", "mail", "mark", "news", "security", "syslog", "user", "uucp" and "local0" through "local7".

Priorities have a specific order, and matching a given priority means "this one or higher" unless an initial = (or !=) is used. Priorities are, in ascending order: "debug", "info", "notice", "warning" or "warn", "err" or "error", "crit", "alert", "emerg" or "panic" (several names have synonyms). "none" means that no priority is indicated.

Both facilities and priorities may accept the * wildcard. Multiple facilities may be comma-separated, and multiple selectors may be semi-colon separated. For example:

# from /etc/syslog.conf
# all kernel mesages
kern.*                    -/var/log/kern.log
# `catch-all' logfile
  mail,news.none          -/var/log/messages
# Emergencies are sent to everybody logged in
*.emerg                   *

Configuring remote logging

To enable remote logging of syslogd messages (really application messages, but handled by syslogd), you must first enable the "syslog" service on both the listening and the sending machines. To do this, you need to add a line to each /etc/services configuration file containing something like:

syslog    514/UDP

To configure the local (sending) syslogd to send messages to a remote host, you specify a regular facility and priority, but give an action beginning with an @ symbol for the destination host. A host may be configured in the usual fashion, it either /etc/hosts or via DNS (it need not be resolved already when syslogd first launches). For example:

# from /etc/syslog.conf
# log all critical messages to master.example.com
*.crit                @master.example.com
# log all mail messages except info level to mail.example.com
mail.*;mail.!=info    @mail.example.com

Rotating log files

Often you will not want to let particular log files grow unboundedly. The utility logrotate may be used to archive older logged information. Usually logrotate is run as a cron job, generally daily. 'logrotate allows automatic rotation, compression, removal, and mailing of log files. Each log file may be handled daily, weekly, monthly, or only when it grows too large.

The behavior of logrotate is controlled by the configuration file /etc/logrotate.conf (or some other file, if specified). The configuration file may contain both global options and file-specific options. Generally, archived logs are saved for a finite time period, and are given sequential backup names. For example, one system of mine contains the following files due to its rotation schedule.

-rw-r-----  1 root adm 4135 2005-08-10 04:00 /var/log/syslog
-rw-r-----  1 root adm 6022 2005-08-09 07:36 /var/log/syslog.0
-rw-r-----  1 root adm  883 2005-08-08 07:35 /var/log/syslog.1.gz
-rw-r-----  1 root adm  931 2005-08-07 07:35 /var/log/syslog.2.gz
-rw-r-----  1 root adm  888 2005-08-06 07:35 /var/log/syslog.3.gz
-rw-r-----  1 root adm 9494 2005-08-05 07:35 /var/log/syslog.4.gz
-rw-r-----  1 root adm 8931 2005-08-04 07:35 /var/log/syslog.5.gz

Packaging Software

In the beginning was the tarball

For custom software distribution on Linux, there is actually much less needed than you might think. Linux has a fairly clean standard about where files of various types should reside, and installing custom software, at heart, need not involve much more than putting the right files in the right places.

The Linux tool tar ("tape archive"; though it need not, and usually does not, actually utilize tapes) is perfectly adequate to create an archive of files with specified filesystem locations. For distribution, you generally want to compress a tar archive with gzip (or bzip2) as well. See the final section of this tutorial on Backup for more information on these utilities. A compressed tar archive is generally named with the extensions .tar.gz or .tgz' (or .tar.bz2).

Early Linux distributions--and some current ones like Slackware--use simple tarballs as their distribution mechanism. For a custom distribution of in-house software to centrally maintained Linux systems, this continues to be the simplest approach

Custom archive formats

Many programming languages and other tools come with custom distribution systems that are neutral between Linux distributions, and usually between altogether different operating systems. Python has its distutils tools and archive format; Perl has CPAN archives; Java has .jar files; Ruby has gems. Many non-language applications have a standard system for distributing plug-ins or other enhancements to a base application as well.

While you can perfectly well use an package format like DEB or RPM to distribute, e.g. a Python package, it often makes more sense to follow the packaging standard of the tool the package is created for. Of course, for system level utilities and applications, or for most compiled userland applications, that standard is on the Linux distribution packaging standards. But for custom tools written in specific programming languages, something different might promote easier reuse of your tools across distributions and platforms (whether in-house or external users are the intended target).

The "big two" package formats

There are two main package formats used by Linux distributions: Redhat Package Manager (RPM) and Debian (DEB). Both are similar in purpose, but different in details. In general, either one is a format for an "enhanced" archive of files. The enhancements provided by these package formats include annotations for version numbers, dependencies of one application upon other applications or libraries, human readable descriptions of packaged tools, and a general mechanism for managing the installation, upgrade and deinstallation of packaged tools.

Under DEB files, the nested configuration file control contains most of the package metadata. For RPM files, the file spec plays this role. The full details of creating good packages in either format is beyond this tutorial, but we will outline the basics here.

What is in a .deb file?

A DEB package is created with the archive tool, and cousin of tar, ar (or by some higher-level tool that utilizes ar). Therefore we can use ar to see just what is really inside a .deb file. Normally we would use higher-level tools like dpkg, dpkg-deb or apt-get to actually work with a DEB package. For example:

% ar tv unzip_5.51-2ubuntu1.1_i386.deb
rw-r--r-- 0/0      4 Aug  1 07:23 2005 debian-binary
rw-r--r-- 0/0   1007 Aug  1 07:23 2005 control.tar.gz
rw-r--r-- 0/0 133475 Aug  1 07:23 2005 data.tar.gz

The file debian-binary simply contains the DEB version (currently 2.0). The tarball data.tar.gz contains the actually application files--executables, documentation, manual pages, configuration files, and so on.

The tarball control.tar.gz is the most interesting, from a packaging perspective. Let us look at the example DEB package we chose:

% tar tvfz control.tar.gz
drwxr-xr-x root/root         0 2005-08-01 07:23:43 ./
-rw-r--r-- root/root       970 2005-08-01 07:23:43 ./md5sums
-rw-r--r-- root/root       593 2005-08-01 07:23:43 ./control

As you might expect, md5sums contains cryptographic hashes of all the distributed files for verification purposes. The file control is where the metadata lives. In some cases you might also find or wish to include scripts called postinst and prerm in control.tar.gz to take special steps after installation or before removal, respectively.

Creating a DEB control file

The installation scripts can do anything a shell script might. Look at some examples in existing packages for examples. But those scripts are optional, and often not needed or included. Required for a .deb package is its control file. The format of this file contains various metadata fields, and is best illustrated by showing an example:

% cat control
Package: unzip
Version: 5.51-2ubuntu1.1
Section: utils
Priority: optional
Architecture: i386
Depends: libc6 (>= 2.3.2.ds1-4)
Suggests: zip
Conflicts: unzip-crypt (<< 5.41)
Replaces: unzip-crypt (<< 5.41)
Installed-Size: 308
Maintainer: Santiago Vila <sanvila@debian.org>
Description: De-archiver for .zip files
 InfoZIP's unzip program. With the exception of multi-volume archives
 (ie, .ZIP files that are split across several disks using PKZIP's /& option),
 this can handle any file produced either by PKZIP, or the corresponding
 InfoZIP zip program.
 This version supports encryption.

Basically, except the custom data values, you should make your control file look just like this one. For non-CPU specific packages--either scripts, pure documentation, or source code use Architecture: all.

Making a DEB package

Creating a DEB package is performed with the tool dpkg-deb. We cannot cover all the intricacies of good packaging, but the basic idea is to create a working directory ./debian/ and put the necessary contents into it before running dpkg-deb. You will also want to set permissions on your files to match their intended state when installed. For example:

% mkdir -p ./debian/usr/bin/
% cp foo-util ./debian/usr/bin                # copy executable/script
% mkdir -p ./debian/usr/share/man/man1
% cp foo-util.1 ./debian/usr/share/man/man1   # copy the manpage
% gzip --best ./debian/usr/share/man/man1/foo-util.1
% find ./debian -type d | xarg chmod 755      # set dir permissions
% mkdir -p ./debian/DEBIAN
% cp control ./debian/DEBIAN   # first create a matching 'control'
% dpkg-deb --build debian      # create the archive
% mv debian.deb foo-util_1.3-1all.deb  # rename to final package name

More on DEB package creation

In the last panel you can see that our local directory structure underneath ./debian/ is meant to match the intended installation structure. A few more points on creating a good package are worth observing.

* Generally you should create a file as part of your distribution

called ./debian/usr/share/doc/foo-util/copyright (adjust for package name).

* It is also good practice to create the files

./debian/usr/share/doc/foo-util/changelog.gz and ./debian/usr/share/doc/foo-utils/changelog.Debian.gz.

* The tool lintian will check for questionable features in a DEB

package. Not everything lintian complains about is strictly necessary to fix; but if you intend wider distribution, it is a good idea to fix all issues it raises.

* The tool fakeroot is helpful for packaging with the right owner.

Usually a destination wants tools installed as root, not as the individual user who happened to generate the package (lintian will warn about this). You can accomplish this with, e.g.:
% fakeroot dpkg-deb --build debian

What is in an .rpm file?

RPM takes a slightly different strategy than DEB does in creating packages. Its configuration file is called spec rather than control, but the spec file also does more work than a control file does. All the details of steps needed for preinstallation, postinstallation, preremoval, and installation itself, are contained as embedded script files in a spec configuration. In fact, the spec format even comes with macros for common actions.

Once you create an RPM package, you do so with the rpm -b utility. For example:

% rpm -ba foo-util-1.3.spec  # peform all build steps

This package build process does not rely on specific named directories as with DEB, but rather on directives in the more complex spec file.

Creating RPM metadata

The basic metadata in an RPM is much like that in a DEB. For example, foo-util-1.3.spec might contain something like:

# spec file for foo-util 1.3
Summary: A utility that fully foos
Name: foo-util
Version: 1.3
Release: 1
Copyright: GPL
Group: Application/Misc
Source: ftp://example.com/foo-util.tgz
URL: http://example.com/about-foo.html
Distribution: MyLinux
Vendor: Acme Systems
Packager: John Doe <jdoe@acme.example.com>

The foo-util program is an advanced fooer that combines the
capabilities of OneTwo's foo-tool and those in the GPL bar-util.

Scripting in an RPM

Several sections of an RPM spec file may contain mini shell scripts. These include:

* %prep: steps to perform to get the build ready, such as clean out

earlier (partial) builds. Often the following macro is helpful and sufficient:
%prep %setup

* %build: steps to actually build the tool. If you use the make

facility, this might amount to:
%build make

* %install: steps to install the tool. Again, if you use make this

might mean:
%install make install

* %files: you must include a list of files that are part of the

package. Even though your Makefile might use these files, the package manager program (rpm) will not know about them unless you include them here, e.g.:
%files %doc README /usr/bin/foo-util /usr/share/man/man1/foo-util.1

Backup Operations

About backup

The first rule about making backups is: Do it! It is all too easy in server administration--or even just with Linux on the desktop--to neglect backing up, or backing up on the schedule suited to your requirements.

The easiest way to carry out backups in a systematic and schedules way is to perform them on a cron job. See the Topic 213 tutorial for a discussion of configuring crontab. In part, the choice of backup schedule depends on the backup tool and media you choose to use.

Backup to tape is a traditional technique, and tape drives continue to offer the largest capacity of relatively inexpensive media. But recently, writeable or rewriteable CDs and DVD have become ubiquitious, and will often make reasonable removable media for backups.

What to backup

A nice thing about Linux is that it uses a predicable and hierarchical arrangement of files. As a consequence, you rarely need to backup an entire filesystem hierarchy; most of a Linux filesystem hierarchy can be reinstalled from distribution media easily enough. In large environments, a master server image might be used to create a basic Linux system, which can be customized by restoring a few selected files that were backed up elsewhere.

Basically, what you want backed up is /home/, /etc/, /usr/local/, and maybe /root/ and /boot/. Often you will also want to backup some parts of /var/, such as /var/log/ and /var/mail/.

Backup with cp and scp

Perhaps the simplest way to perform a backup is with cp or scp and the -r (recurse) switch. The former copies to "local" media (but including NFS mounts), the latter can copy to remote servers in a securely encrypted fashion. For either, you need a mounted drive with sufficient space to accomodate the files you want to backup, of course. To gain any real hardware protection, you want the partition you copy to to be a different physical device than the drive(s) you are backing up from.

Copying with cp or scp can be part of an incremental backup schedule. The trick here is to use the utility find to figure out which files have been modified recently. Here is a simple example where we copy all the files in /home/ that have been modified in the lst day:

# File: backup-daily.sh
# ++ Run this on a daily cron job ++
#-- first make sure the target directories exist
for d in `find /home -type d` ; do mkdir -p /mnt/backup$d ; done
#-- then copy all the recently modified files (one day)
for f in `find /home -mtime -1` ; do cp $f /mnt/backup$f  ; done

The cp -u flag is somewhat similar, but it depends on the continuity of the target filesystem between backup events. The find approach works fine if you change the mount point of /mnt/backup (e.g. to a different NFS location). And the find system works equally well with scp, once you specify the necessary login information to the remote site.

Backup with tar

Although cp and scp are workable for backup, generally tar sees wider use, being designed specifically for creating "tape archives". Despite the origin of the name, tar is equally capable of creating a simple .tar file as raw data on a tape drive. For example, you might backup to a tape drive using a command like:

% tar -cvf /dev/rmt0 /home    # Archive /home to tape

To direct the output to a file is hardly any different:

% tar -cvf /mnt/backup/2005-08-12.tar /home

In fact, since gzip is streamable, you can easily compress an archive during creation:

% tar -cv /home | gzip - > /mnt/backup/2005-08-12.tgz

You can combine tar with find in the same ways shown for cp or scp. To list the files on a tape drive, you might use:

% tar -tvf /dev/rmt0

To retrieve a particular file:

% tar -xvf /dev/rmt0 file.name

Backup with cpio

The utility cpio is a superset of tar. It handles tar archives, but will also work with several other formats, and has many more options built in. cpio comes with a huge wealth of switches to filter which files are backed up, and even supports remote backup internally (rather than needing to pipe through scp or the like). The main advantage cpio has over tar is that you can both add files to archives, but also remove files back out.

Below are some quick examples of cpio. Create a file archive on a tape device:

% find /home -print |cpio -ocBv  /dev/rmt0

List the entries in a file archive on a tape device:

% cpio -itcvB < /dev/rmt0

Retrieve a file from a tape drive:

% cpio -icvdBum file.name < /dev/rmt0

Backup with dump and restore

A set of tools named dump and restore or related names are sometimes used to backup whole filesystems. Unfortunately, these tools are specific to filesystem types, and are not uniformly available. For example, the original dump and restore are ext2/3 specific, while the tools xfsdump and xfsrestore are used for XFS filesystems. Not every filesystem type has the equivalent tools, and even if they do, switches are not necessarily uniform.

It is good to be aware of these utilities, but they are not very uniform across Linux systems. For some purposes--such as if you use entirely XFS partitions--the performance of dump and restore can be a great boost over a simple tar or cpio.

Incremental backup with rsync

rsync is utility that provides fast incremental file transfer. For automated remote backups, rsync is often the best tool for the job. A nice feature of rsync over other tools is that it can optionally enforce two-way syncronization. That is, rather than simply backup newer or changed files, it can also automatically remove locally deleted files from the remote backup.

To get a sense of the options, the moderately complex script at http://samba.anu.edu.au/rsync/examples.html is useful to look at:

# This script does personal backups to a rsync backup server. You will
# end up with a 7 day rotating incremental backup. The incrementals will
# go into subdirs named after the day of the week, and the current
# full backup goes into a directory called "current"
# tridge@linuxcare.com
# directory to backup
# excludes file - this contains a wildcard pats of files to exclude
# the name of the backup machine
# your password on the backup server
BACKUPDIR=`date +%A`
OPTS="--force --ignore-errors --delete-excluded --exclude-from=$EXCLUDES
  --delete --backup --backup-dir=/$BACKUPDIR -a"
export PATH=$PATH:/bin:/usr/bin:/usr/local/bin
# the following line clears the last weeks incremental directory
[ -d $HOME/emptydir ] || mkdir $HOME/emptydir
rsync --delete -a $HOME/emptydir/ $BSERVER::$USER/$BACKUPDIR/
rmdir $HOME/emptydir
# now the actual transfer
rsync $OPTS $BDIR $BSERVER::$USER/current