Linux Administration I - System and Users

Authors tuxcademy

Plaintext

Version 4.0

Linux Administration I
System and Users

$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

tuxcademy – Linux and Open Source learning materials for everyone
www.tuxcademy.org ⋅ info@tuxcademy.org
This training manual is designed to correspond to the objectives of the LPI-101 (LPIC-1, version
4.0) certiﬁcation exam promulgated by the Linux Professional Institute. Further details are
available in Appendix B.

The Linux Professional Institute does not endorse speciﬁc exam preparation materials or tech-
niques. For details, refer to info@lpi.org .

The tuxcademy project aims to supply freely available high-quality training materials on
Linux and Open Source topics – for self-study, school, higher and continuing education
and professional training.
Please visit http://www.tuxcademy.org/ ! Do contact us with questions or suggestions.

Linux Administration I System and Users
Revision: adm1:067839db3bb6bd7f:2015-08-08
adm1:33e55eeadba676a3:2015-08-08 1–13, B
adm1:HWye5vypKrdhKjDySGhqPI

© 2015 Linup Front GmbH Darmstadt, Germany
© 2015 tuxcademy (Anselm Lingnau) Darmstadt, Germany
http://www.tuxcademy.org ⋅ info@tuxcademy.org
Linux penguin “Tux” © Larry Ewing (CC-BY licence)

All representations and information contained in this document have been com-
piled to the best of our knowledge and carefully tested. However, mistakes cannot
be ruled out completely. To the extent of applicable law, the authors and the tux-
cademy project assume no responsibility or liability resulting in any way from the
use of this material or parts of it or from any violation of the rights of third parties.
Reproduction of trade marks, service marks and similar monikers in this docu-
ment, even if not specially marked, does not imply the stipulation that these may
be freely usable according to trade mark protection laws. All trade marks are used
without a warranty of free usability and may be registered trade marks of third
parties.

This document is published under the “Creative Commons-BY-SA 4.0 Interna-
tional” licence. You may copy and distribute it and make it publically available as
long as the following conditions are met:
Attribution You must make clear that this document is a product of the tux-
cademy project.

Share-Alike You may alter, remix, extend, or translate this document or modify
or build on it in other ways, as long as you make your contributions available
under the same licence as the original.
Further information and the full legal license grant may be found at
http://creativecommons.org/licenses/by- sa/4.0/

Authors: Thomas Erker, Anselm Lingnau
Technical Editor: Anselm Lingnau ⟨anselm@tuxcademy.org ⟩
English Translation: Anselm Lingnau
Typeset in Palatino, Optima and DejaVu Sans Mono
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

Contents

1 System Administration 13
1.1 Introductory Remarks . . . . . . . . . . . . . . . . . . 14
1.2 The Privileged root Account . . . . . . . . . . . . . . . . 14
1.3 Obtaining Administrator Privileges . . . . . . . . . . . . . 16
1.4 Distribution-speciﬁc Administrative Tools . . . . . . . . . . . 18

2 User Administration 21
2.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . 22
2.1.1 Why Users? . . . . . . . . . . . . . . . . . . . . 22
2.1.2 Users and Groups . . . . . . . . . . . . . . . . . 23
2.1.3 People and Pseudo-Users . . . . . . . . . . . . . . . 25
2.2 User and Group Information . . . . . . . . . . . . . . . . 25
2.2.1 The /etc/passwd File . . . . . . . . . . . . . . . . . 25
2.2.2 The /etc/shadow File . . . . . . . . . . . . . . . . . 28
2.2.3 The /etc/group File . . . . . . . . . . . . . . . . . 30
2.2.4 The /etc/gshadow File . . . . . . . . . . . . . . . . . 31
2.2.5 The getent Command . . . . . . . . . . . . . . . . 32
2.3 Managing User Accounts and Group Information . . . . . . . . 32
2.3.1 Creating User Accounts . . . . . . . . . . . . . . . 33
2.3.2 The passwd Command . . . . . . . . . . . . . . . . 34
2.3.3 Deleting User Accounts . . . . . . . . . . . . . . . 36
2.3.4 Changing User Accounts and Group Assignment . . . . . . 36
2.3.5 Changing User Information Directly—vipw . . . . . . . . . 37
2.3.6 Creating, Changing and Deleting Groups . . . . . . . . . 37

3 Access Control 41
3.1 The Linux Access Control System . . . . . . . . . . . . . . 42
3.2 Access Control For Files And Directories . . . . . . . . . . . 42
3.2.1 The Basics . . . . . . . . . . . . . . . . . . . . 42
3.2.2 Inspecting and Changing Access Permissions. . . . . . . . 43
3.2.3 Specifying File Owners and Groups—chown and chgrp . . . . . 44
3.2.4 The umask . . . . . . . . . . . . . . . . . . . . 45
3.3 Access Control Lists (ACLs) . . . . . . . . . . . . . . . . 47
3.4 Process Ownership . . . . . . . . . . . . . . . . . . . 47
3.5 Special Permissions for Executable Files . . . . . . . . . . . . 47
3.6 Special Permissions for Directories . . . . . . . . . . . . . 48
3.7 File Attributes . . . . . . . . . . . . . . . . . . . . . 50

4 Process Management 53
4.1 What Is A Process? . . . . . . . . . . . . . . . . . . . 54
4.2 Process States . . . . . . . . . . . . . . . . . . . . . 55
4.3 Process Information—ps . . . . . . . . . . . . . . . . . 56
4.4 Processes in a Tree—pstree . . . . . . . . . . . . . . . . 57
4.5 Controlling Processes—kill and killall . . . . . . . . . . . . 58
4.6 pgrep and pkill . . . . . . . . . . . . . . . . . . . . . 59
4.7 Process Priorities—nice and renice . . . . . . . . . . . . . . 61
4 Contents

4.8 Further Process Management Commands—nohup and top . . . . . 61

5 Hardware 63
5.1 Fundamentals . . . . . . . . . . . . . . . . . . . . . 64
5.2 Linux and PCI (Express) . . . . . . . . . . . . . . . . . 65
5.2.1 USB. . . . . . . . . . . . . . . . . . . . . . . 67
5.3 Peripherals . . . . . . . . . . . . . . . . . . . . . . 69
5.3.1 Overview . . . . . . . . . . . . . . . . . . . . 69
5.3.2 Devices and Drivers . . . . . . . . . . . . . . . . . 70
5.3.3 The /sys Directory . . . . . . . . . . . . . . . . . 72
5.3.4 udev . . . . . . . . . . . . . . . . . . . . . . 73
5.3.5 Device Integration and D-Bus . . . . . . . . . . . . . 74

6 Hard Disks (and Other Secondary Storage) 77
6.1 Fundamentals . . . . . . . . . . . . . . . . . . . . . 78
6.2 Bus Systems for Mass Storage . . . . . . . . . . . . . . . 78
6.3 Partitioning . . . . . . . . . . . . . . . . . . . . . . 81
6.3.1 Fundamentals . . . . . . . . . . . . . . . . . . . 81
6.3.2 The Traditional Method (MBR) . . . . . . . . . . . . . 82
6.3.3 The Modern Method (GPT) . . . . . . . . . . . . . . 83
6.4 Linux and Mass Storage . . . . . . . . . . . . . . . . . 84
6.5 Partitioning Disks. . . . . . . . . . . . . . . . . . . . 86
6.5.1 Fundamentals . . . . . . . . . . . . . . . . . . . 86
6.5.2 Partitioning Disks Using fdisk . . . . . . . . . . . . . 88
6.5.3 Formatting Disks using GNU parted . . . . . . . . . . . 91
6.5.4 gdisk . . . . . . . . . . . . . . . . . . . . . . 92
6.5.5 More Partitioning Tools . . . . . . . . . . . . . . . 93
6.6 Loop Devices and kpartx . . . . . . . . . . . . . . . . . 93
6.7 The Logical Volume Manager (LVM) . . . . . . . . . . . . . 95

7 File Systems: Care and Feeding 99
7.1 Creating a Linux File System . . . . . . . . . . . . . . . . 100
7.1.1 Overview . . . . . . . . . . . . . . . . . . . . 100
7.1.2 The ext File Systems . . . . . . . . . . . . . . . . . 102
7.1.3 ReiserFS . . . . . . . . . . . . . . . . . . . . . 110
7.1.4 XFS . . . . . . . . . . . . . . . . . . . . . . . 111
7.1.5 Btrfs . . . . . . . . . . . . . . . . . . . . . . 113
7.1.6 Even More File Systems . . . . . . . . . . . . . . . 114
7.1.7 Swap space . . . . . . . . . . . . . . . . . . . . 115
7.2 Mounting File Systems . . . . . . . . . . . . . . . . . . 116
7.2.1 Basics . . . . . . . . . . . . . . . . . . . . . . 116
7.2.2 The mount Command . . . . . . . . . . . . . . . . . 116
7.2.3 Labels and UUIDs . . . . . . . . . . . . . . . . . 118
7.3 The dd Command . . . . . . . . . . . . . . . . . . . . 120
7.4 Disk Quotas . . . . . . . . . . . . . . . . . . . . . . 121
7.4.1 Basics . . . . . . . . . . . . . . . . . . . . . . 121
7.4.2 User Quotas (ext and XFS) . . . . . . . . . . . . . . 121
7.4.3 Group Quotas (ext and XFS) . . . . . . . . . . . . . . 123

8 Booting Linux 127
8.1 Fundamentals . . . . . . . . . . . . . . . . . . . . . 128
8.2 GRUB Legacy . . . . . . . . . . . . . . . . . . . . . 131
8.2.1 GRUB Basics . . . . . . . . . . . . . . . . . . . 131
8.2.2 GRUB Legacy Conﬁguration . . . . . . . . . . . . . . 132
8.2.3 GRUB Legacy Installation . . . . . . . . . . . . . . . 133
8.2.4 GRUB 2 . . . . . . . . . . . . . . . . . . . . . 134
8.2.5 Security Advice . . . . . . . . . . . . . . . . . . 135
5

8.3 Kernel Parameters . . . . . . . . . . . . . . . . . . . 135
8.4 System Startup Problems . . . . . . . . . . . . . . . . . 137
8.4.1 Troubleshooting . . . . . . . . . . . . . . . . . . 137
8.4.2 Typical Problems . . . . . . . . . . . . . . . . . . 137
8.4.3 Rescue systems and Live Distributions . . . . . . . . . . 139

9 System-V Init and the Init Process 141
9.1 The Init Process . . . . . . . . . . . . . . . . . . . . 142
9.2 System-V Init . . . . . . . . . . . . . . . . . . . . . 142
9.3 Upstart . . . . . . . . . . . . . . . . . . . . . . . 148
9.4 Shutting Down the System . . . . . . . . . . . . . . . . 150

10 Systemd 155
10.1 Overview. . . . . . . . . . . . . . . . . . . . . . . 156
10.2 Unit Files . . . . . . . . . . . . . . . . . . . . . . . 157
10.3 Unit Types . . . . . . . . . . . . . . . . . . . . . . 161
10.4 Dependencies . . . . . . . . . . . . . . . . . . . . . 162
10.5 Targets. . . . . . . . . . . . . . . . . . . . . . . . 164
10.6 The systemctl Command . . . . . . . . . . . . . . . . . 166
10.7 Installing Units. . . . . . . . . . . . . . . . . . . . . 169

11 Dynamic (AKA Shared) Libraries 171
11.1 Compiling and Installing Software . . . . . . . . . . . . . 172
11.2 Dynamic Libraries In Practice . . . . . . . . . . . . . . . 174
11.3 Installing and Locating Dynamic Libraries . . . . . . . . . . . 176
11.4 Dynamic Library Versioning . . . . . . . . . . . . . . . . 177

12 Software Package Management Using Debian Tools 181
12.1 Overview. . . . . . . . . . . . . . . . . . . . . . . 182
12.2 The Basis: dpkg . . . . . . . . . . . . . . . . . . . . . 182
12.2.1 Debian Packages . . . . . . . . . . . . . . . . . . 182
12.2.2 Package Installation . . . . . . . . . . . . . . . . . 183
12.2.3 Deleting Packages . . . . . . . . . . . . . . . . . 184
12.2.4 Debian Packages and Source Code . . . . . . . . . . . 185
12.2.5 Package Information. . . . . . . . . . . . . . . . . 185
12.2.6 Package Veriﬁcation . . . . . . . . . . . . . . . . . 188
12.3 Debian Package Management: The Next Generation . . . . . . . 189
12.3.1 APT . . . . . . . . . . . . . . . . . . . . . . 189
12.3.2 Package Installation Using apt-get . . . . . . . . . . . . 189
12.3.3 Information About Packages . . . . . . . . . . . . . . 191
12.3.4 aptitude . . . . . . . . . . . . . . . . . . . . . 192
12.4 Debian Package Integrity . . . . . . . . . . . . . . . . . 194
12.5 The debconf Infrastructure . . . . . . . . . . . . . . . . 195
12.6 alien : Software From Diﬀerent Worlds . . . . . . . . . . . . 196

13 Package Management with RPM and YUM 199
13.1 Introduction. . . . . . . . . . . . . . . . . . . . . . 200
13.2 Package Management Using rpm . . . . . . . . . . . . . . . 201
13.2.1 Installation and Update . . . . . . . . . . . . . . . 201
13.2.2 Deinstalling Packages . . . . . . . . . . . . . . . . 201
13.2.3 Database and Package Queries . . . . . . . . . . . . . 202
13.2.4 Package Veriﬁcation . . . . . . . . . . . . . . . . . 204
13.2.5 The rpm2cpio Program . . . . . . . . . . . . . . . . 204
13.3 YUM . . . . . . . . . . . . . . . . . . . . . . . . 205
13.3.1 Overview . . . . . . . . . . . . . . . . . . . . 205
13.3.2 Package Repositories . . . . . . . . . . . . . . . . 205
13.3.3 Installing and Removing Packages Using YUM . . . . . . . 206
13.3.4 Information About Packages . . . . . . . . . . . . . . 208
13.3.5 Downloading Packages. . . . . . . . . . . . . . . . 210
6 Contents

A Sample Solutions 211

B LPIC-1 Certiﬁcation 219
B.1 Overview. . . . . . . . . . . . . . . . . . . . . . . 219
B.2 Exam LPI-101 . . . . . . . . . . . . . . . . . . . . . 219
B.3 Exam LPI-102 . . . . . . . . . . . . . . . . . . . . . 220
B.4 LPI Objectives In This Manual . . . . . . . . . . . . . . . 221

C Command Index 229

Index 233
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

List of Tables

3.1 The most important ﬁle attributes . . . . . . . . . . . . . . . . . . . . 50

5.1 USB standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

6.1 Diﬀerent SCSI variants . . . . . . . . . . . . . . . . . . . . . . . . . . 80
6.2 Partition types for Linux (hexadecimal) . . . . . . . . . . . . . . . . 82
6.3 Partition type GUIDs for GPT (excerpt) . . . . . . . . . . . . . . . . 84

10.1 Common targets for systemd (selection) . . . . . . . . . . . . . . . . 164
10.2 Compatibility targets for System-V init . . . . . . . . . . . . . . . . . 165
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

List of Figures

4.1 The relationship between various process states . . . . . . . . . . . 55

5.1 Output of lspci on a typical x86-based PC . . . . . . . . . . . . . . . 66
5.2 The usbview program . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

7.1 The /etc/fstab ﬁle (example) . . . . . . . . . . . . . . . . . . . . . . . 117

9.1 A typical /etc/inittab ﬁle (excerpt) . . . . . . . . . . . . . . . . . . . 143
9.2 Upstart conﬁguration ﬁle for job rsyslog . . . . . . . . . . . . . . . . 149

10.1 A systemd unit ﬁle: console- getty.service . . . . . . . . . . . . . . . . 159

12.1 The aptitude program . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

Preface
This manual is an introduction to Linux system administration. Based on knowl-
edge about using Linux, it conveys the most important theoretical and practical
knowledge for the setup and operation of a standalone Linux-based computer.
The manual is geared towards users who have knowledge and experience using
Linux or Unix systems at a level comparable to the tuxcademy manual Introduction
to Linux for Users and Administrators and are looking for a compact but extensive in-
troduction to system administration. Prerequisites are conﬁdence using the shell
and a text editor as well as experience with the common command line tools of a
Linux system.
This manual covers, after an introduction to the signiﬁcance and problems
of system administration, the basics of process, user account, and access control
management, the management of disk partitions, ﬁle systems, and quotas, com-
mon boot loaders, the system start and shutdown process, PC hardware, and li-
brary and package management.
The successful completion of this course or comparable knowledge are neces-
sary in order to take part in further Linux study and to obtain Linux Professional
Institute certiﬁcation.
This courseware package is designed to support the training course as eﬃ-
ciently as possible, by presenting the material in a dense, extensive format for
reading along, revision or preparation. The material is divided in self-contained
chapters detailing a part of the curriculum; a chapter’s goals and prerequisites chapters
are summarized clearly at its beginning, while at the end there is a summary and goals
(where appropriate) pointers to additional literature or web pages with further prerequisites
information.

B Additional material or background information is marked by the “light-
bulb” icon at the beginning of a paragraph. Occasionally these paragraphs
make use of concepts that are really explained only later in the courseware,
in order to establish a broader context of the material just introduced; these
“lightbulb” paragraphs may be fully understandable only when the course-
ware package is perused for a second time after the actual course.

A Paragraphs with the “caution sign” direct your attention to possible prob-
lems or issues requiring particular care. Watch out for the dangerous bends!

C Most chapters also contain exercises, which are marked with a “pencil” icon exercises
at the beginning of each paragraph. The exercises are numbered, and sam-
ple solutions for the most important ones are given at the end of the course-
ware package. Each exercise features a level of diﬃculty in brackets. Exer-
cises marked with an exclamation point (“!”) are especially recommended.

Excerpts from conﬁguration ﬁles, command examples and examples of com-
puter output appear in typewriter type . In multiline dialogs between the user and
the computer, user input is given in bold typewriter type in order to avoid misun-
derstandings. The “” symbol appears where part of a command’s output
had to be omitted. Occasionally, additional line breaks had to be added to make
things ﬁt; these appear as “
12 Preface

”. When command syntax is discussed, words enclosed in angle brack-
ets (“⟨Word⟩”) denote “variables” that can assume diﬀerent values; material in
brackets (“[-f ⟨ﬁle⟩]”) is optional. Alternatives are separated using a vertical bar
(“-a |-b ”).
Important concepts Important concepts are emphasized using “marginal notes” so they can be eas-
definitions ily located; deﬁnitions of important terms appear in bold type in the text as well
as in the margin.
References to the literature and to interesting web pages appear as “[GPL91]”
in the text and are cross-referenced in detail at the end of each chapter.
We endeavour to provide courseware that is as up-to-date, complete and error-
free as possible. In spite of this, problems or inaccuracies may creep in. If you
notice something that you think could be improved, please do let us know, e.g.,
by sending e-mail to
info@tuxcademy.org

(For simplicity, please quote the title of the courseware package, the revision ID
on the back of the title page and the page number(s) in question.) Thank you very
much!

LPIC-1 Certification
These training materials are part of a recommended curriculum for LPIC-1 prepa-
ration. Refer to Appendix B for further information.
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

1
System Administration

Contents
1.1 Introductory Remarks . . . . . . . . . . . . . . . . . . 14
1.2 The Privileged root Account . . . . . . . . . . . . . . . . 14
1.3 Obtaining Administrator Privileges . . . . . . . . . . . . . 16
1.4 Distribution-speciﬁc Administrative Tools . . . . . . . . . . . 18

Goals
• Reviewing a system administrator’s tasks
• Being able to log on as the system administrator
• Being able to assess the advantages and disadvantage of (graphical) admin-
istration tools

Prerequisites
• Basic Linux skills
• Administration skills for other operating systems are helpful

adm1-grundlagen.tex (33e55eeadba676a3 )
14 1 System Administration

1.1 Introductory Remarks
As a mere user of a Linux system, you are well oﬀ: You sit down in front of your
computer, everything is conﬁgured correctly, all the hardware is supported and
works. You have no care in the world since you can call upon a system adminis-
trator who will handle all administrative tasks for you promptly and thoroughly
(that’s what we wish your environment is like, anyway).
Should you be (or strive to be) the system administrator yourself—within your
company or the privacy of your home—then you have your work cut out for you:
You must install and conﬁgure the system and connect any peripherals. Having
done that, you need to keep the system running, for example by checking the sys-
tem logs for unusual events, regularly getting rid of old log ﬁles, making backup
copies, installing new software and updating existing programs, and so on.
Today, in the age of Linux distributions with luxurious installation tools, sys-
tem installation is no longer rocket science. However, an ambitious administrator
can spend lots of time mobilising every last resource on their system. In general,
changes system administration mostly takes place when a noticeable change occurs, for
example when new hardware or software is to be integrated, new users arrive or
existing ones disappear, or hardware problems arise.

Tools B Many Linux distributions these days contain specialised tools to facilitate
system administration. These tools perform diﬀerent tasks ranging from
user management and creating ﬁle systems to complete system updates.
Utilities like these can make these tasks a lot easier but sometimes a lot more
diﬃcult. Standard procedures are simpliﬁed but for specialised settings you
should know the exact relationships between system components. Further-
more, most of these tools are only available for certain distributions.

The administration of a Linux system, as of any other computer system, re-
responsibility quires a considerable amount of responsibility and care. You should not see your-
self as a demigod (at least) but as a service provider. No matter whether you are
the only system administrator—say, on your own computer—or working in a team
communication of colleagues to support a company network: communication is paramount. You
should get used to documenting conﬁguration changes and other administrative
decisions in order to be able to retrace them later. The Linux way of directly edit-
ing text ﬁles makes this convenient, since you can comment conﬁguration settings
right where they are made (a luxury not usually enjoyed with graphical adminis-
tration tools). Do so.

1.2 The Privileged root Account
For many tasks, the system administrator needs special privileges. Accordingly,
he can make use of a special user account called root . As root , a user is the so-called
super user super user. In brief: He may do anything.
The normal ﬁle permissions and security precautions do not apply to root . He
unlimited privileges has allowing him nearly unbounded access to all data, devices and system compo-
nents. He can institute system changes that all other users are prohibited from by
the Linux kernel’s security mechanisms. This means that, as root , you can change
every ﬁle on the system no matter who it belongs to. While normal users cannot
wreak damage (e. g., by destroying ﬁle systems or manipulating other users’ ﬁles),
root is not thus constrained.

B In many cases, these extensive system administrator privileges are really
a liability. For example, when making backup copies it is necessary to be
able to read all ﬁles on the system. However, this by no means implies that
the person making the backup (possibly an intern) should be empowered to
open all ﬁles on the system with a text editor, to read them or change them—
or start a network service which might be accessible from anywhere in the
1.2 The Privileged root Account 15

world. There are various ways of giving out administrator privileges only in
controlled circumstances (such as sudo , a system which lets normal users ex- sudo
ecute certain commands using administrator privileges), of selectively giv-
ing particular privileges to individual process rather than operating on an
“all or nothing” principle (cue POSIX capabilities), or of doing away with POSIX capabilities
the idea of an “omnipotent” system administrator completely (for instance,
SELinux—“security-enhanced Linux”—a freely available software package SELinux
by the American intelligence agency, NSA, contains a “role-based” access
control system that can get by without an omnipotent system administra-
tor).

Why does Linux contain security precautions in the ﬁrst place? The most im- Why Security?
portant reason is for users to be able to determine the access privileges that apply
to their own ﬁles. By setting permission bits (using the chmod command), users
can ascertain that certain ﬁles may be read, written to or executed by certain oth-
ers (or no) users. This helps safeguard the privacy and integrity of their data. You
would certainly not approve of other users being able to read your private e-mail
or change the source code of an important program behind your back.
The security mechanisms are also supposed to keep users from damaging the
system. Access to many of the device ﬁles in /dev corresponding to hardware com- Access control for devices
ponents such as hard disks is constrained by the system. If normal users could ac-
cess disk storage directly, all sorts of mayhem might occur (a user might overwrite
the complete content of a disk or, having obtained information about the layout
of the ﬁlesystem on the disk, access ﬁles that are none of his business). Instead,
the system forces normal users to access the disks via the ﬁle system and protects
their data in that way.
It is important to stress that damage is seldom caused on purpose. The system’s
security mechanisms serve mostly to save users from unintentional mistakes and
misunderstandings; only in the second instance are they meant to protect the pri-
vacy of users and data.
On the system, users can be pooled into groups to which you may assign their groups
own access privileges. For example, a team of software developers could have
read and write permission to a number of ﬁles, while other users are not allowed
to change these ﬁles. Every user can determine for their own ﬁles how permissive
or restrictive access to them should be.
The security mechanisms also prevent normal users from performing certain
actions such as the invocation of speciﬁc system calls from a program. For exam- Privileged system calls
ple, there is a system call that will halt the system, which is executed by programs
such as shutdown when the system is to be powered down or rebooted. If normal
users were allowed to invoke this routine from their own programs, they could
inadvertently (or intentionally) stop the system at any time.
The administrator frequently needs to circumvent these security mechanisms
in order to maintain the system or install updated software versions. The root
account is meant to allow exactly this. A good administrator can do his work
without regard for the usual access permissions and other constraints, since these
do not apply to root . The root account is not better than a normal user account
because it has more privileges; the restriction of these privileges to root is a secu-
rity measure. Since the operating system’s reasonable and helpful protection and
security mechanisms do not apply to the system administrator, working as root
is very risky. You should therefore use root to execute only those commands that
really require the privileges.

B Many of the security problems of other popular operating systems can be
traced back to the fact that normal users generally enjoy administrator priv-
ileges. Thus, programs such as “worms” or “Trojan horses”, which users
often execute by accident, ﬁnd it easy to establish themselves on the sys-
tem. With a Linux system that is correctly installed and operated, this is
hardly possible since users read their e-mail without administrator privi-
16 1 System Administration

leges, but administrator privileges are required for all system-wide conﬁg-
uration changes.

B Of course, Linux is not magically immune against malicious pests like
“mail worms”; somebody could write and make popular a mail program
that would execute “active content” such as scripts or binary programs
within messages like some such programs do on other operating systems.
On Linux, such a “malicious” program from elsewhere could remove all
the caller’s ﬁles or try to introduce “Trojan” code to his environment, but
it could not harm other users nor the system itself—unless it exploited a
security vulnerability in Linux that would let a local user gain administrator
privileges “through the back door” (such vulnerabilities are detected now
and again, and patches are promptly published which you should install in
a timely manner).

Exercises
C 1.1 [2] What is the diﬀerence between a user and an administrator? Name
examples for tasks and actions (and suitable commands) that are typically
performed from a user account and the root account, respectively.

C 1.2 [!1] Why should you, as a normal user, not use the root account for your
daily work?

C 1.3 [W]hat about access control on your computer at home? Do you work
from an administrator account?

1.3 Obtaining Administrator Privileges
There are two ways of obtaining administrator privileges:
1. You can log in as user root directly. After entering the correct root password
you will obtain a shell with administrator privileges. However, you should
avoid logging in to the GUI as root , since then all graphical applications in-
cluding the X server would run with root privileges, which is not necessary
and can lead to security problems. Nor should direct root logins be allowed
across the network.

B You can determine which terminals are eligible for direct root login
by listing them in the /etc/securetty ﬁle. The default setting is usually
“all virtual consoles and /dev/ttyS0 ” (the latter for users of the “serial
console”).

2. You can, from a normal shell, use the su command to obtain a new shell with
administrator privileges. su , like login , asks for a password and opens the
root shell only after the correct root password has been input. In GUIs like
KDE there are similar methods.
(See also Introduction to Linux for Users and Administrators.)
Single-user systems, too! Even if a Linux system is used by a single person only, it makes sense to create
a normal account for this user. During everyday work on the system as root , most
of the kernel’s normal security precautions are circumvented. That way errors can
occur that impact on the whole system. You can avoid this danger by logging into
your normal account and starting a root shell via “/bin/su - ” if and when required.

B Using su , you can also assume the identity of arbitrary other users (here hugo )
by invoking it like
1.3 Obtaining Administrator Privileges 17

$ /bin/su - hugo

You need to know the target user’s password unless you are calling su as
user root .
The second method is preferable to the ﬁrst for another reason, too: If you use
the su command to become root after logging in to your own account, su creates a
message like
Apr 1 08:18:21 HOST su: (to root) user1 on /dev/tty2

in the system log (such as /var/log/messages ). This entry means that user user1 suc- system log
cessfully executed su to become root on terminal 2. If you log in as root directly,
no such message is logged; there is no way of ﬁguring out which user has fooled
around with the root account. On a system with several administrators it is often
important to retrace who entered the su command when.
Ubuntu is one of the “newfangled” distributions that deprecate–and, in the
default setup, even disable—logging in as root . Instead, particular users
may use the sudo mechanism to execute individual commands with admin-
istrator privileges. Upon installation, you are asked to create a “normal”
user account, and that user account is automatically endowed with “indi-
rect” administrator privileges.
When installing Debian GNU/Linux, you can choose between assigning a
password to the root account and thereby enabling direct administrator lo-
gins, and declining this and, as on Ubuntu, giving sudo -based administrator
privileges to the ﬁrst unprivileged user account created as part of the instal-
lation process.
On many systems, the shell prompt diﬀers between root and the other users. shell prompt
The classic root prompt contains a hash mark (# ), while other users see a prompt
containing a dollar sign ($ ) or greater-than sign (> ). The # prompt is supposed
to remind you that you are root with all ensuing privileges. However, the shell
prompt is easily changed, and it is your call whether to follow this convention or
not.

B Of course, if you are using sudo , you never get to see a prompt for root .
Like all powerful tools, the root account can be abused. Therefore it is impor- Misuse of root
tant for you as the system administrator too keep the root password secret. It
should only be passed on to users who are trusted both professionally and per-
sonally (or who can be held responsible for their actions). If you are the sole user
of the system this problem does not apply to you.
Too many cooks spoil the broth! This principle also applies to system admin- Administration: alone or by
istration. The main beneﬁt of “private” use of the root account is not that the many
possibility of misuse is minimised (even though this is surely a consequence).
More importantly, root as the sole user of the root account knows the complete
system conﬁguration. If somebody besides the administrator can, for example,
change important system ﬁles, then the system conﬁguration could be changed
without the administrator’s knowledge. In a commercial environment, it is nec-
essary to have several suitably privileged employees for various reasons—for ex-
ample, safeguarding system operation during holidays or sudden severe illness
of the administrator—; this requires close cooperation and communication.
If there is only one system administrator who is responsible for system con-
ﬁguration, you can be sure that one person really knows what is going on on the
system (at least in theory), and the question of accountability also has an obvi- accountability
ous asnwer. The more users have access to root , the greater is the probability that
somebody will commit an error as root at some stage. Even if all users with root
access possess suitable administration skills, mistakes can happen to anybody.
Prudence and thorough training are the only precautions against accidents.
18 1 System Administration

There are a few other useful tools for team-based system administration.
For example, Debian GNU/Linux and Ubuntu support a package called
etckeeper , which allows storing the complete content of the /etc directory in
a revision control system such as Git or Mercurial. Revision control systems
(which we cannot cover in detail here) make it possible to track changes to
ﬁles in a directory hierarchy in a very detailed manner, to comment them
and, if necessary, to undo them. With Git or Mercurial it is even possible to
store a copy of the /etc directory on a completely diﬀerent computer and to
keep it in sync automatically—great protection from accidents.

Exercises
C 1.4 [2] What methods exist to obtain administrator rights? Which method
is better? Why?

C 1.5 [!2] On a conventionally conﬁgured system, how can you recognise
whether you are working as root ?

C 1.6 [2] Log in as a normal user (e. g., test ). Change over to root and back to
test . How do you work best if you frequently need to change between both
these accounts (for example, to check on the results of a new conﬁguration)?

C 1.7 [!2] Log in as a normal user and change to root using su . Where do you
ﬁnd a log entry documenting this change? Look at that message.

1.4 Distribution-specific Administrative Tools
Many Linux distributions try to stand out in the crowd by providing more or less
ingenious tools that are supposed to simplify system administration. These tools
are usually tailored to the distributions in question. Here are a few comments
about typical specimens:

A familiar sight to SUSE administrators is “YaST”, the graphical adminis-
tration interface of the SUSE distributions (it also runs on a text screen). It
allows the extensive conﬁguration of many aspects of the system either by
directly changing the conﬁguration ﬁles concerned or by manipulating ab-
stract conﬁguration ﬁles below /etc/sysconfig which are then used to adapt
the real conﬁguration ﬁles by means of the SuSEconfig tool. For some tasks
such as network conﬁguration, the ﬁles below /etc/sysconfig are the actual
conﬁguration ﬁles.

Unfortunately, YaST is not a silver bullet for all problems of system admin-
istration. Even though many aspects of the system are amenable to YaST-
based administration, important settings may not be accessible via YaST, or
the YaST modules in question simply do not work correctly. The danger
zone starts where you try to administer the computer partly through YaST
and partly through changing conﬁguration ﬁles manually: Yast does exer-
cise some care not to overwrite your changes (which wasn’t the case in the
past—up till SuSe 6 or so, YaST and SuSEconﬁg used to be quite reckless),
but will then not perform its own changes such that they really take eﬀect in
the system. In other places, manual changes to the conﬁguration ﬁles will
actually show up in YaST. Hence you have to have some “insider knowl-
edge” and experience in order to assess which conﬁguration ﬁles you may
change directly and which your grubby ﬁngers had better not touch.

Some time ago, Novell released the YaST source code under the GPL (in
SUSE’s time it used to be available but not under a “free” licence). However,
so far no other distribution of consequence has adapted YaST to its purposes,
let alone made it a standard tool (SUSE fashion).
1.4 Distribution-specific Administrative Tools 19

B The Webmin package by Jamie Cameron (http://www.webmin.com/ ) allows the
convenient administration of various Linux distributions (or Unix versions)
via a web-based interface. Webmin is very extensive and oﬀers special fa-
cilities for administering “virtual” servers (for web hosters and their cus-
tomers). However you may have to install it yourself, since most distribu-
tions do not provide it. Webmin manages its own users, which means that
you can extend administrator privileges to users who do not have interac-
tive system access. (Whether that is a smart idea is a completely diﬀerent
question.)

Most administration tools like YaST and Webmin share the same disadvan-
tages:
• They are not extensive enough to take over all aspects of system administra-
tions, and as an administrator you have to have detailed knowledge of their
limits in order to be able to decide where to intervene manually.
• They make system administration possible for people whose expertise is
not adequate to assess the possible consequences of their actions or to ﬁnd
and correct mistakes. Creating a user account using an administration tool
is certainly not a critical job and surely more convenient than editing four
diﬀerent system ﬁles using vi , but other tasks such as conﬁguring a ﬁre-
wall or mail server are not suitable for laypeople even using a convenient
administration tool. The danger is that inexperienced administrators will
use an administration tool to attempt tasks which do not look more com-
plicated than others but which, without adequate background knowledge,
may endanger the safety and/or reliability of the system.
• They usually do not oﬀer a facility to version control or document any
changes made, and thus complicate teamwork and auditing by requiring
logs to be kept externally.
• They are often intransparent, i. e., they do not provide documentation about
the actual steps they take on the system to perform administrative tasks.
This keeps the knowledge about the necessary procedures buried in the pro-
grams; as the administrator you have no direct way of “learning” from the
programs like you could by observing an experienced administrator. Thus
the adminstration tools keep you artiﬁcially stupid.
• As an extension of the previous point: If you need to administer several
computers, common administration tools force you to execute the same
steps repeatedly on every single machine. Many times it would be more
convenient to write a shell script automating the required procedure, and to
execute it automatically on every computer using, e. g., the “secure shell”,
but the administration tool does not tell you what to put into this shell
script. Therefore, viewed in a larger context, their use is ineﬃcient.
From various practical considerations like these we would like to recommend
against relying too much on the “convenient” administration tools provided by
the distributions. They are very much like training wheels on a bicycle: They
work eﬀectively against falling over too early and provide a very large sense of
achievement very quickly, but the longer the little ones zoom about with them, the
more diﬃcult it becomes to get them used to “proper” bike-riding (here: doing
administration in the actual conﬁguration ﬁles, including all advantages such as
documentation, transparency, auditing, team capability, transportability, …).
Excessive dependence on an administration tool also leads to excessive depen-
dence on the distribution featuring that tool. This may not seem like a real liabil-
ity, but on the other hand one of the more important advantages of Linux is the fact
that there are multiple independent vendors. So, if one day you should be fed up
with the SUSE distributions (for whatever reason) and want to move over to Red
Hat or Debian GNU/Linux, it would be very inconvenient if your administrators
20 1 System Administration

knew only YaST and had to relearn Linux administration from scratch. (Third-
party administration tools like Webmin do not exhibit this problem to the same
degree.)

Exercises
C 1.8 [!2] Does your distribution provide an administration tool (such as
YaST)? What can you do with it?

C 1.9 [3] (Continuation of the previous exercise—when working through the
manual for the second time.) Find out how your administration tool works.
Can you change the system conﬁguration manually so the administration
tool will notice your changes? Only under some circumstances?

C 1.10 [!1] Administration tools like Webmin are potentially accessible to ev-
erybody with a browser. Which advantages and disadvantages result from
this?

Summary
• Every computer installation needs a certain amount of system administra-
tion. In big companies, universities and similar institutions these services
are provided by (teams of) full-time administrators; in smaller companies
or private households, (some) users usually serve as administrators.
• Linux systems are, on the whole, straightforward to administer. Work arises
mostly during the initial installation and, during normal operation, when
the conﬁguration changes noticeably.
• On Linux systems, there usually is a privileged user account called root , to
which the normal security mechanisms do not apply.
• As an administrator, one should not work as root exclusively, but use a nor-
mal user account and assume root privileges only if necessary.
• Administration tools such as YaST or Webmin can help perform some ad-
ministrative duties, but are no substitute for administrator expertise and
may have other disadvantages as well.
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

2
User Administration

Contents
2.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . 22
2.1.1 Why Users? . . . . . . . . . . . . . . . . . . . . 22
2.1.2 Users and Groups . . . . . . . . . . . . . . . . . 23
2.1.3 People and Pseudo-Users . . . . . . . . . . . . . . . 25
2.2 User and Group Information . . . . . . . . . . . . . . . . 25
2.2.1 The /etc/passwd File . . . . . . . . . . . . . . . . . 25
2.2.2 The /etc/shadow File . . . . . . . . . . . . . . . . . 28
2.2.3 The /etc/group File . . . . . . . . . . . . . . . . . 30
2.2.4 The /etc/gshadow File . . . . . . . . . . . . . . . . . 31
2.2.5 The getent Command . . . . . . . . . . . . . . . . 32
2.3 Managing User Accounts and Group Information . . . . . . . . 32
2.3.1 Creating User Accounts . . . . . . . . . . . . . . . 33
2.3.2 The passwd Command . . . . . . . . . . . . . . . . 34
2.3.3 Deleting User Accounts . . . . . . . . . . . . . . . 36
2.3.4 Changing User Accounts and Group Assignment . . . . . . 36
2.3.5 Changing User Information Directly—vipw . . . . . . . . . 37
2.3.6 Creating, Changing and Deleting Groups . . . . . . . . . 37

Goals
• Understanding the user and group concepts of Linux
• Knowing how user and group information is stored on Linux
• Being able to use the user and group administration commands

Prerequisites
• Knowledge about handling conﬁguration ﬁles

adm1-benutzer.tex (33e55eeadba676a3 )
22 2 User Administration

2.1 Basics
2.1.1 Why Users?
Computers used to be large and expensive, but today an oﬃce workplace without
its own PC (“personal computer”) is nearly inconceivable, and a computer is likely
to be encountered in most domestic “dens” as well. And while it may be suﬃcient
for a family to agree that Dad, Mom and the kids will put their ﬁles into diﬀerent
directories, this will no longer do in companies or universities—once shared disk
space or other facilities are provided by central servers accessible to many users,
the computer system must be able to distinguish between diﬀerent users and to
assign diﬀerent access rights to them. After all, Ms Jones from the Development
Division has as little business looking at the company’s payroll data as Mr Smith
from Human Resources has accessing the detailed plans for next year’s products.
And a measure of privacy may be desired even at home—the Christmas present
list or teenage daughter’s diary (erstwhile ﬁtted with a lock) should not be open
to prying eyes as a matter of course.

B We shall be discounting the fact that teenage daughter’s diary may be visible
to the entire world on Facebook (or some such); and even if that is the case,
the entire world should surely not be allowed to write to teenage daughter’s
dairy. (Which is why even Facebook supports the notion of diﬀerent users.)

The second reason for distinguishing between diﬀerent users follows from the
fact that various aspects of the system should not be visible, much less change-
able, without special privileges. Therefore Linux manages a separate user iden-
tity (root ) for the system administrator, which makes it possible to keep informa-
tion such as users’ passwords hidden from “common” users. The bane of older
Windows systems—programs obtained by e-mail or indiscriminate web surﬁng
that then wreak havoc on the entire system—will not plague you on Linux, since
anything you can execute as a common user will not be in a position to wreak
system-wide havoc.

A Unfortunately this is not entirely correct: Every now and then a bug comes
to light that enables a “normal user” to do things otherwise restricted to
administrators. This sort of error is extremely nasty and usually corrected
very quickly after having been found, but there is a considerable chance that
such a bug has remained undetected in the system for an extended period
of time. Therefore, on Linux (as on all other operating systems) you should
strive to run the most current version of critical system parts like the kernel
that your distributor supports.

A Even the fact that Linux safeguards the system conﬁguration from unau-
thorised access by normal users should not entice you to shut down your
brain. We do give you some advice (such as not to log in to the graphical
user interface as root ), but you should keep thinking along. E-mail messages
asking you to view web site 𝑋 and enter your credit card number and PIN
there can reach you even on Linux, and you should disregard them in the
same way as everywhere else.

user accounts Linux distinguishes between diﬀerent users by means of diﬀerent user ac-
counts. The common distributions typically create two user accounts during
installation, namely root for administrative tasks and another account for a “nor-
mal” user. You (as the administrator) may add more accounts later, or, on a client
PC in a larger network, they may show up automatically from a user account
database stored elsewhere.

B Linux distinguishes between user accounts, not users. For example, no one
keeps you from using a separate user account for reading e-mail and surﬁng
the web, if you want to be 100% sure that things you download from the
2.1 Basics 23

Net have no access to your important data (which might otherwise happen
in spite of the user/administrator divide). With a little cunning you can
even display a browser and e-mail program running under your “surﬁng
account” among your “normal” programs1 .

Under Linux, every user account is assigned a unique number, the so-called
user ID (or UID, for short). Every user account also features a textual user name UID
(such as root or joe ) which is easier to remember for humans. In most places where user name
it counts—e. g., when logging in, or in a list of ﬁles and their owners—Linux will
use the textual name whenever possible.

B The Linux kernel does not know anything about textual user names; process
data and the ownership data in the ﬁlesystem use the UID exclusively. This
may lead to diﬃculties if a user is deleted while he still owns ﬁles on the
system, and the UID is reassigned to a diﬀerent user. That user “inherits”
the previous UID owner’s ﬁles.

B There is no technical problem with assigning the same (numerical) UID to
diﬀerent user names. These users have equal access to all ﬁles owned by that
UID, but every user can have his own password. You should not actually
use this (or if you do, use it only with great circumspection).

2.1.2 Users and Groups
To work with a Linux computer you need to log in ﬁrst. This allows the system
to recognise you and to assign you the correct access rights (of which more later).
Everything you do during your session (from logging in to logging out) happens
under your user account. In addition, every user has a home directory, where home directory
only they can store and manage their own ﬁles, and where other users often have
no read permission and very emphatically no write permission. (Only the system
administrator—root —may read and write all ﬁles.)

A Depending on which Linux distribution you use (cue: Ubuntu) it may be
possible that you do not have to log into the system explicitly. This is be-
cause the computer “knows” that it will usually be you and simply assumes
that this is going to be the case. You are trading security for convenience; this
particular deal probably makes sense only where you can stipulate with rea-
sonable certainty that nobody except you will switch on your computer—
and hence should be restricted by rights to the computer in your single-
person household without a cleaner. We told you so.

Several users who want to share access to certain system resources or ﬁles can
form a group. Linux identiﬁes group members either ﬁxedly by name or tran- group
siently by a login procedure similar to that for users. Groups have no “home di-
rectories” like users do, but as the administrator you can of course create arbitrary
directories meant for certain groups and having appropriate access rights.
Groups, too, are identiﬁed internally using numerical identiﬁers (“group IDs”
or GIDs).

B Group names relate to GIDs as user names to UIDs: The Linux kernel only
knows about the former and stores only the former in process data or the
ﬁle system.

Every user belongs to a primary group and possibly several secondary or addi-
tional groups. In a corporate setting it would, for example, be possible to introduce
project-speciﬁc groups and to assign the people collaborating on those projects
to the appropriate group in order to allow them to manage common data in a
directory only accessible to group members.
1 Which of course is slightly more dangerous again, since programs runninig on the same screen

can communicate with one another
24 2 User Administration

For the purposes of access control, all groups carry equivalent weight—every
user always enjoys all rights deriving from all the groups that he is a member of.
The only diﬀerence between the primary and secondary groups is that ﬁles newly
created by a user are usually2 assigned to his primary group.

B Up to (and including) version 2.4 of the Linux kernel, a user could be a mem-
ber of at most 32 additional groups; since Linux 2.6 the number of secondary
groups is unlimited.

You can ﬁnd out a user account’s UID, the primary and secondary groups and
the corresponding GIDs by means of the id program:

$ id
uid=1000(joe) gid=1000(joe) groups=24(cdrom),29(audio),44(video),
1000(joe)
$ id root
uid=0(root) gid=0(root) groups=0(root)

B With the options -u , -g , and -G , id lets itself be persuaded to output just the
account’s UID, the GID of the primary group, or the GIDs of the secondary
groups. (These options cannot be combined.) With the additional option -n
you get names instead of numbers:

$ id -G
1000 24 29 44
$ id -Gn
joe cdrom audio video

B The groups command yields the same result as the ”‘id -Gn ”’ command.
last You can use the last command to ﬁnd who logged into your computer and
when (and, in the case of logins via the network, from where):

$ last
joe pts/1 pcjoe.example.c Wed Feb 29 10:51 still logged in
bigboss pts/0 pc01.example.c Wed Feb 29 08:44 still logged in
joe pts/2 pcjoe.example.c Wed Feb 29 01:17 - 08:44 (07:27)
sue pts/0 :0 Tue Feb 28 17:28 - 18:11 (00:43)

reboot system boot 3.2.0-1-amd64 Fri Feb 3 17:43 - 13:25 (4+19:42)

For network-based sessions, the third column speciﬁes the name of the ssh client
computer. “:0 ” denotes the graphical screen (the ﬁrst X server, to be exact—there
might be more than one).

B Do also note the reboot entry, which tells you that the computer was started.
The third column contains the version number of the Linux operating sys-
tem kernel as provided by “uname -r ”.

With a user name, last provides information about a particular user:

$ last
joe pts/1 pcjoe.example.c Wed Feb 29 10:51 still logged in
joe pts/2 pcjoe.example.c Wed Feb 29 01:17 - 08:44 (07:27)

2 The exception occurs where the owner of a directory has decreed that new ﬁles and subdirectories

within this directory are to be assigned to the same group as the directory itself. We mention this
strictly for completeness.
2.2 User and Group Information 25

B You might be bothered (and rightfully so!) by the fact that this somewhat
sensitive information is apparently made available on a casual basis to arbi-
trary system users. If you (as the administrator) want to protect your users’
privacy somewhat better than you Linux distribution does by default, you
can use the
# chmod o-r /var/log/wtmp

command to remove general read permissions from the ﬁle that last con-
sults for the telltale data. Users without administrator privileges then get to
see something like

$ last
last: /var/log/wtmp: Permission denied

2.1.3 People and Pseudo-Users
Besides “natural” persons—the system’s human users—the user and group con-
cept is also used to allocate access rights to certain parts of the system. This means
that, in addition to the personal accounts of the “real” users like you, there are fur-
ther accounts that do not correspond to actual human users but are assigned to pseudo-users
administrative functions internally. They deﬁne functional “roles” with their own
accounts and groups.
After installing Linux, you will ﬁnd several such pseudo-users and groups in
the /etc/passwd and /etc/group ﬁles. The most important role is that of the root user
(which you know) and its eponymous group. The UID and GID of root are 0 (zero).

B root ’s privileges are tied to UID 0; GID 0 does not confer any additional
access privileges.

Further pseudo-users belong to certain software systems (e. g., news for Usenet
news using INN, or postfix for the Postﬁx mail server) or certain components or
devices (such as printers, tape or ﬂoppy drives). You can access these accounts,
if necessary, like other user accounts via the su command. These pseudo-users pseudo-users for privileges
are helpful as ﬁle or directory owners, in order to ﬁt the access rights tied to ﬁle
ownership to special requirements without having to use the root account. The
same appkies to groups; the members of the disk group, for example, have block-
level access to the system’s disks.

Exercises
C 2.1 [1] How does the operating system kernel diﬀerentiate between various
users and groups?

C 2.2 [2] What happens if a UID is assigned to two diﬀerent user names? Is
that allowed?

C 2.3 [1] What is a pseudo-user? Give examples!

C 2.4 [2] (On the second reading.) Is it acceptable to assign a user to group
disk who you would not want to trust with the root password? Why (not)?

2.2 User and Group Information
2.2.1 The /etc/passwd File
The /etc/passwd ﬁle is the system user database. There is an entry in this ﬁle for
every user on the system—a line consisting of attributes like the Linux user name,
26 2 User Administration

“real” name, etc. After the system is ﬁrst installed, the ﬁle contains entries for
most pseudo-users.
The entries in /etc/passwd have the following format:

⟨user name⟩: ⟨password⟩: ⟨UID⟩: ⟨GID⟩: ⟨GECOS⟩: ⟨home directory⟩: ⟨shell⟩

⟨user name⟩ This name should consist of lowercase letters and digits; the ﬁrst char-
acter should be a letter. Unix systems often consider only the ﬁrst eight
characters—Linux does not have this limitation but in heterogeneous net-
works you should take it into account.

A Resist the temptation to use umlauts, punctuation and similar special
characters in user names, even if the system lets you do so—not all
tools that create new user accounts are picky, and you could of course
edit /etc/passwd by hand. What seems to work splendidly at ﬁrst glance
may lead to problems elsewhere later.

B You should also stay away from user names consisting of only upper-
case letters or only digits. The former may give their owners trouble
logging in (see exercise 2.6), the latter can lead to confusion, especially
if the numerical user name does not equal the account’s numerical
UID. Commands such as ”‘ls -l ”’ will display the UID if there is no
corresponding entry for it in /etc/passwd , and it is not exactly straight-
forward to tell UIDs from purely numerical user names in ls output.

⟨password⟩ Traditionally, this ﬁeld contains the user’s encrypted password. Today,
most Linux distributions use “shadow passwords”; instead of storing the
password in the publically readable /etc/passwd ﬁle, it is stored in /etc/shadow
which can only be accessed by the administrator and some privileged pro-
grams. In /etc/passwd , a “x ” calls attention to this circumstance. Every user
can avail himself of the passwd program to change his password.
⟨UID⟩ The numerical user identiﬁer—a number between 0 and 232 − 1. By con-
vention, UIDs from 0 to 99 (inclusive) are reserved for the system, UIDs
from 100 to 499 are for use by software packages if they need pseudo-user
accounts. With most popular distributions, “real” users’ UIDs start from
500 (or 1000).
Precisely because the system diﬀerentiates between users not by name but
by UID, the kernel treats two accounts as completely identical if they con-
tain diﬀerent user names but the same UID—at least as far as the access
privileges are concerned. Commands that display a user name (e. g., ”‘ls
-l ”’ or id ) show the one used when the user logged in.

primary group ⟨GID⟩ The GID of the user’s primary group after logging in.

The Novell/SUSE distributions (among others) assign a single group
such as users as the shared primary group of all users. This method is
quite established as well as easy to understand.

Many distributions, such as those by Red Hat or Debian GNU/Linux,
create a new group whenever a new account is created, with the GID
equalling the account’s UID. The idea behind this is to allow more
sophisticated assignments of rights than with the approach that puts
all users into the same group users . Consider the following situation:
Jim (user name jim ) is the personal assistant of CEO Sue (user name
sue ). In this capacity he sometimes needs to access ﬁles stored inside
Sue’s home directory that other users should not be able to get at. The
method used by Red Hat, Debian & co., “one group per user”, makes it
straightforward to put user jim into group sue and to arrange for Sue’s
2.2 User and Group Information 27

ﬁles to be readable for all group members (the default case) but not oth-
ers. With the “one group for everyone” approach it would have been
necessary to introduce a new group completely from scratch, and to
reconﬁgure the jim and sue accounts accordingly.

By virtue of the assignment in /etc/passwd , every user must be a member of
at least one group.

B The user’s secondary groups (if applicable) are determined from en-
tries in the /etc/group ﬁle.

⟨GECOS⟩ This is the comment ﬁeld, also known as the “GECOS ﬁeld”.

B GECOS stands for “General Electric Comprehensive Operating Sys-
tem” and has nothing whatever to do with Linux, except that in the
early days of Unix this ﬁeld was added to /etc/passwd in order to keep
compatibility data for a GECOS remote job entry service.

This ﬁeld contains various bits of information about the user, in particular
his “real” name and optional data such as the oﬃce number or telephone
number. This information is used by programs such as mail or finger . The
full name is often included in the sender’s address by news and mail soft-
ware.

B Theoretically there is a program called chfn that lets you (as a user)
change the content of your GECOS ﬁeld. Whether that works in any
particular case is a diﬀerent question, since at least in a corporate set-
ting one does not necessarily want to allow people to change their
names at a whim.

⟨home directory⟩ This directory is that user’s personal area for storing his own ﬁles.
A newly created home directory is by no means empty, since a new user
normally receives a number of “proﬁle” ﬁles as his basic equipment. When
a user logs in, his shell uses his home directory as its current directory, i. e.,
immediately after logging in the user is deposited there.
⟨shell⟩ The name of the program to be started by login after successful authentication—
this is usually a shell. The seventh ﬁeld extends through the end of the line.

B The user can change this entry by means of the chsh program. The
eligible programs (shells) are listed in the /etc/shells ﬁle. If a user is
not supposed to have an interactive shell, an arbitrary program, with
arguments, can be entered here (a common candidate is /bin/true ). This
ﬁeld may also remain empty, in which case the standard shell /bin/sh
will be started.

B If you log in to a graphical environment, various programs will be
started on your behalf, but not necessarily an interactive shell. The
shell entry in /etc/passwd comes into its own, however, when you in-
voke a terminal emulator such as xterm or konsole , since these programs
usually check it to identify your preferred shell.

Some of the ﬁelds shown here may be empty. Absolutely necessary are only the
user name, UID, GID and home directory. For most user accounts, all the ﬁelds
will be ﬁlled in, but pseudo-users might use only part of the ﬁelds.
The home directories are usually located below /home and take their name from home directories
their owner’s user name. In general this is a fairly sensible convention which
makes a given user’s home directory easy to ﬁnd. In theory, a home directory
might be placed anywhere in the ﬁle system under a completely arbitrary name.

B On large systems it is common to introduce one or more additional levels
of directories between /home and the “user name” directory, such as
28 2 User Administration

/home/hr/joe Joe from Human Resources
/home/devel/sue Sue from Development
/home/exec/bob Bob the CEO

There are several reasons for this. On the one hand this makes it easier to
keep one department’s home directory on a server within that department,
while still making it available to other client computers. On the other hand,
Unix (and some Linux) ﬁle systems used to be slow dealing with directories
containing very many ﬁles, which would have had an unfortunate impact
on a /home with several thousand entries. However, with current Linux ﬁle
systems (ext3 with dir_index and similar) this is no longer an issue.

Note that as an administrator you should not really be editing /etc/passwd by
tools hand. There is a number of programs that will help you create and maintain user
accounts.

B In principle it is also possible to store the user database elsewhere than in
/etc/passwd . On systems with very many users (thousands), storing user
data in a relational database is preferable, while in heterogeneous networks
a shared multi-platform user database, e. g., based on an LDAP directory,
might recommend itself. The details of this, however, are beyond the scope
of this course.

2.2.2 The /etc/shadow File
For security, nearly all current Linux distributions store encrypted user passwords
in the /etc/shadow ﬁle (“shadow passwords”). This ﬁle is unreadable for normal
users; only root may write to it, while members of the shadow group may read it in
addition to root . If you try to display the ﬁle as a normal user an error occurs.

B Use of /etc/shadow is not mandatory but highly recommended. However
there may be system conﬁgurations where the additional security aﬀorded
by shadow passwords is nulliﬁed, for example if NIS is used to export user
data to other hosts (especially in heterogeneous Unix environments).

format Again, this ﬁle contains one line for each user, with the following format:

⟨user name⟩: ⟨password⟩: ⟨change⟩: ⟨min⟩: ⟨max⟩
: ⟨warn⟩: ⟨grace⟩: ⟨lock⟩: ⟨reserved⟩

For example:

root:gaY2L19jxzHj5:10816:0:10000::::
daemon:*:8902:0:10000::::
joe:GodY6c5pZk1xs:10816:0:10000::::

Here is the meaning of the individual ﬁelds:
⟨user name⟩ This must correspond to an entry in the /etc/passwd ﬁle. This ﬁeld
“joins” the two ﬁles.
⟨password⟩ The user’s encrypted password. An empty ﬁeld generally means that
the user can log in without a password. An asterisk or an exclamation point
prevent the user in question from logging in. It is common to lock user’s ac-
counts without deleting them entirely by placing an asterisk or exclamation
point at the beginning of the corresponding password.
⟨change⟩ The date of the last password change, in days since 1 January 1970.
2.2 User and Group Information 29

⟨min⟩ The minimal number of days that must have passed since the last password
change before the password may be changed again.
⟨max⟩ The maximal number of days that a password remains valid without hav-
ing to be changed. After this time has elapsed the user must change his
password.
⟨warn⟩ The number of days before the expiry of the ⟨max⟩ period that the user will
be warned about having to change his password. Generally, the warning
appears when logging in.
⟨grace⟩ The number of days, counting from the expiry of the ⟨max⟩ period, after
which the account will be locked if the user does not change his password.
(During the time from the expiry of the ⟨max⟩ period and the expiry of this
grace period the user may log in but must immediately change his pass-
word.)
⟨lock⟩ The date on which the account will be deﬁnitively locked, again in days
since 1 January 1970.
Some brief remarks concerning password encryption are in order. You might password encryption
think that if passwords are encrypted they can also be decrypted again. This would
open all of the system’s accounts to a clever cracker who manages to obtain a copy
of /etc/shadow . However, in reality this is not the case, since password “encryption”
is a one-way street. It is impossible to recover the decrypted representation of a
Linux password from the “encrypted” form because the method used for encryp-
tion prevents this. The only way to “crack” the encryption is by encrypting likely
passwords and checking whether they match what is in /etc/shadow .

B Let’s assume you select the characters of your password from the 95 vis-
ible ASCII characters (uppercase and lowercase letters are distinguished).
This means that there are 95 diﬀerent one-character passwords, 952 = 9025
two-character passwords, and so on. With eight characters you are already
up to 6.6 quadrillion (6.6 ⋅ 1015 ) possibilities. Stipulating that you can trial-
encrypt 10 million passwords per second (not entirely unrealistic on current
hardware), this means you would require approximately 21 years to work
through all possible passwords. If you are in the fortunate position of own-
ing a modern graphics card, another acceleration by a factor of 50–100 is
quite feasible, which makes that about two months. And then of course
there are handy services like Amazon’s EC2, which will provide you (or
random crackers) with almost arbitrary CPU power, or the friendly neigh-
bourhood Russian bot net … so don’t feel too safe.

B There are a few other problems. The traditional method (usually called
“crypt” or “DES”—the latter because it is based on, but not identical to, the
eponymous encryption method3 ) should no longer be used if you can avoid
it. It has the unpleasant property of only looking at the ﬁrst eight characters
of the entered password, and clever crackers can nowadays buy enough disk
space to build a pre-encrypted cache of the 50 million (or so) most common
passwords. To “crack” a password they only need to search their cache for
the encrypted password, which can be done extremely quickly, and read oﬀ
the corresponding clear-text password.

B To make things even more laborious, when a newly entered password is
encrypted the system traditionally adds a random element (the so-called
3 If you must know exactly: The clear-text password is used as the key (!) to encrypt a constant

string (typically a sequence of zero bytes). A DES key is 56 bits, which just happens to be 8 characters
of 7 bits each (as the leftmost bit in each character is ignored). This process is repeated for a total of
25 rounds, with the previous round’s output serving as the new input. Strictly speaking the encryption
scheme used isn’t quite DES but changed in a few places, to make it less feasible to construct a special
password-cracking computer from commercially available DES encryption chips.
30 2 User Administration

“salt”) which selects one of 4096 diﬀerent possibilities for the encrypted
password. The main purpose of the salt is to avoid random hits result-
ing from user 𝑋, for some reason or other, getting a peek at the content
of /etc/shadow and noting that his encrypted password looks just like that
of user 𝑌 (hence letting him log into user 𝑌’s account using his own clear-
text password). For a pleasant side eﬀect, the disk space required for the
cracker’s pre-encrypted dictionary from the previous paragraph is blown
up by a factor of 4096.

B Nowadays, password encryption is commonly based on the MD5 algorithm,
allows for passwords of arbitrary length and uses a 48-bit salt instead of
the traditional 12 bits. Kindly enough, the encryption works much more
slowly than “crypt”, which is irrelevant for the usual purpose (checking a
password upon login—you can still encrypt several hundred passwords per
second) but does encumber clever crackers to a certain extent. (You should
not let yourself be bothered by the fact that cryptographers poo-poo the
MD5 scheme as such due to its insecurity. As far as password encryption is
concerned, this is fairly meaningless.)

A You should not expect too much of the various password administration pa-
rameters. They are being used by the text console login process, but whether
other parts of the system (such as the graphical login screen) pay them any
notice depends on your setup. Nor is there usually an advantage in forc-
ing new passwords on users at short intervals—this usually results in a se-
quence like bob1 , bob2 , bob3 , …, or users alternate between two passwords.
A minimal interval that must pass before a user is allowed to change their
password again is outright dangerous, since it may give a cracker a “win-
dow” for illicit access even though the user knows their password has been
compromised.

The problem you need to cope with as a system administrator is usually not
people trying to crack your system’s passwords by “brute force”. It is much more
promising, as a rule, to use “social engineering”. To guess your password, the
clever cracker does not start at a , b , and so on, but with your spouse’s ﬁrst name,
your kids’ ﬁrst names, your car’s plate number, your dog’s birthday et cetera. (We
do not in any way mean to imply that you would use such a stupid password. No,
no, not you by any means. However, we are not quite so positive about your boss
…) And then there is of course the time-honoured phone call approach: “Hi, this
is the IT department. We’re doing a security systems test and urgently require
your user name and password.”
There are diverse ways of making Linux passwords more secure. Apart from
the improved encryption scheme mentioned above, which by now is used by de-
fault by most Linux distributions, these include complaining about (too) weak
passwords when they are ﬁrst set up, or proactively running software that will
try to identify weak encrypted passwords, just like clever crackers would (Cau-
tion: Do this in your workplace only with written (!) pre-approval from your
boss!). Other methods avoid passwords completely in favour of constantly chang-
ing magic numbers (as in SecurID) or smart cards. All of this is beyond the scope
of this manual, and therefore we refer you to the Linup Front manual Linux Secu-
rity.

2.2.3 The /etc/group File
group database By default, Linux keeps group information in the /etc/group ﬁle. This ﬁle contains
one-line entry for each group in the system, which like the entries in /etc/passwd
consists of ﬁelds separated by colons (: ). More precisely, /etc/group contains four
ﬁelds per line.

⟨group name⟩: ⟨password⟩: ⟨GID⟩: ⟨members⟩
2.2 User and Group Information 31

Their meaning is as follows:
⟨group name⟩ The name of the group, for use in directory listings, etc.
⟨password⟩ An optional password for this group. This lets users who are not mem-
bers of the group via /etc/shadow or /etc/group assume membership of the
group using newgrp . A “* ” as an invalid character prevents normal users
from changing to the group in question. A “x ” refers to the separate pass-
word ﬁle /etc/gshadow .
⟨GID⟩ The group’s numerical group identiﬁer.

⟨Members⟩ A comma-separated list of user names. This list contains all users who
have this group as a secondary group, i. e., who are members of this group
but have a diﬀerent value in the GID ﬁeld of their /etc/passwd entry. (Users
with this group as their primary group may also be listed here but that is
unnecessary.)

A /etc/group ﬁle could, for example, look like this:

root:x:0:root
bin:x:1:root,daemon
users:x:100:
project1:x:101:joe,sue
project2:x:102:bob

The entries for the root and bin groups are entries for administrative groups, sim- administrative groups
ilar to the system’s pseudo-user accounts. Many ﬁles are assigned to groups like
this. The other groups contain user accounts.
Like UIDs, GIDs are counted from a speciﬁc value, typically 100. For a valid GID values
entry, at least the ﬁrst and third ﬁeld (group name and GID) must be ﬁlled in.
Such an entry assigns a GID (which might occur in a user’s primary GID ﬁeld in
/etc/passwd ) a textual name.
The password and/or membership ﬁelds must only be ﬁlled in for groups that
are assigned to users as secondary groups. The users listed in the membership membership list
list are not asked for a password when they want to change GIDs using the new-
grp command. If an encrypted password is given, users without an entry in the group password
membership list can authenticate using the password to assume membership of
the group.

B In practice, group passwords are hardly if ever used, as the administrative
overhead barely justiﬁes the beneﬁts to be derived from them. On the one
hand it is more convenient to assign the group directly to the users in ques-
tion (since, from version 2.6 of the Linux kernel on, there is no limit to the
number of secondary groups a user can join), and on the other hand a single
password that must be known by all group members does not exactly make
for bullet-proof security.

B If you want to be safe, ensure that there is an asterisk (“* ”) in every group
password slot.

2.2.4 The /etc/gshadow File
As for the user database, there is a shadow password extension for the group
database. The group passwords, which would otherwise be encrypted but read-
able for anyone in /etc/group (similar to /etc/passwd ), are stored in the separate ﬁle
/etc/gshadow . This also contains additional information about the group, for ex-
ample the names of the group administrators who are entitled to add or remove
members from the group.
32 2 User Administration

2.2.5 The getent Command
Of course you can read and process the /etc/passwd , /etc/shadow , and /etc/group ﬁles,
like all other text ﬁles, using programs such as cat , less or grep (OK, OK, you need
to be root to get at /etc/shadow ). There are, however, some practical problems:
• You may not be able to see the whole truth: Your user database (or parts of
it) might be stored on an LDAP server, SQL database, or a Windows domain
controller, and there really may not be much of interest in /etc/passwd .
• If you want to look for a speciﬁc user’s entry, it is slightly inconvenient to
type this using grep if you want to avoid “false positives”.
The getent command makes it possible to query the various databases for user and
group information directly. With
$ getent passwd

you will be shown something that looks like /etc/passwd , but has been assembled
from all sources of user information that are currently conﬁgured on your com-
puter. With
$ getent passwd hugo

you can obtain user hugo ’s entry, no matter where it is actually stored. Instead
of passwd , you may also specify shadow , group , or gshadow to consult the respective
database. (Naturally, even with getent you can only access shadow and gshadow as
user root .)

B The term “database” is understood as “totality of all sources from where
the C library can obtain information on that topic (such as users)”. If you
want to know exactly where that information comes from (or might come
from), then read nsswitch.conf (5) and examine the /etc/nsswitch.conf ﬁle on
your system.

B You may also specify several user or group names. In that case, information
on all the named users or groups will be output:

$ getent passwd hugo susie fritz

Exercises
C 2.5 [1] Which value will you ﬁnd in the second column of the /etc/passwd
ﬁle? Why do you ﬁnd that value there?

C 2.6 [2] Switch to a text console (using, e. g., Alt + F1 ) and try logging in but
enter your user name in uppercase letters. What happens?

C 2.7 [2] How can you check that there is an entry in the shadow database for
every entry in the passwd database? (pwconv only considers the /etc/passwd and
/etc/shadow ﬁles, and also rewrites the /etc/shadow ﬁle, which we don’t want.

2.3 Managing User Accounts and Group Information
After a new Linux distribution has been installed, there is often just the root ac-
count for the system administrator and the pseudo-users’ accounts. Any other
user accounts must be created ﬁrst (and most distributions today will gently but
ﬁrmly nudge the installing person to create at least one “normal” user account).
As the administrator, it is your job to create and manage the accounts for all
tools for user management required users (real and pseudo). To facilitate this, Linux comes with several tools
for user management. With them, this is mostly a straightforward task, but it is
important that you understand the background.
2.3 Managing User Accounts and Group Information 33

2.3.1 Creating User Accounts
The procedure for creating a new user account is always the same (in principle)
and consists of the following steps:
1. You must create entries in the /etc/passwd (and possibly /etc/shadow ) ﬁles.

2. If necessary, an entry (or several) in the /etc/group ﬁle is necessary.
3. You must create the home directory, copy a basic set of ﬁles into it, and
transfer ownership of the lot to the new user.
4. If necessary, you must enter the user in further databases, e. g., for disk quo-
tas (section 7.4), database access privilege tables and special applications.

All ﬁles involved in adding a new account are plain text ﬁles. You can perform
each step manually using a text editor. However, as this is a job that is as tedious
as it is elaborate, it behooves you to let the system help you, by means of the useradd
program.
In the simplest case, you pass useradd merely the new user’s user name. Op- useradd
tionally, you can enter various other user parameters; for unspeciﬁed parameters
(typically the UID), “reasonable” default values will be chosen automatically. On
request, the user’s home directory will be created and endowed with a basic set of
ﬁles that the program takes from the /etc/skel directory. The useradd command’s
syntax is:

useradd [⟨options⟩] ⟨user name⟩

The following options (among others) are available:
-c ⟨comment⟩ GECOS ﬁeld entry
-d ⟨home directory⟩ If this option is missing, /home/ ⟨user name⟩ is assumed

-e ⟨date⟩ On this date the account will be deactivated automatically (format
“YYYY-MM-DD”)
-g ⟨group⟩ The new user’s primary group (name or GID). This group must exist.
-G ⟨group⟩[,⟨group⟩]… Supplementary groups (names or GIDs). These groups
must also exist.
-s ⟨shell⟩ The new user’s login shell
-u ⟨UID⟩ The new user’s numerical UID. This UID must not be already in use,
unless the “-o ” option is given

-m Creates the home directory and copies the basic set of ﬁles to it. These ﬁles
come from /etc/skel , unless a diﬀerent directory was named using “-k
⟨directory⟩”.
For instance, the

# useradd -c "Joe Smith" -m -d /home/joe -g devel \
> -k /etc/skel.devel

command creates an account by the name of joe for a user called Joe Smith, and
assigns it to the devel group. joe ’s home directory is created as /home/joe , and the
ﬁles from /etc/skel.devel are being copied into it.

B With the -D option (on SUSE distributions, --show-defaults ) you may set de-
fault values for some of the properties of new user accounts. Without addi-
tional options, the default values are displayed:
34 2 User Administration

# useradd -D
GROUP=100
HOME=/home
INACTIVE=-1
EXPIRE=
SHELL=/bin/sh
SKEL=/etc/skel
CREATE_MAIL_SPOOL=no

You can change these values using the -g , -b , -f , -e , and -s options, respec-
tively:

# useradd -D -s /usr/bin/zsh zsh as the default shell

The ﬁnal two values in the list cannot be changed.

B useradd is a fairly low-level tool. In real life, you as an experienced adminis-
trator will likely not be adding new user accounts by means of useradd , but
through a shell script that incorporates your local policies (just so you don’t
have to remember them all the time). Unfortunately you will have to come
up with this shell script by yourself—at least unless you are using Debian
GNU/Linux or one of its derivatives (see below).
Watch out: Even though every serious Linux distribution comes with a program
called useradd , the implementations diﬀer in their details.
The Red Hat distributions include a fairly run-of-the-mill version of useradd ,
without bells and whistles, which provides the features discussed above.
The SUSE distributions’ useradd is geared towards optionally adding users to
a LDAP directory rather than the /etc/passwd ﬁle. (This is why the -D option
cannot be used to query or set default values like it can elsewhere—it is
already spoken for to do LDAPy things.) The details are beyond the scope
of this manual.
On Debian GNU/Linux and Ubuntu, useradd does exist but the recom-
mended method to create new user accounts is a program called adduser
(thankfully this is not confusing). The advantage of adduser is that it plays
according to Debian GNU Linux’s rules, and furthermore makes it possible
to execute arbitrary other actions for a new account besides creating the
actual account. For example, one might create a directory in a web server’s
document tree so that the new user (and nobody else) can publish ﬁles
there, or the user could automatically be authorised to access a database
server. You can ﬁnd the details in adduser (8) and adduser.conf (5).
After it has been created using useradd , the new account is not yet accessible;
password the system administrator must ﬁrst set up a password. We shall be explaining this
presently.

2.3.2 The passwd Command
The passwd command is used to set up passwords for users. If you are logged in as
root , then

# passwd joe

asks for a new password for user john (You must enter it twice as it will not be
echoed to the screen).
The passwd command is also available to normal users, to let them change their
own passwords (changing other users’ passwords is root ’s prerogative):
2.3 Managing User Accounts and Group Information 35

$ passwd
Changing password for joe.
(current) UNIX password: secret123
Enter new UNIX password: 321terces
Retype new UNIX password: 321terces
passwd: password updated successfully

Normal users must enter their own password correctly once before being allowed
to set a new one. This is supposed to make life diﬃcult for practical jokers that
play around on your computer if you had to step out very urgently and didn’t
have time to engage the screen lock.
On the side, passwd serves to manage various settings in /etc/shadow . For exam-
ple, you can look at a user’s “password state” by calling the passwd command with
the -S option:

# passwd -S bob
bob LK 10/15/99 0 99999 7 0

The ﬁrst ﬁeld in the output is (once more) the user name, followed by the password
state: “PS ” or “P ” if a password is set, “LK ” or “L ” for a locked account, and “NP ” for
an account with no password at all. The other ﬁelds are, respectively, the date of
the last password change, the minimum and maximum interval for changing the
password, the expiry warning interval and the “grace period” before the account
is locked completely after the password has expired. (See also Section 2.2.2.)
You can change some of these settings by means of passwd options. Here are a
few examples:

# passwd -l joe Lock the account
# passwd -u joe Unlock the account
# passwd -n 7 joe Password change at most every 7 days
# passwd -x 30 joe Password change at least every 30 days
# passwd -w 3 joe 3 days grace period before password expires

E Locking and unlocking accounts by means of -l and -u works by putting
a “! ” in front of the encrypted password in /etc/shadow . Since “! ” cannot
result from password encryption, it is impossible to enter something upon
login that matches the “encrypted password” in the user database—hence
access via the usual login procedure is prevented. Once the “! ” is removed,
the original password is back in force. (Astute, innit?) However, you should
keep in mind that users may be able to gain access to the system by other
means that do not refer to the encrypted password in the user database,
such as the secure shell with public-key authentication.

Changing the remaining settings in /etc/shadow requires the chage command:

# chage -E 2009-12-01 joe Lock account from 1 Dec 2009
# chage -E -1 joe Cancel expiry date
# chage -I 7 joe Grace period 1 week from password expiry
# chage -m 7 joe Like passwd -n (Grr.)
# chage -M 7 joe Like passwd -x (Grr, grr.)
# chage -W 3 joe Like passwd -w (Grr, grr, grr.)

(chage can change all settings that passwd can change, and then some.)

B If you cannot remember the option names, invoke chage with the name of
a user account only. The program will present you with a sequence of the
current values to change or conﬁrm.
36 2 User Administration

You cannot retrieve a clear-text password even if you are the administrator.
Even checking /etc/shadow doesn’t help, since this ﬁle stores all passwords already
encrypted. If a user forgets their password, it is usually suﬃcient to reset their
password using the passwd command.

B Should you have forgotten the root password and not be logged in as root by
any chance, your last option is to boot Linux to a shell, or boot from a rescue
disk or CD. (See Chapter 8.) After that, you can use an editor to clear the
⟨password⟩ ﬁeld of the root entry in /etc/passwd .

Exercises
C 2.8 [3] Change user joe ’s password. How does the /etc/shadow ﬁle change?
Query that account’s password state.

C 2.9 [!2] The user dumbo has forgotten his password. How can you help him?

C 2.10 [!3] Adjust the settings for user joe ’s password such that he can change
his password after at least a week, and must change it after at most two
weeks. There should be a warning two days before the two weeks are up.
Check the settings afterwards.

2.3.3 Deleting User Accounts
To delete a user account, you need to remove the user’s entries from /etc/passwd and
/etc/shadow , delete all references to that user in /etc/group , and remove the user’s
home directory as well as all other ﬁles created or owned by that user. If the
user has, e. g., a mail box for incoming messages in /var/mail , that should also be
removed.
userdel Again there is a suitable command to automate these steps. The userdel com-
mand removes a user account completely. Its syntax:

userdel [-r ] ⟨user name⟩

The -r option ensures that the user’s home directory (including its content) and
his mail box in /var/mail will be removed; other ﬁles belonging to the user—e. g.,
crontab ﬁles—must be delete manually. A quick way to locate and remove ﬁles
belonging to a certain user is the

find / -uid ⟨UID⟩ -delete

command. Without the -r option, only the user information is removed from the
user database; the home directory remains in place.

2.3.4 Changing User Accounts and Group Assignment
User accounts and group assignments are traditionally changed by editing the
/etc/passwd and /etc/group ﬁles. However, many systems contain commands like
usermod and groupmod for the same purpose, and you should prefer these since they
are safer and—mostly—more convenient to use.
usermod The usermod program accepts mostly the same options as useradd , but changes
existing user accounts instead of creating new ones. For example, with

usermod -g ⟨group⟩ ⟨user name⟩

you could change a user’s primary group.
Changing UIDs Caution! If you want to change an existing user account’s UID, you could edit
the ⟨UID⟩ ﬁeld in /etc/passwd directly. However, you should at the same time trans-
fer that user’s ﬁles to the new UID using chown : “chown -R tux /home/tux ” re-confers
2.3 Managing User Accounts and Group Information 37

ownership of all ﬁles below user tux ’s home directory to user tux , after you have
changed the UID for that account. If “ls -l ” displays a numerical UID instead of
a textual name, this implies that there is no user name for the UID of these ﬁles.
You can ﬁx this using chown .

2.3.5 Changing User Information Directly—vipw
The vipw command invokes an editor (vi or a diﬀerent one) to edit /etc/passwd di-
rectly. At the same time, the ﬁle in question is locked in order to keep other users
from simultaneously changing the ﬁle using, e. g., passwd (which changes would
be lost). With the -s option, /etc/shadow can be edited.

B The actual editor that is invoked is determined by the value of the VISUAL
environment variable, alternatively that of the EDITOR environment variable;
if neither exists, vi will be launched.

Exercises
C 2.11 [!2] Create a user called test . Change to the test account and create a
few ﬁles using touch , including a few in a diﬀerent directory than the home
directory (say, /tmp ). Change back to root and change test ’s UID. What do
you see when listing user test ’s ﬁles?

C 2.12 [!2] Create a user called test1 using your distribution’s graphical tool (if
available), test2 by means of the useradd command, and another, test3 , man-
ually. Look at the conﬁguration ﬁles. Can you work without problems using
any of these three accounts? Create a ﬁle using each of the new accounts.

C 2.13 [!2] Delete user test2 ’s account and ensure that there are no ﬁles left on
the system that belong to that user.

C 2.14 [2] Change user test1 ’s UID. What else do you need to do?

C 2.15 [2] Change user test1 ’s home directory from /home/test1 to /home/user/
test1 .

2.3.6 Creating, Changing and Deleting Groups
Like user accounts, you can create groups using any of several methods. The
“manual” method is much less tedious here than when creating new user ac-
counts: Since groups do not have home directories, it is usually suﬃcient to edit
the /etc/group ﬁle using any text editor, and to add a suitable new line (see be-
low for vigr ). When group passwords are used, another entry must be added to
/etc/gshadow .
Incidentally, there is nothing wrong with creating directories for groups.
Group members can place the fruits of their collective labour there. The approach
is similar to creating user home directories, although no basic set of conﬁguration
ﬁles needs to be copied.
For group management, there are, by analogy to useradd , usermod , and userdel ,
the groupadd , groupmod , and groupdel programs that you should use in favour of edit-
ing /etc/group and /etc/gshadow directly. With groupadd you can create new groups groupadd
simply by giving the correct command parameters:

groupadd [-g ⟨GID⟩] ⟨group name⟩

The -g option allows you to specify a given group number. As mentioned be-
fore, this is a positive integer. The values up to 99 are usually reserved for system
groups. If -g is not speciﬁed, the next free GID is used.
You can edit existing groups with groupmod without having to write to /etc/group groupmod
directly:
38 2 User Administration

groupmod [-g ⟨GID⟩] [-n ⟨name⟩] ⟨group name⟩

The “-g ⟨GID⟩” option changes the group’s GID. Unresolved ﬁle group assign-
ments must be adjusted manually. The “-n ⟨name⟩” option sets a new name for the
group without changing the GID; manual adjustments are not necessary.
There is also a tool to remove group entries. This is unsurprisingly called
groupdel groupdel :

groupdel ⟨group name⟩

Here, too, it makes sense to check the ﬁle system and adjust “orphaned” group
assignments for ﬁles with the chgrp command. Users’ primary groups may not be
removed—the users in question must either be removed beforehand, or they must
be reassigned to a diﬀerent primary group.
gpasswd The gpasswd command is mainly used to manipulate group passwords in a way
similar to the passwd command. The system administrator can, however, delegate
group administrator the administration of a group’s membership list to one or more group adminis-
trators. Group administrators also use the gpasswd command:

gpasswd -a ⟨user⟩ ⟨group⟩

adds the ⟨user⟩ to the ⟨group⟩, and

gpasswd -d ⟨user⟩ ⟨group⟩

removes him again. With

gpasswd -A ⟨user⟩,… ⟨group⟩

the system administrator can nominate users who are to serve as group adminis-
trators.

The SUSE distributions haven’t included gpasswd for some time. Instead
there are modiﬁed versions of the user and group administration tools that
can handle an LDAP directory.

As the system administrator, you can change the group database directly using
vigr the vigr command. It works like vipw , by invoking an editor for “exclusive” access
to /etc/group . Similarly, “vigr -s ” gives you access to /etc/gshadow .

Exercises
C 2.16 [2] What are groups needed for? Give possible examples.

C 2.17 [1] Can you create a directory that all members of a group can access?

C 2.18 [!2] Create a supplementary group test . Only user test1 should be a
member of that group. Set a group password. Log in as user test1 or test2
and try to change over to the new group.
2.3 Managing User Accounts and Group Information 39

Commands in this Chapter
adduser Convenient command to create new user accounts (Debian)
adduser (8) 34
chfn Allows users to change the GECOS ﬁeld in the user database
chfn (1) 27
getent Gets entries from administrative databases getent (1) 32
gpasswd Allows a group administrator to change a group’s membership and up-
date the group password gpasswd (1) 38
groupadd Adds user groups to the system group database groupadd (8) 37
groupdel Deletes groups from the system group database groupdel (8) 38
groupmod Changes group entries in the system group database groupmod (8) 37
groups Displays the groups that a user is a member of groups (1) 24
id Displays a user’s UID and GIDs id (1) 24
last List recently-logged-in users last (1) 24
useradd Adds new user accounts useradd (8) 33
userdel Removes user accounts userdel (8) 36
usermod Modiﬁes the user database usermod (8) 36
vigr Allows editing /etc/group or /etc/gshadow with “ﬁle locking”, to avoid con-
ﬂicts vipw (8) 38

Summary
• Access to the system is governed by user accounts.
• A user account has a numerical UID and (at least) one textual user name.
• Users can form groups. Groups have names and numerical GIDs.
• “Pseudo-users” and “pseudo-groups” serve to further reﬁne access rights.
• The central user database is (normally) stored in the /etc/passwd ﬁle.
• The users’ encrypted passwords are stored—together with other password
parameters—in the /etc/shadow ﬁle, which is unreadable for normal users.
• Group information is stored in the /etc/group and /etc/gshadow ﬁles.
• Passwords are managed using the passwd program.
• The chage program is used to manage password parameters in /etc/shadow .
• User information is changed using vipw or—better—using the specialised
tools useradd , usermod , and userdel .
• Group information can be manipulated using the groupadd , groupmod , groupdel
and gpasswd programs.
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

3
Access Control

Contents
3.1 The Linux Access Control System . . . . . . . . . . . . . . 42
3.2 Access Control For Files And Directories . . . . . . . . . . . 42
3.2.1 The Basics . . . . . . . . . . . . . . . . . . . . 42
3.2.2 Inspecting and Changing Access Permissions. . . . . . . . 43
3.2.3 Specifying File Owners and Groups—chown and chgrp . . . . . 44
3.2.4 The umask . . . . . . . . . . . . . . . . . . . . 45
3.3 Access Control Lists (ACLs) . . . . . . . . . . . . . . . . 47
3.4 Process Ownership . . . . . . . . . . . . . . . . . . . 47
3.5 Special Permissions for Executable Files . . . . . . . . . . . . 47
3.6 Special Permissions for Directories . . . . . . . . . . . . . 48
3.7 File Attributes . . . . . . . . . . . . . . . . . . . . . 50

Goals
• Understanding the Linux access control/privilege mechanisms
• Being able to assign access permissions to ﬁles and directories
• Knowing about the “umask”, SUID, SGID and the “sticky bit”
• Knowing about ﬁle attributes in the ext ﬁle systems

Prerequisites
• Knowledge of Linux user and group concepts (see Chapter 2)
• Knowledge of Linux ﬁles and directories

adm1-rechte.tex (33e55eeadba676a3 )
42 3 Access Control

3.1 The Linux Access Control System
Whenever several users have access to the same computer system there must be
access control system an access control system for processes, ﬁles and directories in order to ensure that
user 𝐴 cannot access user 𝐵’s private ﬁles just like that. To this end, Linux imple-
ments the standard system of Unix privileges.
In the Unix tradition, every ﬁle and directory is assigned to exactly one user
separate privileges (its owner) and one group. Every ﬁle supports separate privileges for its owner,
the members of the group it is assigned to (“the group”, for short), and all other
users (“others”). Read, write and execute privileges can be enabled individually
for these three sets of users. The owner may determine a ﬁle’s access privileges.
The group and others may only access a ﬁle if the owner confers suitable privileges
access mode to them. The sum total of a ﬁle’s access permissions is also called its access mode.
In a multi-user system which stores private or group-internal data on a gen-
erally accessible medium, the owner of a ﬁle can keep others from reading or
access control modifying his ﬁles by instituting suitable access control. The rights to a ﬁle can be
determined separately and independently for its owner, its group and the others.
Access permissions allow users to map the responsibilities of a group collabora-
tive process to the ﬁles that the group is working with.

3.2 Access Control For Files And Directories
3.2.1 The Basics
For each ﬁle and each directory in the system, Linux allows separate access rights
for each of the three classes of users—owner, members of the ﬁle’s group, others.
These rights include read permission, write permission, and execute permission.
file permissions As far as ﬁles are concerned, these permissions control approximately what
their names suggest: Whoever has read permission may look at the ﬁle’s content,
whoever has write permission is allowed to change its content. Execute permis-
sion is necessary to launch the ﬁle as a process.

B Executing a binary “machine-language program” requires only execute per-
mission. For ﬁles containing shell scripts or other types of “interpreted”
programs, you also need read permission.

directory permissions For directories, things look somewhat diﬀerent: Read permission is required
to look at a directory’s content—for example, by executing the ls command. You
need write permission to create, delete, or rename ﬁles in the directory. “Execute”
permission stands for the possibility to “use” the directory in the sense that you
can change into it using cd , or use its name in path names referring to ﬁles farther
down in the directory tree.

B In directories where you have only read permission, you may read the ﬁle
names but cannot ﬁnd out anything else about the ﬁles. If you have only “ex-
ecute permission” for a directory, you can access ﬁles as long as you know
their names.

Usually it makes little sense to assign write and execute permission to a directory
separately; however, it may be useful in certain special cases.

A It is important to emphasise that write permission on a ﬁle is completely
immaterial if the ﬁle is to be deleted—you need write permission to the direc-
tory that the ﬁle is in and nothing else! Since “deleting” a ﬁle only removes
a reference to the actual ﬁle information (the inode) from the directory, this
is purely a directory operation. The rm command does warn you if you’re
trying to delete a ﬁle that you do not have write permission for, but if you
conﬁrm the operation and have write permission to the directory involved,
nothing will stand in the way of the operation’s success. (Like any other
3.2 Access Control For Files And Directories 43

Unix-like system, Linux has no way of “deleting” a ﬁle outright; you can
only remove all references to a ﬁle, in which case the Linux kernel decides
on its own that no one will be able to access the ﬁle any longer, and gets rid
of its content.)

B If you do have write permission to the ﬁle but not its directory, you cannot
remove the ﬁle completely. You can, however, truncate it down to 0 bytes
and thereby remove its content, even though the ﬁle itself still exists in prin-
ciple.
For each user, Linux determines the “most appropriate” access rights. For ex-
ample, if the members of a ﬁle’s group do not have read permission for the ﬁle
but “others” do, then the group members may not read the ﬁle. The (admittedly
enticing) rationale that, if all others may look at the ﬁle, then the group members,
who are in some sense also part of “all others”, should be allowed to read it as
well, does not apply.

3.2.2 Inspecting and Changing Access Permissions
You can obtain information about the rights, user and group assignment that ap- information
ply to a ﬁle using “ls -l ”:

$ ls -l
-rw-r--r-- 1 joe users 4711 Oct 4 11:11 datei.txt
drwxr-x--- 2 joe group2 4096 Oct 4 11:12 testdir

The string of characters in the ﬁrst column of the table details the access permis-
sions for the owner, the ﬁle’s group, and others (the very ﬁrst character is just the
ﬁle type and has nothing to do with permissions). The third column gives the
owner’s user name, and the fourth that of the ﬁle’s group.
In the permissions string, “r ”, “w ”, and “x ” signify existing read, write, and
execute permission, respectively. If there is just a “- ” in the list, then the corre-
sponding category does not enjoy the corresponding privilege. Thus, “rw-r--r-- ”
stands for “read and write permission for the owner, but read permission only for
group members and others”.
As the ﬁle owner, you may set access permissions for a ﬁle using the chmod com- chmod command
mand (from “change mode”). You can specify the three categories by means of the
abbreviations “u ” (user) for the owner (yourself), “g ” (group) for the ﬁle’s group’s
members, and “o ” (others) for everyone else. The permissions themselves are
given by the already-mentioned abbreviations “r ”, “w ”, and “x ”. Using “+ ”, “- ”,
and “= ”, you can specify whether the permissions in question should be added to
any existing permissions, “subtracted” from the existing permissions, or used to
replace whatever was set before. For example:

$ chmod u+x file Execute permission for owner
$ chmod go+w file sets write permission for group and others
$ chmod g+rw file sets read and write permission for group
$ chmod g=rw,o=r file sets read and write permission,
removes group execute permission;
sets just read permission for others
$ chmod a+w file equivalent to ugo+w

B In fact, permission speciﬁcations can be considerably more complex. Con-
sult the info documentation for chmod to ﬁnd out all the details.
A ﬁle’s owner is the single user (apart from root ) who is allowed to change a
ﬁle’s or directory’s access permissions. This privilege is independent of the actual
permissions; the owner may take away all their own permissions, but that does
not keep them from giving them back later.
The general syntax of the chmod command is
44 3 Access Control

chmod [⟨options⟩] ⟨permissions⟩ ⟨name⟩ …

You can give as many ﬁle or directory names as desired. The most important
options include:

-R If a directory name is given, the permissions of ﬁles and directories inside this
directory will also be changed (and so on all the way down the tree).
--reference= ⟨name⟩ Uses the access permissions of ﬁle ⟨name⟩. In this case no
⟨permissions⟩ must be given with the command.

B You may also specify a ﬁle’s access mode “numerically” instead of “symbol-
ically” (what we just discussed). In practice this is very common for setting
all permissions of a ﬁle or directory at once, and works like this: The three
permission triples are represented as a three-digit octal number—the ﬁrst
digit describes the owner’s rights, the second those of the ﬁle’s group, and
the third those that apply to “others”. Each of these digits derives from
the sum of the individual permissions, where read permission has value 4,
write permission 2, and execute permission 1. Here are a few examples for
common access modes in “ls -l ” and octal form:
rw-r--r-- 644
r-------- 400
rwxr-xr-x 755

B Using numerical access modes, you can only set all permissions at once—
there is no way of setting or removing individual rights while leaving the
others alone, like you can do with the “+ ” and “- ” operators of the symbolic
representation. Hence, the command

$ chmod 644 file

is equivalent to the symbolic

$ chmod u=rw,go=r file

3.2.3 Specifying File Owners and Groups—chown and chgrp
The chown command lets you set the owner and group of a ﬁle or directory. This
command takes the desired owner’s user name and/or group name and the ﬁle
or directory name the change should apply to. It is called like

chown ⟨user name⟩[: ][⟨group name⟩] ⟨name⟩ …
chown : ⟨group name⟩ ⟨name⟩ …

If both a user and group name are given, both are changed; if just a user name is
given, the group remains as it was; if a user name followed by a colon is given,
then the ﬁle is assigned to the user’s primary group. If just a group name is given
(with the colon in front), the owner remains unchanged. For example:

# chown joe:devel letter.txt
# chown www-data foo.html new user www-data
# chown :devel /home/devel new group devel

B chown also supports an obsolete syntax where a dot is used in place of the
colon.
3.2 Access Control For Files And Directories 45

To “give away” ﬁles to other users or arbitrary groups you need to be root . The
main reason for this is that normal users could otherwise annoy one another if
the system uses quotas (i.e., every user can only use a certain amount of storage
space).
Using the chgrp command, you can change a ﬁle’s group even as a normal
user—as long as you own the ﬁle and are a member of the new group:
chgrp ⟨group name⟩ ⟨name⟩ …

B Changing a ﬁle’s owner or group does not change the access permissions
for the various categories.
chown and chgrp also support the -R option to apply changes recursively to part
of the directory hierarchy.

B Of course you can also change a ﬁle’s permissions, group, and owner using
most of the popular ﬁle browsers (such as Konqueror or Nautilus).

Exercises
C 3.1 [!2] Create a new ﬁle. What is that ﬁle’s group? Use chgrp to assign the
ﬁle to one of your secondary groups. What happens if you try to assign the
ﬁle to a group that you are not a member of?

C 3.2 [4] Compare the mechanisms that various ﬁle browsers (like Konqueror,
Nautilus, …) oﬀer for setting a ﬁle’s permissions, owner, group, … Are there
notable diﬀerences?

3.2.4 The umask
New ﬁles are usually created using the (octal) access mode 666 (read and write
permission for everyone). New directories are assigned the access mode 777.
Since this is not always what is desired, Linux oﬀers a mechanism to remove cer-
tain rights from these access modes. This is called “umask”.

B Nobody knows exactly where this name comes from—even though there
are a few theories that all sound fairly implausible.
The umask is an octal number whose complement is ANDed bitwise to the
standard access mode—666 or 777—to arrive at the new ﬁle’s or directory’s actual
access mode. In other words: You can consider the umask an access mode contain- umask interpretation
ing exactly those rights that the new ﬁle should not have. Here’s an example—let
the umask be 027:
1. Umask value: 027 ----w-rwx
2. Complement of umask value: 750 rwxr-x---
3. A new ﬁle’s access mode: 666 rw-rw-rw-
4. Result (2 and 3 ANDed together): 640 rw-r-----
The third column shows the octal value, the fourth a symbolic representation. The
AND operation in step 4 can also be read oﬀ the fourth column of the second and
third lines: In the fourth line ther e is a letter in each position that had a letter in
the second and the third line—if there is just one dash (“- ”), the result will be a
dash.

B If you’d rather not bother with the complement and AND, you can simply
imagine that each digit of the umask is subtracted from the corresponding
digit of the actual access mode and negative results are considered as zero
(so no “borrowing” from the place to the left). For our example—access
mode 666 and umask 027—this means something like
666 ⊖ 027 = 640,
since 6 ⊖ 0 = 6, 6 ⊖ 4 = 2, and 6 ⊖ 7 = 0.
46 3 Access Control

umask shell command The umask is set using the umask shell command, either by invoking it di-
rectly or via a shell startup ﬁle—typically ~/.profile , ~/.bash_profile , or ~/.bashrc .
process attribute The umask is a process attribute similar to the current directory or the process
environment, i. e., it is passed to child processes, but changes in a child process do
not modify the parent process’s settings.
syntax The umask command takes a parameter specifying the desired umask:

umask [-S |⟨umask⟩]

symbolic representation The umask may be given as an octal number or in a symbolic representation sim-
ilar to that used by chmod —deviously enough, the symbolic form contains the per-
missions that should be left (rather than those to be taken away):

$ umask 027 … is equivalent to …
$ umask u=rwx,g=rx,o=

This means that in the symbolic form you must give the exact complement of the
value that you would specify in the octal form—exactly those rights that do not
occur in the octal speciﬁcation.
If you specify no value at all, the current umask is displayed. If the -S option
is given, the current umask is displayed in symbolic form (where, again, the re-
maining permissions are set):

$ umask
0027
$ umask -S
u=rwx,g=rx,o=

execute permission? Note that you can only remove permissions using the umask. There is no way
of making ﬁles executable by default.
umask and chmod Incidentally, the umask also inﬂuences the chmod command. If you invoke chmod
with a “+ ” mode (e. g., “chmod +w file ”) without referring to the owner, group or oth-
ers, this is treated like “a+ ”, but the permissions set in the umask are not modiﬁed.
Consider the following example:

$ umask 027
$ touch file
$ chmod +x file
$ ls -l file
-rwxr-x--- 1 tux users 0 May 25 14:30 file

The “chmod +x ” sets execute permission for the user and group, but not the others,
since the umask contains the execute bit for “others”. Thus with the umask you
can take precautions against giving overly excessive permissions to ﬁles.

B Theoretically, this also works for the chmod operators “- ” and “= ”, but this
does not make a lot of sense in practice.

Exercises
C 3.3 [!1] State a numerical umask that leaves the user all permissions, but
removes all permissions from group members and others? What is the cor-
responding symbolic umask?

C 3.4 [2] Convince yourself that the “chmod +x ” and “chmod a+x ” commands in-
deed diﬀer from each other as advertised.
3.3 Access Control Lists (ACLs) 47

3.3 Access Control Lists (ACLs)
As mentioned above, Linux allows you to assign permissions for a ﬁle’s owner,
group, and all others separately. For some applications, though, this three-tier
system is too simple-minded, or the more sophisticated permission schemes of
other operating systems must be mapped to Linux. Access control lists (ACLs)
can be used for this.
On most ﬁle systems, Linux supports “POSIX ACLs” according to IEEE 1003.1e
(draft 17) with some Linux-speciﬁc extensions. This lets you specify additional
groups and users for ﬁles and directories, who then can be assigned read, write,
and execute permissions that diﬀer from those of the ﬁle’s group and “others”.
Other rights, such as that to assign permissions, are still restricted to a ﬁle’s owner
(or root ) and cannot be delegated even wiht ACLs. The setfacl and getfacl com-
mands are used to set and query ACLs.
ACLs are a fairly new and rarely-used addition to Linux, and their use is subject
to certain restrictions. The kernel does oversee compliance with them, but, for
instance, not every program is able to copy ACLs along with a ﬁle’s content—you
may have to use a specially-adapted tar (star ) for backups of a ﬁle system using
ACLs. ACLs are supported by Samba, so Windows clients get to see the correct
permissions, but if you export ﬁle systems to other (proprietary) Unix systems, it
may be possible that your ACLs are ignored by Unix clients that do not support
ACLs.

B You can read up on ACLs on Linux on http://acl.bestbits.at/ and in acl (5)
as well as getfacl (1) and setfacl (1).

Detailed knowledge of ACLs is not required for the LPIC-1 exams.

3.4 Process Ownership
Linux considers not only the data on a storage medium as objects that can be
owned. The processes on the system have owners, too.
Many commands create a process in the system’s memory. During normal use,
there are always several processes that the system protects from each other. Every
process together with all data within its virtual address space is assigned to a Processes have owners
user, its owner. This is most often the user who started the process—but processes
created using administrator privileges may change their ownership, and the SUID
mechanism (section 3.5) can also have a hand in this.
The owners of processes are displayed by the ps program if it is invoked using
the -u option.

# ps -u
USER PID %CPU %MEM SIZE RSS TTY STAT START TIME COMMAND
bin 89 0.0 1.0 788 328 ? S 13:27 0:00 rpc.portmap
test1 190 0.0 2.0 1100 28 3 S 13:27 0:00 bash
test1 613 0.0 1.3 968 24 3 S 15:05 0:00 vi XF86.tex
nobody 167 0.0 1.4 932 44 ? S 13:27 0:00 httpd
root 1 0.0 1.0 776 16 ? S 13:27 0:03 init [3]
root 2 0.0 0.0 0 0 ? SW 13:27 0:00 (kflushd)

3.5 Special Permissions for Executable Files
When listing ﬁles using the “ls -l ” command, you may sometimes encounter per-
mission sets that diﬀer from the usual rwx , such as

-rwsr-xr-x 1 root shadow 32916 Dec 11 20:47 /usr/bin/passwd
48 3 Access Control

What does that mean? We have to digress here for a bit:
Assume that the passwd program carries the usual access mode:

-rwxr-xr-x 1 root shadow 32916 Dec 11 20:47 /usr/bin/passwd

A normal (unprivileged) user, say joe , wants to change his password and invokes
the passwd program. Next, he receives the message “permission denied”. What is
the reason? The passwd process (which uses joe ’s privileges) tries to open the /etc/
shadow ﬁle for writing and fails, since only root may write to that ﬁle—this cannot
be diﬀerent since otherwise, everybody would be able to manipulate passwords
arbitrarily and, for example, change the root password.
SUID bit By means of the set-UID bit (frequently called “SUID bit”, for short) a program
can be caused to run not with the invoker’s privileges but those of the ﬁle owner—
here, root . In the case of passwd , the process executing passwd has write permission
to /etc/shadow , even though the invoking user, not being a system administrator,
generally doesn’t. It is the responsibility of the author of the passwd program to en-
sure that no monkey business goes on, e. g., by exploiting programming errors to
change arbitrary ﬁles except /etc/shadow , or entries in /etc/shadow except the pass-
word ﬁeld of the invoking user. On Linux, by the way, the set-UID mechanism
works only for binary programs, not shell or other interpreter scripts.

B Bell Labs used to hold a patent on the SUID mechanism, which was invented
by Dennis Ritchie [SUID]. Originally, AT&T distributed Unix with the
caveat that license fees would be levied after the patent had been granted;
however, due to the logistical diﬃculties of charging hundreds of Unix in-
stallations small amounts of money retroactively, the patent was released
into the public domain.

SGID bit By analogy to the set-UID bit there is a SGID bit, which causes a process to be
executed with the program ﬁle’s group and the corresponding privileges (usually
to access other ﬁles assigned to that group) rather than the invoker’s group setting.
chmod syntax The SUID and SGID modes, like all other access modes, can be changed using
the chmod program, by giving symbolic permissions such as u+s (sets the SUID bit)
or g-s (deletes the SGID bit). You can also set these bits in octal access modes by
adding a fourth digit at the very left: The SUID bit has the value 4, the SGID bit
the value 2—thus you can assign the access mode 4755 to a ﬁle to make it readable
and executable to all users (the owner may also write to it) and to set the SUID bit.
ls output You can recognise set-UID and set-GID programs in the output of “ls -l ” by
the symbolic abbreviations “s ” in place of “x ” for executable ﬁles.

3.6 Special Permissions for Directories
There is another exception from the principle of assigning ﬁle ownership accord-
ing to the identity of its creator: a directory’s owner can decree that ﬁles created
in that directory should belong to the same group as the directory itself. This can
SGID for directories be speciﬁed by setting the SGID bit on the directory. (As directories cannot be
executed, the SGID bit is available to be used for such things.)
A directory’s access permissions are not changed via the SGID bit. To create a
ﬁle in such a directory, a user must have write permission in the category (owner,
group, others) that applies to him. If, for example, a user is neither the owner of a
directory nor a member of the directory’s group, the directory must be writable for
“others” for him to be able to create ﬁles there. A ﬁle created in a SGID directory
then belongs to that directory’s group, even if the user is not a member of that
group at all.

B The typical application for the SGID bit on a directory is a directory that is
used as ﬁle storage for a “project group”. (Only) the members of the project
group are supposed to be able to read and write all ﬁles in the directory, and
3.6 Special Permissions for Directories 49

to create new ﬁles. This means that you need to put all users collaborating
on the project into a project group (a secondary group will suﬃce):

# groupadd project Create new group
# usermod -a -G project joe joe into the group
# usermod -a -G project sue sue too

Now you can create the directory and assign it to the new group. The owner
and group are given all permissions, the others none; you also set the SGID
bit:
# cd /home/project
# chgrp project /home/project
# chmod u=rwx,g=srwx /home/project

Now, if user hugo creates a ﬁle in /home/project , that ﬁle should be assigned
to group project :

$ id
uid=1000(joe) gid=1000(joe) groups=101(project),1000(joe)
$ touch /tmp/joe.txt Test: standard directory
$ ls -l /tmp/joe.txt
-rw-r--r-- 1 joe joe 0 Jan 6 17:23 /tmp/joe.txt
$ touch /home/project/joe.txt project directory
$ ls -l /home/project/joe.txt
-rw-r--r-- 1 joe project 0 Jan 6 17:24 /home/project/joe.txt

There is just a little ﬂy in the ointment, which you will be able to discern by
looking closely at the ﬁnal line in the example: The ﬁle does belong to the
correct group, but other members of group project may nevertheless only
read it. If you want all members of group project to be able to write to it as
well, you must either apply chmod after the fact (a nuisance) or else set the
umask such that group write permission is retained (see Exercise 3.6).

The SGID mode only changes the system’s behaviour when new ﬁles are cre-
ated. Existing ﬁles work just the same as everywhere else. This means, for in-
stance, that a ﬁle created outside the SGID directory keeps its existing group as-
signment when moved into it (whereas on copying, the new copy would be put
into the directory’s group).
The chgrp program works as always in SGID directories, too: the owner of a
ﬁle can assign it to any group he is a member of. Is the owner not a member of
the directory’s group, he cannot put the ﬁle into that group using chgrp —he must
create it afresh within the directory.

B It is possible to set the SUID bit on a directory—this permission does not
signify anything, though.

Linux supports another special mode for directories, where only a ﬁle’s owner
may delete or remove ﬁles within that directory:

drwxrwxrwt 7 root root 1024 Apr 7 10:07 /tmp

This t mode, the “sticky bit”, can be used to counter a problem which arises when
public directories are in shared use: Write permission to a directory lets a user
delete other users’ ﬁles, regardless of their access mode and owner! For example,
the /tmp directories are common ground, and many programs create their tempo-
rary ﬁles there. To do so, all users have write permission to that directory. This
implies that any user has permission to delete ﬁles there.
50 3 Access Control

Table 3.1: The most important ﬁle attributes

Attribute Meaning
A atime is not updated (interesting for mobile computers)
a (append-only) The ﬁle can only be appended to
c The ﬁle’s content is compressed transparently (not implemented)
d The ﬁle will not be backed up by dump
i (immutable) The ﬁle cannot be changed at all
j Write operations to the ﬁle’s content are passed through the journal
(ext3 only)
s File data will be overwritten with zeroes on deletion (not imple-
mented)
S Write operations to the ﬁle are performed “synchronously”, i. e.,
without buﬀering them internally
u The ﬁle may be “undeleted” after deletion (not implemented)

Usually, when deleting or renaming a ﬁle, the system does not consider that
ﬁle’s access permissions. If the “sticky bit” is set on a directory, a ﬁle in that di-
rectory can subsequently be deleted only by its owner, the directory’s owner, or
root . The “sticky bit” can be set or removed by specifying the symbolic +t and -t
modes; in the octal representation it has value 1 in the same digit as SUID and
SGID.

B The “sticky bit” derives its name from an additional meaning it used to have
in earlier Unix systems: At that time, programs were copied to swap space
in their entirety when started, and removed completely after having ter-
minated. Program ﬁles with the sticky bit set would be left in swap space
instead of being removed. This would accelerate subsequent invocations of
those programs since no copy would have to be done. Like most current
Unix systems, Linux uses demand paging, i. e., it fetches only those parts
of the code from the program’s executable ﬁle that are really required, and
does not copy anything to swap space at all; on Linux, the sticky bit never
had its original meaning.

Exercises
C 3.5 [2] What does the special “s ” privilege mean? Where do you ﬁnd it?
Can you set this privilege on a ﬁle that you created yourself?

C 3.6 [!1] Which umask invocation can be used to set up a umask that would, in
the project directory example above, allow all members of the project group
to read and write ﬁles in the project directory?

C 3.7 [2] What does the special “t ” privilege mean? Where do you ﬁnd it?

C 3.8 [4] (For programmers.) Write a C program that invokes a suitable com-
mand (such as id ). Set this program SUID root (or SGID root ) and observe
what happens when you execute it.

C 3.9 [I]f you leave them alone for a few minutes with a root shell, clever users
might try to stash a SUID root shell somewhere in the system, in order to
assume administrator privileges when desired. Does that work with bash ?
With other shells?

3.7 File Attributes
file attributes Besides the access permissions, the ext2 and ext3 ﬁle systems support further ﬁle
3.7 File Attributes 51

attributes enabling access to special ﬁle system features. The most important ﬁle
attributes are summarised in Table 3.1.
Most interesting are perhaps the “append-only” and “immutable” attributes, a and i attributes
which you can use to protect log ﬁles and conﬁguration ﬁles from modiﬁcation;
only root may set or reset these attributes, and once set they also apply to processes
running as root .

B In principle, an attacker who has gained root privileges may reset these at-
tributes. However, attackers do not necessarily consider that they might be
set.

The A attribute may also be useful; you can use it on mobile computers to ensure A attribute
that the disk isn’t always running, in order to save power. Usually, whenever
a ﬁle is read, its “atime”—the time of last access—is updated, which of course
entails an inode write operation. Certain ﬁles are very frequently looked at in
the background, such that the disk never gets to rest, and you can help here by
judiciously applying the A attribute.

B The c , s and u attributes sound very nice in theory, but are not (yet) sup-
ported by “normal” kernels. There are some more or less experimental en-
hancements making use of these attributes, and in part they are still pipe
dreams.

You can set or reset attributes using the chattr command. This works rather chattr
like chmod : A preceding “+ ” sets one or more attributes, “- ” deletes one or more
attributes, and “= ” causes the named attributes to be the only enabled ones. The
-R option, as in chmod , lets chattr operate on all ﬁles in any subdirectories passed
as arguments and their nested subdirectories. Symbolic links are ignored in the
process.

# chattr +a /var/log/messages Append only
# chattr -R +j /data/important Data journaling …
# chattr -j /data/important/notso … with exception

With the lsattr command, you can review the attributes set on a ﬁle. The pro- lsattr
gram behaves similar to “ls -l ”:

# lsattr /var/log/messages
-----a----------- /var/log/messages

Every dash stands for a possible attribute. lsattr supports various options such
as -R , -a , and -d , which generally behave like the eponymous options to ls .

Exercises
C 3.10 [!2] Convince yourself that the a and i attributes work as advertised.

C 3.11 [2] Can you make all dashes disappear in the lsattr output for a given
ﬁle?
52 3 Access Control

Commands in this Chapter
chattr Sets ﬁle attributes for ext2 and ext3 ﬁle systems chattr (1) 51
chgrp Sets the assigned group of a ﬁle or directory chgrp (1) 44
chmod Sets access modes for ﬁles and directories chmod (1) 43
chown Sets the owner and/or assigned group of a ﬁle or directory
chown (1) 44
getfacl Displays ACL data getfacl (1) 47
lsattr Displays ﬁle attributes on ext2 and ext3 ﬁle systems lsattr (1) 51
setfacl Enables ACL manipulation setfacl (1) 47
star POSIX-compatible tape archive with ACL support star (1) 47

Summary
• Linux supports ﬁle read, write and execute permissions, where these per-
missions can be set separately for a ﬁle’s owner, the members of the ﬁle’s
group and “all others”.
• The sum total of a ﬁle’s permissions is also called its access mode.
• Every ﬁle (and directory) has an owner and a group. Access rights—read,
write, and execute permission—are assigned to these two categories and
“others” separately. Only the owner is allowed to set access rights.
• Access rights do not apply to the system administrator (root ). He may read
or write all ﬁles.
• File permissions can be manipulated using the chmod command.
• Using chown , the system administrator can change the user and group as-
signment of arbitrary ﬁles.
• Normal users can use chgrp to assign their ﬁles to diﬀerent groups.
• The umask can be used to limit the standard permissions when ﬁles and
directories are being created.
• The SUID and SGID bits allow the execution of programs with the privileges
of the ﬁle owner or ﬁle group instead of those of the invoker.
• The SGID bit on a directory causes new ﬁles in that directory to be assigned
the directory’s group (instead of the primary group of the creating user).
• The “sticky bit” on a directory lets only the owner (and the system admin-
istrator) delete ﬁles.
• The ext ﬁle systems support special additional ﬁle attributes.

Bibliography
SUID Dennis M. Ritchie. “Protection of data ﬁle contents”. US patent 4,135,240.
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

4
Process Management

Contents
4.1 What Is A Process? . . . . . . . . . . . . . . . . . . . 54
4.2 Process States . . . . . . . . . . . . . . . . . . . . . 55
4.3 Process Information—ps . . . . . . . . . . . . . . . . . 56
4.4 Processes in a Tree—pstree . . . . . . . . . . . . . . . . 57
4.5 Controlling Processes—kill and killall . . . . . . . . . . . . 58
4.6 pgrep and pkill . . . . . . . . . . . . . . . . . . . . . 59
4.7 Process Priorities—nice and renice . . . . . . . . . . . . . . 61
4.8 Further Process Management Commands—nohup and top . . . . . 61

Goals
• Knowing the Linux process concept
• Using the most important commands to query process information
• Knowing how to signal and stop processes
• Being able to inﬂuence process priorities

Prerequisites
• Linux commands

adm1-prozesse.tex (33e55eeadba676a3 )
54 4 Process Management

4.1 What Is A Process?
A process is, in eﬀect, a “running program”. Processes have code that is executed,
and data on which the code operates, but also various attributes the operating uses
to manage them, such as:
process number • The unique process number (PID or “process identity”) serves to identify
the process and can only be assigned to a single process at a time.
parent process number • All processes know their parent process number, or PPID. Every process can
spawn others (“children”) that then contain a reference to their procreator.
The only process that does not have a parent process is the “pseudo” process
with PID 0, which is generated during system startup and creates the “init”
process with a PID of 1, which in turn is the ancestor of all other processes
in the system.

user • Every process is assigned to a user and a set of groups. These are important
groups to determine the access the process has to ﬁles, devices, etc. (See Section 3.4.)
Besides, the user the process is assigned to may stop, terminate, or other-
wise inﬂuence the process. The owner and group assignments are passed
on to child processes.

• The system splits the CPU time into little chunks (“time slices”), each of
which lasts only for a fraction of a second. The current process is entitled to
such a time slice, and afterwards the system decides which process should
be allowed to execute during the next time slice. This decision is made by
priority the appropriate “scheduler” based on the priority of a process.

B In multi-processor systems, Linux also takes into account the particu-
lar topology of the computer in question when assigning CPU time to
processes—it is simple to run a process on any of the diﬀerent cores
of a multi-core CPU which share the same memory, while the “migra-
tion” of a process to a diﬀerent processor with separate memory entails
a noticeable administrative overhead and is therefore less often worth-
while.

other attributes • A process has other attributes—a current directory, a process environment,
…—which are also passed on to child processes.
process file system You can consult the /proc ﬁle system for this type of information. This process ﬁle
system is used to make available data from the system kernel which is collected at
run time and presented by means of directories and ﬁles. In particular, there are
various directories using numbers as names; every such directory corresponds to
one process and its name to the PID of that process. For example:

dr-xr-xr-x 3 root root 0 Oct 16 11:11 1
dr-xr-xr-x 3 root root 0 Oct 16 11:11 125
dr-xr-xr-x 3 root root 0 Oct 16 11:11 80

In the directory of a process, there are various “ﬁles” containing process informa-
tion. Details may be found in the proc (5) man page.

job control B The job control available in many shells is also a form of process management—
a “job” is a process whose parent process is a shell. From the corresponding
shell, its jobs can be controlled using commands like jobs , bg , and fg , as well
as the key combinations Ctrl + z and Ctrl + c (among others). More in-
formation is available from the manual page of the shell in question, or
from the Linup Front training manual, Introduction to Linux for Users and
Administrators.
4.2 Process States 55

Process
Process
is runnable operating
terminates
created

sleeping

Figure 4.1: The relationship between various process states

Exercises
C 4.1 [3] How can you view the environment variables of any of your pro-
cesses? (Hint: /proc ﬁle system.)

C 4.2 [2] (For programmers.) What is the maximum possible PID? What hap-
pens when this limit is reached? (Hint: Look for the string “PID_MAX ” in the
ﬁles below /usr/include/linux .)

4.2 Process States
Another important property of a process is its process state. A process in mem- process state
ory waits to be executed by the CPU. This state is called “runnable”. Linux uses
pre-emptive multitasking, i. e., a scheduler distributes the available CPU time to pre-emptive multitasking
waiting processes in pieces called “time slices”. If a process is actually execut-
ing on the CPU, this state is called “operating”, and after its time slice is over the
process reverts to the “runnable” state.

B From an external point of view, Linux does not distinguish between these
two process states; the process in question is always marked “runnable”.

It is quite possible that a process requires further input or needs to wait for
peripheral device operations to complete; such a process cannot be assigned CPU
time, and its state is considered to be “sleeping”. Processes that have been stopped
by means of Ctrl + z using the shell’s job control facility are in state “stopped”.
Once the execution of a process is over, it terminates itself and makes a return return code
code available, which it can use to signal, for example, whether it completed suc-
cessfully or not (for a suitable deﬁnition of “success”).
Once in a while processes appear who are marked as zombies using the “Z” zombies
state. These “living dead” usually exist only for a brief instant. A process becomes
a zombie when it ﬁnishes and dies for good once its parent process has queried
its return code. If a zombie does not disappear from the process table this means
that its parent should really have picked up the zombie’s return code but didn’t.
A zombie cannot be removed from the process table. Because the original pro-
cess no longer exists and cannot take up neither RAM nor CPU time, a zombie
has no impact on the system except for an unattractive entry in the system state.
Persistent or very numerous zombies usually indicate programming errors in the
parent process; when the parent process terminates they should do so as well.

B Zombies disappear when their parent process disappears because “or-
phaned” processes are “adopted” by the init process. Since the init process
56 4 Process Management

spends most of its time waiting for other processes to terminate so that it
can collect their return code, the zombies are then disposed of fairly quickly.

B Of course, zombies take up room in the process table that might be required
for other processes. If that proves a problem, look at the parent process.

Exercises
C 4.3 [2] Start a xclock process in the background. In the $! shell variable you
will ﬁnd the PID of that process (it always contains the PID of the most re-
cently launched background process). Check the state of that process by
means of the “grep ^State: /proc/$!/status ” command. Stop the xclock by
moving it to the foreground and stopping it using Ctrl + z . What is the
process state now? (Alternatively, you may use any other long-running pro-
gram in place of xclock .)

C 4.4 [4] (When going over this manual for the second time.) Can you create
a zombie process on purpose?

4.3 Process Information—ps
You would normally not access the process information in /proc directly but use
the appropriate commands to query it.
The ps (“process status”) command is available on every Unix-like system.
Without any otions, all processes running on the current terminal are output. The
resulting list contains the process number PID , the terminal TTY , the process state
STAT , the CPU time used so far TIME and the command being executed.

$ ps
PID TTY STAT TIME COMMAND
997 1 S 0:00 -bash
1005 1 R 0:00 ps
$ _

There are two processes currently executing on the tty1 terminal: Apart from the
bash with PID 997, which is currently sleeping (state “S ”), a ps command is executed
using PID 1005 (state “R ”). The “operating” state mentioned above is not being
displayed in ps output.
The syntax of ps is fairly confusing. Besides Unix98-style options (like -l ) and
GNU-style long options (such as --help ), it also allows BSD-style options without
a leading dash. Here is a selection out of all possible parameters:
a (“all”) displays all processes with a terminal
--forest displays the process hierarchy
l (“long”) outputs extra information such as the priority
r (“running”) displays only runnable processes
T (“terminal”) displays all processes on the current terminal
U ⟨name⟩ (“user”) displays processes owned by user ⟨name⟩
x also displays processes without a terminal

B The unusual syntax of ps derives from the fact that AT&T’s ps traditionally
used leading dashes on options while BSD’s didn’t (and the same option
can have quite diﬀerent results in both ﬂavours). When the big reuniﬁcation
came in System V Release 4, one could hang on to most options with their
customary meaning.
4.4 Processes in a Tree—pstree 57

If you give ps a PID, only information pertaining to the process in question will
be displayed (if it exists):

$ ps 1
PID TTY STAT TIME COMMAND
1 ? Ss 0:00 init [2]

With the -C option, ps displays information about the process (or processes) based
on a particular command:

$ ps -C konsole
PID TTY TIME CMD
4472 ? 00:00:10 konsole
13720 ? 00:00:00 konsole
14045 ? 00:00:14 konsole

(Alternatively, grep would help here as well.)

Exercises
C 4.5 [!2] What does the information obtainable with the ps command mean?
Invoke ps without an option, then with the a option, and ﬁnally with the ax
option. What does the x option do?

C 4.6 [3] The ps command allows you to determine the output format your-
self by means of the -o option. Study the ps (1) manual page and specify a
ps command line that will output the PID, PPID, the process state and the
command.

4.4 Processes in a Tree—pstree
If you do not want to obtain every bit of information about a process but are rather
interested in the relationships between processes, the pstree command is helpful. pstree
pstree displays a process tree in which the child processes are shown as depending
on their parent process. The processes are displayed by name:

Identical processes are collected in brackets and a count and “*” are displayed.
The most important options of pstree include:
-p displays PIDs along with process names
-u displays process owners’ user name
-G makes the display prettier by using terminal graphics characters—whether this
is in fact an improvement depends on your terminal

B You can also obtain an approximated tree structure using “ps --forest ”. The
tree structure is part of the COMMAND column in the output.

4.5 Controlling Processes—kill and killall
signals The kill command sends signals to selected processes. The desired signal can be
speciﬁed either numerically or by name; you must also pass the process number
in question, which you can ﬁnd out using ps :

$ kill -15 4711 Send signal SIGTERM to process 4711
$ kill -TERM 4711 Same thing
$ kill -SIGTERM 4711 Same thing again
$ kill -s TERM 4711 Same thing again
$ kill -s SIGTERM 4711 Same thing again
$ kill -s 15 4711 Guess what

Here are the most important signals with their numbers and meaning:
SIGHUP (1, “hang up”) causes the shell to terminate all of its child processes that
use the same controlling terminal as itself. For background processes with-
out a controlling terminal, this is frequently used to cause them to re-read
their conﬁguration ﬁles (see below).

SIGINT (2, “interrupt”) Interrupts the process; equivalent to the Ctrl + c key com-
bination.
SIGKILL (9, “kill”) Terminates the process and cannot be ignored; the “emergency
brake”.
SIGTERM (15, “terminate”) Default for kill and killall ; terminates the process.
SIGCONT (18, “continue”) Lets a process that was stopped using SIGSTOP continue.
SIGSTOP (19, “stop”) Stops a process temporarily.

SIGTSTP (20, “terminal stop”) Equivalent to the Ctrl + z key combination.

A You shouldn’t get hung up on the signal numbers, which are not all guaran-
teed to be the same on all Unix versions (or even Linux platforms). You’re
usually safe as far as 1, 9, or 15 are concerned, but for everything else you
should rather be using the names.
4.6 pgrep and pkill 59

Unless otherwise speciﬁed, the signal SIGTERM (“terminate”) will be sent, which
(usually) ends the process. Programs can be written such that they “trap” signals
(handle them internally) or ignore them altogether. Signals that a process neither
traps nor ignores usually cause it to crash hard. Some (few) signals are ignored
by default.
The SIGKILL and SIGSTOP signals are not handled by the process but by the kernel
and hence cannot be trapped or ignored. SIGKILL terminates a process without
giving it a chance to object (as SIGTERM would), and SIGSTOP stops the process such
that it is no longer given CPU time.
kill does not always stop processes. Background processes which provide sys-
tem services without a controlling terminal—daemons—usually reread their con- daemons
ﬁguration ﬁles without a restart if they are sent SIGHUP (“hang up”).
You can apply kill , like many other Linux commands, only to processes that
you actually own. Only root is not subject to this restriction.
Sometimes a process will not even react to SIGKILL . The reason for this is ei-
ther that it is a zombie (which is already dead and cannot be killed again) or else
blocked in a system call. The latter situation occurs, for example, if a process waits
for a write or read operation on a slow device to ﬁnish.
An alternative to the kill command is the killall command. killall acts just killall
like kill —it sends a signal to the process. The diﬀerence is that the process must
be named instead of addressed by its PID, and that all processes of the same name
are signalled. If no signal is speciﬁed, it sends SIGTERM by default (like kill ). killall
outputs a warning if there was nothing to signal to under the speciﬁed name.
The most important options for killall include:
-i killall will query you whether it is actually supposed to signal the process in
question.
-l outputs a list of all available signals.
-w waits whether the process that was signalled actually terminates. killall
checks every second whether the process still exists, and only terminates
once it is gone.

A Be careful with killall if you get to use Solaris or BSD every now and then.
On these systems, the command does exactly what its name suggests—it
kills all processes.

Exercises
C 4.7 [2] Which signals are being ignored by default? (Hint: signal (7))

4.6 pgrep and pkill
As useful as ps and kill are, as diﬃcult can it be sometimes to identify exactly the
right processes of interest. Of course you can look through the output of ps using
grep , but to make this “foolproof” and without allowing too many false positives
is at least inconvenient, if not tricky. Nicely enough, Kjetil Torgrim Homme has
taken this burden oﬀ us and developed the pgrep program, which enables us to
search the process list conveniently. A command like

$ pgrep -u root sshd

will, for example, list the PIDs of all sshd processes belonging to root .

B By default, pgrep restricts itself to outputting PIDs. Use the -l option to get it
to show the command name, too. With -a it will list the full command line.

B The -d option allows you to specify a separator (the default is “\n ”):
60 4 Process Management

$ pgrep -d, -u hugo bash
4261,11043,11601,12289

You can obtain more detailed information on the processes by feeding the
PIDs to ps :

$ ps up $(pgrep -d, -u hugo bash)

(The p option lets you give ps a comma-separated list of PIDs of interest.)
pgrep ’s parameter is really an (extended) regular expression (consider egrep )
which is used to examine the process names. Hence something like

$ pgrep '^([bd]a|t?c|k|z|)sh$'

will look for the common shells.

B Normally pgrep considers only the process name (the ﬁrst 15 characters of the
process name, to be exact). Use the -f option to search the whole command
line.
You can add search criteria by means of options. Here is a small selection:
-G Consider only processes belonging to the given group(s). (Groups can be spec-
iﬁed using names or GIDs.)
-n Only display the newest (most recently started) of the found processes.
-o Only display the oldest (least recently started) of the found processes.
-P Consider only processes whose parent processes have one of the given PIDs.
-t Consider only processes whose controlling terminal is listed. (Terminal names
should be given without the leading “/dev/ ”.)
-u Consider only processes with the given (eﬀective) UIDs.

B If you specify search criteria but no regular expression for the process name,
all processes matching the search criteria will be listed. If you omit both you
will get an error message.
The pkill command behaves like pgrep , except that it does not list the found
processes’ PIDs but sends them a signal directly (by default, SIGTERM ). As in kill
you can specify another signal:

# pkill -HUP syslogd

The --signal option would also work:

# pkill --signal HUP syslogd

B The advantage of pkill compared to killall is that pkill can be much more
speciﬁc.

Exercises
C 4.8 [!1] Use pgrep to determine the PIDs of all processes belonging to user
hugo . (If you don’t have a user hugo , then specify some other user instead.)

C 4.9 [2] Use two separate terminal windows (or text consoles) to start one
“sleep 60 ” command each. Use pkill to terminate (a) the ﬁrst command
started, (b) the second command started, (c) the command started in one
of the two terminal windows.
4.7 Process Priorities—nice and renice 61

4.7 Process Priorities—nice and renice
In a multi-tasking operating system such as Linux, CPU time must be shared
among various processes. This is the scheduler’s job. There is normally more
than one runnable process, and the scheduler must allot CPU time to runnable
processes according to certain rules. The deciding factor for this is the priority priority
of a process. The priority of a process changes dynamically according to its prior
behaviour—“interactive” processes, i. e, ones that do I/O, are favoured over those
that just consume CPU time.
As a user (or administrator) you cannot set process priorities directly. You can
merely ask the kernel to prefer or penalise processes. The “nice value” quantiﬁes
the degree of favouritism exhibited towards a process, and is passed along to child
processes.
A new process’s nice value can be speciﬁed with the nice command. Its syntax nice
is

nice [- ⟨nice value⟩] ⟨command⟩ ⟨parameter⟩ …

(nice is used as a “preﬁx” for another command).
The possible nice values are numbers between −20 and +19. A negative nice possible nice values
value increases the priority, a positive value decreases it (the higher the value, the
“nicer” you are towards the system’s other users by giving your own processes a
lower priority). If no nice value is speciﬁed, the default value of +10 is assumed.
Only root may start processes with a negative nice value (negative nice value are
not generally nice for other users).
The priority of a running process can be inﬂuenced using the renice command. renice
You call renice with the desired new nice value and the PID (or PIDs) of the pro-
cess(es) in question:

renice [- ⟨nice value⟩] ⟨PID⟩ …

Again, only the system administrator may assign arbitrary nice values. Normal
users may only increase the nice value of their own processes using renice —for
example, it is impossible to revert a process started with nice value 5 back to nice
value 0, while it is absolutely all right to change its nice value to 10. (Think of a
ratchet.)

Exercises
C 4.10 [2] Try to give a process a higher priority. This may possibly not work—
why? Check the process priority using ps .

4.8 Further Process Management Commands—nohup
and top
When you invoke a command using nohup , that command will ignore a SIGHUP sig- Ignoring SIGHUP
nal and thus survive the demise of its parent process:

nohup ⟨command⟩ …

The process is not automatically put into the background but must be placed there
by appending a & to the command line. If the program’s standard output is a ter-
minal and the user has not speciﬁed anything else, the program’s output together
with its standard error output will be redirected to the nohup.out ﬁle. If the current
directory is not writable for the user, the ﬁle is created in the user’s home directory
instead.
62 4 Process Management

top top uniﬁes the functions of many process management commands in a single
program. It also provides a process table which is constantly being updated. You
can interactively execute various operations; an overview is available using h .
For example, it is possible to sort the list according to several criteria, send signals
to processes ( k ), or change the nice value of a process ( r ).

Commands in this Chapter
kill Terminates a background process bash (1), kill (1) 58
killall Sends a signal to all processes matching the given name killall (1) 59
nice Starts programs with a diﬀerent nice value nice (1) 61
nohup Starts a program such that it is immune to SIGHUP signals nohup (1) 61
pgrep Searches processes according to their name or other criteria
pgrep (1) 59
pkill Signals to processes according to their name or other criteria
pkill (1) 60
ps Outputs process status information ps (1) 56
pstree Outputs the process tree pstree (1) 57
renice Changes the nice value of running processes renice (8) 61
top Screen-oriented tool for process monitoring and control top (1) 61

Summary
• A process is a program that is being executed.
• Besides a program text and the corresponding data, a process has attributes
such as a process number (PID), parent process number (PPID), owner,
groups, priority, environment, current directory, …
• All processes derive from the init process (PID 1).
• ps can be used to query process information.
• The pstree command shows the process hierarchy as a tree.
• Processes can be controlled using signals.
• The kill and killall commands send signals to processes.
• The nice and renice commands are used to inﬂuence process priorities.
ulimit limits the resource usage of a process.
• top is a convenient user interface for process management.
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

5
Hardware

Contents
5.1 Fundamentals . . . . . . . . . . . . . . . . . . . . . 64
5.2 Linux and PCI (Express) . . . . . . . . . . . . . . . . . 65
5.2.1 USB. . . . . . . . . . . . . . . . . . . . . . . 67
5.3 Peripherals . . . . . . . . . . . . . . . . . . . . . . 69
5.3.1 Overview . . . . . . . . . . . . . . . . . . . . 69
5.3.2 Devices and Drivers . . . . . . . . . . . . . . . . . 70
5.3.3 The /sys Directory . . . . . . . . . . . . . . . . . 72
5.3.4 udev . . . . . . . . . . . . . . . . . . . . . . 73
5.3.5 Device Integration and D-Bus . . . . . . . . . . . . . 74

Goals
• Knowing the basics of Linux hardware support on PCs
• Being familiar with terms like BIOS, UEFI, PCI, USB and others
• Mastering the basics of dynamic peripheral support via udev

Prerequisites
• Basic Linux knowledge
• A basic familiarity with PC hardware and its support by other operating
systems is helpful

adm1-hardware.tex (33e55eeadba676a3 )
64 5 Hardware

5.1 Fundamentals
Linux supports a very broad spectrum of system architectures and platforms: The
Linux kernel is available for all of today’s common microprocessors, and Linux
runs on computers from modest PDAs to the largest mainframes. Linux is also an
important part of many devices that one would not associate with it at ﬁrst sight—
digital cameras and camcorders, routers, television sets and set-top boxes, GPS
navigation devices and many more—and that sometimes use “unusual” hard-
ware. The most common Linux systems by far that are used as “computers” in
the proper sense of the word are based on the x86 PC architecture founded by
IBM and Intel.

The x86 PC architecture by IBM and Intel is also the only one relevant to
LPIC-1 certiﬁcation, although it should be said that architecture-speciﬁc
questions form a very small part of the exam today (they used to be much
more prevalent).

Architecture Computers today possess one or more processors (CPUs), memory for running
programs and data (RAM), and secondary storage for ﬁles (disks, SSDs). Added
to this are various peripheral devices for input (keyboard, mouse, graphics tablet,
webcam, …) and output (graphics “card”—today often part of the CPU—, audio,
…) and much more. These peripherals are either part of the “motherboard” or
are added via various interfaces (PCIe, USB, SATA, SCSI, …). Ethernet or WLAN
are used for networking. To be able to boot, a computer also needs “ﬁrmware” in
read-only memory (ROM)1

B On PCs, the ﬁrmware is either called “BIOS” (on older systems) or “UEFI”.
In former times—during the age of MS-DOS—the BIOS served as the inter-
face between the operating system proper (DOS) and the computer’s hard-
ware. Today it is only used to initialise the computer at boot and to ﬁnd
and launch a “real” operating system. Modern operating systems tend to
disregard the BIOS completely.

B UEFI is a more modern implementation of the same idea, which does away
with some nasty shortcomings of the BIOS—which is after all a remnant of
the 1980s—and oﬀers more extensive options, e. g., when dealing with sev-
eral operating systems (or operating system versions) on the same computer
or attempting to secure the system’s software against malicious manipula-
tion.

B New computers are based on UEFI, but can frequently pretend to have a tra-
ditional BIOS if desired, so that older operating systems can be supported.

While in former times extensive BIOS conﬁguration might have been required
before a computer would run Linux eﬃciently, this is usually no longer a problem
today. You may only have to perform very few settings, e. g., to set the date and
time.

hardware clock B You will need to decide whether the hardware clock should be set to “zone
time” (such as Paciﬁc Standard Time for the west coast of North America)
or “universal time” (UTC). Linux—which consults the hardware clock only
when starting—can handle both, but must know where it is at. UTC is
preferable on machines running only Linux, while zone time may be useful
if the computer is occasionally running other alternative operating systems.

Hard disks B On BIOS-based computers, the BIOS must know about the hard disk the
system is to be booted from (at minimum). You can generally tell the BIOS
about the properties of the various disk drives in the system, and whether
1 Technically speaking, most computers today use “ﬂash ROM”, which is rewritable using appro-

priate tools, instead of “true” ROM. This doesn’t change the principle.
5.2 Linux and PCI (Express) 65

the BIOS should see them at all—these settings have no bearing on the later
use of the disk(s) by Linux, hence it is possible to trick old-fashioned BIOS
implementations which cannot handle modern large disk drives, simply by
deactivating the disk drive in the BIOS. Linux can access it later, but you
won’t be able to boot from that disk.

B Within the BIOS setup, you can enter a hard disks>geometrydisk geome- disk geometry
try for each disk, i. e., the number of read/write heads, cylinders per head,
and sectors per cylinder of the disk. Today’s disks have more sectors on
the outside cylinders than on those near the centre of the platters, and no
longer bother with the “cylinder/head/sector” or “CHS” model of address-
ing disk sectors but simply count sectors oﬀ sequentially from 0 to …—the
LBA mode (“linear block access”). Even so, every disk still claims a com- LBA mode
pletely fabricated geometry matching the disk capacity in order to make
older BIOSes and DOS happy. With modern BIOS implementations, you
can put disks into LBA mode explicitly, but as long as the system boots us-
ing the default values this is not required.

The BIOS can also be used to enable or disable various types of peripherals: peripherals

• Assignment of serial ports from the operating system’s point of view (usu-
ally COM1: or some such) to existing serial ports, IrDA ports, etc.
• Support for USB, in particular USB keyboards and mice
• Internal graphics and sound support
• Energy and status management (APM, ACPI)
Within the scope of this course it is not feasible (and, fortunately, not necessary
for the purposes of LPIC-1) to give more detailed advice. Keep in mind that these
settings exist and that, if some device persistently refuses to get to work on Linux,
you should check what the BIOS has to say on the matter. It is possible for the
device in question to be disabled at the BIOS level, or for a BIOS-enabled device
to conﬂict with the device that is really wanted.

5.2 Linux and PCI (Express)
During the 30 years or more since the ﬁrst IBM PC was marketed, the “internals”
of the hardware have changed radically in almost all aspects. Not least the bus
system connecting the CPU and memory with the most important peripherals
has undergone several metamorphoses since the original “ISA” bus; the current
standard is called PCI Express (PCIe) and has all but ousted its predecessors. PCI Express

B PCIe is vaguely based on PCI and has taken over many PCI concepts (such as
device codes for hardware detection, see below), but its electrical properties
are completely diﬀerent.

B Unlike its predecessors, PCIe is conceptually a serial bus which allows in-
dependent point-to-point connections between diﬀerent devices. Formerly,
all devices shared a parallel bus, which put a cap on the maximum speeds
that could be achieved.

B In the interest of eﬃciency, PCIe allows the transfer of data packets over
several connections (“lanes”), but that does not take away from the general
principle.

A major advantage of PCIe is that automatic hardware detection is possible. hardware detection
Every device connected to the bus reports a code specifying its type, manufac-
turer, and model. You can query this information using the lspci command (Fig-
ure 5.1). At the beginning of each line there is the “PCI ID” of each device, giving
its position on the PCI bus.
66 5 Hardware

# lspci
00:00.0 Host bridge: Intel Corp Core Processor DRAM Controller (rev 12)
00:01.0 PCI bridge: Intel Corp Core Processor PCI Express x16 Root Port
00:16.0 Communication controller: Intel Corporation 5 Series/3400
Series Chipset HECI Controller (rev 06)
00:16.3 Serial controller: Intel Corp 5 Series/3400 Series Chipset KT
Controller (rev 06)
00:19.0 Ethernet controller: Intel Corporation 82577LM Gigabit Network
Connection (rev 06)
00:1a.0 USB controller: Intel Corporation 5 Series/3400 Series
Chipset USB2 Enhanced Host Controller (rev 06)
00:1b.0 Audio device: Intel Corporation 5 Series/3400 Series Chipset
High Definition Audio (rev 06)

01:00.0 VGA compatible controller: NVIDIA Corporation GT218M
[NVS 3100M] (rev a2)
01:00.1 Audio device: NVIDIA Corporation High Definition Audio
Controller (rev a1)

Figure 5.1: Output of lspci on a typical x86-based PC

B Linux uses the device codes to select and conﬁgure drivers for the various
peripheral devices it detects. The kernel and the udev infrastructure collab-
orate on this. We shall look at this in more detail later.

B In former times you would have had to guess, but that could cause prob-
lems including system crashes, if a driver poked the “right” wrong device.
In cases of doubt, the system administrator (you) would have to perform
tedious manual interventions!

lspci supports some interesting options: “lspci -v ” provides more verbose out-
put:

# lspci -v
00:00.0 Host bridge: Intel Corp Core Processor DRAM Controller (rev 12)
Subsystem: Hewlett-Packard Company Device 172b
Flags: bus master, fast devsel, latency 0
Capabilities: [e0] Vendor Specific Information: Len=0c <?>

00:01.0 PCI bridge: Intel Corporation Core Processor PCI Express x16
Root Port (rev 12) (prog-if 00 [Normal decode])
Flags: bus master, fast devsel, latency 0
Bus: primary=00, secondary=01, subordinate=01, sec-latency=0
I/O behind bridge: 00005000-00005fff
Memory behind bridge: d2000000-d30fffff

B In former times, similar information used to be available from the /proc/pci
“ﬁle”, which is no longer provided by default by newer kernels. The oﬃcial
method of obtaining PCI data is lspci .

“lspci -t ” gives a tree-like representation of the connections between the vari-
ous components:

# lspci -t
-+-[0000:ff]-+-00.0
5.2 Linux and PCI (Express) 67

| +-00.1
| +-02.0
| +-02.1
| +-02.2
| \-02.3
\-[0000:00]-+-00.0
+-01.0-[01]--+-00.0
| \-00.1
+-16.0
+-16.3
+-19.0

This tells you, for example, that the chipset’s “PCI bridge” (device 0000:00:01.0 )
provides the connection to the “VGA compatible controller) (device 0000:01:00.0)
and the “audio device” (0000:01:00.1 , more precisely the graphic card’s HDMI au-
dio output). The Ethernet adapter (device 0000:01:19.0 ) is also connected via PCIe.
Finally, “lspci -n ” outputs the device codes directly, without looking up their device codes
meaning in the PCI database:

# lspci -n
00:00.0 0600: 8086:0044 (rev 12)
00:01.0 0604: 8086:0045 (rev 12)
00:16.0 0780: 8086:3b64 (rev 06)
00:16.3 0700: 8086:3b67 (rev 06)
00:19.0 0200: 8086:10ea (rev 06)
00:1a.0 0c03: 8086:3b3c (rev 06)
00:1b.0 0403: 8086:3b56 (rev 06)

Exercises
C 5.1 [!2] Examine the hardware structure of your system using commands
such as “lspci -v ” or “lspci -t ”. What can you ﬁnd out?

5.2.1 USB
The “Universal Serial Bus”, or USB, is nowadays used to connect practically all
external peripheral devices that do not use wireless technology (such as WLAN or
Bluetooth). The goal is a “legacy-free” computer that can do without the erstwhile
multitude of device-speciﬁc connectors for the keyboard, mouse, printer, modem,
…

B USB traditionally uses a “foolproof” cabling concept with diﬀerent sock-
ets for computers (“A”) and peripherals (“B”) and appropriate plugs at the
end of the cables, in order to avoid mistakes. In addition, USB allows un-
plugging and re-plugging of devices when powered up (even though this
is generally not a great idea when dealing with hard disks or USB thumb-
drives).

B The newest fad (2015) are “USB C” sockets, which are identical between
computers and peripherals and do not care which way the plug is inserted
(which is a constant nuisance especially with USB “A” plugs—Pro tip, the
USB logo on the plug is always supposed to point towards you). USB C con-
nections aren’t just useful for USB, but also many other things (like powerful
charging currents, graphics signals, and what else one might think of), but
the system is still fairly new.
68 5 Hardware

Table 5.1: USB standards

Version since Max speed Devices
1.1 1998 1,5 MiBit/s Keyboards, mice, modems, audio de-
vices, …
12 MiBit/s 10-MiBit ethernet, disks, …
2.0 2000 480 MiBit/s disks, 100-MiBit ethernet, video …
3.0 2008 5 GiBit/s disks, graphics, …
3.1 2013 10 GiBit/s PCIe

USB is now part of all new PCs. This means that every PC contains one or more
USB controllers USB controllers, each of which can manage up to 127 USB devices.

B Since PCs do not feature that many USB sockets, USB allows the tree-like
USB hubs connection of devices via USB hubs. You can therefore use one USB socket in
your computer to connect a hub with several additional sockets; the devices
connected to the hub will share the connection to your computer. You can
even connect further hubs to a hub (although this may not always be a bright
idea in actual practice).

Since its introduction in 1996, the USB standard has continuously been im-
proved, and newer versions pushed the maximum possible transmission speed
ever higher. The current incarnation (2015) is version 3.1, which basically makes
it possible to drive external devices at PCIe bus speeds (which is nice, e. g., for
external graphics cards).

B Newer USB implementations are, in principle, backwards-compatible. You
can connect USB-2.0 devices to a USB-1.1 bus and operate them at USB-1.1
speeds. Conversely, you can connect USB-1.1 devices to a USB-2.0 hub or
controller, but they won’t be any faster—but they will not slow down “real”
USB-2.0 devices, either.

B USB-3.𝑥 devices actually support USB 2.0 and USB 3.𝑥 simultaneously on
separate leads in the cable and separate contacts in the connectors. You
can plug a USB-2.0 plug into a USB-3.𝑥 socket on your computer, and that
will yield a USB-2.0 connection; conversely, you can plug a USB-3.𝑥 plug
into a USB-2.0 socket, which will also lead to a USB-2.0 connection. For
“genuine” USB 3.𝑥, your computer must support USB 3.𝑥, and you need
USB-3.𝑥 devices and matching cables.

Enumeration When a USB device is connected to the bus, it is assigned a number between 1
descriptor and 127 as its device number, and the computer reads the device’s “descriptor”.
The descriptor speciﬁes the type of device, the type of USB port it supports, and
class so on. In particular, each device belongs to a class of devices that are handled in
a similar fashion by the computer—such as the “human interface devices”, i. e.,
devices that serve as input devices for people: keyboards, mice, joysticks, but also
switches, dials, phone keyboards, steering wheels, data gloves and so on. Classes
subclass may have subclasses that narrow the type of device down even more.
You can check out which USB devices are connected to your system by using
the lsusb command. Like lspci , it outputs the device numbers and names:

$ lsusb
Bus 002 Device 002: ID 8087:0020 Intel Corp. Integrated Rate
Matching Hub
Bus 002 Device 001: ID 1d6b:0002 Linux Foundation 2.0 root hub
Bus 001 Device 006: ID 04f2:b15e Chicony Electronics Co., Ltd
Bus 001 Device 008: ID 046d:0807 Logitech, Inc. Webcam B500
Bus 001 Device 010: ID 046a:0023 Cherry GmbH CyMotion Master Linux
5.3 Peripherals 69

Keyboard G230
Bus 001 Device 009: ID 046d:c52b Logitech, Inc. Unifying Receiver
Bus 001 Device 007: ID 05e3:0608 Genesys Logic, Inc. USB-2.0 4-Port HUB

Using the -v option, the program becomes noticeable more “chatty”:

# lsusb -v
Bus 002 Device 002: ID 8087:0020 Intel Corp. Integrated Rate
Matching Hub
Device Descriptor:
bLength 18
bDescriptorType 1
bcdUSB 2.00
bDeviceClass 9 Hub
bDeviceSubClass 0 Unused
bDeviceProtocol 1 Single TT
bMaxPacketSize0 64
idVendor 0x8087 Intel Corp.
idProduct 0x0020 Integrated Rate Matching Hub
bcdDevice 0.00
iManufacturer 0
iProduct 0
iSerial 0
bNumConfigurations 1
Configuration Descriptor:
bLength 9
bDescriptorType 2
wTotalLength 25
bNumInterfaces 1
bConfigurationValue 1
iConfiguration 0
bmAttributes 0xe0
Self Powered
Remote Wakeup
MaxPower 0mA
Interface Descriptor:
bLength 9
bDescriptorType 4
bInterfaceNumber 0
bAlternateSetting 0

B The output of “lsusb -v ” is quite extensive but also fairly unreadable. The
usbview program presents the data rather more nicely (Figure 5.2).

B If you would like to know where lsusb gets its encyclopedic knowledge of
USB peripherals: Take a look at the /var/lib/usbutils/usb.ids ﬁle.

5.3 Peripherals
5.3.1 Overview
The interesting question that remains is how Linux can handle almost arbitrary
peripheral devices—both those built in to the computer as well as those that can
come and go while the machine is running.
70 5 Hardware

Figure 5.2: The usbview program

Talking to peripheral devices is the Linux kernel’s natural business. Its job
is to ﬁnd and initialise connected peripherals and to allow userspace programs
controlled access to them. This of course includes appropriate rights assignment,
such that only suitably privileged users can access peripherals directly.

B In real life, the Linux kernel likes to share the device management work
with userspace programs. This makes sense, because it both keeps the ker-
nel lean and also avoids the hard-wiring of too much functionality inside
the kernel if it is better placed in userspace programs, where it is easier to
program and customise.

B By way of an example: Simple printers are nowadays connected to the com-
puter via USB. If a new printer is connected, the Linux kernel ﬁnds out about
this and can use the USB manufacturer and model codes to ﬁgure out that
the newly recognised peripheral is a printer, and which model of which
manufacturer it is. The result of this initialisation process is a device ﬁle be-
low /dev which userspace programs can use to contact the printer.—Which
doesn’t mean that arbitrary user applications get to talk to the printer di-
rectly. Direct access to its device ﬁle is restricted, via suitable access rights, to
the printer subsystem (CUPS). The CUPS conﬁguration routine can access
the Linux kernel’s information on the printer’s make and model, and se-
lect an appropriate printer driver. User applications—such as LibreOﬃce—
print documents by passing them to CUPS in a suitable format (PostScript or
PDF). CUPS then converts the job into a format that the printer understands,
and passes it via the device ﬁle to the Linux kernel, which then routes it to
the actual printer via USB.

5.3.2 Devices and Drivers
Linux supports the wide variety of peripheral devices available for PCs by means
loadable modules of drivers which are usually furnished as loadable modules. These loadable mod-
ules can be found in subdirectories of /lib/modules and can be loaded into the ker-
nel manually using the modprobe command:

# modprobe foo

for example locates and loads the foo.ko . Should the desired module depend on
the functionality of other modules which are not currently loaded, modprobe ar-
ranges for them to be loaded as well.
5.3 Peripherals 71

B Linux uses loadable modules not just to implement device drivers in the
proper sense of the word, but also for network protocols, the packet ﬁlter
infrastructure, and many more system features that are not required on ev-
ery system or at all times. Modules also make kernel development easier
and are therefore popular with programmers.

The lsmod command gives an overview of the currently-loaded modules.

$ lsmod
Module Size Used by
nfs 213896 1
lockd 54248 1 nfs
nfs_acl 2912 1 nfs
sunrpc 162144 10 nfs,lockd,nfs_acl
tun 8292 1
michael_mic 2304 4
arc4 1824 4

You can also try to remove a loaded module using “modprobe -r ”:

# modprobe -r foo

(Whether this actually works depends on your kernel.) If the module to be re-
moved is the only one depending on another module, modprobe tries to remove the
other module, too.
You will only rarely have to run modprobe manually, as the kernel can mostly take Automatic loading on demand
care of loading modules on demand. The basis for this is the /etc/modprobe.conf ﬁle
(or the ﬁles in the /etc/modprobe.d directory), which contains entries like

alias block-major-3-* ide_generic Block device
alias char-major-10-1 psmouse Character-oriented device
alias net-pf-16-proto-8 scsi_transport_iscsi Network protocol

(and various others). These entries connect device ﬁles in /dev such as

$ ls -l /dev/hda1 /dev/psaux
brw-rw---- 1 root disk 3, 1 Jan 22 02:03 /dev/hda1
crw-rw---- 1 root root 10, 1 Jan 22 02:03 /dev/psaux

to the corresponding drivers—block-major-3-* means nothing other that for block-
oriented device ﬁles with a major device number of 3 and arbitrary minor device
number (such as /dev/hda1 in our example), the ide_generic driver module is appro-
priate, while access to /dev/psaux , a character-oriented device ﬁle with numbers
(10, 1) is managed through the driver named by char-major-10-1 , namely psmouse .
Whenever the device ﬁle in question is ﬁrst accessed, the Linux kernel tries to
locate and enable the corresponding driver module.

B By now you are surely asking yourself how a driver like ide_generic can be
read on demand from an IDE disk, if Linux requires it to talk to the disk
in the ﬁrst place. The answer to that is that this driver, when booting from
IDE disk, is not really read from /lib/modules , but that the kernel obtained it
earlier on from the “initial RAM disk”, which the boot loader fetched from
disk together with the kernel itself (by means of the BIOS, hence without
involving Linux). (More detail is in Chapter 8).

B With modprobe and the corresponding conﬁguration ﬁles, you can do many
other strange and wonderful things that are far beyond the scope of this
discussion. You will ﬁnd more about these topics from the Linup Front
training manual, Linux System Adaptation.
72 5 Hardware

5.3.3 The /sys Directory
Up to and including Linux 2.4, the /proc directory represented the only way to
access details of the kernel and system conﬁguration. However, the kernel devel-
opers disliked the uncontrolled growth of entries under /proc , in particular those
whose purpose did not have anything whatsoever to do with processes (the orig-
inal intent of the directory). For this reason, the kernel developers decided to
move, in the medium to long term, those aspects of /proc that didn’t have any-
sysfs thing to do with process management to a new pseudo ﬁle system, sysfs , where
stricter rules should apply. The sysfs is usually mounted on /sys , and is available
from the Linux kernel version 2.6 onwards.
connection type In particular, /sys/bus allows accessing devices depending on their connection
type (“bus”; pci , usb , scsi , …). File /sys/devices/ also allows accessing devices,
only the sort order is dﬀerent (device type, e. g., PCI bus address). This redun-
dancy is implemented by means of symbolic links:

$ ls -ld /sys/devices/pci0000:00/0000:00:07.2/usb1
drwxr-xr-x 6 root root 0 Jun 27 19:35
/sys/devices/pci0000:00/0000:00:1a.0/usb1
$ ls -ld /sys/bus/usb/devices/usb1
lrwxrwxrwx 1 root root 0 Jun 27 19:35
/sys/bus/usb/devices/usb1

In the example you see the directory allowing access to the data of the ﬁrst USB
connector. This is, in fact, the same directory. Such a directory contains various
ﬁles with information about the device in question:

$ ls /sys/bus/usb/devices/usb1
1-0:1.0 bmAttributes devpath remove
1-1 bMaxPacketSize0 driver serial
authorized bMaxPower ep_00 speed
authorized_default bNumConfigurations idProduct subsystem
avoid_reset_quirk bNumInterfaces idVendor uevent
bcdDevice busnum manufacturer urbnum
bConfigurationValue configuration maxchild version
bDeviceClass descriptors power
bDeviceProtocol dev product
bDeviceSubClass devnum quirks
$ cat /sys/bus/usb/devices/usb1/product
EHCI Host Controller

One disadvantage of the former kernel concept was the uncertainty about the
assignment of interfaces assignment of interfaces to devices, such as the device ﬁles. The “numbering” of
devices (as in, sda , sdb , sdc , …, or eth0 , eth1 , …) was not easy to reproduce. sysfs ,
on the other hand, lets us assign interfaces to devices in an unambiguous fashion.
You will ﬁnd the interfaces in the directories /sys/block (block devices) and /sys/
class (character devices, more or less):

$ ls /sys/block/
dm-0 dm-2 dm-4 dm-6 loop1 loop3 loop5 loop7 sr0
dm-1 dm-3 dm-5 loop0 loop2 loop4 loop6 sda
$ ls /sys/class/
ata_device dmi mem regulator tty
ata_link firmware misc rfkill vc
ata_port graphics mmc_host rtc video4linux
backlight hidraw net scsi_device vtconsole
bdi hwmon pci_bus scsi_disk wmi
block i2c-adapter pcmcia_socket scsi_host
bluetooth ieee80211 power_supply sound
5.3 Peripherals 73

bsg input ppdev spi_master
dma leds printer thermal

An advantage: This lists only the devices that are actually on the system.
Interfaces are assigned to block devices by means of symbolic links; there is a
ﬁle called device , like

$ ls -l /sys/block/sda/device
lrwxrwxrwx 1 root root 0 Jun 27 19:45 /sys/block/sda/device
-> ../../0:0:0:0

5.3.4 udev
In former times, Linux distributors used to create, as part of the installation pro-
cess, a /dev directory that was ﬁlled with device ﬁles for all the peripherals under
the sun. In the meantime, this approach is deprecated in favour of the idea of dy-
namically creating device ﬁles only for those devices that are actually available.
The Linux infrastructure for this is called udev , short for “userspace /dev ”. It
was implemented by Greg Kroah-Hartman and Kay Sievers (among others) and
is available in its current form since kernel 2.6.13. udev consists mainly of a library
called namedev , which deals with assigning names to devices, and a daemon called
udevd which does (or delegates) the actual work. udevd
With udev , the /dev directory on disk only contains the essential entries. Fairly
soon after the system is started, a tmpfs ﬁle system (i. e., a RAM disk, see Sec-
tion 7.1.6) is mounted over /dev , in which udev places the appropriate device ﬁles.
To this end, the udevd is started, which listens to “uevents”—event reports by the uevents
kernel concerning devices that have been added or removed. Each such report is
compared to a set of rules which decides under which name the device should ap-
pear in the /dev directory managed by udev , and which can also execute additional
actions, like uploading “ﬁrmware” available as a binary ﬁle to the device in order firmware
to initialise it. These rules can be found in /etc/udev .

B There is nothing wrong with having a device appear in /dev under several
names. Your digital camera, for example, might be assigned the name /dev/
camera in addition to /dev/sd “whatever”. It all comes down to a question of
rules.

B udev can also deal with devices that do not correspond to device ﬁles in /dev ,
such as network cards.

An interesting side eﬀect of this approach is that it was originally designed to
deal with devices that come and go at runtime—to be exact, at some point after
udevd was started (“hotplugging”). The same method, however, is also used un-
der the guise of “coldplugging” when the system is originally started, to initialise
the devices that are already available at that time. Of course the uevents the ker-
nel sends to register these devices cannot be processed before udevd is running.
Still, the kernel makes it possible to re-send these uevents on demand, and ex-
actly that happens after udevd has become available, in order to add the devices
to the dynamic /dev directory. Accordingly, there is no diﬀerence between device
initialisation using “hotplugging” and “coldplugging”.

B To allow this, there is a device called uevent in each device’s entry below /sys .
You just need to write the string add to that ﬁle to trigger the corresponding
uevent.

B udev isn’t the ﬁrst attempt to include such an infrastructure in Linux. The
devfs proposed by Richard Gooch did a similar thing but inside the kernel
rather than outside (like udev ). devfs never gathered a large following in the
community, since various matters of policy concerning, e. g., device naming
74 5 Hardware

had been hard-wired into devfs (and hence in the kernel) and not all devel-
opers agreed with that. udev exposes its rules by means of editable conﬁgu-
ration ﬁles and is therefore much more ﬂexible. There used to be an infras-
tructure called hotplug to handle dynamic registration and de-registration of
devices, but this has been superseded by udev .

5.3.5 Device Integration and D-Bus
application programs The ﬁnal piece of the puzzle brings in application programs. How do you set
things up such that plugging in your digital camera will start your photo man-
agement program to download any new pictures? Or how does the icon for your
USB disk get placed on the backdrop of your graphical desktop environment?
One possible answer to this is “udisks”.

B Formerly this used an infrastructure called HAL (short for “hardware ab-
straction layer”). HAL, however, proved too complicated and limited and is
no longer being developed further.

B Udisks is one of the successor projects, and deals exclusively with stor-
age devices such as external disks and USB thumb drives; there are other
projects for other types of device. Large parts of HAL have also been assim-
ilated into udev .

HAL The idea behind HAL is to aggregate information from various sources: The
kernel (via udev ) reports that a new device is available and which one. The device
itself can be asked, as can default settings for the device that the user once made
in their graphical environment.

B This aggregation is necessary because the kernel may not actually know
everything worthwhile about a device. Many digital cameras, for exam-
ple, register with the system as “hard disks”, and the information that they
should really be treated as cameras must come from elsewhere.

D-Bus Udisks and similar project are based on D-Bus, a simple system for inter-
process communication that is supported by most desktop environments. Pro-
grams can tell D-Bus about services that they want to oﬀer to other programs.
Programs can also wait for events. (It doesn’t matter if several programs wait for
the same event; all interested parties are notiﬁed.) D-Bus has two major applica-
tion areas:
• Communication between programs in the same graphical environment. A
possible application might be for your telephony program to automatically
mute your music player when a call comes in.
• Communication between the operating system and your graphical environ-
ment.
Udisks consists of two components, a background service (udisksd ) and a user-
accessible tool (udisksctl ) which can communicate with the background service.
The background service is accessible through the D-Bus and will be started on
demand whenever a program issues a D-Bus query to udisks. This way programs
can ﬁnd out which storage devices are available as well as execute operations such
as mounting or unmounting devices. This includes appropriate privilege checks;
users may, for example, be allowed to plug USB thumb drives into their own com-
puters but will be denied access to USB thumb drives on other “seats”.
Here is an example of a udisksctl query about a USB thumb drive:

$ udisksctl status
MODEL REVISION SERIAL DEVICE
---------------------------------------------------------------------
ST9320423AS 0002SDM1 5VH5TBTC sda
5.3 Bibliography 75

hp DVDRAM GT30L mP04 KZKA2LB3306 sr0
TOSHIBA TransMemory 1.00 7427EAB351F9CE110FE8E23E sdb
$ udisksctl info -b /dev/sdb
/org/freedesktop/UDisks2/block_devices/sdb:
org.freedesktop.UDisks2.Block:
Configuration: []
CryptoBackingDevice: '/'
Device: /dev/sdb
DeviceNumber: 2064
Drive: '/org/freedesktop/UDisks2/drives/
TOSHIBA_TransMemory_7427EAB351F9CE110FE8E23E'
HintAuto: true
HintIconName:
HintIgnore: false
HintName:
HintPartitionable: true

Many of these data really derive from the udev database and may be inﬂuenced by
the administrator by way of suitable rules.

B In fairness we should also add that udisks represents a minimal solution.
The common graphical desktop environments like KDE or GNOME include
similar infrastructure, which also takes care of suitably representing storage
devices in the graphical environment.

Commands in this Chapter
lsmod Lists loaded kernel modules lsmod (8) 71
lspci Displays information about devices on the PCI bus lspci (8) 65
lsusb Lists all devices connected to the USB lsusb (8) 68
modprobe Loads kernel modules, taking dependencies into account
modprobe (8) 70
udevd Kernel uevent management daemon udevd (8) 73

Summary
• The ﬁrmware (BIOS/UEFI) organises the early startup of a Linux system,
but will subsequently not be used (apart from exceptional cases).
• Typical BIOS settings relevant to Linux include, for example, the date and
time, disk parameters, boot order, and the assignment of some system in-
terfaces to actual peripherals.
• The lspci command sheds light on a system’s PCI bus and the devices con-
nected to it.

Bibliography
Hardware-HOWTO04 Steven Pritchard. “Linux Hardware Compatibility
HOWTO”, January 2004. http://www.tldp.org/HOWTO/Hardware- HOWTO/

Linux-USB-Subsystem Brad Hards. “The Linux USB sub-system”.
http://www.linux- usb.org/USB- guide/book1.html

PCI-HOWTO01 Michael Will. “Linux PCI-HOWTO”, June 2001.
http://www.tldp.org/HOWTO/PCI- HOWTO.html
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

6
Hard Disks (and Other Secondary
Storage)

Contents
6.1 Fundamentals . . . . . . . . . . . . . . . . . . . . . 78
6.2 Bus Systems for Mass Storage . . . . . . . . . . . . . . . 78
6.3 Partitioning . . . . . . . . . . . . . . . . . . . . . . 81
6.3.1 Fundamentals . . . . . . . . . . . . . . . . . . . 81
6.3.2 The Traditional Method (MBR) . . . . . . . . . . . . . 82
6.3.3 The Modern Method (GPT) . . . . . . . . . . . . . . 83
6.4 Linux and Mass Storage . . . . . . . . . . . . . . . . . 84
6.5 Partitioning Disks. . . . . . . . . . . . . . . . . . . . 86
6.5.1 Fundamentals . . . . . . . . . . . . . . . . . . . 86
6.5.2 Partitioning Disks Using fdisk . . . . . . . . . . . . . 88
6.5.3 Formatting Disks using GNU parted . . . . . . . . . . . 91
6.5.4 gdisk . . . . . . . . . . . . . . . . . . . . . . 92
6.5.5 More Partitioning Tools . . . . . . . . . . . . . . . 93
6.6 Loop Devices and kpartx . . . . . . . . . . . . . . . . . 93
6.7 The Logical Volume Manager (LVM) . . . . . . . . . . . . . 95

Goals
• Understanding how Linux deals with secondary storage devices based on
ATA, SATA, and SCSI.
• Understanding MBR-based and GPT-based partitioning
• Knowing about Linux partitioning tools and how to use them
• Being able to use loop devices

Prerequisites
• Basic Linux knowledge
• Knowledge about Linux hardware support (see chapter 5)

adm1-platten.tex (33e55eeadba676a3 )
78 6 Hard Disks (and Other Secondary Storage)

6.1 Fundamentals
RAM is fairly cheap these days, but even so not many computers can get by with-
out the permanent storage of programs and data on mass storage devices. These
include (among others):
• Hard disks with rotating magnetic platters
• “Solid-state disks” (SSDs) that look like hard disks from the computer’s
point of view, but use ﬂash memory internally
• USB thumb drives, SD cards, CF cards, and other interchangeable media
based on ﬂash memory
• RAID systems that aggregate hard disks and present them as one big storage
medium
• SAN devices which provide “virtual” disk drives on the network
• File servers that oﬀer ﬁle access at an abstract level (CIFS, NFS, …)
In this chapter we shall explain the basics of Linux support for the ﬁrst three en-
tries in the list—hard disks, SSDs and ﬂash-based portable media like USB thumb
drives. RAID sstems and SAN are discussed in the Linup Front training man-
ual, Linux Storage and File Systems, ﬁle servers are discussed in Linux Infrastructure
Services.

6.2 Bus Systems for Mass Storage
IDE, ATA and SATA Until not so long ago, hard disks and optical drives such
IDE as CD-ROM and DVD readers and writers used to be connected via a “IDE con-
troller”, of which self-respecting PCs had at least two (with two “channels” each).

B “IDE” is really an abbreviation of “Integrated Drive Electronics”. The “inte-
grated drive electronics” alluded to here lets the computer see the disk as a
sequence of numbered blocks without having to know anything about sec-
tors, cylinders, and read/write heads, even though this elaborate charade
is still kept up between the BIOS, disk controller, and disks. However, for a
long time this description has applied to all hard disks, not just those with
the well-known “IDE” interface, which by now is oﬃcially called “ATA”,
short for “AT Attachment”1 .

Computers bought new these days usually still contain IDE interfaces, but the
method of choice to connect hard disks and optical drives today is a serial ver-
Serial ATA sion of the ATA standard, imaginatively named “Serial ATA” (SATA, for short).
Since SATA (i. e., approximately 2003), traditional ATA (or “IDE”) is commonly
called “P-ATA”, an abbreviation of “parallel ATA”. The diﬀerence applies to the
cable, which for traditional ATA is an inconvenient-to-place and electrically not
100% fortunate 40- or 80-lead ribbon cable which can transfer 16 data bits in par-
allel, and which links several devices at once to the controller. SATA, on the other
hand, utilises narrow ﬂat seven-lead cables for serial transmission, one per device
(usually produced in a cheerful orange colour).

B SATA cables and connectors are much more sensitive mechanically than the
corresponding P-ATA hardware, but they make up for that through other
advantages: They are less of an impediment to air ﬂow in a PC case, cannot
be installed wrongly due to their diﬀerent connectors on each end, which
furthermore cannot be plugged in the wrong way round. In addition, the il-
logical separation between 2.5-inch and 3.5-inch diskdrives, which required
diﬀerent connectors for P-ATA, goes away.
1 Anyone remember the IBM PC/AT?
6.2 Bus Systems for Mass Storage 79

B Interestingly, serial ATA allows considerably faster data transfers than tra-
ditional ATA, even though with the former all bits are transferred in “single
ﬁle” rather than 16 at a go in parallel. This is due to the electrical proper-
ties of the interface, which uses diﬀerential transmission and a signalling
voltage of only 0.5 V instead of 5 V. (This is why cables may be longer, too—
1 m instead of formerly 45 cm.) Current SATA interfaces can theoretically
transfer up to 16 GiBit/s (SATA 3.2) which due to encoding and other im-
pediments comes out as approximately 2 MiB/s—rather more than single
disk drives can keep up with at sustained rates, but useful for RAID sys-
tems that access multiple disk drives at the same time, and for fast SSDs. It
is unlikely that SATA speeds will evolve further, since the trend with SSDs
is towards connecting them directly via PCIe2 .

B Besides the higher speed and more convenient cabling, SATA also oﬀers
the advantage of “hot-swapping”: It is possible to disconnect a SATA disk
drive and connect another one in its place, without having to shut down the
computer. This of course presupposes that the computer can do without
the data on the drive in question—typically because it is part of a RAID-1
or RAID-5, where the the data on the new drive can be reconstructed based
on other drives in the system. With traditional ATA, this was impossible (or
only possible by jumping through hoops).

B External SATA (“eSATA”) is a derivative of SATA for use with external eSATA
drives. It has diﬀerent connectors and electrical speciﬁcations, which are
much more robust mechanically and better suited for hot-swapping. In
the meantime, it has been almost completely ousted from the market by
USB 3.𝑥, but can still be found in older hardware.

SCSI and SAS The “Small Computer System Interface” or SCSI (customary pro-
nunciation: “SCUZ-zy”) has served for more than 25 years to connect hard disks,
tape drives and other mass storage devices, but also peripherals such as scanners,
to “small” computers3 . SCSI buses and connectors exist in a confusing variety,
beginning with the “traditional” 8-bit bus and ranging from the fast 16-bit vari-
eties to new, even faster serial implementations (see below). They also diﬀer in
the maximum number of devices per bus, and in physical parameters such as the
maximum cable length and the allowable distances between devices on the cable.
Nicely enough, most of the variants are compatible or can be made compatible
(possibly with loss of eﬃciency!). Varieties such as FireWire (IEEE-1394) or Fiber-
Channel are treated like SCSI by Linux.

B Nowadays, most work goes into the serial SCSI implementations, most no-
tably “Serial Attached SCSI” (SAS). As with SATA, data transfer is poten-
tially faster (at the moment, SAS is slightly slower than the fastest parallel
SCSI version, Ultra-640 SCSI) and electrically much less intricate. In partic-
ular, the fast parallel SCSI versions are plagued by clocking problems that
derive from the electrical properties of the cables and termination, and that
do not exist with SAS (where the pesky termination is no longer necessary
at all).

B SAS and SATA are fairly closely related; the most notable diﬀerences are that
SAS allows things like accessing a drive via several cable paths for redun-
dancy (“multipath I/O”; SATA requires jumping through hoops for this),
supports more extensive diagnosis and logging functions, and is based on
a higher signalling voltage, which allows for longer cables (up to 8 m) and
physically larger servers.
2 SATA in a strict sense allows speeds up to 6 GiBit/s; the higher speed of SATA 3.2 is already

achieved by means of PCIe. This “SATA Express” speciﬁcation deﬁnes an interface that can carry
SATA signals as well as PCIe, such that compatible devices can be connected not only to SATA Express
controllers, but also to older hosts which support “only” SATA with up to 6 GiBit/s.
3 Nobody has ever deﬁned the meaning of “small” in this context, but it must be something like

“can be bodily lifted by at most two people”
80 6 Hard Disks (and Other Secondary Storage)

Table 6.1: Diﬀerent SCSI variants

Name Width Transfer rate Devices Explanation
SCSI-1 8 bit ≤ 5 MiB/s 8 “Ancestor”
SCSI-2 »Fast« 8 bit 10 MiB/s 8
SCSI-2 »Wide« 16 bit 20 MiB/s 16
SCSI-3 »Ultra« 8 bit 20 MiB/s 8
SCSI-3 »Ultrawide« 16 bit 40 MiB/s 16
Ultra2 SCSI 16 bit 80 MiB/s 16 LVD busa
Ultra-160 SCSIb 16 bit 160 MiB/s 16 LVD bus
Ultra-320 SCSIc 16 Bit 320 MiB/s 16 LVD bus
Ultra-640 SCSI 16 Bit 640 MiB/s 16 LVD bus

B SATA and SAS are compatible to an extent where you can use SATA disk
drives on a SAS backplane (but not vice-versa).

Vorkommen “Pure-bred” SCSI, as far as PCs are concerned, is found mostly in servers; work
stations and “consumer PCs” tend to use IDE or SATA for mass storage and USB
(qv. section 5.2.1) for other devices. Devices based on IDE and USB are much
cheaper to manufacture than SCSI-based devices—IDE disks, for example, cost
about a third or a fourth of the price of comparatively large SCSI disks.

B We do need to mention that SCSI disks are usually designed especially for
use in servers, and are therefore optimised for high speed and longevity.
SATA disks for workplace PCs do not come with the same warranties, are
not supposed to rival a jet ﬁghter for noise, and should support fairly fre-
quent starting and stopping.

As a Linux administrator, you should know about SCSI even if you do not run
any SCSI-based systems, since from the point of view of the Linux kernel, in ad-
dition to SATA many USB or FireWire devices are accessed like SCSI devices and
use the same infrastrucure.

SCSI ID B Every device on a SCSI bus requires a unique “SCSI ID”. This number
between 0 and 7 (15 on the wide buses) is used to address the device.
Most “true” SCSI devices sport jumpers or a switch to select it; with Fibre-
Channel, USB, or SATA devices that are accessed via the SCSI infrastructure,
the system arranges for suitable unique SCSI IDs to be assigned.

host adapter B To use SCSI, a PC needs at least one host adapter (or “host”). Motherboard-
SCSI BIOS based and better expansion card host adapters contain a SCSI BIOS which
lets the system boot from a SCSI device. You can also use this to check which
SCSI IDs are available and which are used, and which SCSI device, if any,
should be used for booting.

B The host adapter counts as a device on the SCSI bus—apart from itself you
can connect 7 (or 15) other devices.

boot order B If your BIOS can boot from SCSI devices, you can also select in the boot order
whether the ATA disk C: should be preferred to any (potentially) bootable
SCSI devices.

B Most important for the correct function of a parallel SCSI system is appro-
termination priate termination of the SCSI bus. This can either be ensured via a special
plug (“terminator”) or switched on or oﬀ on individual devices. Erroneous
termination is the possible origin of all sorts of SCSI problems. If you do
experience diﬃculties with SCSI, always check ﬁrst that termination is in
order. SAS does not require termination.
6.3 Partitioning 81

USB With the new fast USB variants (Section 5.2.1), few if any compromises will
be needed when connecting mass storage devices—reading and writing speeds
are bounded by the storage device, not (as with USB 1.1 and USB 2.0) by the bus.
Linux manages USB-based storage devices exactly like SCSI devices.

Exercises
C 6.1 [1] How many hard disks or SSDs does your computer contain? What
is their capacity? How are they connected to the computer (SATA, …)?

6.3 Partitioning
6.3.1 Fundamentals
Mass storage devices such as hard disks or SSDs are commonly “partitioned”, i. e.,
subdivided into several logical storage devices that the operating system can then
access independently. This does not only make it easier to use data structures that
are appropriate to the intended use—sometimes partitioning is the only way to
make a very large storage medium fully accessible, if limits within the operating
system preclude the use of the medium “as a whole” (even though this sort of
problem tends to be rare today).
Advantages of partitioning include the following:
• Logically separate parts of the system may be separated. For example, you
could put your users’ data on a diﬀerent partition from that used by the op-
erating system itself. This makes it possible to reinstall the operating system
from scratch without endangering your users’ data. Given the often rudi-
mentary “upgrade” functionality of even current distributions this is very
important. Furthermore, if inconsistencies occur in a ﬁle system then only
one partition may be impacted at ﬁrst.

• The structure of the ﬁle system may be adapted to the data to be stored.
Most ﬁle systems keep track of data by means of ﬁxed-size “blocks”, where
every ﬁle, no matter how small, occupies at least a single block. With a 4 KiB
block size this implies that a 500-byte ﬁle only occupies 1/8 of its block—the
rest goes to waste. If you know that a directory will contain mostly small
ﬁles (cue: mail server), it may make sense to put this directory on a parti-
tion using smaller blocks (1 or 2 KiB). This can reduce waste considerably.
Some database servers, on the other hand, like to work on “raw” partitions
(without any ﬁle system) that they manage themselves. An operating sys-
tem must make that possible, too.
• “Runaway” processes or incautious users can use up all the space available
on a ﬁle system. At least on important server systems it makes sense to
allow user data (including print jobs, unread e-mail, etc.) only on partitions
that may get ﬁlled up without getting the system itself in trouble, e.g., by
making it impossible to append information to important log ﬁles.
There are currently two competing methods to partition hard disks for PCs.
The traditional method goes back to the 1980s when the ﬁrst hard disks (with
awesome sizes like 5 or 10 MB) appeared. Recently a more modern method was
introduced; this does away with various limitations of the traditional approach,
but in some cases requires special tools.

B Hard disks are virtually always partitioned, even though at times only one
partition will be created. With USB thumb drives, one sometimes eschews
partitioning altogether.
82 6 Hard Disks (and Other Secondary Storage)

Table 6.2: Partition types for Linux (hexadecimal)

Type Description
81 Linux data
82 Linux swap space
86 RAID super block (old style)
8E Linux LVM
E8 LUKS (encrypted partition)
EE “Protective partition” for GPT-partitioned disk
FD RAID super block with autodetection
FE Linux LVM (old style)

6.3.2 The Traditional Method (MBR)
The traditional method stores partitioning information inside the “master boot
record” (MBR), the ﬁrst sector (number 0) of a hard disk. (Traditionally, PC hard
disk sectors are 512 bytes long, but see below.) The space there—64 bytes starting
primary partitions at oﬀset 446—is suﬃcient for four primary partitions. If you want to create more
extended partition than four partitions, you must use one of these primary partitions as an extended
logical partitions partition. An extended partition may contain further logical partitions.

B The details about logical partitions are not stored inside the MBR, but at the
start of the partition (extended or logical) in question, i. e., they are scattered
around the hard disk.

Partition entries today usually store the starting sector number of the partition on
the disk as well as the length of the partition in question in sectors4 . Since these
values are 32-bit numbers, given the common 512-byte sectors this results in a
maximum partition size of 2 TiB.

B There are hard disks available today which are larger than 2 TiB. Such disks
cannot be made fully accessible using MBR partitioning. One common ruse
consists in using disks whose sectors are 4096 bytes long instead of 512. This
will let you have 16-TiB disks even with MBR, but not every operating sys-
tem supports such “4Kn” drives (Linux from kernel 2.6.31, Windows from
8.1 or Server 2012).

B 4-KiB sectors are useful on hard disks even without considering partitions.
The larger sectors are more eﬃcient for storing larger ﬁles and allow better
error correction. Therefore the market oﬀers “512e” disks which use 4-KiB
sectors internally but pretend to the outside that they really have 512-byte
sectors. This means that if a single 512-byte sector needs to be rewritten, the
adjoining 7 sectors must be read and also rewritten (a certain, but usually
bearable, loss of eﬃciency, since data is most often written in larger chunks).
When partitioning, you will have to pay attention that the 4-KiB blocks that
Linux uses internally for hard disk access coincide with the disk’s internal
4-KiB sectors—if that is not the case, then to write one 4-KiB Linux block two
4-KiB disk sectors might have to be read and rewritten, and that would not
be good. (Fortunately, the partitioning tools help you watch out for this.)

Besides the starting address and length of the (primary) partitions, the parti-
partition type tion table contains a partition type which loosely describe the type of data man-
agement structure that might appear on the partition. A selection of Linux parti-
tion types appears in table 6.2.
4 In former times, partitions used to be described in terms of the cylinder, head, and sector addresses

of the sectors in question, but this has been deprecated for a very long time.
6.3 Partitioning 83

6.3.3 The Modern Method (GPT)
In the late 1990s, Intel developed a new partitioning method that should do away
with the limitations of the MBR approach, namely “GUID Partition Table” or GPT.

B GPT was developed hand-in-hand with UEFI and is part of the UEFI spec-
iﬁcation today. You can, however, use a BIOS-based Linux system to access
GPT-partitioned disks and vice-versa.

B GPT uses 64-bit sector addresses and thus allows a maximum disk size of
8 ZiB—zebibyte, in case you haven’t run into that preﬁx. 1 ZiB are 270 bytes,
or, roughly speaking, about one million million tebibytes. This should last
even the NSA for a while. (Disk manufactures, who prefer to talk powers of
ten rather than powers of two, will naturally sell you an 8-ZiB disk as a 9.4
zettabyte disk.)
With GPT, the ﬁrst sector of the disk remains reserved as a “protective MBR”
which designates the whole disk as partitioned from a MBR point of view. This
avoids problems if a GPT-partitioned disk is connected to a computer that can’t
handle GPT.
The second sector (address 1) contains the “GPT header” which stores man-
agement information for the whole disk. Partitioning information is usually con-
tained in the third and subsequent sectors.

B The GPT header points to the partitioning information, and therefore they
could be stored anywhere on the disk. It is, however, reasonable to place
them immediately after the GPT header. The UEFI header stipulates a min-
imum of 16 KiB for partitioning information (regardless of the disk’s sector
size).

B On a disk with 512-byte sectors, with a 16 KiB space for partitioning infor-
mation the ﬁrst otherwise usable sector on the disk is the one at address 34.
You should, however, avoid placing the disk’s ﬁrst partition at that address
because that will get you in trouble with 512e disks. The next correctly-
aligned sector is the one at address 40.

B For safety reasons, GPT replicates the partitioning information at the end of
the disk.
Traditionally, partition boundaries are placed at the start of a new “track” on
the disk. Tracks, of course, are a relic from the hard disk paleolithic, since con-
temporary disks are addressed linearly (in other words, the sectors are numbered
consecutively from the start of the disk to the end)—but the idea of describing a
disk by means of a combination of a number of read/write heads, a number of
“cylinders”, and a number of sectors per “track” (a track is the concentric circle a
single head describes on a given cylinder) has continued to be used for a remark-
ably long time. Since the maximum number of sectors per track is 63, this means
that the ﬁrst partition would start at block 63, and that is, of course, disastrous for
512e disks.

B Since Windows Vista it is common to have the ﬁrst partition start 1 MiB after
the start of the disk (with 512-byte sectors, at sector 2048). This isn’t a bad
idea for Linux, either, since the ample free space between the partition table
and the ﬁrst partition can be used to store the GRUB boot loader. (The space
between the MBR and sector 63 was quite suﬃcient earlier, too.)
Partition table entries are at least 128 bytes long and, apart from 16-byte GUIDs
for the partition type and the partition itself and 8 bytes each for a starting and
ending block number, contains 8 bytes for “attributes” and 72 bytes for a partition
name. It is debatable whether 16-byte GUIDs are required for partition types, but
on the one hand the scheme is called “GUID partition table” after all, and on the
other hand this ensures that we won’t run out of partition types anytime soon. A
selection is displayed in table 6.3.
84 6 Hard Disks (and Other Secondary Storage)

GUID Description
00000000-0000-0000-0000-000000000000 Unused entry
C12A7328-F81F-11D2-BA4B-00A0C93EC93B EFI system partition (ESP)
21686148-6449-6E6F-744E-656564454649 BIOS boot partition
0FC63DAF-8483-4772-8E79-3D69D8477DE4 Linux ﬁle system
A19D880F-05FC-4D3B-A006-743F0F84911E Linux RAID partition
0657FD6D-A4AB-43C4-84E5-0933C84B4F4F Linux swap space
E6D6D379-F507-44C2-A23C-238F2A3DF928 Linux LVM
933AC7E1-2EB4-4F13-B844-0E14E2AEF915 /home partition
3B8F8425-20E0-4F3B-907F-1A25A76F98E8 /srv partition
7FFEC5C9-2D00-49B7-8941-3EA10A5586B7 dm-crypt partition
CA7D7CCB-63ED-4C53-861C-1742536059CC LUKS partition

Table 6.3: Partition type GUIDs for GPT (excerpt)

B Linux can use GPT-partitioned media. This needs the “EFI GUID Partition
support” option enabled in the kernel, but with current distributions this
is the case. Whether the installation procedure allows you to create GPT-
partitioned disks is a diﬀerent question, just like the question of whether
the boot loader will be able to deal with them. But that is neither here nor
there.

6.4 Linux and Mass Storage
If a mass storage device is connected to a Linux computer, the Linux kernel tries
to locate any partitions. It then creates block-oriented device ﬁles in /dev for the
device itself and its partitions. You can subsequently access the partitions’ device
ﬁles and make the directory hierarchies there available as part of the computer’s
ﬁle system.

B A new mass storage device may have no partitions at all. In this case you
can use suitable tools to create partitions. This will be explained later in this
chapter. The next step after partitioning consists of generating ﬁle systems
on the partitions. This is explained in detail in chapter 7.

The device names for mass storage are usually /dev/sda , /dev/sdb , …, in the order
the devices are recognised. Partitions are numbered, the /dev/sda device therefore
contains partitions like /dev7sda1 , /dev/sda2 , … A maximum of 15 partitions per de-
vice is possible. If /dev/sda is partitioned according to the MBR scheme, /dev/sda1
to /dev/sda4 correspond to the primary partitions (possibly including an extended
partition), while any logical partitions are numbered starting with /dev/sda5 (re-
gardless of whether there are four primary partitions on the disk or fewer).

B The “s ” in /dev/sda derives from “SCSI”. Today, almost all mass storage de-
vices in Linux are managed as SCSI devices.

B For P-ATA disks there is another, more speciﬁc mechanism. This accesses
the IDE controllers inside the computer directly—the two drives connected
to the ﬁrst controller are called /dev/hda and /dev/hdb , the ones connected to
the second /dev/hdc and /dev/hdd . (These names are used independently of
whether the drives actually exist or not—if you have a single hard disk and
a CD-ROM drive on your system, you do well to connect the one to one
controller and the other to the other so they will not be in each other’s way.
Therefore you will have /dev/hda for the disk and /dev/hdc for the CD-ROM
drive.) Partitions on P-ATA disks are, again, called /dev/hda1 , /dev/hda2 and
so on. In this scheme, 63 (!) partitions are allowed.
6.4 Linux and Mass Storage 85

B If you still use a computer with P-ATA disks, you will notice that in the
vast majority of cases the SCSI infrastructure is used for those, too (note the
/dev/sda style device names). This is useful for convenience and consistency.
Some very few P-ATA controllers are not supported by the SCSI infrastruc-
ture, and must use the old P-ATA speciﬁc infrastructure.

B The migration of an existing Linux system from “traditional” P-ATA drivers
to the SCSI infrastructure should be well-considered and involve changing
the conﬁguration in /etc/fstab such that ﬁle systems are not mounted via
their device ﬁles but via volume labels or UUIDs that are independent of
the partitions’ device ﬁle names. (See section 7.2.3.)

The Linux kernel’s mass storage subsystem uses a three-tier architecture. At architecture
the bottom there are drivers for the individual SCSI host adapters, SATA or USB
controllers and so on, then there is a generic “middle layer”, on top of which there
are drivers for the various devices (disks, tape drives, …) that you might encounter
on a SCSI bus. This includes a “generic” driver which is used to access devices
without a specialised driver such as scanners or CD-ROM recorders. (If you can
still ﬁnd any of those anywhere.)

B Every SCSI host adapter supports one or more buses (“channels”). Up to
7 (or 15) other devices can be connected to each bus, and every device can
support several “local unit numbers” (or LUNs), such as the individual CDs LUNs
in a CD-ROM changer (rarely used). Every SCSI device in the system can
thus be describe by a quadrupel (⟨host⟩, ⟨channel⟩, ⟨ID⟩, ⟨LUN⟩). Usually
(⟨host⟩, ⟨channel⟩, ⟨ID⟩) are suﬃcient.

B In former times you could ﬁnd information on SCSI devices within the /proc/
scsi/scsi directory. This is no longer available on current systems unless the
kernel was compiled using “Legacy /proc/scsi support”.

B Nowadays, information about “SCSI controllers” is available in /sys/class/
scsi_host (one directory per controller). This is unfortunately not quite as
accessible as it used to be. You can still snoop around:

# cd /sys/class/scsi_host/host0/device
# ls
power scsi_host subsystem target0:0:0 uevent
# cd target0:0:0; ls
0:0:0:0 power subsystem uevent
# ls 0:0:0:0/block
sda

A peek into /sys/bus/scsi/devices will also be instructive:

# ls /sys/bus/scsi/devices
0:0:0:0 10:0:0:0 host1 host2 host4 target0:0:0 target10:0:0
1:0:0:0 host0 host10 host3 host5 target1:0:0

Device names such as /dev/sda , /dev/sdb , etc. have the disadvantage of not being
very illuminating. In addition, they are assigned to devices in the order of their
appearance. So if today you connect ﬁrst your MP3 player and then your digital
camera, they might be assigned the device ﬁles /dev/sdb and /dev/sdc ; if tomorrow
you start with the digital camera and continue with the MP3 player, the names
might be the other way round. This is of course a nuisance. Fortunately, udev
assigns some symbolic names on top of the traditional device names. These can
be found in /dev/block :
86 6 Hard Disks (and Other Secondary Storage)

# ls -l /dev/block/8:0
lrwxrwxrwx 1 root root 6 Jul 12 14:02 /dev/block/8:0 -> ../sda
# ls -l /dev/block/8:1
lrwxrwxrwx 1 root root 6 Jul 12 14:02 /dev/block/8:1 -> ../sda1
# ls -l /dev/disk/by-id/ata-ST9320423AS_5VH5TBTC
lrwxrwxrwx 1 root root 6 Jul 12 14:02 /dev/disk/by-id/
ata-ST9320423AS_5VH5TBTC -> ../../sda
# ls -l /dev/disk/by-id/ata-ST9320423AS_5VH5TBTC-part1
lrwxrwxrwx 1 root root 6 Jul 12 14:02 /dev/disk/by-id/
ata-ST9320423AS_5VH5TBTC-part1 -> ../../sda1
# ls -l /dev/disk/by-path/pci-0000:00:1d.0-usb-
0:1.4:1.0-scsi-0:0:0:0
lrwxrwxrwx 1 root root 6 Jul 12 14:02 /dev/disk/by-path/
pci-0000:00:1d.0-usb-0:1.4:1.0-scsi-0:0:0:0 -> ../../sdb
# ls -l /dev/disk/by-uuid/c59fbbac-9838-4f3c-830d-b47498d1cd77
lrwxrwxrwx 1 root root 10 Jul 12 14:02 /dev/disk/by-uuid/
c59fbbac-9838-4f3c-830d-b47498d1cd77 -> ../../sda1
# ls -l /dev/disk/by-label/root
lrwxrwxrwx 1 root root 10 Jul 12 14:02 /dev/disk/by-label/root
-> ../../sda1

These device names are derived from data such as the (unique) serial number of
the disk drive, its position on the PCIe bus, or the UUID or name of the ﬁle system,
and are independent of the name of the actual device ﬁle.

Exercises
C 6.2 [!2] On your ssytem there are two SATA hard disks. The ﬁrst disk has
two primary and two logical partitions. The second disk has one primary
and three logical partitions. Which are the device names for these partitions
on Linux?

C 6.3 [!1] Examine the /dev directory on your system. Which storage media are
available and what are the corresponding device ﬁles called? (Also check
/dev/block and /dev/disk .)

C 6.4 [1] Plug a USB thumb drive into your computer. Check whether new
device ﬁles have been added to /dev . If so, which ones?

6.5 Partitioning Disks
6.5.1 Fundamentals
Before you partition the (possibly sole) disk on a Linux system, you should brieﬂy
consider what a suitable partitioning scheme might look like and how big the
partitions ought to be. Changes after the fact are tedious and inconvenient at best
and may at worst necessitate a complete re-install of the system (which would be
exceedingly tedious and inconvenient). (See section 6.7 for an alternative, much
less painful approach.)
Here are a few basic suggestions for partitioning:
• Apart from the partition with the root directory / , you should provide at
least one spearate partition for the ﬁle system containing the /home directory.
This lets you cleanly separate the operating system from your own data, and
facilitates distribution upgrades or even switching from one Linux distribu-
tion to a completely diﬀerent one.
6.5 Partitioning Disks 87

B If you follow this approach, you should probably also use symbolic
links to move the /usr/local and /opt directories to (for example) /home/
usr- local and /home/opt . This way, these directories, which also contain
data provided by you, are on “your” partition and can more easily be
included in regular backups.

• It is absolutely possible to ﬁt a basic Linux system into a 2 GiB partition, but,
considering today’s (low) costs per gigabyte for hard disk storage, there is
little point in scrimping and saving. With something like 30 GiB, you’re sure
to be on the safe side and will have enough room for log ﬁles, downloaded
distribution packages during a larger update, and so on.
• On server systems, it may make sense to provide separate partitions for /tmp ,
/var , and possibly /srv . The general idea is that arbitrary users can put data
into these directories (besides outright ﬁles, this could include unread or
unsent e-mail, queued print jobs, and so on). If these directories are on
separate partitions, users cannot ﬁll up the system in general and thereby
create problems.
• You should provide swap space of approximately the same size as the com-
puter’s RAM, up to a maximum of 8 GiB or thereabouts. Much more is
pointless, but on workstations and mobile computers you may want to avail
yourself of the possibility to “suspend” your computer instead of shutting
it down, in order to speed up a restart and end up exactly where you were
before—and the infrastructures enabling this like to use the swap space to
save the RAM content.

B There used to be a rule of thumb saying that the swap space should be
about twice or three times the available RAM size. This rule of thumb
comes from traditional Unix systems, where RAM works as “cache”
for the swap space. Linux doesn’t work that way, instead RAM and
swap space are added—on a computer with 4 GiB of RAM and 2 GiB
of swap space, you get to run processes to the tune of 6 GiB or so. With
8 GiB of RAM, providing 16 to 24 GiB of swap space would be absurd.

B You should dimension the RAM of a computer (especially a server) to
be big enough that practically no swap space is necessary during nor-
mal operations.; on an 8-GiB server, you won’t usually need 16 GiB of
swap space, but a gigabyte or two to be on the safe side will certainly
not hurt (especially considering today’s prices for disk storage). That
way, if RAM gets tight, the computer will slow down before processes
crash outright because they cannot get memory from the operating sys-
tem.

• If you have several (physical) hard disks, it can be useful to spread the sys-
tem across the available disks in order to increase the access speeed to indi-
vidual components.

B Traditionally, one would place the root ﬁle system (/ with the essential
subdirectories /bin , /lib , /etc , and so on) on one disk and the /usr direc-
tory with its subdirectories on a separate ﬁle system on another disk.
However, the trend on Linux is decisively away from the (artiﬁcial)
separation between /bin and /usr/bin or /lib and /usr/lib and towards
a root ﬁle system which is created as a RAM disk on boot. Whether the
traditional separation of / and /usr will gain us a lot in the future is up
for debate.

B What will certainly pay oﬀ is to spread swap space across several disks.
Linux always uses the least-used disk for swapping.
88 6 Hard Disks (and Other Secondary Storage)

Provided that there is enough empty space on the medium, new partitions
can be created and included (even while the system is running). This procedure
consists of the following steps:
1. Backup the current boot sectors and data on the hard disk in question
2. Partition the disk using fdisk (or a similar program)
3. Possibly create ﬁle systems on the new partitions (“formatting”)
4. Making the new ﬁle systems accessible using mount or /etc/fstab

B Items 3 and 4 on this list will be considered in more detail in chapter 7.
Data and boot-sector contents can be saved using the dd program (among others).

# dd if=/dev/sda of=/dev/st0

will, for example, save all of the sda hard disk to magnetic tape.
You should be aware that the partitioning of a storage medium has nothing to
do with the data stored on it. The partition table simply speciﬁes where on the
disk the Linux kernel can ﬁnd the partitions and hence the ﬁle structures. Once the
Linux kernel has located a partition, the content of the partition table is irrelevant
until it searches again, possibly during the next system boot. This gives you—
if you are courageous (or foolhardy)—far-reaching opportunities to tweak your
system while it is running. For example, you can by all means enlarge partitions
(if after the end of the partition there is unused space or a partition whose contents
you can do without) or make them smaller (and possibly place another partition
in the space thus freed up). As long as you exercise appropriate care this will be
reasonably safe.

B This should of course in no way discourage you from making appropriate
backup copies before doing this kind of open-heart surgery.

B In addition, the ﬁle systems on the disks must play along with such shenani-
gans (many Linux ﬁle systems can be enlarged without loss of data, and
some of them even be made smaller), or you must be prepared to move the
data out of the way, generate a new ﬁle system, and then fetch the data back.

6.5.2 Partitioning Disks Using fdisk
fdisk fdisk is an interactive program for manipulating disk partition tables. It can also
create “foreign” partition types such as DOS partitions. Drives are addressed us-
ing the corresponding device ﬁles (such as /dev/sda for the ﬁrst disk).

B fdisk conﬁnes itself to entering a partition into the partition table and setting
the correct partition type. If you create a DOS or NTFS partition using fdisk ,
this means just that the partition exists in the partition table, not that you can
boot DOS or Windows NT now and write ﬁles to that partition. Before doing
that, you must create a ﬁle system, i. e., write the appropriate management
data structures to the disk. Using Linux-based tools you can do this for
many but not all non-Linux ﬁle systems.

After invoking “fdisk ⟨device⟩”, the program comes back with a succinct prompt
of
# fdisk /dev/sdb Neue (leere) Platte
Welcome to fdisk (util-linux 2.25.2).
Changes will remain in memory only, until you decide to write them.
Be careful before using the write command.

Device does not contain a recognized partition table.
6.5 Partitioning Disks 89

Created a new DOS disklabel with disk identifier 0x68d97339.

Command (m for help): _

The m command will display a list of the available commands.

B fdisk lets you partition hard disks according to the MBR or GPT schemes.
It recognises an existing partition table and adjusts itself accordingly. On
an empty (unpartitioned) disk fdisk will by default create an MBR partition
table, but you can change this afterwards (we’ll show you how in a little
while).

You can create a new partition using the “n ” command:

Command (m for help): n
Partition type
p primary (0 primary, 0 extended, 4 free)
e extended (container for logical partitions)
Select (default p): p
Partition number (1-4, default 1): 1
First sector (2048-2097151, default 2048): ↩
Last sector, +sectors or +sizeK,M,G,T,P (2048-2097151,
default 2097151): +500M

Created a new partition 1 of type 'Linux' and of size 500 MiB.

Command (m for help): _

The p command displays the current partition table. This could look like this:

Command (m for help): p
Disk /dev/sdb: 1 GiB, 1073741824 bytes, 2097152 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disklabel type: dos
Disk identifier: 0x68d97339

Device Boot Start End Sectors Size Id Type
/dev/sdb1 2048 1026047 1024000 500M 83 Linux

B You can change the partition type using the t command. You must select the
desired partition and can then enter the code (as a hexadecimal number). L
displays the whole list.

You can delete a partition you no longer want by means of the d command. When
you’re done, you can write the partition table to disk and quit the program using
w . With q , you can quit the program without rewriting the partition table.

B After storing the partition table, fdisk tries to get the Linux kernel to reread
the new partition table; this works well with new or unused disks, but fails
if a partition on the disk is currently in use (as a mounted ﬁle system, active
swap space, …). This lets you repartition the disk with the / ﬁle system only
with a system reboot. One of the rare occasions when a Linux system needs
to be rebooted …

Like all Linux commands, fdisk supports a number of command-line options. Options
The most important of those include:
90 6 Hard Disks (and Other Secondary Storage)

-l displays the partition table of the selected disk and then terminates the pro-
gram.
-u (“units”) lets you select the units of measure used when displaying partition
tables. The default is “sectors”; if you specify “-u=cylinders ”, cylinders will
be used instead (but there is no good reason for that today).

B If you use fdisk in MBR mode, it tries to observe the usual conventions and
arrange the partitions such that they work properly on 4Kn and 512e hard
disks. You should follow the program’s suggestions wherever possible, and
not deviate from them unless there are very compelling reasons.

If you partition a hard disk according to the GPT standard and there is no GPT-
style partition table on the disk yet, you can generate one using the g command
(Warning: A possibly existing MBR partition table will be overwritten in the pro-
cess):

Command (m for help): g
Created a new GPT disklabel (GUID: C2A556FD-7C39-474A-B383-963E09AA7269)

(The GUID shown here applies to the disk as a whole.) Afterwards you can use the
n command to create partitions in the usual way, even if the dialog looks slightly
diﬀerent:
Command (m for help): n
Partition number (1-128, default 1): 1
First sector (2048-2097118, default 2048): ↩
Last sector, +sectors or +sizeK,M,G,T,P (2048-2097118, default
2097118): +32M

Created a new partition 1 of type 'Linux filesystem' and of size 32 MiB.

The partition type selection is diﬀerent, too, because it is about GUIDs rather than
two-digit hexadecimal numbers:

Command (m for help): t
Selected partition 1
Partition type (type L to list all types): L
1 EFI System C12A7328-F81F-11D2-BA4B-00A0C93EC93B

14 Linux swap 0657FD6D-A4AB-43C4-84E5-0933C84B4F4F
15 Linux filesystem 0FC63DAF-8483-4772-8E79-3D69D8477DE4
16 Linux server data 3B8F8425-20E0-4F3B-907F-1A25A76F98E8
17 Linux root (x86) 44479540-F297-41B2-9AF7-D131D5F0458A
18 Linux root (x86-64) 4F68BCE3-E8CD-4DB1-96E7-FBCAF984B709

Partition type (type L to list all types): _

Exercises
C 6.5 [!2] Create an empty 1 GiB ﬁle using the
# dd if=/dev/zero of=$HOME/test.img bs=1M count=1024

command. Use fdisk to “partition” the ﬁle according to the MBR scheme:
Create two Linux partitions of 256 MiB and 512 MiB, respectively, and create
a swap space partitions using the balance.

C 6.6 [!2] Repeat the following exercise, but create a GPT partition table in-
stead. Assume that the 512-MiB partition will contain a /home directory.
6.5 Partitioning Disks 91

6.5.3 Formatting Disks using GNU parted
Another popular program for partitioning storage media is the GNU project’s
parted . Featurewise, it is roughly comparable with fdisk , but it has a few useful
features.

B Unlike fdisk , parted does not come pre-installed with most distributions, but
can generally be added after the fact from the distribution’s servers.

Similar to fdisk , parted must be started with the name of the medium to be
partitioned as a parameter.

# parted /dev/sdb
GNU Parted 3.2
Using /dev/sdb
Welcome to GNU Parted! Type 'help' to view a list of commands.
(parted) _

You can create a new partition using mkpart . This works either interactively …

(parted) mkpart
Partition name? []? Test
File system type? [ext2]? ext4
Start? 211MB
End? 316MB

… or directly when the command is invoked:

(parted) mkpart primary ext4 211MB 316MB

B You can abbreviate the commands down to an unambiguous preﬁx. Hence,
mkp will work instead of mkpart (mk would collide with the mklabel command).

B The ﬁle system type will only be used to guess a partition type. You will
still need to manually create a ﬁle system on the partition later on.

You can examine the partition table using the print command:

(parted) p
Model: ATA VBOX HARDDISK (scsi)
Disk /dev/sdb: 1074MB
Sector size (logical/physical): 512B/512B
Partition Table: gpt
Disk Flags:

Number Start End Size File system Name Flags
1 1049kB 106MB 105MB
2 106MB 211MB 105MB
3 211MB 316MB 105MB ext4 primary

(parted) _

(This also shows you where the magic numbers “211MB ” and “316MB ” came from,
earlier on.)

B print has a few interesting subcommands: “print devices ” lists all available
block devices, “print free ” displays free (unallocated) space, and “print all ”
outputs the partition tables of all block devices.

You can get rid of unwanted partitions using rm . Use name to give a name to a
partition (only for GPT). The quit command stops the program.
92 6 Hard Disks (and Other Secondary Storage)

A Important: While fdisk updates the partition table on the disk only once you
leave the program, parted does it on an ongoing basis. This means that the
addition or removal of a partition takes eﬀect on the disk immediately.

If you use parted on a new (unpartitioned) disk, you must ﬁrst create a partition
table.
(parted) mklabel gpt

creates a GPT-style partition table, and

(parted) mklabel msdos

one according to the MBR standard. There is no default value; without a partition
table, parted will refuse to execute the mkpart command.
If you inadvertently delete a partition that you would rather have kept, parted
can help you ﬁnd it again. You will just need to tell it approximately where on the
disk the partition used to be:

(parted) rm 3 Oops.
(parted) rescue 200MB 350MB
Information: A ext4 primary partition was found at 211MB -> 316MB.
Do you want to add it to the partition table?

Yes/No/Cancel? yes

For this to work, there must be a ﬁle system on the partition, because parted looks
for data blocks that appear to be the start of a ﬁle system.
In addition to the interactive mode, parted allows you to pass commands im-
mediately on the Linux command line, like this:

# parted /dev/sdb mkpart primary ext4 316MB 421MB
Information: You may need to update /etc/fstab.

Exercises
C 6.7 [!2] Repeat exercise 6.5 using parted rather than fdisk , for the MBR as well
as the GPT scheme.

C 6.8 [2] (If you have worked through chapter 7 already.) Generate Linux ﬁle
systems on the partitions on the “disk” from the earlier exercises. Remove
these partitions. Can you restore them using parted ’s rescue command?

6.5.4 gdisk
The gdisk program specialises in GPT-partitioned disks and can do a few useful
things the previously explained programs can’t. You may however have to install
it specially.
The elementary functions of gdisk correspond to those of fdisk and parted , and
we will not reiterate those (read the documentation and do a few experiments). A
few features of gdisk , however, are of independent interest:
• You can use gdisk to convert an MBR-partitioned medium to a GPT-partitioned
medium. This presupposes that there is enough space at the start and the
end of the medium for GPT partition tables. With media where, according
to current conventions, the ﬁrst partition starts at sector 2048, the former is
not a problem, but the latter may be. You will possibly have to ensure that
the last 33 sectors on the medium are not assigned to a partition.
6.6 Loop Devices and kpartx 93

For the conversion it is usually suﬃcient to start gdisk with the device ﬁle
name of the medium in question as a parameter. You will either receive
a warning that no GPT partition table was found and disk used the MPT
partition table instead (at this point you can quit the program using w and
you’re done), or that an intact MBR, but a damaged GPT partition table was
found (then you tell gdisk to follow the MBR, and can then quit the program
using w and you’re done).
• The other direction is also possible. To do this, you must use the r command
in gdisk to change to the menu for “recovery/transformation commands”,
and select the g command there (“convert GPT into MBR and exit”). After-
wards you can quit the program using w and convert the storage medium
this way.

Exercises
C 6.9 [!2] Repeat exercise 6.5 using gdisk rather than fdisk and generate a GPT
partition table.

C 6.10 [2] Create (e. g., using fdisk ) an MBR-partitioned “disk” and use gdisk
to convert it to GPT partitioning. Make sure that a correct “protective MBR”
was created.

6.5.5 More Partitioning Tools
Most distributions come with alternative ways of partitioning disks. Most of them distributions
oﬀer the cfdisk program as an alternative to fdisk . This is screen-oriented and thus
somewhat more convenient to use. Even easier to use are graphical programs,
such as SUSE’s YaST or “DiskDruid” on Red Hat distributions.

B Also worth mentioning is sfdisk , a completely non-interactive partitioning
program. sfdisk takes partitioning information from an input ﬁle and is
therefore useful for unattended partitioning, e. g., as part of an automated
installation procedure. Besides, you can use sfdisk to create a backup copy
of your partitioning information and either print it as a table or else store it
on a disk or CD as a ﬁle. If the worst happens, this copy can then be restored
using sfdisk .

B sfdisk only works for MBR-partitioned disks. There is a corresponding pro-
gram called sgdisk which does an equivalent job for GPT-partitioned disks.
However, sfdisk and sgdisk are not compatible—their option structures are
completely diﬀerent.

6.6 Loop Devices and kpartx
Linux has the useful property of being able to treat ﬁles like storage media. This
means that if you have a ﬁle you can partition it, generate ﬁle systems, and gener-
ally treat the “partitions” on that ﬁle as if they were partitions on a “real” hard
disk. In real life, this can be useful if you want to access CD-ROMs or DVDs
without having a suitable drive in your computer (it is also faster). For learn-
ing purposes, it means that you can perform various experiments without having
to obtain extra hard disks or mess with your computer.
A CD-ROM image can be created straightforwardly from an existing CD-ROM CD-ROM image
using dd :

# dd if=/dev/cdrom of=cdrom.iso bs=1M

You can subsequently make the image directly accessible:
94 6 Hard Disks (and Other Secondary Storage)

# mount -o loop,ro cdrom.iso /mnt

In this example, the content of the CD-ROM will appear within the /mnt directory.
You can also use the dd command to create an empty ﬁle:

# dd if=/dev/zero of=disk.img bs=1M count=1024

That ﬁle can then be “partitioned” using one of the common partitioning tools:

# fdisk disk.img

Before you can do anything with the result, you will have to ensure that there are
device ﬁles available for the partitions (unlike with “real” storage media, this is
not automatic for simulated storage media in ﬁles). To do so, you will ﬁrst need a
device ﬁle for the ﬁle as a whole. This—a so-called “loop device”—can be created
using the losetup command:

# losetup -f disk.img
# losetup -a
/dev/loop0: [2050]:93 (/tmp/disk.img)

losetup uses device ﬁle names of the form “/dev/loop 𝑛”. The “-f ” option makes the
program search for the ﬁrst free name. “losetup -a ” outputs a list of the currently
active loop devices.
Once you have assigned your disk image to a loop device, you can create device
ﬁles for its partitions. This is done using the kpartx command.

B You may have to install kpartx ﬁrst. On Debian and Ubuntu, the package is
called kpartx .
The command to create device ﬁles for the partitions on /dev/loop0 is

# kpartx -av /dev/loop0
add map loop0p1 (254:0): 0 20480 linear /dev/loop0 2048
add map loop0p2 (254:1): 0 102400 linear /dev/loop0 22528

(without the “-v ” command, kpartx keeps quiet). The device ﬁles then appear in
the /dev/mapper directory:

# ls /dev/mapper
control loop0p1 loop0p2

Now nothing prevents you from, e. g., creating ﬁle systems on these “partitions”
and mounting them into your computer’s directory structure. See chapter 7.
When you no longer need the device ﬁles for the partitions, you can remove
them again using the

# kpartx -dv /dev/loop0
del devmap : loop0p2
del devmap : loop0p1

command. An unused loop device can be released using

# losetup -d /dev/loop0

B The
# losetup -D

command releases all loop devices.
6.7 The Logical Volume Manager (LVM) 95

Exercises
C 6.11 [!2] Use the test “disk” from exercise 6.5. Assign it a loop device using
losetup and make its partitions accessible with kpartx . Convince yourself that
the correct device ﬁles show up in /dev/mapper . Then release the partitions
and the loop device again.

6.7 The Logical Volume Manager (LVM)
Partitioning a disk and creating ﬁle systems on it seems like a simple and obvious
thing to do. On the other hand, you are committing yourself: The partition scheme
of a hard disk can only be changed with great diﬃculty and, if the disk in question
contains the root ﬁle system, may even involve the use of a “rescue system”. In
addition, there is no compelling reason why you should be constrained in your
system architecture by trivialities such as the limited capacity of hard disks and
the fact that ﬁle system can be at most as large as the partitions they are sitting on.
One method to transcend these limitations is the use of the “Logical Volume
Manager” (LVM). LVM provides an abstraction layer between disks (and disk par-
titions) and ﬁle systems—instead of creating ﬁle systems directly on partitions,
you can contribute partitions (or whole disks) to a “pool” of disk space and then
allocate space from that pool to create ﬁle systems. Single ﬁle systems can freely
use space which is located on more than one physical disk.
In LVM terminology, disks and disk partitions are considered “physical vol-
umes” (PV) which are made part of a “volume group” (VG). There may be more
than one VG on the same computer. The space within a VG is available for cre-
ating “logical volumes” (LV), which can then hold arbitrary ﬁle systems or swap
space.

B When creating LVs, you can cause their storage space to be spread deviously
across several physical disks (“striping”) or multiple copies of their data to
be stored in several places within the VG at the same time (“mirroring”).
The former is supposed to decrease retrieval time (even if there is a danger
of losing data whenever any of the disks in the volume group fail), the latter
is supposed to reduce the risk of losing data (even if you are paying for this
with increased access times). In real life, you will probably prefer to rely on
(hardware or software) RAID instead of using LVM’s features.
One of the nicer properties of LVM is that LVs can be changed in size while the
system is running. If a ﬁle system is running out of space, you can ﬁrst enlarge
the underlying LV (as long as your VG has unused space available—otherwise you
would ﬁrst need to install another disk and add it to the VG). Afterwards you can
enlarge the ﬁle system on the LV in question.

B This presumes that the ﬁle system in question enables size changes after the
fact. With the popular ﬁle systems, e. g., ext3 or ext4 , this is the case. They
even allow their size to be increased while the ﬁle system is mounted. (You
will need to unmount the ﬁle system to reduce the size.)

B If you use a ﬁle system that does not let itself be enlarged, you will have
to bite the bullet, copy the data elsewhere, recreate the ﬁle system with the
new size, and copy the data back.
If a disk within your VG should start acting up, you can migrate the LVs from
that disk to another within the VG (if you still have or can make enough space).
After that, you can withdraw the ﬂaky disk from the VG, install a new disk, add
that to the VG and migrate the LVs back.

B You can do that, too, while the system is running and with your users none
the wiser—at least as long as you have invested enough loose change into
making your hard disks “hot-swappable”.
96 6 Hard Disks (and Other Secondary Storage)

“snapshots” Also nice are “snapshots”, which you can use for backup copies without hav-
ing to take your system oﬄine for hours (which would otherwise be necessary
to ensure that nothing changes while the backup is being performed). You can
“freeze” the current state of an LV on another (new) LV—which takes a couple of
seconds at most—and then make a copy of that new LV in your own time while
normal operations continue on the old LV.

B The “snapshot” LV only needs to be big enough to hold the amount of
changes to the original LV you expect while the backup is being made (plus
a sensible safety margin), since only the changes are being stored inside the
new LV. Hence, nobody prevents you from making a snapshot of your 10 TB
ﬁle system even if you don’t have another 10 TB of free disk space: If you
only expect 10 GB of data to be changed while you’re writing the copy to
tape, a snapshot LV of 20–30 GB should be fairly safe.

B As a matter of fact it is now possible to create writable snapshots. This is
useful, for example, if you are working with “virtual machines” that share
a basic OS installation but diﬀer in details. Writable snapshots make it pos-
sible to make the basic installation in one LV for all virtual machines and
then store the conﬁguration speciﬁc to each virtual machine in one LV with
a writable snapshot each. (You shouldn’t overstretch this approach, though;
if you change the LV with the basic installation the virtual machines won’t
notice.)

On Linux, LVM is a special application of the “device mapper”, a system com-
ponent enabling the ﬂexible use of block devices. The device mapper also pro-
vides other useful features such as encrypted disks or space-saving storage provi-
sioning for “virtual servers”. Unfortunately we do not have room in this training
manual to do LVM and the device mapper justice, and refer you to the manual,
Linux Storage and File Systems (STOR).

Commands in this Chapter
cfdisk Character-screen based disk partitioner cfdisk (8) 93
gdisk Partitioning tool for GPT disks gdisk (8) 92
kpartx Creates block device maps from partition tables kpartx (8) 94
losetup Creates and maintains loop devices losetup (8) 94
sfdisk Non-interactive hard disk partitioner sfdisk (8) 93
sgdisk Non-interactive hard disk partitioning tool for GPT disks sgdisk (8) 93
6.7 Bibliography 97

Summary
• Linux supports all notable types of mass storage device—magnetic hard
disks (SATA, P-ATA, SCSI, SAS, Fibre Channel, USB, …), SSDs, USB thumb
drives, SD cards, …
• Storage media such as hard disks may be partitioned. Partitions allow the
independent management of parts of a hard disk, e. g., with diﬀerent ﬁle
systems or operating systems.
• Linux can deal with storage media partitioned according to the MBR and
GPT schemes.
• Linux manages most storage media like SCSI devices. There is an older
infrastructure for P-ATA disks which is only rarely used.
• Linux oﬀers various tools for partitioning such as fdisk , parted , gdisk , cfdisk ,
or sfdisk . Various distributions also provide their own tools.
• Loop devices make block-oriented devices from ﬁles. Partitions on loop de-
vices can be made accessible using kpartx .
• The Logical Volume Manager (LVM) decouples physical storage space on
media from logical storage structures. It enables the ﬂexible management
of mass storage, e. g., to create ﬁle systems which are larger than a single
physical storage medium. Snapshots help create backup copies and provi-
sion storage space for virtual machines.

Bibliography
SCSI-2.4-HOWTO03 Douglas Gilbert. “The Linux 2.4 SCSI subsystem HOWTO”,
May 2003. http://www.tldp.org/HOWTO/SCSI- 2.4- HOWTO/
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

7
File Systems: Care and Feeding

Contents
7.1 Creating a Linux File System . . . . . . . . . . . . . . . . 100
7.1.1 Overview . . . . . . . . . . . . . . . . . . . . 100
7.1.2 The ext File Systems . . . . . . . . . . . . . . . . . 102
7.1.3 ReiserFS . . . . . . . . . . . . . . . . . . . . . 110
7.1.4 XFS . . . . . . . . . . . . . . . . . . . . . . . 111
7.1.5 Btrfs . . . . . . . . . . . . . . . . . . . . . . 113
7.1.6 Even More File Systems . . . . . . . . . . . . . . . 114
7.1.7 Swap space . . . . . . . . . . . . . . . . . . . . 115
7.2 Mounting File Systems . . . . . . . . . . . . . . . . . . 116
7.2.1 Basics . . . . . . . . . . . . . . . . . . . . . . 116
7.2.2 The mount Command . . . . . . . . . . . . . . . . . 116
7.2.3 Labels and UUIDs . . . . . . . . . . . . . . . . . 118
7.3 The dd Command . . . . . . . . . . . . . . . . . . . . 120
7.4 Disk Quotas . . . . . . . . . . . . . . . . . . . . . . 121
7.4.1 Basics . . . . . . . . . . . . . . . . . . . . . . 121
7.4.2 User Quotas (ext and XFS) . . . . . . . . . . . . . . 121
7.4.3 Group Quotas (ext and XFS) . . . . . . . . . . . . . . 123

Goals
• Knowing the most important ﬁle systems for Linux and their properties
• Being able to generate ﬁle systems on partitions and storage media
• Knowing about ﬁle system maintenance tools
• Being able to manage swap space
• Being able to mount local ﬁle systems
• Knowing how to set up disk quotas for users and groups

Prerequisites
• Competent use of the commands to handle ﬁles and directories
• Knowledge of mass storage on Linux and partitioning (Chapter 6)
• Existing knowledge about the structure of PC disk storage and ﬁle systems
is helpful

adm1-dateisysteme.tex (33e55eeadba676a3 )
100 7 File Systems: Care and Feeding

7.1 Creating a Linux File System
7.1.1 Overview
After having created a new partition, you must “format” that partition, i. e., write
the data structures necessary to manage ﬁles and directories onto it. What these
data structures look like in detail depends on the “ﬁle system” in question.

B Unfortunately, the term “ﬁle system” is overloaded on Linux. It means, for
example:
1. A method to arrange data and management information on a medium
(“the ext3 ﬁle system”)
2. Part of the ﬁle hierarchy of a Linux system which sits on a particular
medium or partition (“the root ﬁle system”, “the /var ﬁle system”)
3. All of the ﬁle hierarchy, across media boundaries

The ﬁle systems (meaning 1 above) common on Linux may diﬀer considerably.
On the one hand, there are ﬁle systems that were developed speciﬁcally for Linux,
such as the “ext ﬁlesystems” or the Reiser ﬁle system, and on the other hand there
are ﬁle systems that originally belonged to other operating systems but that Linux
supports (to a greater or lesser degree) for compatibility. This includes the ﬁle
systems of DOS, Windows, OS X, and various Unix variants as well as “network
ﬁle systems” such as NFS or SMB which allow access to ﬁle servers via the local
network.
Many ﬁle systems “native” to Linux are part of the tradition of ﬁle systems com-
mon on Unix, like the Berkeley Fast Filesystem (FFS), and their data structures are
based on those. However, development did not stop there; more modern inﬂu-
ences are constantly being integrated in order to keep Linux current with the state
of the art.

B Btrfs (pronounced “butter eﬀ ess”) by Chris Mason (Fusion-IO) is widely
considered the answer to the famous ZFS of Solaris. (The source code for
ZFS is available but cannot be integrated in the Linux directly, due to li-
censing considerations.) Its focus is on “fault tolerance, repairs and simple
administration”. By now it seems to be mostly usable, at least some distri-
butions rely on it.

superblock With Linux ﬁle systems it is common to have a superblock at the beginning
of the ﬁle system. This contains information pertaining to the ﬁle system as a
whole—such as when it was last mounted or unmounted, whether it was un-
mounted “cleanly” or because of a system crash, and so on. The superblock nor-
mally also points to other parts of the management data structures, like where the
inodes or free/occupied block lists are to be found and which parts of the medium
are available for data.

B It is usual to keep spare copies of the superblock elsewhere on the ﬁle sys-
tem, in case something happens to the original. This is what the ext ﬁle
systems do.

B On disk, there is usually a “boot sector” in front of the superblock, into
which you can theoretically put a boot loader (chapter 8). This makes it
possible to, e. g., install Linux on a computer alongside Windows and use
the Windows boot manager to start the system.

mkfs On Linux, ﬁle systems (meaning 2 above) are created using the mkfs command.
mkfs is independent of the actual ﬁle system (meaning 1) desired; it invokes the real
routine speciﬁc to the ﬁle system (meaning 1) in question, mkfs. ⟨ﬁle system name⟩.
You can select a ﬁle system type by means of the -t option—with “mkfs -t ext2 ”,
for example, the mkfs.ext2 program would be started (another name for mke2fs ).
7.1 Creating a Linux File System 101

When the computer has been switched oﬀ inadvertently or crashed, you have
to consider that the ﬁle system might be in an inconsistent state (even though this inconsistent state
happens very rarely in real life, even on crashes). File system errors can occur
because write operations are cached inside the computer’s RAM and may be lost
if the system is switched oﬀ before they could be written to disk. Other errors can
come up when the system gives up the ghost in the middle of an unbuﬀered write
operation.
Besides data loss, problems can include errors within the ﬁle system manage- structural errors
ment structure. These can be located and repaired using suitable programs and
include

• Erroneous directory entries
• Erroneous inode entries
• Files that do not occur in any directory
• Data blocks belonging to several diﬀerent ﬁles

Most but not all such problems can be repaired automatically without loss of data;
generally, the ﬁle system can be brought back to a consistent state.

B On boot, the system will ﬁnd out whether it has not been shut down cor-
rectly by checking a ﬁle system’s state. During a regular shutdown, the ﬁle
systems are unmounted and the “valid ﬂag” in every ﬁle system’s super valid flag
block will be set. On boot, this super block information may be used to au-
tomatically check these possibly-erroneous ﬁle systems and repair them if
necessary—before the system tries to mount a ﬁle system whose valid ﬂag
is not set, it tries to do a ﬁle system check.

B With all current Linux distributions, the system initialisation scripts exe-
cuted by init after booting contain all necessary commands to perform a
ﬁle system check.

If you want to check the consistency of a ﬁle system you do not need to wait
for the next reboot. You can launch a ﬁle system check at any time. Should a ﬁle file system check
contain errors, however, it can only be repaired if it is not currently mounted. This
restriction is necessary so that the kernel and the repair program do not “collide”.
This is another argument in favour of the automatic ﬁle system checks during
booting.
Actual consistency checks are performed using the fsck command. Like mkfs ,
depending on the type of the ﬁle system to be checked this command uses a spe-
ciﬁc sub-command called fsck. ⟨type⟩—e.g., fsck.ext2 for ext2 . fsck identiﬁes the
required sub-command by examining the ﬁle system in question. Using the

# fsck /dev/sdb1

command, for example, you can check the ﬁle system on /dev/sdb1 .

B The simple command
# fsck

checks all ﬁle systems listed in /etc/fstab with a non-zero value in the sixth
(last) column in sequence. (If several diﬀerent values exist, the ﬁle systems
are checked in ascending order.) /etc/fstab is explained in more detail in
section 7.2.2.

B fsck supports a -t option which at ﬁrst sight resembles mkfs but has a diﬀer-
ent meaning: A command like
102 7 File Systems: Care and Feeding

# fsck -t ext3

checks all ﬁle systems in /etc/fstab that are marked as type ext3 there.

options The most important options of fsck include:
-A (All) causes fsck to check all ﬁle systems mentioned in /etc/fstab .

B This obeys the checking order in the sixth column of the ﬁle. If several
ﬁle systems share the same value in that column, they are checked in
parallel if they are located on diﬀerent physical disks.

-R With -A , the root ﬁle system is not checked (which is useful if it is already
mounted for writing).
-V Outputs verbose messages about the check run.
-N Displays what fsck would do without actually doing it.
-s Inhibits parallel checking of multiple ﬁle systems. The “fsck ” command with-
out any parameters is equivalent to “fsck -A -s ”.
Besides its own options, you can pass additional options to fsck which it will
forward to the speciﬁc checking program. These must occur after the name of the
ﬁle system(s) to be checked and possibly a “-- ” separator. The -a , -f , -p and -v
options are supported by most such programs. Details may be found within the
documentation for the respective programs. The

# fsck /dev/sdb1 /dev/sdb2 -pv

for example would check the ﬁle systems on the /dev/sdb1 and /dev/sdb2 partitions
automatically, ﬁx any errors without manual intervention and report verbosely on
its progress.

B At program termination, fsck passes information about the ﬁle system state
to the shell:

0 No error was found in the ﬁle system
1 Errors were found and corrected
2 Severe errors were found and corrected. The system should be rebooted
4 Errors were found but not corrected
8 An error occurred while the program was executed
16 Usage error (e. g., bad command line)
128 Error in a shared library function
It is conceivable to analyse these return values in an init script and deter-
mine how to continue with the system boot. If several ﬁle systems are being
checked (using the -A option), the return value of fsck is the logical OR of
the return values of the individual checking programs.

7.1.2 The ext File Systems
History and Properties The original “extended ﬁle system” for Linux was imple-
mented in April, 1992, by Rémy Card. It was the ﬁrst ﬁle system designed speciﬁ-
cally for Linux (although it did take a lot of inspiration from general Unix ﬁle
systems) and did away with various limitations of the previously popular Minix
ﬁle system.
7.1 Creating a Linux File System 103

B The Minix ﬁle system had various nasty limits such as a maximum ﬁle sys-
tem size of 64 MiB and ﬁle names of at most 14 characters. (To be fair, Minix
was introduced when the IBM PC XT was considered a hot computer and
64 MiB, for PCs, amounted to an unimaginably vast amount of disk storage.
By 1990, that assumption had begun to crumble.) ext allowed ﬁle systems
of up to 2 GiB—quite useful at the time, but naturally somewhat ridiculous
today.

B The arrival of the ext ﬁle system marks another important improvement
to the Linux kernel, namely the introduction of the “virtual ﬁle system
switch”, or VFS. The VFS abstracts ﬁle system operations such as the open-
ing and closing of ﬁles or the reading and writing of data, and as such
enables the coexistence of diﬀerent ﬁle system implementations in Linux.

A The original ext ﬁle system is no longer used today. From here on, when we
talk about “the ext ﬁle systems”, we refer to ext2 and everything newer than
that.
The subsequent version, ext2 (the “second extended ﬁle system”), which was
begun by Rémy Card in January, 1993, amounted to a considerable rework of the
original “extended ﬁle system”. The development of ext2 made use of many ideas
from the BSD “Berkeley Fast Filesystem”. ext2 is still being maintained and makes
eminent sense for certain applications.

B Compared to ext , ext2 pushes various size limits—with the 4 KiB block size
typical for Intel-based Linux systems, ﬁle systems can be 16 TiB and single
ﬁles 2 TiB in size. Another important improvement in ext2 was the intro-
duction of separate timestamps for the last access, last content modiﬁcation
and last inode modiﬁcation, which achieved compatibility to “traditional”
Unix in this respect.

B From the beginning, ext2 was geared towards continued development and
improvement: Most data structures contained surplus space which was
later used for important extensions. These include ACLs and “extended
attributes”.
Since the end of the 1990s, Stephen Tweedie worked on a successor to ext2 ,
which was made part of the Linux kernel at the end of 2001 under the name of
ext3 . (That was Linux 2.4.15.) The most important diﬀerences between ext2 and
ext3 include:

• ext3 supports Journaling.
• ext3 allows enlarging ﬁle systems while they are mounted.
• ext3 supports more eﬃcient internal data structures for directories with
many entries.
Even so it is largely compatible with ext2 . It is usually possible to access ext3 ﬁle
systems as ext2 ﬁle systems (which implies that the new features cannot be used)
and vice-versa.

B “Journaling” solves a problem that can be very tedious with the increasing
size of ﬁle systems, namely that an unforeseen system crash makes it neces-
sary to do a complete consistency check of the ﬁle system. The Linux kernel
does not perform write operations immediately, but buﬀers the data in RAM
and writes them to disk when that is convenient (e. g., when the read/write
head of the disk drive is in the appropriate place). In addition, many write
operations involve writing data to various places on the disk, e. g., one or
more data blocks, the inode table, and the list of available blocks on the
disk. If the power fails in the right (or wrong) moment, such an operation
can remain only half-done—the ﬁle system is “inconsistent” in the sense
that a data block can be assigned to a ﬁle in the inode, but not marked used
in the free-block list. This can lead to serious problems later on.
104 7 File Systems: Care and Feeding

B A journaling ﬁle system like ext3 considers every write access to the disk
as a “transaction” which must be performed completely or not at all. By
deﬁnition, the ﬁle system is consistent before and after a transaction is per-
formed. Every transaction is ﬁrst written into a special area of the ﬁle sys-
tem called the journal. If it has been entirely written, it is marked “complete”
and, as such, it is oﬃcial. The Linux kernel can do the actual write opera-
tions later.—If the system crashes, a journaling ﬁle system does not need to
undergo a complete ﬁle system check, which with today’s ﬁle system sizes
could take hours or even days. Instead, the journal is considered and any
transactions marked “complete” are transferred to the actual ﬁle system.
Transactions not marked “complete” are thrown out.

A Most journaling ﬁle systems use the journal to log changes to the ﬁle sys-
tem’s “metadata”, i. e., directories, inodes, etc. For eﬃciency, the actual ﬁle
data are normally not written to the journal. This means that after a crash
and reboot you will have a consistent ﬁle system without having to spend
hours or days on a complete consistency check. However, your ﬁle contents
may have been scrambled—for example, a ﬁle might contain obsolete data
blocks because the updated ones couldn’t be written before the crash. This
problem can be mitigated by writing the data blocks to disk ﬁrst and then
the metadata to the journal, but even that is not without risk. ext3 gives
you the choice between three operating modes—writing everything to the
journal (mount option data=journal ), writing data blocks directly and then
metadata to the journal (data=ordered ), or no restrictions (data=writeback ). The
default is data=ordered .

B Writing metadata or even ﬁle data twice—once to the journal, and then later
to the actual ﬁle system—involves a certain loss of performance compared
to ﬁle systems like ext2 , which ignore the problem. One approach to ﬁx
this consists of log-structured ﬁle systems, in which the journal makes up the
actual ﬁle system. Within the Linux community, this approach has so far not
prevailed. Another approach is exempliﬁed by “copy-on-write ﬁlesystems”
like Btrfs.

A Using a journaling ﬁle system like ext3 does not absolve you from having to
perform complete consistency checks every so often. Errors in a ﬁle system’s
data structures might arise through disk hardware errors, cabling problems,
or the dreaded cosmic rays (don’t laugh) and might otherwise remain un-
noticed until they wreak havoc. For this reason, the ext ﬁle systems force a
ﬁle system check every so often when the system is booted (usually when
you can least aﬀord it). You will see how to tweak this later in this chapter.

A With server systems that are rarely rebooted and that you cannot simply
take oﬄine for a few hours or days for a prophylactic ﬁle system check, you
may have a big problem. We shall also come back to this.
The apex of ext ﬁle system evolution is currently represented by ext4 , which has
been developed since 2006 under the guidance of Theodore Ts’o. This has been
considered stable since 2008 (Kernel version 2.6.28). Like ext3 and ext2 , backward
compatibility was an important goal: ext2 and ext3 ﬁle systems can be mounted
as ext4 ﬁle systems and will proﬁt from some internal improvements in ext4 . On
the other hand, the ext4 code introduces some changes that result in ﬁle systems
no longer being accessible as ext2 and ext3 . Here are the most important improve-
ments in ext4 as compared to ext3 :
• Instead of maintaining the data blocks of individual ﬁles as lists of block
numbers, ext4 uses “extents”, i. e., groups of physically contiguous blocks
on disk. This leads to a considerable simpliﬁcation of space management
and to greater eﬃciency, but makes ﬁle systems using extents incompatible
to ext3 . It also avoids fragmentation, or the wild scattering of blocks belong-
ing to the same ﬁle across the whole ﬁle system.
7.1 Creating a Linux File System 105

• When data is written, actual blocks on the disk are assigned as late as pos-
sible. This also helps prevent fragmentation.
• User programs can advise the operating system how large a ﬁle is going
to be. Again, this can be used to assign contiguous ﬁle space and mitigate
fragmentation.
• Ext4 uses checksums to safeguard the journal. This increases reliability and
avoids some hairy problems when the journal is replayed after a system
crash.
• Various optimisations of internal data structures increase the speed of con-
sistency checks.
• Timestamps now carry nanosecond resolution and roll over in 2242 (rather
than 2038).
• Some size limits have been increased—directories may now contain 64,000
or more subdirectories (previously 32,000), ﬁles can be as large as 16 TiB,
and ﬁle systems as large as 1 EiB.
In spite of these useful improvements, according to Ted Ts’o ext4 is not to be con-
sidered an innovation, but rather a stopgap until even better ﬁle systems like Btrfs
become available.
All ext ﬁle systems include powerful tools for consistency checks and ﬁle sys-
tem repairs. This is very important for practical use.

Creating ext file systems To create a ext2 or ext3 ﬁle system, it is easiest to use the
mkfs command with a suitable -t option:

# mkfs -t ext2 /dev/sdb1 ext2 ﬁle system
# mkfs -t ext3 /dev/sdb1 ext3 ﬁle system
# mkfs -t ext4 /dev/sdb1 ext4 ﬁle system

After the -t option and its parameter, you can specify further parameters which
will be passed to the program performing the actual operation—in the case of the
ext ﬁle systems, the mke2fs program. (In spite of the e2 in its name, it can also create
ext3 and ext4 ﬁle systems.)

B The following commands would also work:
# mkfs.ext2 /dev/sdb1 ext2 ﬁle system
# mkfs.ext3 /dev/sdb1 ext3 ﬁle system
# mkfs.ext4 /dev/sdb1 ext4 ﬁle system

These are exactly the commands that mkfs would invoke. All three com-
mands are really symbolic links referring to mke2fs ; mke2fs looks at the name
used to call it and behaves accordingly.

B You can even call the mke2fs command directly: mke2fs

# mke2fs /dev/sdb1

(Passing no options will get you a ext2 ﬁle system.)

The following options for mke2fs are useful (and possibly important for the
exam):
-b ⟨size⟩ determines the block size. Typical values are 1024, 2048, or 4096. On
partitions of interesting size, the default is 4096.
106 7 File Systems: Care and Feeding

-c checks the partition for damaged blocks and marks them as unusable.

B Current hard disks can notice “bad blocks” and replace them by blocks
from a “secret reserve” without the operating system even noticing (at
least as long as you don’t ask the disk directly). While this is going on,
“mke2fs -c ”) does not provide an advantage. The command will only
ﬁnd bad blocks when the secret reserve is exhausted, and at that point
you would do well to replace the disk, anyway. (A completely new
hard disk would at this point be a warranty case. Old chestnuts are
only ﬁt for the garbage.)

-i ⟨count⟩ determines the “inode density”; an inode is created for every ⟨count⟩
bytes of space on the disk. The value must be a multiple of the block size
(option b ); there is no point in selecting a ⟨count⟩ that is less than the block
size. The minimum value is 1024, the default is the current block size.
-m ⟨percentage⟩ sets the percentage of data blocks reserved for root (default: 5%)
-S causes mke2fs to rewrite just the super blocks and group descriptors and leave
the inodes intact
-j creates a journal and, hence, an ext3 or ext4 ﬁle system.

B It is best to create an ext4 ﬁle system using one of the precooked calls
like “mkfs -t ext4 ”, since mke2fs then knows what it is suppsed to do. If
you must absolutely do it manually, use something like
# mke2fs -j -O extents,uninit_bg,dir_index /dev/sdb1

The ext ﬁle systems (still) need at least one complete data block for every ﬁle, no
matter how small. Thus, if you create an ext ﬁle system on which you intend
to store many small ﬁles (cue: mail or Usenet server), you may want to select a
internal fragmentation smaller block size in order to avoid internal fragmentation. (On the other hand,
disk space is really quite cheap today.)

B The inode density (-i option) determines how many ﬁles you can create on
the ﬁle system—since every ﬁle requires an inode, there can be no more
ﬁles than there are inodes. The default, creating an inode for every single
data block on the disk, is very conservative, but from the point of view of
the developers, the danger of not being able to create new ﬁles for lack of
inodes seems to be more of a problem than wasted space due to unused
inodes.

B Various ﬁle system objects require inodes but no data blocks—such as de-
vice ﬁles, FIFOs or short symbolic links. Even if you create as many inodes
as data blocks, you can still run out of inodes before running out of data
blocks.

B Using the mke2fs -F option, you can “format” ﬁle system objects that are not
block device ﬁles. For example, you can create CD-ROMs containing an ext2
ﬁle system by executing the command sequence

# dd if=/dev/zero of=cdrom.img bs=1M count=650
# mke2fs -F cdrom.img
# mount -o loop cdrom.img /mnt
# … copy stuﬀ to /mnt …
# umount /mnt
# cdrecord -data cdrom.img

(/dev/zero is a “device” that produces arbitrarily many zero bytes.) The re-
sulting CD-ROMs contain “genuine” ext2 ﬁle systems with all permissions,
attributes, ACLs etc., and can be mounted using
7.1 Creating a Linux File System 107

# mount -t ext2 -o ro /dev/scd0 /media/cdrom

(or some such command); you should replace /dev/scd0 by the device name
of your optical drive. (It is best to avoid using an ext3 ﬁle system here, since
the journal would be an utter waste of space. An ext4 ﬁle system, though,
can be created without a journal.)

Repairing ext file systems e2fsck is the consistency checker for ext ﬁle systems. e2fsck
There are usually symbolic links such as fsck.ext2 so it can be invoked from fsck .

B Like mke2fs , e2fsck also works for ext3 and ext4 ﬁle systems.

B You can of course invoke the program directly, which might save you a little
typing when passing options. On the other hand, you can only specify the
name of one single partition (strictly speaking, one single block device).

The most important options for e2fsck include: options

-b ⟨number⟩ reads the super block from block ⟨number⟩ of the partition (rather
than the ﬁrst super block)

-B ⟨size⟩ gives the size of a block group between two copies of the super block;
with the ext ﬁle systems, backup copies of the super block are usually placed
every 8192 blocks, on larger disks every 32768 blocks. (You can query this
using the tune2fs command explained below; look for “blocks per group” in
the output of “tune2fs -l ”.)

-f forces a ﬁle system to be checked even if its super block claims that it is clean
-l ⟨ﬁle⟩ reads the list of bad blocks from the ⟨ﬁle⟩ and marks these blocks as “used”
-c (“check”) searches the ﬁle system for bad blocks
-p (“preen”) causes errors to be repaired automatically with no further user in-
teraction
-v (“verbose”) outputs information about the program’s execution status and the
ﬁle system while the program is running
The device ﬁle speciﬁes the partition whose ﬁle system is to be checked. If that
partition does not contain an ext ﬁle system, the command aborts. e2fsck performs
the following steps: steps

1. The command line arguments are checked
2. The program checks whether the ﬁle system in question is mounted

3. The ﬁle system is opened
4. The super block is checked for readability
5. The data blocks are checked for errors
6. The super block information on inodes, blocks and sizes are compared with
the current system state
7. Directory entries are checked against inodes
8. Every data block that is marked “used” is checked for existence and whether
it is referred to exactly once by some inode

9. The number of links within directories is checked with the inode link coun-
ters (must match)
108 7 File Systems: Care and Feeding

10. The total number of blocks must equal the number of free blocks plus the
number of used blocks

exit code B e2fsck returns an exit code with the same meaning as the standard fsck exit
codes..

It is impossible to list all the ﬁle system errors that e2fsck can handle. Here are
a few important examples:
• Files whose inodes are not referenced from any directory are placed in the
ﬁle system’s lost+found directory using the inode number as the ﬁle name
and can be moved elsewhere from there. This type of error can occur, e. g.,
if the system crashes after a ﬁle has been created but before the directory
entry could be written.
• An inode’s link counter is greater than the number of links pointing to this
inode from directories. e2fsck corrects the inode’s link counter.

• e2fsck ﬁnds free blocks that are marked used (this can occur, e. g., when the
system crashes after a ﬁle has been deleted but before the block count and
bitmaps could be updated).
• The total number of blocks is incorrect (free and used blocks together are
diﬀerent from the total number of blocks).

complicated errors Not all errors are straightforward to repair. What to do if the super block is
unreadable? Then the ﬁle system can no longer be mounted, and e2fsck often fails
as well. You can then use a copy of the super block, one of which is included with
every block group on the partition. In this case you should boot a rescue system
and invoke fsck from there. With the -b option, e2fsck can be forced to consider a
particular block as the super block. The command in question then becomes, for
example:

# e2fsck -f -b 8193 /dev/sda2

B If the ﬁle system cannot be automatically repaired using fsck , it is pos-
sible to modify the ﬁle system directly. However, this requires very de-
tailed knowledge of ﬁle system structures which is beyond the scope of
this course.—There are two useful tools to aid with this. First, the dumpe2fs
program makes visible the internal management data structures of a ext
ﬁle system. The interpretation of its output requires the aforementioned
detailed knowledge. An ext ﬁle system may be repaired using the debugfs
ﬁle system debugger.

A You should keep your hands oﬀ programs like debugfs unless you know ex-
actly what you are doing. While debugfs enables you to manipulate the ﬁle
system’s data structures on a very low level, it is easy to damage a ﬁle sys-
tem even more by using it injudiciously. Now that we have appropriately
warned you, we may tell you that

# debugfs /dev/sda1

will open the ext ﬁle system on /dev/sda1 for inspection (debugfs , reasonably,
enables writing to the ﬁle system only if it was called with the -w option).
debugfs displays a prompt; “help ” gets you a list of available commands.
These are also listed in the documentation, which is in debugfs (8).
7.1 Creating a Linux File System 109

Querying and Changing ext File System Parameters If you have created a parti-
tion and put an ext ﬁle system on it, you can change some formatting parameters changing format parameters
after the fact. This is done using the tune2fs command, which should be used with
utmost caution and should never be applied on a ﬁle system mounted for writing:

tune2fs [⟨options⟩] ⟨device⟩

The following options are important:
-c ⟨count⟩ sets the maximum number of times the ﬁle system may be mounted
between two routine ﬁle system checks. The default value set by mke2fs is a
random number somewhere around 30 (so that not all ﬁle systems are pre-
emptively checked at the same time). The value 0 means “inﬁnitely many”.
-C ⟨count⟩ sets the current “mount count”. You can use this to cheat fsck or (by
setting it to a larger value than the current maximum set up using -c ) force
a ﬁle system check during the next system boot.

-e ⟨behaviour⟩ determines the behaviour of the system in case of errors. The fol-
lowing possibilities exist:
continue Go on as normal
remount-ro Disallow further writing to the ﬁle system
panic Force a kernel panic

In every case, a ﬁle system consistency check will be performed during the
next reboot.
-i ⟨interval⟩⟨unit⟩ sets the maximum time between two routine ﬁle system checks.
⟨interval⟩ is an integer; the ⟨unit⟩ is d for days, w for weeks and m for months.
The value 0 means “inﬁnitely long”.
-l displays super block information.
-m ⟨percent⟩ sets the percentage of data blocks reserved for root (or the user speci-
ﬁed using the -u option). The default value is 5%.

-L ⟨name⟩ sets a partition name (up to 16 characters). Commands like mount and
fsck make it possible to refer to partitions by their names rather than the
names of their device ﬁles.
To upgrade an existing ext3 ﬁle system to an ext4 ﬁle system, you need to exe-
cute the commands

# tune2fs -O extents,uninit_bg,dir_index /dev/sdb1
# e2fsck -fDp /dev/sdb1

(stipulating that the ﬁle system in question is on /dev/sdb1 ). Make sure to change
/etc/fstab such that the ﬁle system is mounted as ext4 later on (see section 7.2).

B Do note, though, that all existing ﬁles will still be using ext3 structures—
improvements like extents will only be used for ﬁles created later. The
e4defrag defragmentation tool is supposed to convert older ﬁles but is not
quite ready yet.

B If you have the wherewithal, you should not upgrade a ﬁle system “in place”
but instead backup its content, recreate the ﬁle system as ext4 , and the re-
store the content. The performance of ext4 is considerably better on “native”
ext4 ﬁle systems than on converted ext3 ﬁle systems—this can amount to a
factor of 2.
110 7 File Systems: Care and Feeding

B If you have ext2 ﬁle systems lying around that you would like to convert into
ext3 ﬁle systems: This is easily done by creating a journal. tune2fs will do
that for you, too:

# tune2fs -j /dev/sdb1

Again, you will have to adjust /etc/fstab if necessary.

Exercises
C 7.1 [!2] Generate an ext4 ﬁle system on a suitable medium (hard disk parti-
tion, USB thumb drive, ﬁle created using dd ).

C 7.2 [2] Change the maximum mount count of the ﬁlesystem created in ex-
ercise 7.1 to 30. In addition, 30% of the space available on the ﬁle system
should be reserved for user test .

7.1.3 ReiserFS
Overview ReiserFS is a Linux ﬁle system meant for general use. It was developed
by a team under the direction of Hans Reiser and debuted in Linux 2.4.1 (that was
in 2001). This made it the ﬁrst journaling ﬁle system available for Linux. ReiserFS
also contained some other innovations that the most popular Linux ﬁle system at
the time, ext2 , did not oﬀer:
• Using a special tool, ReiserFS ﬁle systems could be changed in size. Enlarge-
ment was even possible while the ﬁle system was mounted.
• Small ﬁles and the ends of larger ﬁles could be packed together to avoid
“internal fragmentation” which arises in ﬁle systems like ext2 because space
on disk is allocated based on the block size (usually 4 KiB). With ext2 and
friends, even a 1-byte ﬁle requires a full 4-KiB block, which could be consid-
ered wasteful (a 4097-byte ﬁle requires two data blocks, and that is almost
as bad). With ReiserFS, several such ﬁles could share one data block.

B There is nothing in principle that would keep the ext developers to add
this “tail packing” feature to the ext ﬁle systems. This was discussed
and the consensus was that by now, disk space is cheap enough that
the added complexity would be worth the trouble.

• Inodes aren’t pregenerated when the ﬁle system is created, but are allocated
on demand. This avoids a pathological problem possible with the ext ﬁle
systems, where there are blocks available in the ﬁle system but all inodes
are occupied and no new ﬁles can be generated.

B The ext ﬁle systems mitigate this problem by allocating one inode per
data block per default (the inode density corresponds to the block size).
This makes it diﬃcult to provoke the problem.

• ReiserFS uses trees instead of lists (like ext2 ) for its internal management
data structures. This makes it more eﬃcient for directories with many ﬁles.

B Ext3 and in particular ext4 can by now do that too.
As a matter of fact, ReiserFS uses the same tree structure not just for di-
rectory entries, but also for inodes, ﬁle metadata and ﬁle block lists, which
leads to a performance increase in places but to a decrease in others.

For a long time, ReiserFS used to be the default ﬁle system for the SUSE
distributions (and SUSE contributed to the project’s funding). Since 2006,
Novell/SUSE has moved from ReiserFS to ext3 ; very new SLES versions use
Btrfs for their root ﬁle system.
7.1 Creating a Linux File System 111

A In real life you should give the Reiser ﬁle system (and its designated succes-
sor, Reiser4) a wide berth unless you need to manage older systems using
it. This is less to do with the fact that Hans Reiser was convicted of his
wife’s murder (which of course does not speak in his favour as a human
being, but things like these do happen not just among Linux kernel devel-
opers), but more with the fact that the Reiser ﬁle system does have its good
points but is built on a fairly brittle base. For example, certain directory
operations in ReiserFS break basic assumptions that are otherwise univer-
sally valid for Unix-like ﬁle systems. This means, for instance, that mail
servers storing mailboxes on a ReiserFS ﬁle system are less resilient against
system crashes than ones using diﬀerent ﬁle systems. Another grave prob-
lem, which we will talk about brieﬂy later on, is the existence of technical
ﬂaws in the ﬁle system repair program. Finally—and that may be the most
serious problem—nobody seems to maintain the code any longer.

Creating ReiserFS file systems mkreiserfs serves to create a ReiserFS ﬁle system. mkreiserfs
The possible speciﬁcation of a logical block size is currently ignored, the size is
always 4 KiB. With dumpreiserfs you can determine information about ReiserFS dumpreiserfs
ﬁle systems on your disk. resize_reiserfs makes it possible to change the size of resize_reiserfs
currently-unused ReiserFS partitions. Mounted partitions may be resized using a
command like “mount -o remount,resize= ⟨block count⟩ ⟨mount point⟩”.

Consistency Checks for ReiserFS For the Reiser ﬁle system, too, there is a check- Reiser file system
ing and repair program, namely reiserfsck .
reiserfsck performs a consistency check and tries to repair any errors found,
much like e2fsck . This program is only necessary if the ﬁle system is really dam-
aged. Should a Reiser ﬁle system merely have been unmounted uncleanly, the
kernel will automatically try to restore it according to the journal.

A reiserfsck has some serious issues. One is that when the tree structure needs
to be reconstructed (which may happen in certain situations) it gets com-
pletely mixed up if data ﬁles (!) contain blocks that might be misconstrued
as another ReiserFS ﬁle system’s superblock. This will occur if you have
an image of a ReiserFS ﬁle system in a ﬁle used as a ReiserFS-formatted
“virtual” hard disk for a virtualisation environment such as VirtualBox or
VMware. This eﬀectively disqualiﬁes the ReiserFS ﬁle system for serious
work. You have been warned.

Exercises
C 7.3 [!1] What is the command to create a Reiser ﬁle system on the ﬁrst logical
partition of the second disk?

7.1.4 XFS
The XFS ﬁle system was donated to Linux by SGI (the erstwhile Silicon Graphics, XFS
Inc.); it is the ﬁle system used by SGI’s Unix variant, IRIX, which is able to handle
very large ﬁles eﬃciently. All Linux distributions of consequence oﬀer XFS sup-
port, even though few deploy it by default; you may have to install the XFS tools
separately.

B In some circles, “XFS” is the abbreviation of “X11 Font Server”. This can
occur in distribution package names. Don’t let yourself be confused.

You can create an XFS ﬁle system on an empty partition (or ﬁle) using the

# mkfs -t xfs /dev/sda2
112 7 File Systems: Care and Feeding

command (insert the appropriate device name). Of course, the real work is done
by a program called mkfs.xfs . You can control it using various options; consult the
documentation (xfs (5) and mkfs.xfs (8)).

B If performance is your goal, you can, for example, create the journal on an-
other (physical) storge medium by using an option like “-l logdev=/dev/sdb1,size=10000b ”.
(The actual ﬁle system should of course not be on /dev/sdb , and the partition
for the journal should not otherwise be used.)

The XFS tools contain a fsck.xfs (which you can invoke using “fsck -t xfs ”), but
this program doesn’t really do anything at all—it is merely there to give the sys-
tem something to call when “all” ﬁle systems are to be checked (which is easier
than putting a special exception for XFS into fsck ). In actual fact, XFS ﬁle sys-
tems are checked automatically on mounting if they have not been unmounted
cleanly. If you want to check the consistency of an XFS or have to repair one, use
the xfs_repair (8) program—“xfs_repair -n ” checks whether repairs are required;
without the option, any repairs will be performed outright.

B In extreme cases xfs_repair may not be able to repair the ﬁle system. In such a
situation you can use xfs_metadump to create a dump of the ﬁlesystem’s meta-
data and send that to the developers:

# xfs_metadump /dev/sdb1 sdb1.dump

(The ﬁle system must not be mounted when you do this.) The dump is a
binary ﬁle that does not contain actual ﬁle data and where all ﬁle names
have been obfuscated. Hence there is no risk of inadvertently passing along
conﬁdential data.

B A dump that has been prepared using xfs_metadump can be written back
to a ﬁle system (on a “real” storage medium or an image in a ﬁle) using
xfs_mdrestore . This will not include ﬁle contents as these aren’t part of the
dump to begin with. Unless you are an XFS developer, this command will
not be particularly interesting to you.

The xfs_info command outputs information about a (mounted) XFS ﬁle system:

# xfs_info /dev/sdb1
meta-data=/dev/sdb1 isize=256 agcount=4, agsize=16384 blks
= sectsz=512 attr=2, projid32bit=1
= crc=0 finobt=0
data = bsize=4096 blocks=65536, imaxpct=25
= sunit=0 swidth=0 blks
naming =version 2 bsize=4096 ascii-ci=0 ftype=0
log =Intern bsize=4096 blocks=853, version=2
= sectsz=512 sunit=0 blks, lazy-count=1
realtime =keine extsz=4096 blocks=0, rtextents=0

You can see, for example, that the ﬁle system consists of 65536 blocks of 4 KiB each
(bsize and blocks in the data section), while the journal occupies 853 4-KiB blocks
in the same ﬁle system (Intern , bsize and blocks in the log section).

B The same information is output by mkfs.xfs after creating a new XFS ﬁle
system.

You should avoid copying XFS ﬁle systems using dd (or at least proceed very
cautiously). This is because every XFS ﬁle system contains a unique UUID, and
programs like xfsdump (which makes backup copies) can get confused if they run
into two independent ﬁle systems using the same UUID. To copy XFS ﬁle systems,
use xfsdump and xfsrestore or else xfs_copy instead.
7.1 Creating a Linux File System 113

7.1.5 Btrfs
Btrfs is considered the up-and-coming Linux ﬁle system for the future. It com-
bines the properties traditionally associated with a Unix-like ﬁle system with some
innovative ideas that are partly based on Solaris’s ZFS. Besides some features oth-
erwise provided by the Logical Volum Manager (LVM; Section 6.7)—such as the
creation of ﬁle systems that span several physical storage media—or provided
by the Linux kernel’s RAID support—such as the redundant storage of data on
several physical media—this includes transparent data compression, consistency
checks on data blocks by means of checksums, and various others. The “killer
feature” is probably snapshots that can provide views of diﬀerent versions of ﬁles
or complete ﬁle hierarchies simultaneously.

B Btrfs is several years younger than ZFS, and its design therefore contains a
few neat ideas that hadn’t been invented yet when ZFS was ﬁrst introduced.
ZFS is currently considered the “state of the art” in ﬁle systems, but it is to
be expected that some time in the not-too-distant future it will be overtaken
by Btrfs.

B Btrfs is based, in principle, on the idea of “copy on write”. This means that
if you create a snapshot of a Btrfs ﬁle system, nothing is copied at all; the
system only notes that a copy exists. The data is accessible both from the
original ﬁle system and the snapshot, and as long as data is just being read,
the ﬁle systems can share the complete storage. Once write operations hap-
pen either in the original ﬁle system or the snapshot, only the data blocks
being modiﬁed are copied. The data itself is stored in eﬃcient data struc-
tures called B-trees.

Btrfs ﬁle systems are created with mkfs , as usual:

# mkfs -t btrfs /dev/sdb1

B You can also mention several storage media, which will all be made part
of the new ﬁle system. Btrfs stores metadata such as directory information
redundantly on several media; by default, data is spread out across various
disks (“striping”) in order to accelerate access1 . You can, however, request
other storage arrangements:

# mkfs -t btrfs -L MyBtrfs -d raid1 /dev/sdb1 /dev/sdc1

This example generates a Btrfs ﬁle system which encompasses the /dev/sdb1
and /dev/sdc1 disks and is labeled “MyBtrfs”. Data is stored redundantly on
both disks (“-d raid1 ”).

B Within Btrfs ﬁle systems you can create “subvolumes”, which serve as a type
of partition at the ﬁle system level. Subvolumes are the units of which you
will later be able to make snapshots. If your system uses Btrfs as its root ﬁle
system, the command

# btrfs subvolume create /home

would, for instance, allow you to keep your own data within a separate sub-
volume. Subvolumes do not take a lot of space, so you should not hesitate
to create more of them rather than fewer—in particular, one for every direc-
tory of which you might later want to create independent snapshots, since
it is not possible to make directories into subvolumes after the fact.

1 In other words, Btrfs uses RAID-1 for metadata and RAID-0 for data.
114 7 File Systems: Care and Feeding

B You can create a snapshot of a subvolume using
# btrfs subvolume snapshot /mnt/sub /mnt/sub-snap

The snapshot (here, /mnt/sub- snap ) is at ﬁrst indistinguishable from the origi-
nal subvolume (here, /mnt/sub ); both contain the same ﬁles and are writable.
At ﬁrst no extra storage space is being used—only if you change ﬁles in the
original or snapshot or create new ones, the system copies whatever is re-
quired.

Btrfs makes on-the-ﬂy consistency checks and tries to ﬁx problems as they are
detected. The “btrfs scrub start ” command starts a house-cleaning operation that
recalculates the checksums on all data and metadata on a Btrfs ﬁle system and
repairs faulty blocks according to a diﬀerent copy if required. This can, of course,
take a long time; with “btrfs scrub status ” you can query how it is getting on, with
“btrfs scrub cancel ” you can interrupt it, and restart it later with “btrfs scrub resume ”.
There is a fsck.btrfs program, but it does nothing beyond outputting a message
that it doesn’t do anything. The program is required because something needs
to be there to execute when all ﬁle systems are checked for consistency during
startup. To really check or repair Btrfs ﬁle systems there is the “btrfs check ” com-
mand. By default this does only a consistency check, and if it is invoked with the
“--repair ” it tries to actually repair any problems it found.
Btrfs is very versatile and complex and we can only give you a small glimpse
here. Consult the documentation (starting at btrfs (8)).

Exercises
C 7.4 [!1] Generate a Btrfs ﬁle system on an empty partition, using “mkfs -t
btrfs ”.

C 7.5 [2] Within your Btrfs ﬁle system, create a subvolume called sub0 . Create
some ﬁles within sub0 . Then create a snapshot called snap0 . Convince your-
self that sub0 and snap0 have the same content. Remove or change a few ﬁles
in sub0 and snap0 , and make sure that the two subvolumes are independent
of each other.

7.1.6 Even More File Systems
tmpfs tmpfs is a ﬂexible implementation of a “RAM disk ﬁle system”, which stores ﬁles
not on disk, but in the computer’s virtual memory. They can thus be accessed
more quickly, but seldom used ﬁles can still be moved to swap space. The size of
a tmpfs is variable up to a set limit. There is no special program for generating a
tmpfs , but you can create it simply by mounting it: For example, the

# mount -t tmpfs -o size=1G,mode=0700 tmpfs /scratch

command creates a tmpfs of at most 1 GiB under the name of /scratch , which can
only be accessed by the owner of the /scratch directory. (We shall be coming back
to mounting ﬁle systems in section 7.2.)
A popular ﬁle system for older Windows PCs, USB sticks, digital cameras, MP3
players and other “storage devices” without big ideas about eﬃciency and ﬂexi-
VFAT bility is Microsoft’s venerable VFAT ﬁle system. Naturally, Linux can mount, read,
and write media formatted thusly, and also create such ﬁle systems, for example
by

# mkfs -t vfat /dev/mcblk0p1
7.1 Creating a Linux File System 115

(insert the appropriate device name again). At this point you will no longer be sur-
prised to hear that mkfs.vfat is just another name for the mkdosfs program, which
can create all sorts of MS-DOS and Windows ﬁle systems—including the ﬁle sys-
tem used by the Atari ST of blessed memory. (As there are Linux variants running
on Atari computers, this is not quite as far-fetched as it may sound.)

B mkdosfs supports various options allowing you to determine the type of ﬁle
system created. Most of these are of no practical consequence today, and
mkdosfs will do the Right Thing in most cases, anyway. We do not want to
disgress into a taxonomy of FAT ﬁle system variants and restrict ourselves to
pointing out that the main diﬀerence between FAT and VFAT is that ﬁle sys-
tems of the latter persuasion allow ﬁle names that do not follow the older,
strict 8 + 3 scheme. The “ﬁle allocation table”, the data structure that re-
members which data blocks belong to which ﬁle and that gave the ﬁle sys-
tem its name, also exists invarious ﬂavours, of which mkdosfs selects the one
most suitable to the medium in question—ﬂoppy disks are endowed with a
12-bit FAT, and hard disk (partitions) or (today) USB sticks of considerable
capacity get 32-bit FATs; in the latter case the resulting ﬁle system is called
“VFAT32”.

NTFS, the ﬁle system used by Windows NT and its successors including Win- NTFS
dows Vista, is a bit of an exasperating topic. Obviously there is considerable
interest in enabling Linux to handle NTFS partitions—everywhere but on Mi-
crosoft’s part, where so far one has not deigned to explain to the general public
how NTFS actually works. (It is well-known that NTFS is based on BSD’s “Berke-
ley Fast Filesystem”, which is reasonably well understood, but in the meantime
Microsoft butchered it into something barely recognisable.) In the Linux com-
munity there have been several attempts to provide NTFS support by trying to
understand NTFS on Windows, but complete success is still some way oﬀ. At
the moment there is a kernel-based driver with good support for reading, but
questionable support for writing, and another driver running in user space which
according to the grapevine works well for reading and writing. Finally, there are
the “ntfsprogs”, a package of tools for managing NTFS ﬁle systems, which also
allow rudimentary access to data stored on them. Further information is available
from http://www.linux- ntfs.org/ .

7.1.7 Swap space
In addition to the ﬁle system partitions, you should always create a swap parti- swap partition
tion. Linux can use this to store part of the content of system RAM; the eﬀective
amount of working memory available to you is thus greater than the amount of
RAM in your computer.
Before you can use a swap partition you must “format” it using the mkswap com-
mand:
# mkswap /dev/sda4

This writes some administrative data to the partition.
When the system is started, it is necessary to “activate” a swap partition. This
corresponds to mounting a partition with a ﬁle system and is done using the swapon
command:
# swapon /dev/sda4

The partition should subsequently be mentioned in the /proc/swaps ﬁle:

# cat /proc/swaps
Filename Type Size Used Priority
/dev/sda4 partition 2144636 380 -1
116 7 File Systems: Care and Feeding

After use the swap partition can be deactivated using swapoff :

# swapoff /dev/sda4

B The system usually takes care of activating and deactivating swap parti-
tions, as long as you put them into the /etc/fstab ﬁle. See section 7.2.2.

You can operate up to 32 swap partitions (up to and including kernel version
2.4.10: 8) in parallel; the maximum size depends on your computer’s architecture
and isn’t documented anywhere exactly, but “stupendously gigantic” is a reason-
able approximation. It used to be just a little less than 2 GiB for most Linux plat-
forms.

B If you have several disks, you should spread your swap space across all of
them, which should increase speed noticeably.

B Linux can prioritise swap space. This is worth doing if the disks containing
your swap space have diﬀerent speeds, because Linux will prefer the faster
disks. Read up on this in swapon (8).

B Besides partitions, you can also use ﬁles as swap space. Since Linux 2.6 this
isn’t even any slower! This allows you to temporarily provide space for rare
humongous workloads. You must initially create a swap ﬁle as a ﬁle full of
zeros, for instance by using

# dd if=/dev/zero of=swapfile bs=1M count=256

before preparing it using the mkswap command and activating it with swapon .
(Desist from tricks using dd or cp ; a swap ﬁle may not contain “holes”.)

B You can ﬁnd information about the currently active swap areas in the /proc/
swaps ﬁle.

7.2 Mounting File Systems
7.2.1 Basics
To access data stored on a medium (hard disk, USB stick, ﬂoppy, …), it would in
principle be possible to access the device ﬁles directly. This is in fact being done,
for example when accessing tape drives. However, the well-known ﬁle manage-
ment commands (cp , mv , and so on) can only access ﬁles via the directory tree.
To use these commands, storage media must be made part of the directory tree
(“mounted”) using their device ﬁles. This is done using the mount command.
The place in the directory tree where a ﬁle system is to be mounted is called a
mount point mount point. This can be any directory; it does not even have to be empty, but you
will not be able to access the original directory content while another ﬁle system
is mounted “over” it.

B The content reappears once the ﬁle system is unmounted using umount . Even
so you should restrain yourself from mounting stuﬀ on /etc and other im-
portant system directories …

7.2.2 The mount Command
The mount command mounts ﬁle systems into the directory tree. It can also be
used to display the currently mounted ﬁle systems, simply by calling it without
parameters:
7.2 Mounting File Systems 117

proc /proc proc defaults 0 0
/dev/sda2 / ext3 defaults,errors=remount-ro 0 1
/dev/sda1 none swap sw 0 0
/dev/sda3 /home ext3 defaults,relatime 0 1
/dev/sr0 /media/cdrom0 udf,iso9660 ro,user,exec,noauto 0 0
/dev/sdb1 /media/usb auto user,noauto 0 0
/dev/fd0 /media/floppy auto user,noauto,sync 0 0

Figure 7.1: The /etc/fstab ﬁle (example)

$ mount
/dev/sda2 on / type ext3 (rw,relatime,errors=remount-ro)
tmpfs on /lib/init/rw type tmpfs (rw,nosuid,mode=0755)
proc on /proc type proc (rw,noexec,nosuid,nodev)
sysfs on /sys type sysfs (rw,noexec,nosuid,nodev)

To mount a medium, for example a hard disk partition, you must specify its
device ﬁle and the desired mount point:

# mount -t ext2 /dev/sda5 /home

It is not mandatory to specify the ﬁle system type using the -t option, since the
kernel can generally ﬁgure it out for itself. If the partition is mentioned in /etc/
fstab , it is suﬃcient to give either the mount point or the device ﬁle:

# mount /dev/sda5 One possibility …
# mount /home … and another

Generally speaking, the /etc/fstab ﬁle describes the composition of the whole /etc/fstab
ﬁle system structure from various ﬁle systems that can be located on diﬀerent
partitions, disks etc. In addition to the device names and corresponding mount
points, you can specify various options used to mount the ﬁle systems. The allow-
able options depend on the ﬁle system; many options are to be found in mount (8).
A typical /etc/fstab ﬁle could look similar to ﬁgure 7.1. The root partition usu-
ally occupies the ﬁrst line. Besides the “normal” ﬁle systems, pseudo ﬁle systems
such as devpts or proc and the swap areas are mentioned here.
The third ﬁeld describes the type of the ﬁle system in question. Entries like ext3 type
and iso9660 speak for themselves (if mount cannot decide what to do with the type
speciﬁcation, it tries to delegate the job to a program called /sbin/mount. ⟨type⟩), swap
refers to swap space (which does not require mounting), and auto means that mount
should try to determine the ﬁle system’s type.

B To guess, mount utilises the content of the /etc/filesystems ﬁle, or, if that ﬁle
does not exist, the /proc/filesystems ﬁle. (/proc/filesystems is also read if /etc/
filesystems ends with a line containing just an asterisk.) In any case, mount
processes only those lines that are not marked nodev . For your ediﬁcation,
here is a snippet from a typical /proc/filesystems ﬁle:

nodev sysfs
nodev rootfs

nodev usbfs
ext3
nodev nfs
vfat
118 7 File Systems: Care and Feeding

xfs

B The kernel generates /proc/filesystems dynamically based on those ﬁle sys-
tems for which it actually contains drivers. /etc/filesystems is useful if you
want to specify an order for mount ’s guesswork that deviates from the one
resulting from /proc/filesystems (which you cannot inﬂuence).

B Before mount refers to /etc/filesystems , it tries its luck with the libblkid and
libvolume_id libraries, both of which are (among other things) able to deter-
mine which type of ﬁle system exists on a medium. You can experiment
with these libraries using the command line programs blkid and vol_id :

# blkid /dev/sdb1
/dev/sdb1: LABEL="TESTBTRFS" UUID="d38d6bd1-66c3-49c6-b272-eabdae
877368" UUID_SUB="3c093524-2a83-4af0-8290-c22f2ab44ef3"
TYPE="btrfs" PARTLABEL="Linux filesystem"
PARTUUID="ade1d2db-7412-4bc1-8eab-e42fdee9882b"

options The fourth ﬁeld contains the options, including:
defaults Is not really an option, but merely a place holder for the standard options
(see mount (8)).
noauto Opposite of auto , keeps a ﬁle system from being mounted automatically
when the system is booted.
user In principle, only root can mount storage devices (normal users may only
use the simple mount command to display information), unless the user op-
tion is set. In this case, normal users may say “mount ⟨device⟩” or “mount
⟨mount point⟩”; this will mount the named device on the designated mount
point. The user option will allow the mounting user to unmount the device
(root , too); there is a similar option users that allows any user to unmount
the device.
sync Write operations are not buﬀered in RAM but written to the medium directly.
The end of the write operation is only signaled to the application program
once the data have actually been written to the medium. This is useful for
ﬂoppies or USB thumb drives, which might otherwise be inadvertently re-
moved from the drive while unwritten data is still buﬀered in RAM.
ro This ﬁle system is mounted for reading only, not writing (opposite of rw )
exec Executable ﬁles on this ﬁle system may be invoked. The opposite is noexec ;
exec is given here because the user option implies the noexec option (among
others).
As you can see in the /dev/sdb entry, later options can overwrite earlier ones: user
implies the noexec option, but the exec farther on the right of the line overwrites
this default.

7.2.3 Labels and UUIDs
We showed you how to mount ﬁle systems using device names such as /dev/hda1 .
This has the disadvantage, though, that the correspondence between device ﬁles
and actual devices is not necessarily ﬁxed: As soon as you remove or repartition a
disk or add another, the correspondence may change and you will have to adjust
the conﬁguration in /etc/fstab . With some device types, such as USB media, you
cannot by design rely on anything. This is where labels and UUIDs come in.
label A label is a piece of arbitrary text of up to 16 characters that is placed in a ﬁle
system’s super block. If you have forgotten to assign a label when creating the
7.2 Mounting File Systems 119

ﬁle system, you can add one (or modify an existing one) at any time using e2label .
The command
# e2label /dev/sda3 home

(for example) lets you refer to /dev/sda3 as LABEL=home , e. g., using

# mount -t ext2 LABEL=home /home

The system will then search all available partitions for a ﬁle system containing this
label.

B You can do the same using the -L option of tune2fs :
# tune2fs -L home /dev/sda3

B The other ﬁle systems have their ways and means to set labels, too. With
Btrfs, for example, you can either specify one when the ﬁle system is gener-
ated (option “-L ”) or use

# btrfs filesystem label /dev/sdb1 MYLABEL

If you have very many disks or computers and labels do not provide the re-
quired degree of uniqueness, you can fall back to a “universally unique identiﬁer”
or UUID. An UUID typically looks like UUID

$ uuidgen
bea6383f-22a7-453f-8ef5-a5b895c8ccb0

and is generated automatically and randomly when a ﬁle system is created. This
ensures that no two ﬁle systems share the same UUID. Other than that, UUIDs
are used much like labels, except that you now need to use UUID=bea6383f-22a7-
453f-8ef5-a5b895c8ccb0 (Gulp.) You can also set UUIDs by means of tune2fs , or create
completely new ones using

# tune2fs -U random /dev/hda3

This should seldom prove necessary, though, for example if you replace a disk or
have cloned a ﬁle system.

B Incidentally, you can determine a ﬁle system’s UUID using
# tune2fs -l /dev/hda2 | grep UUID
Filesystem UUID: 4886d1a2-a40d-4b0e-ae3c-731dd4692a77

B With other ﬁle systems (XFS, Btrfs) you can query a ﬁle system’s UUID (blkid
is your friend) but not necessarily change it.

B The
# lsblk -o +UUID

command gives you an overview of all your block devices and their UUIDs.

B You can also access swap partitions using labels or UUIDs:
# swapon -L swap1
# swapon -U 88e5f06d-66d9-4747-bb32-e159c4b3b247
120 7 File Systems: Care and Feeding

You can ﬁnd the UUID of a swap partition using blkid or lsblk , or check the
/dev/disk/by- uuid directory. If your swap partition does not have a UUID nor
a label, you can use mkswap to assign one.
You can also use labels and UUIDs in the /etc/fstab ﬁle (one might indeed claim
that this is the whole point of the exercise). Simply put

LABEL=home

or
UUID=bea6383f-22a7-453f-8ef5-a5b895c8ccb0

into the ﬁrst ﬁeld instead of the device name. Of course this also works for swap
space.

Exercises
C 7.6 [!2] Consider the entries in ﬁles /etc/fstab and /etc/mtab . How do they
diﬀer?

7.3 The dd Command
dd is a command for copying ﬁles “by block”. It is used with particular preference
to create “images”, that is to say complete copies of ﬁle systems—for example,
when preparing for the complete restoration of the system in case of a catastrophic
disk failure.
dd (short for “copy and convert”2 ) reads data block by block from an input ﬁle
and writes it unchanged to an output ﬁle. The data’s type is of no consequence.
Neither does it matter to dd whether the ﬁles in question are regular ﬁles or device
ﬁles.
Using dd , you can create a quickly-restorable backup copy of your system par-
tition as follows:
# dd if=/dev/sda2 of=/data/sda2.dump

This saves the second partition of the ﬁrst SCSI disk to a ﬁle called /data/sda2.
dump —this ﬁle should of course be located on another disk. If your ﬁrst disk is
damaged, you can easily and very quickly restore the original state after replacing
it with an identical (!) drive:

# dd if=/data/sda2.dump of=/dev/sda2

(If /dev/sda is your system disk, you must of course have booted from a rescue or
live system.)
For this to work, the new disk drive’s geometry must match that of the old one.
partition table In addition, the new disk drive needs a partition table that is equivalent to the old
one. You can save the partition table using dd as well (at least for MBR-partitioned
disks):

# dd if=/dev/sda of=/media/floppy/mbr_sda.dump bs=512 count=1

Used like this, dd does not save all of the hard disk to ﬂoppy disk, but writes every-
thing in chunks of 512 bytes (bs=512 )—one chunk (count=1 ), to be exact. In eﬀect, all
of the MBR is written to the ﬂoppy. This kills two birds with the same stone: the
boot loader’s stage 1 also ends up back on the hard disk after the MBR is restored:
2 Seriously! The dd command is inspired by a corresponding command on IBM mainframes (hence

the parameter syntax, which according to Unix standards is quite quaint), which was called CC (as in
“copy and convert”), but on Unix the cc name was already spoken for by the C compiler.
7.4 Disk Quotas 121

# dd if=/media/floppy/mbr_sda.dump of=/dev/sda

You do not need to specify a chunk size here; the ﬁle is just written once and is
(hopefully) only 512 bytes in size.

A Caution: The MBR does not contain partitioning information for logical par-
titions! IF you use logical partitions, you should use a program like sfdisk
to save all of the partitioning scheme—see below.

B To save partitioning information for GPT-partitioned disks, use, for exam-
ple, gdisk (the b command).

B dd can also be used to make the content of CD-ROMs or DVDs permanently
accessible from hard disk. The “dd if=/dev/cdrom of=/data/cdrom1.iso ” places
the content of the CD-ROM on disk. Since the ﬁle is an ISO image, hence
contains a ﬁle system that the Linux kernel can interpret, it can also be
mounted. After “mount -o loop,ro /data/cdrom.iso /mnt ” you can access the
image’s content. You can of course make this permanent using /etc/fstab .

7.4 Disk Quotas
7.4.1 Basics
Linux makes it possible to limit the maximum number of used inodes or data
blocks per user or per group. Two limits are distinguished: The soft quota may soft quota
not be exceeded in the long term, but you can allow users an “overdraft” by setting
the hard quota to a higher value. Users may then, for a short time, occupy space hard quota
up to the hard quota, but within a certain period of time they must reduce the
used space to below the soft quota again. As soon as the hard quota is reached or
the grace period is over, further write operations will fail.

B It is suﬃcient to dip below the soft quota very brieﬂy. After that you will
again be able to use space up to the hard quota, and the grace period will
start from the beginning.
Quotas can be assigned per ﬁle system. You could, for example, set a limit Quota assignment
for the size of the incoming mail box in /var/mail without constraining the users’
home directories or vice-versa. Linux supports quotas on all common Linux ﬁle
systems.

B As far as quotas are concerned, XFS does its own thing—but the current
Linux tools for quota management can also handle XFS. (Should you be
forced to deal with a fairly old Linux, it may be possible that you need to
use the xfs_quota command instead.)

B Btrfs has its own infrastructure for quotas, the so-called “quota groups” or
“qgroups”. In principle, you can set up quotas for individual subvolumes,
which will then be managed hierarchically. The ﬁle system ensures that a
ﬁle system never uses more space than the “quota group” of its subvolume,
that of the enclosing subvolume, etc. specify. Snapshots do not count as long
as they aren’t changed, since snapshots do not take up additional space at
ﬁrst. Study btrfs-qgroup (8).

7.4.2 User Quotas (ext and XFS)
To set up user quotas, you must ﬁrst install the quota software (included with
most distributions). Then you can mark those ﬁle systems where quotas should
be enforced by including the usrquota mount option, either ad hoc for a mounted
ﬁle system as in
122 7 File Systems: Care and Feeding

# mount -o remount,usrquota /home

or (preferably) by permanently including the option in the ﬁle system’s entry in
/etc/fstab :

/dev/hda5 /home ext2 defaults,usrquota 0 2

The quota database is initialised using

# quotacheck -avu

(watch for possible warnings), after the ﬁle system has been mounted using the
usrquota option. The quotacheck program creates the aquota.user database ﬁle in the
partition’s root directory, in this case /home/aquota.user . It is advisable to execute
quotacheck periodically during normal system operations in order to “clean” the
database.
Afterwards you should start the quota system, so that the database created
using quotacheck is brought up to date with every ﬁle system operation:

# quotaon -avu

Your distribution will probably contain an init script which will perform this step
during system boot. Accordingly, the quota system will be deactivated on shut-
down using “quotaoff -auv ”.
Setting quotas You can set quotas for various users using the edquota command. edquota starts
your favourite editor (according to the EDITOR environment variable) with a “tem-
plate” where you can ﬁll in the soft and hard quotas for the ﬁle systems in ques-
tion:
# edquota -u hugo Quotas for hugo

The grace period in days is set using “edquota -t ”.
You can also refer to the quotas of a “prototype user”:

# edquota -p tux hugo

sets hugo ’s quotas to those deﬁned for tux . This makes it easy to assign quotas to
newly created users without having to enter them manually using edquota .
Querying quotas Users can query their quota status with the quota command. “quota -q ” outputs
a brief message mentioning only those ﬁle systems whose soft quota has been
exceeded. This command is suitable for ﬁles such as ~/.profile .
The repquota command generates a tabular overview of the disk usage of vari-
ous users, together with possibly active quotas:

# repquota -a
Block limits File limits
User used soft hard grace used soft hard grace
root -- 166512 0 0 19562 0 0
tux -- 2304 10000 12000 806 1000 2000
hugo -- 1192 5000 6000 389 500 1000

The grace column contains the remaining number of days of the grace period if
the soft quota has been exceeded.
7.4 Disk Quotas 123

7.4.3 Group Quotas (ext and XFS)
You can also set up group quotas applying to all members of a group together. For group quotas
this to become eﬀective, you must limit all groups that the users in question are
members of; otherwise they could circumvent their quota by changing groups.
The mount option for group quotas is grpquota , and the database ﬁle is called
aquota.group . To enable group quotas, you must use the aforementioned com-
mands while substituting or augmenting the -u option by -g . Hence,

# quotaon -auvg

activates all types of quota (user and group) on every ﬁle system, while

$ quota -vg

displays group quotas for those groups that you are a member of.

Exercises
C 7.7 [2] Set up disk quotas as described: a ﬁle system with user quotas,
one with group quotas, and one with both. Watch out for the output of
quotacheck , quotaon , and quota .

C 7.8 [2] Set up quotas for a user and a group (containing that user). How
does quota ’s output change when you reach the soft quota (the hard quota)?
Is a warning displayed?
124 7 File Systems: Care and Feeding

Commands in this Chapter
blkid Locates and prints block device attributes blkid (8) 118
dd “Copy and convert”, copies ﬁles or ﬁle systems block by block and does
simple conversions dd (1) 120
debugfs File system debugger for ﬁxing badly damaged ﬁle systems. For gurus
only! debugfs (8) 108
dumpe2fs Displays internal management data of the ext2 ﬁle system. For gurus
only! dumpe2fs (8) 108
dumpreiserfs Displays internal management data of the Reiser ﬁle system. For
gurus only! dumpreiserfs (8) 111
e2fsck Checks ext2 and ext3 ﬁle systems for consistency e2fsck (8) 107
e2label Changes the label on an ext2/3 ﬁle system e2label (8) 118
edquota Tool for entering and adjusting disk quotas edquota (8) 122
fsck Organises ﬁle system consistency checks fsck (8) 101
lsblk Lists available block devices lsblk (8) 119
mkdosfs Creates FAT-formatted ﬁle systems mkfs.vfat (8) 114
mke2fs Creates ext2 or ext3 ﬁle systems mke2fs (8) 105
mkfs Manages ﬁle system creation mkfs (8) 100
mkfs.vfat Creates FAT-formatted ﬁle systems mkfs.vfat (8) 114
mkfs.xfs Creates XFS-formatted ﬁle systems mkfs.xfs (8) 111
mkreiserfs Creates Reiser ﬁle systems mkreiserfs (8) 111
mkswap Initialises a swap partition or ﬁle mkswap (8) 115
mount Includes a ﬁle system in the directory tree mount (8), mount (2) 116
quota Reports on a user’s quota status quota (1) 122
reiserfsck Checks a Reiser ﬁle system for consistency reiserfsck (8) 111
repquota Summarises ﬁlesystem usage and quota usage for many users
repquota (8) 122
resize_reiserfs Changes the size of a Reiser ﬁle system resize_reiserfs (8) 111
swapoff Deactivates a swap partition or ﬁle swapoff (8) 115
swapon Activates a swap partition or ﬁle swapon (8) 115
tune2fs Adjusts ext2 and ext3 ﬁle system parameters tunefs (8) 108, 119
vol_id Determines ﬁle system types and reads labels and UUIDs
vol_id (8) 118
xfs_mdrestore Restores an XFS metadata dump to a ﬁlesystem image
xfs_mdrestore (8) 112
xfs_metadump Produces metadata dumps from XFS ﬁle systems
xfs_metadump (8) 112

Summary
• After partitioning, a ﬁle system must be created on a new partition before
it can be used. To do so, Linux provides the mkfs command (with a number
of ﬁle-system-speciﬁc auxiliary tools that do the actual work).
• Improperly unmounted ﬁle systems may exhibit inconsistencies. If Linux
notes such ﬁle systems when it boots, these will be checked automatically
and, if possible, repaired. These checks can also be triggered manually us-
ing programs such as fsck and e2fsck .
• The mount command serves to integrate ﬁle systems into the directory tree.
• With dd , partitions can be backed up at block level.
• To enable disk quotas, you must mount the ﬁle systems accordingly, ini-
tialise the database, and activate the quota system.
• Quotas are speciﬁed using edquota and supervised using quota or repquota .
7.4 Bibliography 125

Bibliography
Quota-Mini-HOWTO03 Ralf van Dooren. “Quota mini-HOWTO”, August 2003.
http://tldp.org/HOWTO/Quota.html
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

8
Booting Linux

Contents
8.1 Fundamentals . . . . . . . . . . . . . . . . . . . . . 128
8.2 GRUB Legacy . . . . . . . . . . . . . . . . . . . . . 131
8.2.1 GRUB Basics . . . . . . . . . . . . . . . . . . . 131
8.2.2 GRUB Legacy Conﬁguration . . . . . . . . . . . . . . 132
8.2.3 GRUB Legacy Installation . . . . . . . . . . . . . . . 133
8.2.4 GRUB 2 . . . . . . . . . . . . . . . . . . . . . 134
8.2.5 Security Advice . . . . . . . . . . . . . . . . . . 135
8.3 Kernel Parameters . . . . . . . . . . . . . . . . . . . 135
8.4 System Startup Problems . . . . . . . . . . . . . . . . . 137
8.4.1 Troubleshooting . . . . . . . . . . . . . . . . . . 137
8.4.2 Typical Problems . . . . . . . . . . . . . . . . . . 137
8.4.3 Rescue systems and Live Distributions . . . . . . . . . . 139

Goals
• Knowing the GRUB Legacy and GRUB 2 boot loaders and how to conﬁgure
them
• Being able to diagnose and ﬁx system start problems

Prerequisites
• Basic knowledge of the PC startup procedure
• Handling of conﬁguration ﬁles

adm1-boot.tex (33e55eeadba676a3 )
128 8 Booting Linux

8.1 Fundamentals
When you switch on a Linux computer, an interesting and intricate process takes
place during which the computer initialises and tests itself before launching the
actual operating system (Linux). In this chapter, we consider this process in some
detail and explain how to adapt it to your requirements and to ﬁnd and repair
problems if necessary.

B The word “to boot” is short for “to pull oneself up by one’s bootstraps”.
While, as Newton tells us, this is a physical impossibility, it is a good image
for what goes on, namely that the computer gets itself started from the most
basic beginnings.

Immediately after the computer is switched on, its ﬁrmware—depending on
the computer’s age, either the “basic input/output system” (BIOS) or “uniﬁed
extensible ﬁrmware interface” (UEFI) takes control. What happens next depends
on the ﬁrmware.

BIOS startup On BIOS-based systems, the BIOS searches for an operating system
on media like CD-ROM or hard disk, depending on the boot order speciﬁed in the
BIOS setup. On disks (hard or ﬂoppy), the ﬁrst 512 bytes of the boot medium will
be read. These contain special information concerning the system start. Generally,
boot sector this area is called the boot sector; a hard disk’s boot sector is also called the master
master boot record boot record (MBR).

B We already came across the MBR when discussing the eponymous disk par-
titioning scheme in chapter 6. We’re now looking at the part of the MBR that
does not contain partitioning information.

The ﬁrst 446 bytes of the MBR contain a minimal startup program which in
boot loader turn is responsible for starting the operating system—the boot loader. The rest
is occupied by the partition table. 446 bytes are not enough for the complete boot
loader, but they suﬃce for a small program which can fetch the rest of the boot
loader from disk using the BIOS. In the space between the MBR and the start of
the ﬁrst partition—at least sector 63, today more likely sector 2048 there is enough
room for the rest of the boot loader. (We shall come back to that topic presently.)
Modern boot loaders for Linux (in particular, the “Grand Uniﬁed Boot loader”
GRUB or GRUB) can read common Linux ﬁle systems and are therefore able to ﬁnd the
operating system kernel on a Linux partition, load it into RAM and start it there.

boot manager B GRUB serves not just as a boot loader, but also as a boot manager. As such,
it can, according to the user’s preferences, launch various Linux kernels or
even other operating systems.

B Bootable CD-ROMs or DVDs play an important role for the installation or
update of Linux systems, or as the basis of “live systems” that run directly
from read-only media without having to be installed on disk. To boot a
Linux computer from CD, you must in the simplest case ensure that the
CD-ROM drive is ahead of the ﬁrmware’s boot order than the hard disk,
and start the computer while the desired CD is in the drive.

B In the BIOS tradition, booting oﬀ CD-ROMs follows diﬀerent rules than
booting oﬀ hard disk (or ﬂoppy disk). The “El Torito” standard (which
speciﬁes these rules) basically deﬁnes two approaches: One method is to
include an image of a bootable ﬂoppy disk on the CD-ROM (it may be as big
as 2.88 MiB), which the BIOS ﬁnds and boots; the other method is to boot
directly oﬀ the CD-ROM, which requires a specialised boot loader (such as
ISOLINUX for Linux).
8.1 Fundamentals 129

B With suitable hardware and software (usually part of the ﬁrmware today),
a PC can boot via the network. The kernel, root ﬁle system, and everything
else can reside on a remote server, and the computer itself can be diskless
and hence ear-friendly. The details would be a bit too involved and are irrel-
evant for LPIC-1 in any case; if necessary, look for keywords such as “PXE”
or “Linux Terminal Server Project”.

UEFI boot procedure UEFI-based systems do not use boot sectors. Instead, the
UEFI ﬁrmware itself contains a boot manager which exploits information about
the desired operating system which is held in non-volatile RAM (NVRAM). Boot
loaders for the diﬀerent operating systems on the computer are stored as regular
ﬁles on an “EFI system partition” (ESP), where the ﬁrmware can read and start
them. The system either ﬁnds the name of the desired boot loader in NVRAM, or
else falls back to the default name, /EFI/BOOT/BOOTX64.EFI . (The X64 here stands for
“64-bit Intel-style PC”. Theoretically, UEFI also works for 32-bit systems, but that
doesn’t mean it is a great idea.) The operating-system speciﬁc boot loader then
takes care of the rest, as in the BIOS startup procedure.

B The ESP must oﬃcially contain a FAT32 ﬁle system (there are Linux distri-
butions that use FAT16, but that leads to problems with Windows 7, which
requires FAT32). A size of 100 MiB is generally suﬃcient, but some UEFI
implementations have trouble with FAT32 ESPs which are smaller than
512 MiB, and the Linux mkfs command will default to FAT16 for partitions
of up to 520 MiB. With today’s prices for hard disks, there is little reason
not to play it safe and create an ESP of around 550 MiB.

B In principle it is possible to simply write a complete Linux kernel as BOOTX64.
EFI on the ESP and thus manage without any boot loader at all. PC-based
Linux distributions don’t usually do this, but this approach is interesting for
embedded systems.

B Many UEFI-based systems also allow BIOS-style booting from MBR-parti-
tioned disks, i. e., with a boot sector. This is called “compatibility support
module” or CSM. Sometimes this method is used automatically if a tradi-
tional MBR is found on the ﬁrst recognised hard disk. This precludes an
UEFI boot from an ESP on an MBR-partitioned disk and is not 100% ideo-
logically pure.

B UEFI-based systems boot from CD-ROM by looking for a ﬁle called /EFI/
BOOT/BOOTX64.EFI —like they would for disks. (It is feasible to produce CD-
ROMs that boot via UEFI on UEFI-based systems and via El Torito on BIOS-
based systems.)

“UEFI Secure Boot” is supposed to prevent computers being infected with UEFI Secure Boot
“root kits” that usurp the startup procedure and take over the system before the
actual operating system is being started. Here the ﬁrmware refuses to start boot
loaders that have not been cryptographically signed using an appropriate key. Ap-
proved boot loaders, in turn, are responsible for only launching operating system
kernels that have been cryptographically signed using an appropriate key, and
approved operating system kernels are expected to insist on correct digital sig-
natures for dynamically loadable drivers. The goal is for the system to run only
“trusted” software, at least as far as the operating system is concerned.

B A side eﬀect is that this way one gets to handicap or exclude potentially un-
desirable operating systems. In principle, a company like Microsoft could
exert pressure on the PC industry to only allow boot loaders and operating
systems signed by Microsoft; since various anti-trust agencies would take a
dim view to this, it is unlikely that such a step would become part of oﬃ-
cial company policy. It is more likely that the manufacturers of PC mother-
boards and UEFI implementations concentrate their testing and debugging
130 8 Booting Linux

eﬀorts on the “boot Windows” application, and that Linux boot loaders will
be diﬃcult or impossible to get to run simply due to inadvertent ﬁrmware
bugs.

Linux supports UEFI Secure Boot in various ways. There is a boot loader called
Shim “Shim” (developed by Matthew Garrett) which a distributor can have signed by
Microsoft. UEFI starts Shim and Shim then starts another boot loader or operating
system kernel. These can be signed or unsigned; the security envisioned by UEFI
Secure Boot is, of course, only obtainable with the signatures. You can install your
own keys and then sign your own (self-compiled) kernels.

B The details for this would be carrying things too far. Consult the Linup
Front training manual Linux System Customisation

PreLoader An alternative to Shim is “PreLoader” (by James Bottomley, distributed by the
Linux Foundation). PreLoader is simpler than Shim and makes it possible to ac-
credit a (possibly unsigned) subsequent boot loader with the system, and boot it
later without further enquiries.

Hard disks: MBR vs. GPT The question of which partitioning scheme a hard
disk is using and the question of whether the computer boots via the BIOS (or
CSM) or UEFI really don’t have a lot to do with each other. At least with Linux it
is perfectly possible to boot a BIOS-based system from a GPT-partitioned disk or
a UEFI-based system from an MBR-partitioned disk (the latter possibly via CSM).

B To start a BIOS-based system from a GPT-partitioned disk it makes sense to
create a “BIOS boot partition” to hold that part of the boot loader that does
not ﬁt into the MBR. The alternative—using the empty space between the
MBR and the start of the ﬁrst partition—is not reliable for GPT-partitioned
disks, since the GPT partition table takes up at least part of this space and/
or the ﬁrst partition might start immediately after the GPT partition table.
The BIOS boot partition does not need to be huge at all; 1 MiB is probably
amply enough.

After the boot loader The boot loader loads the Linux operating system kernel
and passes the control to it. With that, it is itself extraneous and can be removed
from the system; the ﬁrmware, too, will be ignored from now on—the kernel is
left to its own devices. In particular, it must be able to access all drivers required
to initialise the storage medium containing the root ﬁle system, as well as that ﬁle
system itself (the boot loader used the ﬁrmware to access the disk), typically at
least a driver for an IDE, SATA, or SCSI controller and the ﬁle system in question.
These drivers must be compiled into the kernel or—the preferred method today—
will be taken from “early userspace”, which can be conﬁgured without having to
recompile the kernel. (As soon as the root ﬁle system is available, everything is
peachy because all drivers can be read from there.) The boot loader’s tasks also
include reading the early-userspace data.

B The “early userspace” used to be called an “initial RAM disk”, because the
data was read into memory en bloc as a (usually read-only) medium, and
treated by the kernel like a block-oriented disk. There used to be special
compressed ﬁle systems for this application. The method most commonly
used today stipulates that the early-userspace data is available as a cpio
archive which the kernel extracts directly into the disk block cache, as if
you had read each ﬁle in the archive directly from a (hypothetical) storage
medium. This makes it easier to get rid of the early userspace once it is no
longer required.

B The kernel uses cpio instead of tar because cpio archives in the format used
by the kernel are better-standardised and easier to unpack than tar archives.
8.2 GRUB Legacy 131

As soon as the “early userspace” is available, a program called /init is invoked.
This is in charge of the remaining system initialisation, which includes tasks such
as the identiﬁcation of the storage medium that should be made available as the
root ﬁle system, the loading of any required drivers to access that medium and the
ﬁle system (these drivers, of course, also come from early userspace), possibly the
(rudimentary) conﬁguration of the network in case the root ﬁle system resides on
a remote ﬁle server, and so on. Subsequently, the early userspace puts the desired
root ﬁle system into place at “/ ” and transfers control to the actual init program—
today most often either System-V init (chapter 9) or systemd (chapter 10), in each
case under the name of /sbin/init . (You can juse the kernel command line option
init= to pick a diﬀerent program.)

B If no early userspace exists, the operating system kernel makes the storage
medium named on its command line using the root= option available as the
root ﬁle system, and starts the program given by the init= option, by default
/sbin/init .

Exercises
C 8.1 [2] Whereabouts on an MBR-partitioned hard disk may a boot loader
reside? Why?

8.2 GRUB Legacy
8.2.1 GRUB Basics
Many distributions nowadays use GRUB as their standard boot loader. It has var-
ious advantages compared to LILO, most notably the fact that it can handle the
common Linux ﬁle systems. This means that it can read the kernel directly from a
ﬁle such as /boot/vmlinuz , and is thus immune against problems that can develop if
you install a new kernel or make other changes to your system. Furthermore, on
the whole GRUB is more convenient—for example oﬀering an interactive GRUB GRUB shell
shell featuring various commands and thus allowing changes to the boot setup
for special purposes or in case of problems.

A The GRUB shell allows access to the ﬁle system without using the usual
access control mechanism. It should therefore never be made available to
unauthorised people, but be protected by a password (on important com-
puters, at least). See also Section 8.2.5.

Right now there are two widespread versions of GRUB: The older version
(“GRUB Legacy”) is found in older Linux distributions—especially those with an
“enterprise” ﬂavour’—, while the newer distributions tend to rely on the more
modern version GRUB 2 (section 8.2.4).
The basic approach taken by GRUB Legacy follows the procedure outlined in
section 8.1. During a BIOS-based startup, the BIOS ﬁnds the ﬁrst part (“stage 1”)
of the boot loader in the MBR of the boot disk (all 446 bytes of it). Stage 1 is able
to ﬁnd the next stage based on sector lists stored inside the program (as part of
the 446 bytes) and the BIOS disk access functions1 .
The “next stage” is usually stage 1.5, which is stored in the otherwise un-
used space immediately after the MBR and before the start of the ﬁrst partition.
Stage 1.5 has rudimentary support for Linux ﬁle systems and can ﬁnd GRUB’s
“stage 2” within the ﬁle system (normally below /boot/grub ). Stage 2 may be any-
where on the disk. It can read ﬁle systems, too, and it fetches its conﬁguration
ﬁle, displays the menu, and ﬁnally loads and starts the desired operating system
(in the case of Linux, possibly including the “early userspace”).
1 At least as long as the next stage can be found within the ﬁrst 1024 “cylinders” of the disk. There

are historical reasons for this and it can, if necessary, be enforced through appropriate partitioning.
132 8 Booting Linux

B Stage 1 could read stage 2 directly, but this would be subject to the same
restrictions as reading stage 1.5 (no ﬁle system access and only within the
ﬁrst 1024 cylinders). This is why things aren’t usually arranged that way.

B GRUB can directly load and start most Unix-like operating systems for x86
computers, including Linux, Minix, NetBSD, GNU Hurd, Solaris, Reac-
tOS, Xen, and VMware ESXi2 . The relevant standard is called “multiboot”.
GRUB starts multiboot-incompatible systems (notably Windows) by invok-
ing the boot loader of the operating system in question—a procedure called
“chain loading”.

To make GRUB Legacy work with GPT-partitioned disks, you need a BIOS boot
partition to store its stage 1.5. There is a version of GRUB Legacy that can deal with
UEFI systems, but for UEFI boot you are generally better oﬀ using a diﬀerent boot
loader.

8.2.2 GRUB Legacy Configuration
/boot/grub/menu.lst The main conﬁguration ﬁle for GRUB Legacy is usually stored as /boot/grub/menu.
lst . It contains basic conﬁguration as well as the settings for the operating systems
to be booted. This ﬁle might look as follows:

default 1
timeout 10

title linux
kernel (hd0,1)/boot/vmlinuz root=/dev/sda2
initrd (hd0,1)/boot/initrd
title failsafe
kernel (hd0,1)/boot/vmlinuz.bak root=/dev/sda2 apm=off acpi=off
initrd (hd0,1)/initrd.bak
title someothersystem
root (hd0,2)
makeactive
chainloader +1
title floppy
root (fd0)
chainloader +1

The individual parameters have the following meaning:
default Denotes the default system to be booted. Caution: GRUB counts from 0!
Thus, by default, the conﬁguration above launches the failsafe entry.
timeout This is how many seconds the GRUB menu will be displayed before the
default entry will be booted.
title Opens an operating system entry and speciﬁes its name, which will be dis-
played within the GRUB menu.
kernel Speciﬁes the Linux kernel to be booted. (hd0, 1)/boot/vmlinuz , for example,
means that the kernel is to be found in /boot/vmlinuz on the ﬁrst partition of
the zeroth hard disk, thus in our example, for linux , on /dev/hda2 . Caution:
The zeroth hard disk is the ﬁrst hard disk in the BIOS boot order! There is
no distinction between IDE and SCSI! And: GRUB starts counting at 0 …
Incidentally, GRUB takes the exact mapping of the individual drives from
the device.map ﬁle.
After the kernel location, arbitrary kernel parameters can be passed. This
includes the boot= entry.
2 The “U” in GRUB must stand for something, after all.
8.2 GRUB Legacy 133

initrd Denotes the location of the cpio archive used for the “early userspace”.
root Determines the system partition for foreign operating systems. You can also
specify media that only occasionally contain something bootable, such as
the ﬂoppy disk drive—this will let you boot from ﬂoppy even though the
ﬂoppy disk is disabled in the BIOS boot order.
chainloader +1 Denotes the boot loader to be loaded from the foreign system’s sys-
tem partition. Generally this is the content of that partition’s boot loader.
makeactive Marks the speciﬁed partition temporarily as “bootable”. Some operat-
ing systems (not Linux) require this in order to be able to boot oﬀ the par-
tition in question. By the way: GRUB supports a few more such directives,
for example map , which makes it possible to fool a system into believing it
is installed on a diﬀerent hard disk (than, e. g., the often disdained second
disk) than it actually is.

8.2.3 GRUB Legacy Installation
Here “installation” does not refer to the installation of an RPM package but the
installation of the GRUB boot sector, or stage 1 (and very likely the stage 1.5). This
is very seldom required, for example during the original installation of the system
(where the installation procedure of your distribution will do it for you).
The installation is done using the grub command, which invokes the GRUB
shell. It is most convenient to use a “batch” ﬁle, since otherwise you would have to
start from the very beginning after an erroneous input. Some distributions (e. g.,
those by SUSE/Novell) already come with a suitable ﬁle. In this case, the instal-
lation procedure might look like

# grub --batch --device-map=/boot/grub/device.map < /etc/grub.inst

The --device-map option creates a device.map ﬁle under the speciﬁed name, if none
exists already.
The /etc/grub.inst ﬁle could have the following content: /etc/grub.inst

root (hd0,1)
setup (hd0)
quit

Here, root denotes the partition containing GRUB’s “home directory” (usually
/boot/grub —the other parts of GRUB necessary for the installation will be looked
for in this directory).

A The partition you specify using root here has nothing to do with the partition
containing your Linux distribution’s root directory, which you specify using
root= in your Linux kernels’ menu entries. At least not necessarily. See also
Section 8.3.
setup installs GRUB on the speciﬁed device, here in hd0 ’s MBR. GRUB’s setup
command is a simpliﬁed version of a more general command called install , which
should work in most cases.

B Alternatively, you may use the grub-install script to install the GRUB com- grub-install
ponents. This comes with some distributions.
Inside the GRUB shell it is straightforward to ﬁgure out how to specify a hard disk specification
disk in the root or kernel directives. The GRUB shell command find is useful here:
# grub

grub> find /boot/vmlinuz
(hd0,1)
134 8 Booting Linux

8.2.4 GRUB 2
new implementation GRUB 2 is a completely new implementation of the boot loader that did not make
particular concessions to GRUB-Legacy compatibility. GRUB 2 was oﬃcially re-
leased in June 2012, even though various distributions used earlier versions by
default.

The LPIC-1 certiﬁcate requires knowledge of GRUB 2 from version 3.5 of the
exam (starting on 2 July 2012).

As before, GRUB 2 consists of several stages that build on each other:
• Stage 1 (boot.img ) is placed inside the MBR (or a partition’s boot sector) on
BIOS-based systems. It can read the ﬁrst sector of stage 1.5 by means of the
BIOS, and that in turn will read the remainder of stage 1.5.
• Stage 1.5 (core.img ) goes either between the MBR and the ﬁrst partition
(on MBR-partitioned disks) or else into the BIOS boot partition (on GPT-
partitioned disks). Stage 1.5 consists of a ﬁrst sector which is tailored to
the boot medium (disk, CD-ROM, network, …) as well as a “kernel” that
provides rudimentary functionality like device and ﬁle access, processing
a command line, etc., and an arbitrary list of modules.

B This modular structure makes it easy to adapt stage 1.5 to size restric-
tions.

• GRUB 2 no longer includes an explicit stage 2; advanced functionality will
be provided by modules and loaded on demand by stage 1.5. The modules
can be found in /boot/grub , and the conﬁguration ﬁle in /boot/grub/grub.cfg .

B On UEFI-based systems, the boot loader sits on the ESP in a ﬁle called
EFI/ ⟨operating system⟩/grubx64.efi , where ⟨operating system⟩ is something
like debian or fedora . Have a look at the /boot/efi/EFI directory on your
UEFI”=based Linux system.

B Again, the “x64 ” in “grubx64.efi ” stands for “64-bit PC”.
configuration file The conﬁguration ﬁle for GRUB 2 looks markedly diﬀerent from that for GRUB
Legacy, and is also rather more complicated (it resembles a bash script more than
a GRUB Legacy conﬁguration ﬁle). The GRUB 2 authors assume that system man-
agers will not create and maintain this ﬁle manually. Instead there is a command
grub-mkconfig called grub-mkconfig which can generate a grub.cfg ﬁle. To do so, it makes use of
a set of auxiliary tools (shell scripts) in /etc/grub.d , which, e. g., search /boot for
Linux kernels to add to the GRUB boot menu. (grub-mkconfig writes the new con-
update-grub ﬁguration ﬁle to its standard output; the update-grub command calls grub-mkconfig
and redirects its output to /boot/grub/grub.cfg .)
You should therefore not modify /boot/grub/grub.cfg directly, since your distri-
bution is likely to invoke update-grub after, e. g., installing a kernel update, which
would overwrite your changes to grub.cfg .
Usually you can, for instance, add more items to the GRUB 2 boot menu by
editing the /etc/grub.d/40_custom ﬁle. grub-mkconfig will copy the content of this ﬁle
verbatim into the grub.cfg ﬁle. As an alternative, you could add your conﬁguration
settings to the /boot/grub/custom.cfg ﬁle, which will be read by grub.cfg if it exists.
For completeness’ sake, here is an excerpt from a typical grub.cfg ﬁle. By anal-
ogy to the example in Section 8.2.2, a menu entry to start Linux might look like
this for GRUB 2:
menuentry 'Linux' --class gnu-linux --class os {
insmod gzio
insmod part_msdos
insmod ext2
8.3 Kernel Parameters 135

set root='(hd0,msdos2)'
linux /boot/vmlinuz root=/dev/hda2
initrd /boot/initrd.img
}

(grub-mkconfig usually produces more complicated stuﬀ.) Do note that the GRUB
modules for decompression (gzio ), for MS-DOS-like partitioning support (part_msdos )
and the ext2 ﬁle system must be loaded explicitly. With GRUB 2, partition num-
bering starts at 1 (it used to be 0 for GRUB Legacy), so (hd0,msdos2) refers to the
second MS-DOS partition on the ﬁrst hard disk. Instead of kernel , linux is used to
start a Linux kernel.

8.2.5 Security Advice
The GRUB shell oﬀers many features, in particular access to the ﬁle system with-
out the root password! Even entering boot parameters may prove dangerous since boot parameters
it is easy to boot Linux directly into a root shell. GRUB makes it possible to close
these loopholes by requiring a password. password
For GRUB Legacy, the password is set in the menu.lst ﬁle. Here, the entry
“password --md5 ⟨encrypted password⟩” must be added to the global section. You
can obtain the encrypted password via the grub-md5-crypt command (or “md5crypt ”
within the GRUB shell) and then use, e. g., the GUI to “copy and paste” it to the ﬁle.
Afterwards, the password will need to be input whenever something is changed
interactively in the GRUB menu.

B You can also prevent particular systems from being booted by adding the
lock option to the appropriate speciﬁc section within menu.lst . GRUB will
query for the password when that system is to be booted. All other systems
can still be started without a password.

Exercises
C 8.2 [2] Which ﬁle contains your boot loader’s conﬁguration? Create a new
entry that will launch another operating system. Make a backup copy of the
ﬁle ﬁrst.

C 8.3 [!3] Prevent a normal user from circumventing init and booting directly
into a shell. How do you generate a password request when a particular
operating system is to be booted?

8.3 Kernel Parameters
Linux can accept a command line from the boot loader and evaluate it during the
kernel start procedure. The parameters on this command line can conﬁgure de-
vice drivers and change various kernel options. This mechanism for Linux kernel Linux kernel runtime configura-
runtime conﬁguration is particularly helpful with the generic kernels on Linux tion
distribution boot disks, when a system with problematic hardware needs to be
booted. To do this, LILO supports the append= …option, while GRUB lets you ap-
pend parameters to the kernel speciﬁcation.
Alternatively, you can enter parameters interactively as the system is being
booted. You may have to grab GRUB’s attention quickly enough (e. g., by press-
ing a cursor or shift key while the boot menu or splash screen is displayed). Af-
terwards you can navigate to the desired menu entry and type e . GRUB then
presents you with the desired entry, which you can edit to your heart’s content
before continuing the boot operation.
There are various types of parameters. The ﬁrst group overwrites hardcoded
defaults, such as root or rw . Another group of parameters serves to conﬁgure de- configuring device drivers
136 8 Booting Linux

vice drivers. If one of these parameters occurs on the command line, the initial-
isation function for the device driver in question is called with the arguments
speciﬁed there rather than the built-in default values.

B Nowadays most Linux distributions use modular kernels that have only
very few device drivers built in. Modular device drivers cannot be con-
ﬁgured from the kernel command line.

B During booting, if there are problems with a device driver that is built into
the kernel, you can usually disable this driver by specifying the number 0
as the parameter for the corresponding boot parameter.

general settings Finally, there are parameters governing general settings. These include, e. g.,
init or reserve .We shall be discussing some typical parameters from the multitude
of possible settings. Further parameters can be found within the kernel sources’
documentation area. Speciﬁc details for particular hardware must be researched
in the manual or on the Internet.
ro This causes the kernel to mount the root partition read-only
rw This causes the kernel to mount the root partition with writing enabled, even if
the kernel executable or the boot loader conﬁguration ﬁle specify otherwise
init= ⟨program⟩ Runs ⟨program⟩ (e. g., /bin/bash ) instead of the customary /sbin/init
⟨runlevel⟩ Boots into runlevel ⟨runlevel⟩, where ⟨runlevel⟩ is generally a number
between 1 and 5. Otherwise the initial runlevel is taken from /etc/inittab .
(Irrelevant for computers running systemd.)
single Boots to single-user mode.
maxcpus= ⟨number⟩ On a multi-processor (or, nowadays, multi-core) system, use
only as many CPUs as speciﬁed. This is useful for troubleshooting or per-
formance measurements.
mem= ⟨size⟩ Speciﬁes the amount of memory to be used. On the one hand, this is
useful if the kernel cannot recognise the correct size by itself (fairly unlikely
these days) or you want to check how the system behaves with little mem-
ory. The ⟨size⟩ is a number, optionally followed by a unit (“TokenG” for
gibibytes, “M ” for mebibytes, or “K ” for kibibytes).

A A typical mistake is something like mem=512 . Linux is thrifty about sys-
tem resources, but even it can’t quite squeeze itself into 512 bytes (!) of
RAM.

panic= ⟨seconds⟩ Causes an automatic reboot after ⟨seconds⟩ in case of a catastrophic
system crash (called a “kernel panic” in the patois, Edsger Dijkstra’s dictum,
“The use of anthropomorphic terminology when dealing with computing
systems is a symptom of professional immaturity”, notwithstanding).
hd 𝑥=noprobe Causes the kernel to ignore the disk-like device /dev/hd 𝑥 (IDE disk, CD-
ROM, …) completely. It is not suﬃcient to disable the device in the BIOS, as
Linux will ﬁnd and access it even so.
noapic and similar parameters like nousb , apm=off , and acpi=off tell Linux not to use
certain kernel features. These options can help getting Linux to run at all
on unusual computers, in order to analyse problems in these areas more
thoroughly and sort them out.
A complete list of all parameters available on the kernel command line is given in
the ﬁle Documentation/kernel- parameters.txt , which is part of the Linux source code.
(However, before you install kernel sources just to get at this ﬁle, you should prob-
ably look for it on the Internet.)
8.4 System Startup Problems 137

B Incidentally, if the kernel notices command-line options that do not corre-
spond to kernel parameters, it passes them to the init process as environ- init environment variables
ment variables.

8.4 System Startup Problems
8.4.1 Troubleshooting
Usually things are simple: You switch on the computer, stroll over to the coﬀee
machine (or not—see Section 9.1), and when you come back you are greeted by
the graphical login screen. But what to do if things don’t work out that way?
The diagnosis of system startup problems sometimes isn’t all that easy—all
sorts of messages zoom by on the screen or (with some distributions) are not dis-
played at all, but hidden behind a nice little picture. The system logging service
(syslogd ) is also started only after a while. Fortunately, though, Linux does not
leave you out in the cold if you want to look at the kernel boot messages at leisure.
For logging purposes, the system startup process can be divided into two
phases. The “early” phase begins with the ﬁrst signs of life of the kernel and
continues up to the instant where the system logging service is activated. The
“late” phase begins just then and ﬁnishes in principle when the computer is shut
down.
The kernel writes early-phase messages into an internal buﬀer that can be dis-
played using the dmesg command. Various distributions arrange for these messages
to be passed on to the system logging service as soon as possible so they will show
up in the “oﬃcial” log.
The system logging service, which we are not going to discuss in detail here,
runs during the “late” phase. It will be covered in the Linup Front training man-
ual, Linux Administration II (and the LPI-102 exam). For now it will be suﬃcient
to know that most distribution place most messages sent to the system logging
service into the /var/log/messages ﬁle. This is also where messages from the boot
process end up if they have been sent after the logging service was started.

On Debian GNU/Linux, /var/log/messages contains only part of the system
messages, namely anything that isn’t a grave error message. If you would
like to see everything you must look at /var/log/syslog —this contains all mes-
sages except (for privacy reasons) those dealing with authentication. The
“early phase” kernel messages, too, incidentally.

B Theoretically, messages sent after init was started but before the system log-
ging service was launched might get lost. This is why the system logging
service is usually among the ﬁrst services started after init .

8.4.2 Typical Problems
Here are some of the typical snags you might encounter on booting:
The computer does not budge at all If your computer does nothing at all, it
probably suﬀers from a hardware problem. (If you’re diagnosing such a
case by telephone, then do ask the obvious questions such as “Is the power
cable plugged into the wall socket?”—perhaps the cleaning squad was des-
perate to ﬁnd a place to plug in their vacuum cleaner—, and “Is the power
switch at the back of the case switched to On?”. Sometimes the simple
things will do.) The same is likely when it just beeps or ﬂashes its LEDs
rhythmically but does not appear to actually start booting.

B The beeps or ﬂashes can allow the initiated to come up with a rough di-
agnosis of the problem. Details of hardware troubleshooting, though,
are beyond the scope of this manual.
138 8 Booting Linux

Things go wrong before the boot loader starts The ﬁrmware performs various
self-tests and outputs error messages to the screen if things are wrong (such
as malfunctioning RAM chips). We shall not discuss how to ﬁx these prob-
lems. If everything works ﬁne, your computer ought to identify the boot
disk and launch the boot loader.

The boot loader does not ﬁnish This could be because the operating system can-
not ﬁnd it (e. g., because the drive it resides on does not show up in the
ﬁrmware boot order) or it is damaged. In the former case you should ensure
that your ﬁrmware does the Right Thing (not our topic). In the latter case
you should receive at least a rudimentary error message, which together
with the boot loader’s documentation should allow you to come up with an
explanation.

B GRUB as a civilised piece of software produces clear-text error mes-
sages which are explained in more detail in the GRUB info documen-
tation.

The cure for most of the fundamental (as opposed to conﬁguration-related)
boot loader problems, if they cannot be obviously reduced to disk or BIOS
errors, consist of booting the system from CD-ROM—the distribution’s
“rescue system” or a “live distribution” such as Knoppix recommend
themselves—and to re-install the boot loader.

B The same applies to problems like a ruined partition table in the MBR.
Should you ever accidentally overwrite your MBR, you can restore a
backup (you do have one, don’t you?) using dd or re-instate the par-
titioning using sfdisk (you do have a printout of your partition table
stashed away somewhere, don’t you?) and rewrite the boot loader.

B In case of the ultimate worst-case partition table scenario, there are
programs which will search the whole disk looking for places that look
like ﬁle system superblocks, and (help) recover the partition scheme
that way. We’re keeping our ﬁngers crossed on your behalf that you
will never need to run such a program.

The kernel doesn’t start Once the boot loader has done its thing the kernel
should at least start (which generally leads to some activity on the screen).
Distribution kernels are generic enough to run on most PCs, but there may
still be problems, e. g., if you have a computer with extremely modern hard-
ware which the kernel doesn’t yet support (which is fatal if, for example, a
driver for the disk controller is missing) or you have messed with the initial
RAM disk (Shame, if you didn’t know what you were doing!). It may be
possible to reconﬁgure the BIOS (e. g., by switching a SATA disk controller
into a “traditional” IDE-compatible mode) or to deactivate certain parts of
the kernel (see Section 8.3) in order to get the computer to boot. It makes
sense to have another computer around so you can search the Internet for
help and explanations.

B If you are fooling around with the kernel or want to install a new ver-
sion of your distribution kernel, do take care to have a known-working
kernel around. If you always have a working kernel in your boot
loader menu, you can save yourself from the tedious job of slinging
CDs about.

Other problems Once the kernel has ﬁnished its initialisations, it hands control
oﬀ to the “init” process. You will ﬁnd out more about this in Chapter 9.
However, you should be out of the woods by then.
8.4 System Startup Problems 139

8.4.3 Rescue systems and Live Distributions
As a system administrator, you should always keep a “rescue system” for your
distribution handy, since usually you need it exactly when you are least in a posi-
tion to obtain it quickly. (This applies in particular if your Linux machine is your
only computer.) A rescue system is a pared-down version of your distribution
which you can launch from a CD or DVD (formerly a ﬂoppy disk or disks) and
which runs in a RAM disk.

B Should your Linux distribution not come with a separate rescue system on
ﬂoppy disk or CD, then get a “live distribution” such as Knoppix. Live dis-
tributions are started from CD (or DVD) and work on your computer with-
out needing to be installed ﬁrst. You can ﬁnd Knoppix as an ISO image on
http://www.knoppix.de/ or, every so often, as a freebie with computer maga-
zines.

The advantage of rescue systems and live distributions consists in the fact that
they work without involving your hard disk. Thus you can do things like fsck
your root ﬁle system, which are forbidden while your system is running from
hard disk. Here are a few problems and their solutions:
Hosed the kernel? Boot the rescue system and re-install the corresponding pack-
age. In the simplest case, you can enter your installed system’s root ﬁle from
the rescue system like so:

# mount -o rw /dev/sda1 /mnt Device name may diﬀer
# chroot /mnt
# _ We are now seeing the installed distribution

After this you can activate the network interface or copy a kernel package
from a USB key or CD-ROM and install it using the package management
tool of your distribution.

Forgot the root password? Boot the rescue system and change to the installed dis-
tribution as above. Then do
# passwd

(You could of course ﬁx this problem without a rescue system by restarting
your system with “init=/bin/bash rw ” as a kernel parameter.)

B Live distributions such as Knoppix are also useful to check in the computer
store whether Linux supports the hardware of the computer you have been
drooling over for a while already. If Knoppix recognises a piece of hardware,
you can as a rule get it to run with other Linux distributions too. If Knoppix
does not recognise a piece of hardware, this may not be a grave problem
(there might be a driver for it somewhere on the Internet that Knoppix does
not know about) but you will at least be warned.

B If there is a matching live version of your distribution—with Ubuntu, for
example, the live CD and the installation CD are identical—, things are es-
pecially convenient, since the live distribution will typically recognise the
same hardware that the installable distribution does.
140 8 Booting Linux

Commands in this Chapter
dmesg Outputs the content of the kernel message buﬀer dmesg (8) 137
grub-md5-crypt Determines MD5-encrypted passwords for GRUB Legacy
grub-md5-crypt (8) 135

Summary
• A boot loader is a program that can load and start an operating system.
• A boot manager is a boot lader that lets the user pick one of several operating
systems or operating system installations.
• GRUB is a powerful boot manager with special properties—such as the pos-
sibility of accessing arbitrary ﬁles and a built-in command shell.
• The GRUB shell helps to install GRUB as well as to conﬁgure individual boot
procedures.
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

9
System-V Init and the Init Process

Contents
9.1 The Init Process . . . . . . . . . . . . . . . . . . . . 142
9.2 System-V Init . . . . . . . . . . . . . . . . . . . . . 142
9.3 Upstart . . . . . . . . . . . . . . . . . . . . . . . 148
9.4 Shutting Down the System . . . . . . . . . . . . . . . . 150

Goals
• Understanding the System-V Init infrastructure
• Knowing /etc/inittab structure and syntax
• Understanding runlevels and init scripts
• Being able to shut down or restart the system orderly

Prerequisites
• Basic Linux system administration knowledge
• Knowledge of system start procedures (Chapter 8)

adm1-init.tex (33e55eeadba676a3 )
142 9 System-V Init and the Init Process

9.1 The Init Process
After the ﬁrmware, the boot loader, the operating system kernel and (possibly)
the early userspace have done their thing, the “init process” takes over the reins.
Its job is to ﬁnish system initialisation and supervise the ongoing operation of the
/sbin/init system. For this, Linux locates and starts a program called /sbin/init .

B The init process always has process ID 1. If there is an early userspace, it in-
herits this from the process that was created to run /init , and subsequently
goes on to replace its program text by that of the init process.

B Incidentally, the init process enjoys a special privilege: it is the only pro-
cess that cannot be aborted using “kill -9 ”. (It can decide to shuﬄe oﬀ this
mortal coil of its own accord, though.)

B If the init process really quits, the kernel keeps running. There are purists
who start a program as the init process that will set up packet ﬁltering rules
and then exit. Such a computer makes a practically impregnable ﬁrewall,
but is somewhat inconvenient to reconﬁgure without a rescue system …

B You can tell the Linux kernel to execute a diﬀerent program as the init pro-
cess by specifying an option like “init=/sbin/myinit ” on boot. There are no
special properties that this program must have, but you should remember
that, if it ever ﬁnishes, you do not get another one without a reboot.

9.2 System-V Init
Basics The traditional infrastructure that most Linux distributions used to use
System-V init is called “System-V init” (pronounced “sys-ﬁve init”). The “V” is a roman nu-
meral 5, and it takes its name from the fact that it mostly follows the example of
Unix System V, where something very similar showed up for the ﬁrst time. That
was during the 1980s.

B For some time there was the suspicion that an infrastructure designed ap-
proximately 30 years ago was no longer up to today’s demands on a Linux
computer’s init system. (Just as a reminder: When System-V init was new,
the typical Unix system was a VAX with 30 serial terminals.) Modern com-
puters must, for example, be able to deal with frequent changes of hard-
ware (cue USB), and that is something that System-V init ﬁnds relatively
diﬃcult to handle. Hence there were several suggestions for alternatives
to System-V init. One of these—systemd by Lennart Poettering and Kay
Sievers—seems to have won out and is the current or upcoming standard of
practically all Linux distributions of importance (we discuss it in more detail
in chapter 10). Another is Upstart by Scott James Remnant (see section 9.3).

runlevels One of the characteristic features of System-V init are runlevels, which describe
the system’s state and the services that it oﬀers. Furthermore, the init process en-
sures that users can log in on virtual consoles, directly-connected serial terminals,
etc., and manages system access via modems if applicable. All of this is conﬁgured
by means of the /etc/inittab ﬁle
/etc/inittab The syntax of /etc/inittab (Figure 9.1), like that of many other Linux conﬁgu-
ration ﬁles, is somewhat idiosyncratic (even if it is really AT&T’s fault). All lines
that are not either empty or comments— starting with “# ” as usual—consist of
four ﬁelds separated by colons:
Label The ﬁrst ﬁeld’s purpose is to identify the line uniquely. You may pick an
arbitrary combination of up to four characters. (Do yourself a favour and
stick with letters and digits.) The label is not used for anything else.
9.2 System-V Init 143

# Standard runlevel
id:5:initdefault

# First script to be executed
si::bootwait:/etc/init.d/boot

# runlevels
l0:0:wait:/etc/init.d/rc 0
l1:1:wait:/etc/init.d/rc 1
l2:2:wait:/etc/init.d/rc 2
l3:3:wait:/etc/init.d/rc 3
#l4:4:wait:/etc/init.d/rc 4
l5:5:wait:/etc/init.d/rc 5
l6:6:wait:/etc/init.d/rc 6

ls:S:wait:/etc/init.d/rc S
~~:S:respawn:/sbin/sulogin

# Ctrl-Alt-Del
ca::ctrlaltdel:/sbin/shutdown -r -t 4 now

# Terminals
1:2345:respawn:/sbin/mingetty --noclear tty1
2:2345:respawn:/sbin/mingetty tty2
3:2345:respawn:/sbin/mingetty tty3
4:2345:respawn:/sbin/mingetty tty4
5:2345:respawn:/sbin/mingetty tty5
6:2345:respawn:/sbin/mingetty tty6

# Serial terminal
# S0:12345:respawn:/sbin/agetty -L 9600 ttyS0 vt102

# Modem
# mo:235:respawn:/usr/sbin/mgetty -s 38400 modem

Figure 9.1: A typical /etc/inittab ﬁle (excerpt)
144 9 System-V Init and the Init Process

B This is not 100% true for lines describing terminals, where according
to convention the label corresponds to the name of the device ﬁle in
question, but without the “tty ” at the beginning, hence 1 for tty1 or S0
for ttyS0 . Nobody knows exactly why.

Runlevels The runlevels this line applies to. We haven’t yet explained in detail
how runlevels work, so excuse us for the moment for limiting ourselves to
telling you that they are usually named with digits and the line in question
will be considered in all runlevels whose digit appears in this ﬁeld.

B In addition to the runlevels with digits as names there is one called “S ”.
More details follow below.

Action The third ﬁeld speciﬁes how to handle the line. The most important pos-
sibilities include
respawn The process described by this line will immediately be started again
once it has ﬁnished. Typically this is used for terminals which, after the
current user is done with their session, should be presented brand-new
to the next user.
wait The process described by this line is executed once when the system
changes to the runlevel in question, and init waits for it to ﬁnish.
bootwait The process described by this line will be executed once during
system startup. init waits for it to ﬁnish. The runlevel ﬁeld on this line
will be ignored.
initdefault The runlevel ﬁeld of this line speciﬁes which runlevel the system
shoud try to reach after booting.
B With LSB-compliant distributions, this ﬁeld usually says “5 ” if the
system should accept logins on the graphical screen, otherwise
“3 ”. See below for details.
B If this entry (or the whole ﬁle /etc/inittab ) is missing, you will need
to state a run level on the console.
ctrlaltdel Speciﬁes what the system should do if the init process is being
sent a SIGINT —which usually happens if anyone presses the Ctrl + Alt
+ Del combination. Normally this turns out to be some kind of shutdown
(see Section 9.4).

B There are a few other actions. powerwait , powerfail , powerokwait , and
powerfailnow , for example, are used to interface System-V init with
UPSs. The details are in the documentation (init (8) and inittab (5)).

Command The fourth ﬁeld describes the command to be executed. It extends to
the end of the line and you can put whatever you like.
If you have made changes to /etc/inittab , these do not immediately take eﬀect.
You must execute the “telinit q ” command ﬁrst in order to get init to reread the
conﬁguration ﬁle.

The Boot Script With System-V init, the init process starts a shell script, the boot
script, typically /etc/init.d/boot (Novell/SUSE), /etc/rc.d/init.d/boot (Red Hat), or
/etc/init.d/rcS (Debian). (The exact name occurs in /etc/inittab ; look for an entry
whose action is bootwait .)
The boot script performs tasks such as checking and possibly correcting the
ﬁle systems listed in /etc/fstab , initialising the system name and Linux clock, and
other important prerequisites for stable system operation. Next, kernel modules
will be loaded if required, ﬁle systems mounted and so on. The speciﬁc actions
and their exact order depend on the Linux distribution in use.
9.2 System-V Init 145

B Today, boot usually conﬁnes itself to executing the ﬁles in a directory such as
/etc/init.d/boot.d (SUSE) in turn. The ﬁles are executed in the order of their
names. You can put additional ﬁles in this directory in order to execute
custom code during system initialisation.

Exercises
C 9.1 [2] Can you ﬁnd out where your distribution keeps the scripts that the
boot script executes?

C 9.2 [!2] Name a few typical tasks performed by the boot script. In which
order should these be executed?

Runlevels After executing the boot script, the init process attempts to place the
system in one of the various runlevels. Exactly which one is given by /etc/inittab runlevels
or determined at the system’s boot prompt and passed through to init by the
kernel.
The various runlevels and their meaning have by now been standardised across standardised runlevels
most distributions roughly as follows:
1 Single-user mode with no networking

2 Multi-user mode with no network servers
3 Multi-user mode with network servers
4 Unused; may be conﬁgured individually if required

5 As runlevel 3, but with GUI login
6 Reboot
0 System halt

B The system runs through the S (or s ) runlevel during startup, before it
changes over to one out of runlevels 2 to 5. If you put the system into
runlevel 1 you will ﬁnally end up in runlevel S .

When the system is started, the preferred runlevels are 3 or 5—runlevel 5 is typical
for workstations running a GUI, while runlevel 3 makes sense for server systems
that may not even contain a video interface. In runlevel 3 you can always start a
GUI afterwards or redirect graphics output to another computer by logging into
your server from that machine over the network.

These predeﬁned runlevels derive from the LSB standard. Not all distribu-
tions actually enforce them; Debian GNU/Linux, for example, mostly leaves
runlevel assignment to the local administrator.

B You may use runlevels 7 to 9 but you will have to conﬁgure them yourself.
During system operation, the runlevel can be changed using the telinit com- telinit command
mand. This command can only be executed as root : “telinit 5 ” changes imme-
diately to runlevel ⟨runlevel⟩. All currently running services that are no longer
required in the new runlevel will be stopped, while non-running services that are
required in the new runlevel will be started.

B You may use init in place of telinit (the latter is just a symbolic link to the
former, anyway). The program checks its PID when it starts, and if it is not 1,
it behaves like telinit , else init .

The runlevel command displays the previous and current runlevel: runlevel
146 9 System-V Init and the Init Process

# runlevel
N 5

Here the system is currently in runlevel 5, which, as the value “N ” for the “previous
runlevel” suggests, was entered right after the system start. Output such as “5 3 ”
would mean that the last runlevel change consisted of bringing the system from
runlevel 5 to runlevel 3.

B We have concealed some more runlevels from you, namely the “on-demand
runlevels” A , B , and C . You may make entries in /etc/inittab which are meant
for any of these three runlevels and use the ondemand action, such as

xy:AB:ondemand:…

If you say something like

# telinit A

these entries are executed, but the actual runlevel does not change: If you
were in runlevel 3 before the telinit command, you will still be there when
it ﬁnishes. a , b , and c are synonyms for A , B , and C.

Exercises
C 9.3 [!2] Display the current runlevel. What exactly is being output? Change
to runlevel 2. Check the current runlevel again.

C 9.4 [2] Try the on-demand runlevels: Add a line to /etc/inittab which ap-
plies to, e. g., runlevel A . Make init reread the inittab ﬁle. Then enter the
»telinit A « command.

Init Scripts The services available in the various runlevels are started and
stopped using the scripts in the /etc/init.d (Debian, Ubuntu, SUSE) or /etc/
rc.d/init.d (Red Hat) directories. These scripts are executed when changing from
one runlevel to another, but may also be invoked manually. You may also add
init scripts your own scripts. All these scripts are collectively called init scripts.
The init scripts>parametersinit scripts usually support parameters such asstart ,
stop , status , restart , or reload , which you can use to start, stop, …, the correspond-
ing services. The “/etc/init.d/network restart ” command might conceivably deac-
tivate a system’s network cards and restart them with an updated conﬁguration.
Of course you do not need to start all services manually when the system is
runlevel directories started or you want to switch runlevels. For each runlevel 𝑟 there is a rc 𝑟.d di-
rectory in /etc (Debian and Ubuntu), /etc/rc.d (Red Hat), or /etc/init.d (SUSE).
The services for each runlevel and the runlevel transitions are deﬁned in terms of
these directories, which contain symbolic links to the scripts in the init.d direc-
tory. These links are used by a script, typically called /etc/init.d/rc , to start and
stop services when a runlevel is entered or exited.
This is done according to the names of the links, in order to determine the start-
ing and stopping order of the services. There are dependencies between various
services—there would not be much point in starting network services such as the
Samba or web servers before the system’s basic networking support has been ac-
Activating services tivated. The services for a runlevel are activated by calling all symbolic links in
its directory that start with the letter “S ”, in lexicographical order with the start
parameter. Since the link names contain a two-digit number after the “S ”, you
can predetermine the order of invocation by carefully choosing these numbers.
Accordingly, to deactivate the services within a runlevel, all the symbolic links
starting with the letter “K ” are called in lexicographical order with the stop pa-
rameter.
9.2 System-V Init 147

If a running service is also supposed to run in the new run level, an extraneous
restart can be avoided. Therefore, before invoking a K link, the rc script checks
whether there is an S link for the same service in the new runlevel’s directory. If
so, the stopping and immediate restart are skipped.

Debian GNU/Linux takes a diﬀerent approach: Whenever a new runlevel
𝑟 is entered, all symbolic links in the new directory (/etc/rc 𝑟.d ) are executed.
Links beginning with a “K ” are passed stop and links beginning with a “S ”
are passed start as the parameter.

To conﬁgure services in a runlevel or to create a new runlevel, you can in princi- Configuring services
ple manipulate the symbolic links directly. However, most distributions deprecate
this.

The Red Hat distributions use a program called chkconfig to conﬁgure run-
levels. “chkconfig quota 35 ”, for example, inserts the quota service not in run-
level 35, but runlevels 3 and 5. “chkconfig -l ” gives a convenient overview
of the conﬁgured runlevels.

The SUSE distributions use a program called insserv to order the services
in each runlevel. It uses information contained in the init scripts to calcu-
late a sequence for starting and stopping the services in each runlevel that
takes the dependencies into account. In addition, YaST2 oﬀers a graphical
“runlevel editor”, and there is a chkconfig program which however is just a
front-end for insserv .

Nor do you have to create links by hand on Debian GNU/Linux—you
may use the update-rc.d program. However, manual intervention is still
allowed—update-rc.d ’s purpose is really to allow Debian packages to inte-
grate their init scripts into the boot sequence. With the

# update-rc.d mypackage defaults

command, the /etc/init.d/mypackage script will be started in runlevels 2, 3, 4,
and 5 and stopped in runlevels 0, 1 and 6. You can change this behaviour by
means of options. If you do not specify otherwise, update-rc.d uses the se-
quence number 20 to calculate the position of the service—contrary to SUSE
and Red Hat, this is not automated.—The insserv command is available on
Debian GNU/Linux as an optional package; if it is installed, it can man-
age at least those init scripts that do contain the necessary metadata like it
would on the SUSE distributions. However, this has not been implemented
throughout.

Exercises
C 9.5 [!2] What do you have to do to make the syslog service reread its conﬁg-
uration?

C 9.6 [1] How can you conveniently review the current runlevel conﬁgura-
tion?

C 9.7 [!2] Remove the cron service from runlevel 2.

Single-User Mode In single-user mode (runlevel S ), only the system administra- single-user mode
tor may work on the system console. There is no way of changing to other virtual
consoles. The single-user mode is normally used for administrative work, espe-
cially if the ﬁle system needs to be repaired or the quota system set up.
148 9 System-V Init and the Init Process

B You can mount the root ﬁle system read-only on booting, by passing the S
option on the kernel command line. If you boot the system to single-user
mode, you can also disable writing to the root ﬁle system “on the ﬂy”, using
the remount and ro mount options: “mount -o remount,ro / ” remounts the root
partition read-only; “mount -o remount,rw / ” undoes it again.

B To remount a ﬁle system “read-only” while the system is running, no pro-
cess may have opened a ﬁle on the ﬁle system for writing. This means that
all such programs must be terminated using kill . These are likely to be
daemons such as syslogd or cron .

It depends on your distribution whether or not you get to leave single-user
mode, and how.

To leave single-user mode, Debian GNU/Linux recommends a reboot
rather than something like »telinit 2 «. This is because entering single-
user mode kills all processes that are not required in signle-user mode.
This removes some essential background processes that were started when
the system passed through runlevel S during boot, which is why it is unwise
to change from runlevel S to a multi-user runlevel.

Exercises
C 9.8 [!1] Put the system into single-user mode (Hint: telinit ). What do you
need to do to actually enter single-user mode?

C 9.9 [1] Convince yourself that you really are the single user on the system
while single-user mode is active, and that no background processes are run-
ning.

9.3 Upstart
While System-V init traditionally stipulates a “synchronous” approach—the init
system changes its state only through explicit user action, and the steps taken
during a state change, like init scripts, are performed in sequence—, Upstart uses
an “event-based” philosophy. This means that the system is supposed to react to
external events (like plugging in an USB device). This happens “asynchronously”.
Starting and stopping services creates new events, so that—and that is one of the
most important diﬀerences between System-V init and Upstart—a service can be
restarted automatically if it crashes unexpectedly. (System-V init, on the other
hand, wouldn’t be bothered at all.)
Upstart has been deliberately designed to be compatible with System-V init, at
least to a point where init scripts for services can be reused without changes.

Upstart was developed by Scott James Remnant, at the time an employee of
Canonical (the company behind Ubuntu) and accordingly debuted in that
distributon. Since Ubuntu 6.10 (“Edgy Eft”) it is the standard init system on
Ubuntu, although it used to be run in a System-V compatible mode at ﬁrst;
since Ubuntu 9.10 (“Karmic Koala”) it is running in “native” mode.

It turns out that Ubuntu is currently in the process of switching over to sys-
temd (see chapter 10).

Since version 3.5 of the LPIC-1 certiﬁcate exams (as of 2 July 2012) you are
expected to know that Upstart exists and what its major properties are. Con-
ﬁguration and operational details are not required.
9.3 Upstart 149

# rsyslog - system logging daemon
#
# rsyslog is an enhanced multi-threaded replacement for the traditional
# syslog daemon, logging messages from applications

description "system logging daemon"

start on filesystem
stop on runlevel [06]

expect fork
respawn

exec rsyslogd -c4

Figure 9.2: Upstart conﬁguration ﬁle for job rsyslog

B Upstart is also purported to accelerate the boot process by being able to
initialise servides in parallel. In actual practice this isn’t the case, as the
limiting factor during booting is, for the most part, the speed with which
blocks of data can be moved from disk to RAM. At the Linux Plumbers
Conference 2008, Arjan van de Ven and Auke Kok demonstrated that it is
possible to boot an Asus EeePC all the way to a usable desktop (i. e., not a
Windows-like desktop with a churning hard disk in the background) within
5 seconds. This work was based on System-V init rather than Upstart.

Upstart conﬁguration is based on the idea of “Jobs” that take on the role of Jobs
init scripts (although init scripts, as we mentioned, are also supported). Upstart
distinguishes “tasks”—jobs that run for a limited time and then shut themselves
down—and “services”—jobs that run permanently “in the background”.

B Tasks can be long-running, too. The main criterion is that services—think
of a mail, database, or web server—do not terminate of their own accord
while tasks do.

Jobs are conﬁgured using ﬁles within the /etc/init directory. The names of these
ﬁles derive from the job name and the “.conf ” suﬃx. See ﬁgure 9.2 for an example.
One of the main objectives of Upstart is to avoid the large amounts of template-
like code typical for most System-V init scripts. Accordingly, the Upstart conﬁgu-
ration ﬁle conﬁnes itself to stating how the service is to be started (“exec rsyslogd
-c4 ”). In addition, it speciﬁes that the service is to be restarted in case it crashes
(“respawn ”) and how Upstart can ﬁnd out which process to track (“expect fork ” says
that the rsyslog process puts itself into the background by creating a child process
and then exiting—Upstart must then watch out for that child process).—Compare
this to /etc/init.d/syslogd (or similar) on a typical Linux based on System-V init.
While with “classic” System-V init the system administrator assigns a “global”
order in which the init scripts for a particular runlevel are to be executed, with
Upstart the jobs decide “locally” where they want to place themselves within a
network of dependencies. The “start on …” and “stop on …” lines stipulate events
that lead to the job being started or stopped. In our example, rsyslog is started as
soon as the ﬁle system is available, and stopped when the system transitions to
the “runlevels” 0 (halt) or 6 (reboot). System-V init’s runlevel directories with
symbolic links are no longer required.

B Upstart supports runlevels mostly for compatibility with Unix tradition and
to ease the migration of System-V init based systems to Upstart. They are
not required in principle, but at the moment are still necessary to shut down
the system (!).
150 9 System-V Init and the Init Process

B Newer implementations of System-V init also try to provide dependencies
between services in the sense that init script 𝑋 is always executed after init
script 𝑌 and so on. (This amounts to a scheme for automatic assignment
of the priority numbers within the runlevel directories.) This is done using
metadata contained in standardised comments at the beginning of the init
scripts. The facilities that this approach provides do fall short of those of
Upstart, though.
On system boot, Upstart creates the startup event as soon as its own initialisa-
tion is complete. This makes it possible to execute other jobs. The complete boot
sequence derives from the startup event and from events being created through
the execution of further jobs and expected by others.

B For example, on Ubuntu 10.04 the startup event invokes the mountall task
which makes the ﬁle systems available. Once that is ﬁnished, the filesystem
event is created (among others), which in turn triggers the start of the rsyslog
service from Figure 9.2.
With Upstart, the initctl command is used to interact with the init process:
# initctl list Which jobs are running now?
alsa-mixer-save stop/waiting
avahi-daemon start/running, process 578
mountall-net stop/waiting
rc stop/waiting
rsyslog start/running, process 549

# initctl stop rsyslog Stop a job
rsyslog stop/waiting
# initctl status rsyslog What is its status?
rsyslog stop/waiting
# initctl start rsyslog Restart a job
rsyslog start/running, process 2418
# initctl restart rsyslog Stop and start
rsyslog start/running, process 2432

B The “initctl stop ”, “initctl start ”, “initctl status ”, and “initctl stop ” can
be abbreviated to “stop ”, “start ”, ….

9.4 Shutting Down the System
A Linux computer should not simply be powered oﬀ, as that could lead to data
loss—possibly there are data in RAM that ought to be written to disk but are still
waiting for the proper moment in time. Besides, there might be users logged in on
the machine via the network, and it would be bad form to surprise them with an
unscheduled system halt or restart. The same applies to users taking advantage
of services that the computer oﬀers on the Net.

B It is seldom necessary to shut down a Linux machine that should really run
continuously. You can install or remove software with impunity and also re-
conﬁgure the system fairly radically without having to restart the operating
system. The only cases where this is really necessary include kernel changes
(such as security updates) or adding new or replacing defective hardware
inside the computer case.

B The ﬁrst case (kernel changes) is being worked on. The kexec infrastructure
makes it possible to load a second kernel into memory and jump into it
directly (without the detour via a system reboot). Thus it is quite possible
that in the future you will always be able to run the newest kernel without
actually having to reboot your machine.
9.4 Shutting Down the System 151

B With the correct kind of (expensive) hardware you can also mostly sort out
the second case: Appropriate server systems allow you to swap CPUs, RAM
modules, and disks in and out while the computer is running.

There are numerous ways of shutting down or rebooting the system:
• By valiantly pushing the system’s on/oﬀ switch. If you keep it pressed until on/off switch
the computer is audibly shutting down the system will be switched oﬀ. You
should only do this in cases of acute panic (ﬁre in the machine hall or a
sudden water inﬂux).
• Using the shutdown command. This is the cleanest method of shutting down shutdown
or rebooting.
• For System-V init: The “telinit 0 ” command can be used to switch to run-
level 0. This is equivalent to a shutdown.
• Using the halt command. This is really a direct order to the kernel to halt the
system, but many distributions arrange for halt to call shutdown if the system
is not in runlevels 0 or 6 already.

B There is a reboot command for reboots, which like halt usually relies on reboot
shutdown . (In fact, halt and reboot are really the same program.)

The commands are all restricted to the system administrator.

B The key combination Ctrl + Alt + Del may also work if it is conﬁgured ap-
propriately in /etc/inittab (see Section 9.1).

B Graphical display managers often oﬀer an option to shut down or reboot
the system. You may have to conﬁgure whether the root password must be
entered or not.

B Finally, modern PCs may interpret a (short) press on the on/oﬀ switch as
“Please shut down cleanly” rather than “Please crash right now”.

Normally you will be using the second option, the shutdown command. It en-
sures that all logged-in users are made aware of the impending system halt, pre-
vents new logins, and, according to its option, performs any requisite actions to
shut down the system:

# shutdown -h +10

for example will shut down the system in ten minutes’ time. With the -r option,
the system will be restarted. With no option, the system will go to single-user
mode after the delay has elapsed.

B You may also give the time of shutdown/reboot as an absolute time:
# shutdown -h 12:00 High Noon

B For shutdown , the now keyword is a synonym of “+0 ”—immediate action. Do
it only if you are sure that nobody else is using the system.

Here is exactly what happens when the shutdown command is given:
1. All users receive a broadcast message saying that the system will be shut broadcast message
down, when, and why.
2. The shutdown command automatically creates the /etc/nologin ﬁle, which is
checked by login (or, more precisely, the PAM infrastructure); its existence
prevents new user logins (except for root ).
152 9 System-V Init and the Init Process

B For consolation, users that the system turns away are being shown the
content of the /etc/nologin ﬁle.

The ﬁle is usually removed automatically when the system starts up again.
3. The system changes to runlevel 0 or 6. All services will be terminated by
means of their init scripts (more exactly, all services that do not occur in
runlevels 0 or 6, which is usually all of them).
4. All still-running processes are ﬁrst sent SIGTERM . They may intercept this sig-
nal and clean up after themselves before terminating.
5. Shortly afterwards, all processes that still exist are forcibly terminated by
SIGKILL .

6. The ﬁle systems are unmounted and the swap spaces are deactivated.
7. Finally, all system activities are ﬁnished. Then either a warm start is initi-
ated or the computer shut oﬀ using APM or ACPI. If that doesn’t work, the
message “System halted ” is displayed on the console. At that point you can
hit the switch yourself.

B You may pass some text to shutdown after the shut-down delay time, which
is displayed to any logged-in users:

# shutdown -h 12:00 '
System halt for hardware upgrade.
Sorry for the inconvenience!
'

B If you have executed shutdown and then change your mind after all, you can
cancel a pending shutdown or reboot using

# shutdown -c "No shutdown after all"

(of course you may compose your own explanatory message).

By the way: The mechanism that shutdown uses to notify users of an impending
wall system halt (or similar) is available for your use. The command is called wall (short
for “write to all”):

$ wall "Cake in the break room at 3pm!"

will produce a message of the form

Broadcast message from hugo@red (pts/1) (Sat Jul 18 00:35:03 2015):

Cake in the break room at 3pm!

on the terminals of all logged-in users.

B If you send the message as a normal user, it will be received by all users who
haven’t blocked their terminal for such messages using “mesg n ”. If you want
to reach those users, too, you must send the message as root .

B Even if you’re not logged in on a text terminal but are instead using a graphi-
cal environment: Today’s desktop environments will pick up such messages
and show them in an extra window (or something; that will depend on the
desktop environment).

B If you’re root and the parameter of wall looks like the name of an existing
ﬁle, that ﬁle will be read and its content sent as the message:
9.4 Shutting Down the System 153

# echo "Cake in the break room at 3pm!" >cake.txt
# wall cake.txt

You don’t get to do this as an ordinary user, but you can still pass the mes-
sage on wall ’s standard input. (You can do that as root , too, of course.) Don’t
use this for War and Peace.

B If you’re root , you can suppress the header line “Broadcast message …” using
the -n option (short for --nobanner ).

Exercises
C 9.10 [!2] Shut down your system 15 minutes from now and tell your users
that this is simply a test. How do you prevent the actual shutdown (so that
it really is simply a test)?

B What happens if you (as root ) pass wall the name of a non-existent ﬁle as its
parameter?

C 9.11 [2] wall is really a special case of the write command, which you can use
to “chat” with other users of the same computer in an unspeakably primitive
fashion. Try write , in the easiest case between two diﬀerent users in diﬀerent
windows or consoles. (write was a lot more interesting back when one had
a VAX with 30 terminals.)

Commands in this Chapter
chkconfig Starts or shuts down system services (SUSE, Red Hat)
chkconfig (8) 147
halt Halts the system halt (8) 151
initctl Supervisory tool for Upstart initctl (8) 150
insserv Activates or deactivates init scripts (SUSE) insserv (8) 147
reboot Restarts the computer reboot (8) 151
runlevel Displays the previous and current run level runlevel (8) 145
shutdown Shuts the system down or reboots it, with a delay and warnings for
logged-in users shutdown (8) 151
update-rc.d Installs and removes System-V style init script links (Debian)
update-rc.d (8) 147

Summary
• After starting, the kernel initialises the system and then hands oﬀ control to
the /sbin/init program as the ﬁrst userspace process.
• The init process controls the system and takes care, in particular, of acti-
vating background services and managing terminals, virtual consoles, and
modems.
• The system distinguishes various “runlevels” (operating states) that are de-
ﬁned through diﬀerent sets of running services.
• A single-user mode is available for large or intrusive administrative tasks.
• The shutdown command is a convenient way of shutting down or rebooting
the system (and it’s friendly towards other users, too).
• You can use the wall command to send a message to all logged-in users.
• Linux systems seldom need to be rebooted—actually only when a new op-
erating system kernel or new hardware has been installed.
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

10
Systemd

Contents
10.1 Overview. . . . . . . . . . . . . . . . . . . . . . . 156
10.2 Unit Files . . . . . . . . . . . . . . . . . . . . . . . 157
10.3 Unit Types . . . . . . . . . . . . . . . . . . . . . . 161
10.4 Dependencies . . . . . . . . . . . . . . . . . . . . . 162
10.5 Targets. . . . . . . . . . . . . . . . . . . . . . . . 164
10.6 The systemctl Command . . . . . . . . . . . . . . . . . 166
10.7 Installing Units. . . . . . . . . . . . . . . . . . . . . 169

Goals
• Understanding the systemd infrastructure
• Knowing the structure of unit ﬁles
• Understanding and being able to conﬁgure targets

Prerequisites
• Knowledge of Linux system administration
• Knowledge of system start procedures (Chapter 8)
• Knowledge about System-V init (Chapter 9)

adm1-systemd.tex (33e55eeadba676a3 )
156 10 Systemd

10.1 Overview
Systemd, by Lennart Poettering and Kay Sievers, is another alternative to the old-
fashioned System-V init system. Like Upstart, systemd transcends the rigid lim-
itations of System-V init, but implements various concepts for the activation and
control of services with much greater rigour than Upstart.

B Systemd is considered the future standard init system by all mainstream
distributions. On many of them—such as Debian, Fedora, RHEL, CentOS,
openSUSE, and SLES—it is now provided by default. Even Ubuntu, origi-
nally the main instigator of Upstart, has by now declared for systemd.
While System-V init and Upstart use explicit dependencies among services—
for instance, services using the system log service can only be started once that
dependencies service is running—, systemd turns the dependencies around. A service requiring
the system log service doesn’t do this because the log service needs to be running,
but because it itself wants to send messages to the system log. This means it must
access the communication channel that the system log service provides. Hence it
is suﬃcient if systemd itself creates that communication channel and passes it to
the system log service once that becomes available—the service wanting to send
messages to the log will wait until its messages can actually be accepted. Hence,
systemd can in principle create all communication channels ﬁrst and then start
all services simultaneously without regard to any dependencies whatsoever. The
dependencies will sort themselves out without any explicit conﬁguration.

B This approach also works when the system is running: If a service is ac-
cessed that isn’t currently running, systemd can start it on demand.

B The same approach can in principle also be used for ﬁle systems: If a service
wants to open a ﬁle on a ﬁle system that is currently unavailable, the access
is suspended until the ﬁle system can actually be accessed.
units Systemd uses “units” as an abstraction for parts of the system to be managed
targets such as services, communication channels, or devices. “Targets” replace SysV
init’s runlevels and are used to collect related units. For example, there is a target
multiuser.target that corresponds to the traditional runlevel 3. Targets can depend
on the availability of devices—for instance, a bluetooth.target could be requested
when a USB Bluetooth adapter is plugged in, and it could launch the requisite
software. (System-V init starts the Bluetooth software as soon as it is conﬁgured,
irrespective of whether Bluetooth hardware is actually available.)
In addition, systemd oﬀers more interesting properties that System-V init and
Upstart cannot match, including:
• Systemd supports service activation “on demand”, not just depending on
hardware that is recognised (as in the Bluetooth example above), but also
via network connections, D-Bus requests or the availability of certain paths
within the ﬁle system.
• Systemd allows very ﬁne-grained control of the services it launches, con-
cerning, e. g., the process environment, resource limits, etc. This includes
security improvements, e. g., providing only a limited view on the ﬁle sys-
tem for certain services, or providing services with a private /tmp directory
or networking environment.

B With SysV init this can be handled on a case-by-case basis within the
init scripts, but by comparison this is very primitive and tedious.

• Systemd uses the Linux kernel’s cgroups mechanism to ensure, e. g., that
stopping a service actually stops all related processes.
• If desired, systemd handles services’ logging output; the services only need
to write messages to standard output.
10.2 Unit Files 157

• Systemd makes conﬁguration maintenance easier, by cleanly separating dis-
tribution default and local customisations.
• Systemd contains a number of tools in C that handle system initialisation
and do approximately what distribution-speciﬁc “runlevel S” scripts would
otherwise do. Using them can speed up the boot process considerably and
also improves cross-distribution standardisation.
Systemd is designed to oﬀer maximum compatibility with System-V init and other
“traditions”. For instance, it supports the init scripts of System-V init if no na-
tive conﬁguration ﬁle is available for a service, or it takes the ﬁle systems to be
mounted on startup from the /etc/fstab ﬁle.
You can use the systemctl command to interact with a running systemd, e. g.,
to start or stop services explicitly:

# systemctl status rsyslog.service
● rsyslog.service - System Logging Service
Loaded: loaded (/lib/systemd/system/rsyslog.service; enabled)
Active: active (running) since Do 2015-07-16 15:20:38 CEST;
3h 12min ago
Docs: man:rsyslogd(8)
http://www.rsyslog.com/doc/
Main PID: 497 (rsyslogd)
CGroup: /system.slice/rsyslog.service
└─497 /usr/sbin/rsyslogd -n
# systemctl stop rsyslog.service
Warning: Stopping rsyslog.service, but it can still be activated by:
syslog.socket
# systemctl start rsyslog.service

Systemd calls such change requests for the system state “jobs”, and puts them into
a queue.

B Systemd considers status change requests “transactions”. If a unit is being transactions
started or stopped, it (and any units that depend on it) are put into a tempo-
rary transaction. Then systemd checks that the transaction is consistent—in
particular, that no circular dependencies exist. If that isn’t the case, systemd
tries to repair the transaction by removing jobs that are not essential in order
to break the cycle(s). Non-essential jobs that would lead to running services
being stopped are also removed. Finally, systemd checks whether the jobs
within the transaction would conﬂict with other jobs that are already in the
queue, and refuses the transaction if that is the case. Only if the transaction
is consistent and the minimisation of its impact on the system is complete,
will its jobs be entered into the queue.

Exercises
C 10.1 [!1] Use “systemctl status ” to get a picture of the units that are active on
your computer. Check the detailed status of some interesting-looking units.

10.2 Unit Files
One of the more important advantages of systemd is that it uses a uniﬁed ﬁle
format for all conﬁguration ﬁles—no matter whether they are about services to
be started, devices, communication channels, ﬁle systems, or other artefacts that
systemd manages.
158 10 Systemd

B This is in stark contrast to the traditional infrastructure based on System-V
init, where almost every functionality is conﬁgured in a diﬀerent way: per-
manently running background services in /etc/inittab , runlevels and their
services via init scripts, ﬁle systems in /etc/fstab , services that are run on
demand in /etc/inetd.conf , … Every single such ﬁle is syntactically diﬀerent
from all others, while with systemd, only the details of the possible (and
sensible) conﬁguration settings diﬀer—the basic ﬁle format is always the
same.

A very important observation is: Unit ﬁles are “declarative”. This means that
they simply describe what the desired conﬁguration looks like—unlike System V
init’s init scripts, which contain executable code that tries to achieve the desired
conﬁguration.

B Init scripts usually consider of huge amounts of boilerplate code which de-
pends on the distribution in question, but which you still need to read and
understand line-per-line if there is a problem or you want to do something
unusual. For somewhat more complex background services, init scripts of a
hundred lines or more are not unusual. Unit ﬁles for systemd, though, usu-
ally get by with a dozen lines or two, and these lines are generally pretty
straightforward to understand.

B Of course unit ﬁles occasionally contain shell commands, for example to
explain how a speciﬁc service should be started or stopped. These, however,
are generally fairly obvious one-liners.

Syntax The basic syntax of unit ﬁles is explained in systemd.unit (5). You can ﬁnd
an example for a unit ﬁle in ﬁgure 10.1. A typical characteristic is the subdivision
into sections that start with a title in square brackets1 . All unit ﬁles (no matter
what they are supposed to do) can include [Unit] and [Install] sections (see be-
low). Besides, there are sections that are speciﬁc to the purpose of the unit.
As usual, blank lines and comment lines are ignored. Comment lines can start
with a # or ; . Over-long lines can be wrapped with a \ at the end of the line,
which will be replaced by a space character when the ﬁle is read. Uppercase and
lowercase letters are important!
Lines which are not section headers, empty lines, nor comment lines contain
options “options” according to a “⟨name⟩ = ⟨value⟩” pattern. Various options may occur
several times, and systemd’s handling of that depends on the option: Multiple
options often form a list; if you specify an empty value, all earlier settings will be
ignored. (If that is the case, the documentation will say so.)

B Options that are not listed in the documentation will be ﬂagged with a warn-
ing by systemd and otherwise ignored. If a section or option name starts
with “X- ”, it is ignored completely (options in an “X- ” section do not need
their own “X- ” preﬁx).

B Yes/no settings in unit ﬁles can be given in a variety of ways. 1 , true , yes ,
and on stand for “yes”, 0 , false , no , and off for “no”.

B Times can also be speciﬁed in various ways. Simple integers will be inter-
preted as seconds2 . If you append a unit, then that unit applies (allowed
units include us , ms , s , min , h , d , w in increasing sequence from microseconds
to weeks—see systemd.time (7)). You can concatenate several time speciﬁca-
tions with units (as in “10min 30s ”), and these times will be added (here,
630 seconds).
1 The syntax is inspired by the .desktop ﬁles of the “XDG Desktop Entry Speciﬁcation” [XDG-DS14],

which in turn have been inspired by the INI ﬁles of Microsoft Windows.
2 Most of the time, anyway—there are (documented) exceptions.
10.2 Unit Files 159

# This file is part of systemd.
#
# systemd is free software; you can redistribute it and/or modify
# it under the terms of the GNU Lesser General Public License as
# published by the Free Software Foundation; either version 2.1
# of the License, or (at your option) any later version.

[Unit]
Description=Console Getty
Documentation=man:agetty(8)
After=systemd-user-sessions.service plymouth-quit-wait.service
After=rc-local.service
Before=getty.target

[Service]
ExecStart=-/sbin/agetty --noclear --keep-baud console
115200,38400,9600 $TERM
Type=idle
Restart=always
RestartSec=0
UtmpIdentifier=cons
TTYPath=/dev/console
TTYReset=yes
TTYVHangup=yes
KillMode=process
IgnoreSIGPIPE=no
SendSIGHUP=yes

[Install]
WantedBy=getty.target

Figure 10.1: A systemd unit ﬁle: console- getty.service
160 10 Systemd

Searching and finding settings Systemd tries to locate unit ﬁles along a list of
directories that is hard-coded in the program. Directories nearer the front of the
list have precedence over directories nearer the end.

B The details are system-dependent, at least to a certain degree. The usual list
is normally something like

/etc/systemd/system Local conﬁguration
/run/systemd/system Dynamically generated unit ﬁles
/lib/systemd/system Unit ﬁles for distribution packages

Local customisation Systemd oﬀers various clever methods for customising settings without having
to change the unit ﬁles generally provided by your distribution—which would be
inconvenient if the distribution updates the unit ﬁles. Imagine you want to change
a few settings in the example.service ﬁle:
• You can copy the distribution’s example.service ﬁle from /lib/systemd/system
to /etc/systemd/system and make any desired customisations. The unit ﬁle
furnished by the distribution will then not be considered at all.
• You can create a directory /etc/systemd/system/example.service.d containing a
ﬁle—for example, local.conf . The settings in that ﬁle override settings with
the same name in /lib/systemd/system/example.service , but any settings not
mentioned in local.conf stay intact.

B Take care to include any required section titles in local.conf , such that
the options can be identiﬁed correctly.

B Nobody keeps you from putting several ﬁles into /etc/systemd/system/
example.service.d . The only prerequisite is that ﬁle names must end in
.conf . Systemd makes no stipulation about the order in which these
ﬁles are read—it is best to ensure that every option occurs in just one
single ﬁle.

Template unit files Sometimes several services can use the same or a very similar
unit ﬁle. In this case it is convenient not to have to maintain several copies of
the same unit ﬁle. Consider, for example, the terminal deﬁnition lines in /etc/
inittab —it would be good not to have to have one unit ﬁle per terminal.
Systemd supports this by means of unit ﬁles with names like example@.service .
Instantiation You could, for example, have a ﬁle called getty@.service and then conﬁgure a vir-
tual console on /dev/tty2 simply by creating a symbolic link from getty@tty2.service
to getty@.service . When this console is to be activated, systemd reads the getty@
.service ﬁle and replaces the %I key, wherever it ﬁnds it, by whatever comes be-
tween @ and . in the name of the unit ﬁle, i. e., tty2 . The result of that replacement
is then put into force as the conﬁguration.

B In fact, systemd replaces not just %I but also some other sequences (and that
not just in template unit ﬁles). The details may be found in systemd.unit (5),
in the “Speciﬁers” section.

Basic settings All unit ﬁles may contain the [Unit] and [Install] sections. The
former contains general information about the unit, the latter provides details for
its installation (for example, to introduce explicit dependencies—which we shall
discuss later).
Here are some of the more important options from the [Unit] section (the com-
plete list is in systemd.unit (5)):
Description A description of the unit (as free text). Will be used to make user in-
terfaces more friendly.
10.3 Unit Types 161

Documentation A space-separated list of URLs containing documentation for the
unit. The allowed protocol schemes include http: , https: , file: , info: , and
man: (the latter three refer to locally-installed documentation). An empty
value clears the list.
OnFailure A space-separated list of other units which will be activated if this unit
transitions into the failed state.
SourcePath The path name of a conﬁguration ﬁle from which this unit ﬁle has been
generated. This is useful for tools that create unit ﬁles for systemd from
external conﬁguration ﬁles.
ConditionPathExists Checks whether there is a ﬁle (or directory) under the given
absolute path name. If not, the unit will be classed as failed . If there is a
! in front of the path name, then a ﬁle (or directory) with that name must
not exist. (There are loads of other “Condition …” tests—for example, you can
have the execution of units depend on whether the system has a particular
computer architecture, is running in a virtual environment, is running on
AC or battery power or on a computer with a particular name, and so on.
Read up in systemd.unit (5).)

Exercises
C 10.2 [!2] Browse the unit ﬁles of your system under /lib/systemd/system (or
/usr/lib/systemd/system , depending on the distribution). How many diﬀerent
Condition … options can you ﬁnd?

10.3 Unit Types
Systemd supports a wide variety of “units”, or system components that it can
manage. These are easy to tell apart by the extensions of the names of the corre-
sponding unit ﬁles. As mentioned in Section 10.2, all units share the same basic
ﬁle format. Here is a list of the most important unit types:
.service A process on the computer that is executed and managed by systemd.
This includes both background services that stay active for a long time (pos-
sibly until the system is shut down), and processes that are only executed
once (for example when the system is booting).

B When a service is invoked by name (such as example ) but no correspond-
ing unit ﬁle (here, example.service ) can be found, systemd looks for a
System-V init script with the same name and generates a service unit
for that on the ﬂy. (The compatibility is fairly wide-ranging but not
100% complete.)

.socket A TCP/IP or local socket, i. e., a communication end point that client pro-
grams can use to contact a server. Systemd uses socket units to activate
background services on demand.

B Socket units always come with a corresponding service unit which will
be started when systemd notes activity on the socket in question.

.mount A “mount point” on the system, i. e., a directory where a ﬁle system should
be mounted.

B The names of these units are derived from the path name by means
of replacing all slashes (“/ ”) with hyphens (“- ”) and all other non-
alphanumeric (as per ASCII) characters with a hexadecimal replace-
ment such as \x2d (“. ” is only converted if it is the ﬁrst charac-
ter of a path name). The name of the root directory (“/ ”) becomes
162 10 Systemd

“- ”, but slashes at the start or end of all other names are removed.
The directory name /home/lost+found , for instance, becomes home- lost\
textbackslash x2bfound .

B You can try this replacement using the “systemd-escape -p ” command:
$ systemd-escape -p /home/lost+found
-home-lost\x2bfound
$ systemd-escape -pu home-lost\\x2bfound
/home/lost+found

The “-p ” option marks the parameter as a path name. The “-u ” option
undoes the replacement.

.automount Declares that a mount point should be mounted on demand (instead
of prophylactically when the system is booted). The names of these units
result from the same path name transformation. The details of mounting
must be described by a corresponding mount unit.
.swap Describes swap space on the system. The names of these units result from
the path name transformation applied to the device or ﬁle name in question.
.target A “target”, or synchronisation point for other units during system boot
or when transitioning into other system states. Vaguely similar to System-V
init’s runlevels. See section 10.5.
.path Observes a ﬁle or a directory and starts another unit (by default, a service
unit of the same name) when, e. g., changes to the ﬁle have been noticed or
a ﬁle has been added to an otherwise empty directory.

.timer Starts another unit (by default, a service unit of the same name) at a cer-
tain point in time or repeatedly at certain intervals. This makes systemd a
replacement for cron and at .
(There are a few other unit types, but to explain all of them here would be carrying
things too far.)

Exercises
C 10.3 [!2] Look for examples for all of these units on your system. Examine
the unit ﬁles. If necessary, consult the manpages for the various types.

10.4 Dependencies
As we have mentioned before, systemd can mostly get by without explicit depen-
dencies because it is able to exploit implicit dependencies (e. g., on communication
channels). Even so, it is sometimes necessary to specify explicit dependencies.
Various options in the [Unit] section of a service ﬁle (e.g., example.service ) allow
you to do just that. For example:
Requires Speciﬁes a list of other units. If the current unit is activated, the listed
units are also activated. If one of the listed units is deactivated or its acti-
vation fails, then the current unit will also be deactivated. (In other words,
the current unit “depends on the listed units”.)

B The Requires dependencies have nothing to do with the order in which
the units are started or stopped—you will have to conﬁgure that sepa-
rately with After or Before . If no explicit order has been speciﬁed, sys-
temd will start all units at the same time.
10.4 Dependencies 163

B You can specify these dependencies without changing the unit ﬁle, by
creating a directory called /etc/systemd/system/example.service.requires
and adding symbolic links to the desired unit ﬁles to it. A directory
like
# ls -l /etc/systemd/system/example.service.requires
lrwxrwxrwx 1 root root 34 Jul 17 15:56 network.target ->
/lib/systemd/system/network.target
lrwxrwxrwx 1 root root 34 Jul 17 15:57 syslog.service ->
/lib/systemd/system/syslog.service

corresponds to the setting
[Unit]
Requires = network.target syslog.service

in example.service .

Wants A weaker form of Requires . This means that the listed units will be started to-
gether with the current unit, but if their activation fails this has no inﬂuence
on the process as a whole. This is the recommended method of making the
start of one unit depend on the start of another one.

B Here, too, you can specify the dependencies “externally” by creating a
directory called example.service.wants .

Conflicts The reverse of Requires —the units listed here will be stopped when the
current unit is started, and vice versa.

B Like Requires , Conflicts makes no stipulation to the order in which units
are started or stopped.

B If a unit 𝑈 conﬂicts with another unit 𝑉 and both are to be started at
the same time, this operation fails if both units are an essential part of
the operation. If one (or both) units are not essential parts of the op-
eration, the operation is modiﬁed: If only one unit is not mandatory,
that one will not be started, if both are not mandatory, the one men-
tioned in Conflicts will be started and the one whose unit ﬁle contains
the Conflicts option will be stopped.

Before (and After ) These lists of units determine the starting order. If example.
service contains the “Before=example2.service ” option and both units are be-
ing started, the start of example2.service will be delayed until example.service
has been started. After is the converse of Before , i. e., if example2.service con-
tains the option “After=example.service ” and both units are being started, the
same eﬀect results—example2.service will be delayed.

B Notably, this has nothing to do with the dependencies in Requires and
Conflicts . It is common, for example, to list units in both Requires and
After . This means that the listed unit will be started before the one
whose unit ﬁle contains these settings.

When deactivating units, the reverse order is observed. If a unit with a Before
or After dependency on another unit is deactivated, while the other is being
started, then the deactivation takes place before the activation no matter in
which direction the dependency is pointing. If there is no Before or After
dependency between two units, they will be started or stopped simultane-
ously.
164 10 Systemd

Table 10.1: Common targets for systemd (selection)

Target Description
basic.target Basic system startup is ﬁnished (ﬁle systems, swap
space, sockets, timers etc.)
ctrl-alt-del.target Is executed when Ctrl + Alt + Del was pressed. Often
the same as reboot.target .
default.target Target which systemd attempts to reach on sys-
tem startup. Usually either multi-user.target or
graphical.target .
emergency.target Starts a shell on the system console. For emer-
gencies. Is usually activated by means of the
“systemd.unit=emergency.target ” on the kernel command
line.
getty.target Activates the statically-deﬁned getty instances (for ter-
minals). Corresponds to the getty lines in /etc/inittab
on System-V init.
graphical.target Establishes a graphical login prompt. Depends on
multi-user.target .
halt.target Stops the system (without powering it down).
multi-user.target Establishes a multi-user system without a graphical lo-
gin prompt. Used by graphical.target .
network-online.target Serves as a dependency for units that require network
services (not ones that provide network services), such
as mount units for remote ﬁle systems. How exactly the
system determines whether the network is available de-
pends on the method for network conﬁguration.
poweroff.target Stops the system and powers it down.
reboot.target Restarts the system.
rescue.target Performs basic system initialisation and then starts a
shell.

Exercises
C 10.4 [!1] What advantage do we expect from being able to conﬁgure depen-
dencies via symbolic links in directories like example.service.requires instead
of the example.service unit ﬁle?

C 10.5 [2] Check your system conﬁguration for examples of Requires , Wants and
Conflicts dependencies, with or without corresponding Before and After de-
pendencies.

10.5 Targets
Targets in systemd are roughly similar to runlevels in System-V init: a possibil-
ity of conveniently describing a set of services. While System-V init allows only
a relatively small number of runlevels and their conﬁguration is fairly involved,
systemd makes it possible to deﬁne various targets very easily.
Unit ﬁles for targets have no special options (the standard set of options for
[Unit] and [Install] should be enough). Targets are only used to aggregate other
units via dependencies or create standardised names for synchronisation points in
dependencies (local-fs.target , for example, can be used to start units depending
on local ﬁle systems only once these are actually available). An overview of the
most important targets is in Table 10.1.
In the interest of backwards compatibility to System-V init, systemd deﬁnes a
number of targets that correspond to the classical runlevels. Consider table 10.2.
10.5 Targets 165

Table 10.2: Compatibility targets for System-V init

Ziele Äquivalent
runlevel0.target poweroff.target
runlevel1.target rescue.target
runlevel2.target multi-user.target (recommended)
runlevel3.target graphical.target (recommended)
runlevel4.target graphical.target (recommended)
runlevel5.target graphical.target (recommended)
runlevel6.target reboot.target

You can set the default target which systemd will attempt to reach on system default target
boot by creating a symbolic link from /etc/systemd/system/default.target to the de-
sired target’s unit ﬁle:

# cd /etc/systemd/system
# ln -sf /lib/systemd/system/multi-user.target default.target

(This is the moral equivalent to the initdefault line in the /etc/inittab ﬁle of System-
V init.) A more convenient method is the “systemctl set-default ” command:

# systemctl get-default
multi-user.target
# systemctl set-default graphical
Removed symlink /etc/systemd/system/default.target.
Created symlink from /etc/systemd/system/default.target to
/lib/systemd/system/graphical.target.
# systemctl get-default
graphical.target

(As you can see, that doesn’t do anything other than tweak the symbolic link,
either.)
To activate a speciﬁc target (like changing to a speciﬁc runlevel on System-V Activate specific target
init), use the “systemctl isolate ” command:

# systemctl isolate multi-user

(“File*.target” will be appended to the parameter if necessary). This command
starts all units that the target depends upon and stops all other units.

B “systemctl isolate ” works only for units in whose [Unit] sections the “AllowIsolate ”
option is switched on.

To stop the system or to change to the rescue mode (System-V init aﬁcionados
would call this “single-user mode”) there are the shortcuts

# systemctl rescue
# systemctl halt
# systemctl poweroff Like halt , but with power-down
# systemctl reboot

These commands correspond roughly to their equivalents using “systemctl isolate ”,
but also output a warning to logged-in users. You can (and should!) of course
keep using the shutdown command.
You can return to the default operating state using

# systemctl default
166 10 Systemd

Exercises
C 10.6 [!2] Which other services does the multi-user.target depend on? Do
these units depend on other units in turn?

C 10.7 [2] Use “systemctl isolate ” to change your system to the rescue (single-
user) mode, and “systemctl default ” to come back to the standard mode.
(Hint: Do this from a text console.)

C 10.8 [2] Restart your system using “systemctl reboot ” and then once again
with shutdown . Consider the diﬀerence.

10.6 The systemctl Command
The systemctl command is used to control systemd. We have already seen a few
applications, and here is a more systematic list. This is, however, still only a small
excerpt of the complete description.
The general structure of systemctl invocations is

# systemctl ⟨subcommand⟩ ⟨parameters⟩ …

systemctlsupports a fairly large zoo of subcommands. The allowable parameters
(and options) depend on the subcommand in question.

unit names as parameters B Often unit names are expected as parameters. These can be speciﬁed
either with a ﬁle name extension (like, e. g., example.service ) or without
(example ). In the latter case, systemd appends an extension that it considers
appropriate—with the start command, for example, “.service ”, with the
isolate command on the other hand, “.target ”.

Commands for units The following commands deal with units and their man-
agement:
list-units Displays the units systemd knows about. You may specify a unit type
(service , socket , …) or a comma-separated list of unit types using the -t op-
tion, in order to conﬁne the output to units of the type(s) in question. You
can also pass a shell search pattern in order to look for speciﬁc units:

# systemctl list-units "ssh*"
UNIT LOAD ACTIVE SUB DESCRIPTION
ssh.service loaded active running OpenBSD Secure Shell server

LOAD = Reflects whether the unit definition was properly loaded.
ACTIVE = The high-level unit activation state, i.e. generalization
of SUB.
SUB = The low-level unit activation state, values depend on
unit type.

1 loaded units listed. Pass --all to see loaded but inactive units,
too. To show all installed unit files use 'systemctl
list-unit-files'.

B As usual, quotes are a great idea here, so the shell will not vandalise
the search patterns that are meant for systemd.

start Starts one or more units mentioned as parameters.
10.6 The systemctl Command 167

B You can use shell search patterns here, too. The search patterns only
work for units that systemd knows about; inactive units that are not
in a failed state will not be searched, nor will units instantiated from
templates whose exact names are not known before the instantiation.
You should not overtax the search patterns.

stop Stops one or more units mentioned as parameters (again with search pat-
terns).
reload Reloads the conﬁguration for the units mentioned as parameters (if the pro-
grams underlying these units go along). Search patterns are allowed.

B This concerns the conﬁguration of the background services them-
selves, not the conﬁguration of the services from systemd’s point of
view. If you want systemd to reload its own conﬁguration with respect
to the background services, you must use the “systemctl daemon-reload ”
command.

B What exactly happens on a “systemctl reload ” depends on the back-
ground service in question (it usually involves a SIGHUP ). You can con-
ﬁgure this in the unit ﬁle for the service.

restart Restarts the units mentioned as parameters (search patterns are allowed).
If a unit doesn’t yet run, it is simply started.
try-restart Like restart , but units that don’t run are not started.
reload-or-restart (and reload-or-try-restart ) Reloads the conﬁguration of the
named units (as per reload ), if the units allow this, or restarts them (as
per restart or try-restart ) if they don’t.

B Instead of reload-or-try-restart you can say force-reload for convenience
(this is at least somewhat shorter).

isolate The unit in question is started (including its dependencies) and all other
units are stopped. Corresponds to a runlevel change on System-V init.
kill Sends a signal to one or several processes of the unit. You can use the
--kill-who option to specify which process is targeted. (The options include
main , control , and all —the latter is the default—, and main and control are
explained in more detail in systemctl (1).) Using the --signal option (-s for
short) you can determine which signal is to be sent.
status Displays the current status of the named unit(s), followed by its most recent
log entries. If you do not name any units, you will get an overview of all
units (possibly restricted to particular types using the -t option).

B The log excerpt is usually abridged to 10 lines, and long lines will be
shortened. You can change this using the --lines (-n ) and --full (-l )
options.

B “status ” is used for human consumption. If you want output that is
easy to process by other programs, use “systemctl show ”.

cat Displays the conﬁguration ﬁle(s) for one or more units (including fragments
in conﬁguration directories speciﬁcally for that unit). Comments with the
ﬁle names in question are inserted for clarity.
help Displays documentation (such as man pages) for the unit(s) in question: For
example,

$ systemctl help syslog
168 10 Systemd

invokes the manual page for the system log service, regardless of which
program actually provides the log service.

B With most distributions, commands like
# service example start
# service example stop
# service example restart
# service example reload

work independently of whether the system uses systemd or System-V init.
In the next section, there are a few commands that deal with installing and
deinstalling units.

Other commands Here are a few commands that do not speciﬁcally deal with
particular units (or groups of units).
daemon-reload This command causes systemd to reload its conﬁguration. This in-
cludes regenerating unit ﬁles that have been created at runtime from other
conﬁguration ﬁles on the system, and reconstructing the dependency tree.

B Communication channels that systemd manages on behalf of back-
ground services will persist across the reload.
daemon-reexec Restarts the systemd program. This saves systemd’s internal state and
restores it later.

B This command is mostly useful if a new version of systemd has been
installed (or for debugging systemd . Here, too, communication channels
that systemd manages on behalf of background services will persist
across the reload.
is-system-running Outputs the current state of the system. Possible answers in-
clude:
initializing The system is in the early boot stage (the basic.target , rescue.target ,
or emergency.target targets have not yet been reached).
starting The system is in the late boot stage (there are still jobs in the queue).
running The system is running normally.
degraded The system is running normally, but one or more units are in a
failed state.
maintenance One of the rescue.target or emergency.target targets are active.
stopping The system is being shut down.

Exercises
C 10.9 [!2] Use systemctl to stop, start, and restart a suitably innocuous service
(such as cups.service ) and to reload its conﬁguration.

C 10.10 [2] The runlevel command of System-V init outputs the system’s cur-
rent runlevel. What would be an approximate equivalent for systemd?

C 10.11 [1] What is the advantage of
# systemctl kill example.service

versus
# killall example

(or “pkill example ”)?
10.7 Installing Units 169

10.7 Installing Units
To make a new background service available using systemd, you need a unit ﬁle,
for example example.service . (Thanks to backwards compatibility, a System-V init
script would also do, but we won’t go there just now.) You need to place this in a
suitable ﬁle (we recommend /etc/systemd/system . Next, it should appear when you
invoke “systemctl list-unit-files ”:

# systemctl list-unit-files
UNIT FILE STATE
proc-sys-fs-binfmt_misc.automount static
org.freedesktop.hostname1.busname static
org.freedesktop.locale1.busname static

example.service disabled

The disabled state means that the unit is available in principle, but is not being
started automatically.
You can “activate” the unit, or mark it to be started when needed (e. g., during Activating units
system startup or if a certain communication channel is being accessed), by issuing
the “systemctl enable ” command:

# systemctl enable example
Created symlink from /etc/systemd/system/multi-user.target.wants/
example.service to /etc/systemd/system/example.service.

The command output tells you what happens here: A symbolic link to the ser-
vice’s unit ﬁle from the /etc/systemd/system/multi- user.target.wants directory en-
sures that the unit will be started as a dependency of the multi-user.target .

B You may ask yourself how systemd knows that the example unit should be
integrated in the multi-user.target (and not some other target). The answer
to that is: The example.service ﬁle has a section saying

[Install]
WantedBy=multi-user.target

After an enable , systemd does the equivalent of a “systemctl daemon-reload ”. How-
ever, no units will be started (or stopped).

B You could just as well create the symbolic links by hand. You would, how-
ever, have to take care of the “systemctl daemon-reload ” yourself, too.

B If you want the unit to be started immediately, you can either give the
# systemctl start example

command immediately afterwards, or you invoke “systemctl enable ” with the
--now option.

B You can start a unit directly (using “systemctl start ”) without ﬁrst activating
it with “systemctl enable ”. The former actually starts the service, while the
latter only arranges for it to be started at an appropriate moment (e. g., when
the system is booted, or a speciﬁc piece of hardware is connected).

You can deactivate a unit again with “systemctl disable ”. As with enable , sys-
temd does an implicit daemon-reload .
170 10 Systemd

B Here, too, the unit will not be stopped if it is currently running. (You are
just preventing it from being activated later on.) Use the --now option or an
explicit “systemctl stop ”.

B The “systemctl reenable ” command is equivalent to a “systemctl disable ” im-
mediately followed by a “systemctl enable ” for the units in question. This lets
you do a “factory reset” of units.

Masking a unit The “systemctl mask ” command lets you “mask” a unit. This means to block it
completely. This will not only prevent it from starting automatically, but will also
keep it from being started by hand. “systemctl unmask ” reverts that operation.

B Systemd implements this by linking the name of the unit ﬁle in /etc/systemd/
system symbolically to /dev/null . Thus, eponymous ﬁles in directories that
systemd considers later (like /lib/systemd/system ) will be completely ignored.

Exercises
C 10.12 [!2] What happens if you execute “systemctl disable cups ”? (Watch the
commands being output.) Reactivate the service again.

C 10.13 [2] How can you “mask” units whose unit ﬁles are in /etc/systemd/
system ?

Commands in this Chapter
systemctl Main control utility for systemd systemctl (1) 157, 166

Summary
• Systemd is a modern alternative to System-V init.
• “Units” are system components managed by systemd. They are conﬁgured
using unit ﬁles.
• Unit ﬁles bear a vague resemblance to Microsoft Windows INI ﬁles.
• Systemd supports ﬂexible mechanisms for local conﬁguration and the au-
tomatic creation of similar unit ﬁles from “templates”.
• Systemd lets you manage a multitude of diﬀerent units—services, mount
points, timers, …
• Dependencies between units can be expressed in various ways.
• “Targets” are units that vaguely correspond to System-V init’s runlevels.
They are used to group related services and for synchronisation.
• You can use the systemctl command to control systemd.
• Systemd contains powerful tools to install and deinstall units.

Bibliography
systemd “systemd System and Service Manager”.
http://www.freedesktop.org/wiki/Software/systemd/

XDG-DS14 Preston Brown, Jonathan Blandford, Owen Taylor, et al. “Desktop
Entry Speciﬁcation”, April 2014.
http://standards.freedesktop.org/desktop- entry- spec/latest/
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

11
Dynamic (AKA Shared) Libraries

Contents
11.1 Compiling and Installing Software . . . . . . . . . . . . . 172
11.2 Dynamic Libraries In Practice . . . . . . . . . . . . . . . 174
11.3 Installing and Locating Dynamic Libraries . . . . . . . . . . . 176
11.4 Dynamic Library Versioning . . . . . . . . . . . . . . . . 177

Goals
• Being able to identify shared libraries
• Knowing where shared libraries are usually stored
• Being able to manage shared libraries

Prerequisites
• Solid knowledge of the Linux command line
• Programming experience is useful but not essential

adm1-biblio.tex (33e55eeadba676a3 )
172 11 Dynamic (AKA Shared) Libraries

11.1 Compiling and Installing Software
An important principle of programming is not to reinvent the wheel continually.
Most programs share a lot of code—from basic functions like handling ﬁles and
strings to more specialised stuﬀ like dealing with date or time data. Neither does it
make sense to write this code from scratch over and over again, nor is it reasonable
to integrate it bodily into every program. The former because it is a lot of tedious
work and the same mistakes would have to be corrected again and again; the latter
because it takes a lot of space on disk and is extremely inconvenient if a mistake
does need to be corrected.
The answer to the ﬁrst objection—the code should not have to be rewritten
software libraries over and over again—is software libraries. They serve to collect often-required
functionality, to endow it with a standardised and well-documented interface, and
to make it available to other programs. For example, an essential ingredient for a
functioning Linux system is libc , which bundles all sorts of vital functionality for
programs written in the C programming language.

B Practically all programs on a Linux system are written in C, if only indi-
rectly: Many are written in native C, some are written in languages derived
from C such as C++ which usually also use libc , and even shell scripts and
programs in languages such as awk , Python, and Perl use the shell or the awk ,
python , and perl programs—all of which are written in C …

In addition to libc , a typical Linux system contains a few hundred or thousand
other libraries that cater for requirements ranging from support for mathematical
functions or various data formats to providing complete graphical interfaces.
In the ﬁrst instance, libraries are important when programming. In the sim-
linker plest case, a “linker” is used to combine a program’s “object code”—that is, the
output of the compiler, which translates a program written, say, in C to instruc-
tions executable on the CPU—with the object code of the libraries used by this
program to yield an “executable program” that you can then start.

B Consider, by way of illustration, the following tiny C program:
/* sqrttest.c */
#include <math.h>
#include <stdio.h>
int main(void)
{
printf("%g\n", sin(3.141592654/2.0));
return 0;
}

(It serves to answer the nagging question about the value of sin(𝜋/2).) The C
compiler turns this into an assembly language program that starts approx-
imately like

; sqrttest.s
.file "sqrttest.c"
.section .rodata
.LC1:
.string "%g\n"
.align 8
.LC0:
.long 1414677840
.long 1073291771
.text
.globl main
.type main, @function
11.1 Compiling and Installing Software 173

main:
leal 4(%esp), %ecx
andl $-16, %esp
pushl -4(%ecx)
pushl %ebp

and the assembler, in turn, produces an object ﬁle which, displayed “read-
ably”, looks like

7f 45 4c 46 01 01 01 00 00 00 00 00 00 00 00 00
01 00 03 00 01 00 00 00 00 00 00 00 00 00 00 00
14 01 00 00 00 00 00 00 34 00 00 00 00 00 28 00
0b 00 08 00 8d 4c 24 04 83 e4 f0 ff 71 fc 55 89
e5 51 83 ec 14 dd 05 08 00 00 00 dd 1c 24 e8 fc
ff ff ff dd 5c 24 04 c7 04 24 00 00 00 00 e8 fc

(952 bytes in total, if you must know). This ﬁle then needs to be put together
with the libraries libc (for the printf() function) and libm (for the sin() func-
tion) to create an executable program, which is the linker’s job. Fortunately
all of this can be done using a single command such as

$ gcc -static sqrttest.c -lm -o sqrttest-static

On the author’s system, the resulting program weighs in at a mere 560,608 bytes.

In this case, the program and libraries form a permanent combinaton. The
libraries spare us the tedious repetition of code that is useful in several programs, Advantages of libraries
but if a problem is found in a library, we need to re-link all programs using that
library. The fact that a trivial program is half a megabyte long is also vaguely less
than satisfying.
Here is where dynamic libraries come in. The basic idea behind them is that dynamic libraries
the actual program and libraries are not combined permanently during linking,
but are only put together when the program is actually executed. Thus, program
and libraries remain independent of each other. This results in the following ad-
vantages:
• As long as a library’s interface stays the same (i. e., the “signatures” of
the functions concerned—the number and types of their arguments and
results—do not change), the library can easily be replaced without having
to re-link all programs using it. When a program is started, it automatically
fetches the new library version.
• If 𝑛 programs are using the library, it does not require 𝑛 times the space on
disk (as part of each of those programs). One ﬁle is enough.

• In fact it usually suﬃces to have the library in memory just once. All pro-
cesses can share one copy. Another saving! This is why we speak of “shared
libraries”.
Incidentally, you’re paying for this with time: On the one hand it takes a bit longer
to launch a program using dynamic libraries, since the connection between the
actual program and its libraries needs to be made at runtime instead of when the
program is compiled to an executable ﬁle. On the other hand—depending on the
processor and programming model—it may be the case that dynamic libraries
carry a runtime penalty as opposed to “static”, i. e., connected-at-compile-time,
libraries. This penalty usually amounts to a low single-digit percentage and is
more than outweighed by the advantages mentioned above.
174 11 Dynamic (AKA Shared) Libraries

B Our example from above could be linked against dynamic libraries using
$ gcc sqrttest.c -lm -o sqrttest-dynamic

The result is all of 6510 bytes long—a marked improvement from more than
500,000!

B Perhaps you would think that disk space is so cheap nowadays that half a
megabyte here or there is not worth losing sleep over. Dynamic libraries
were introduced to Unix in the 1980s, when disk space was still measured
in megabytes rather than terabytes; libraries weren’t quite as big as they
are today, but one used to notice the diﬀerence rather more in these days.
Anyway, the other advantages are really even more important.

Rehabilitation of stat- To rehabilitate statically linked programs we should mention that they also
ically linked programs have their place. On the one hand, they work fairly reliably even is the system
is all but hosed otherwise—should you ever manage to somehow junk your dy-
namic libc , you will be glad to have a statically-linked version of busybox or sash
on your system, since bash and its friends such as cp or tar will be no good for re-
pairs. (The alternative would be to boot the rescue system, but that is of course a
cop-out for wimps.) On the other hand, statically linked programs are huge and
inconvenient, but you know what you have1 – so if you want to distribute a com-
plex program in executable form, linking statically makes you independent of the
libraries available on the systems the program is supposed to run on later. (This
can of course imply diﬀerent problems, but an exhausting discussion would be
beyond the scope of this manual.)

LSB B A large part of the exertions of the LSB (Linux Standard Base), an agreement
between various distributors as to which environment should be available
for software distributed in executable form by third-party vendors (think
SAP, Oracle and games), concerns itself with prescribing the version and
content of important dynamic libraries.

11.2 Dynamic Libraries In Practice
As we said earlier, most executable programs on a typical Linux system are linked
dynamically. You can make sure using the file command:

$ file sqrttest-dynamic
sqrttest-dynamic: ELF 32-bit LSB executable, Intel 80386,
version 1 (SYSV), dynamically linked (uses shared libs),
for GNU/Linux 2.6.8, not stripped

By way of comparison:

$ file sqrttest-static
sqrttest-static: ELF 32-bit LSB executable, Intel 80386,
version 1 (SYSV), statically linked, for GNU/Linux 2.6.8,
not stripped

B The “not stripped ” refers to the fact that our small (?) C program still con-
tains information supposed to make debugging easier. You can slim it down
to just below 500 kB by executing “strip sqrttest- static ”.

Required libraries You can ﬁnd out in detail which libraries sqrttest- dynamic requires by using the
ldd command:
1 Apologies to the Persil man.
11.2 Dynamic Libraries In Practice 175

$ ldd sqrttest-dynamic
linux-gate.so.1 => (0xb7f51000)
libm.so.6 => /lib/i686/cmov/libm.so.6 (0xb7f0b000)
libc.so.6 => /lib/i686/cmov/libc.so.6 (0xb7db0000)
/lib/ld-linux.so.2 (0xb7f52000)

Its output tells you the names of the libraries as well as the names of the ﬁles
corresponding to them—if possible (otherwise the ﬁle name is empty).

B We have already discussed libc and libm . /lib/ld- linux.so.2 is the “dynamic
linker”, the program that causes the actual program to be connected to its
dynamic libraries when it is started (we shall come back to this presently).

B So what is linux-gate.so.1 , and what is it good for? This question is great
for ﬁnding out how much your buddies really know about Linux. The most
obvious observation is that ldd didn’t ﬁnd a ﬁle corresponding to this li-
brary. Thus by rights the program should not even run. If you look you
will even ﬁnd that there is no ﬁle on the system by this name at all. However,
this doesn’t matter, as the kernel takes care of making this “virtual dynamic
library” available to all programs. linux-gate.so.1 is used to speed up sys-
tem calls on modern x86 processors (which, in 2009, basically means “all of
them”). The details are too unsavoury to explain in detail; read [Pet08] but
not immediately before or after a meal.

B On Intel- and AMD-based 64-bit systems, the linux-vdso.so.1 “library”
serves the same purpose as linux-gate.so.1 on 32-bit systems. Don’t let
yourself become confused.

Typical places where Linux distributors put libraries (static or dynamic ones) Libraries: where?
include the /lib and /usr/lib directories. As usual, the former is mostly meant
for libraries that need to be available immediately after the system is booted, and
the latter caters for those that are only needed after all ﬁle systems have been
mounted.
How do you recognise a dynamic library, anyway? By its name—the names of Recognising dynamic libraries
dynamic libraries usually end in .so , usually followed by a version number. (The
names of static libraries end in .a , and a version number does not apply there.) If
you want to be anal-retentive, you can can also trot out file :

$ file /lib/libc-2.7.so
/lib/libc-2.7.so: ELF 32-bit LSB shared object, Intel 80386,
version 1 (SYSV), dynamically linked (uses shared libs),
for GNU/Linux 2.6.8, stripped

Look closely—we have

ELF 32-bit LSB shared object

instead of
ELF 32-bit LSB executable

—a subtle diﬀerence!

B With static libraries, the following happens:
$ file /usr/lib/libc.a
/usr/lib/libc.a: current ar archive
176 11 Dynamic (AKA Shared) Libraries

It does take a little experience to conclude “library!” from this—take a look
at ar (1).

File names for dynamic libraries In “real life”, the ﬁle name of a dynamic library could look approximately like
this:
$ ls -l /usr/lib/libcurl.*
lrwxrwxrwx root root 16 Jan 22 01:23 /usr/lib/libcurl.so.4
-> libcurl.so.4.1.0
-rw-r--r-- root root 271364 Dec 27 14:33 /usr/lib/libcurl.so.4.1.0

The symbolic link is used to relate actual programs and the library. A program
actually asks for a speciﬁc version of a library—in our example, libcurl.so.4 . It
depends on the system whether this amounts to /usr/lib/libcurl.so.4.1.0 or /usr/
lib/libcurl.so.4.2.1 , whichever is installed. (The basic assumption is that all ﬁles
purporting to be libcurl.so.4 implement the same interface.) More information
about versioning dynamic libraries is in Section 11.4.

B Incidentally, dynamic libraries may depend on other dynamic libraries, too.
You may ﬁnd out about this using ldd (of course):

$ ldd /usr/lib/libcurl.so.4.1.0
linux-gate.so.1 => (0xb7f6f000)
libidn.so.11 => /usr/lib/libidn.so.11 (0xb7edb000)
libssh2.so.1 => /usr/lib/libssh2.so.1 (0xb7eba000)
liblber-2.4.so.2 => /usr/lib/liblber-2.4.so.2 (0xb7eac000)

(What you get to see if you apply ldd to a program is the transitive closure
over all dynamic libraries that the program together with all of its dynamic
libraries depends on.)

Exercises
C 11.1 [!1] Find a statically-linked executable on your system (apart from
sqrttest- static ).

C 11.2 [1] What does ldd say if you apply it to a statically-linked program?

C 11.3 [2] (For shell programmers.) Which program in /usr/bin on your system
is linked against the greatest number of dynamic libraries?

11.3 Installing and Locating Dynamic Libraries
dynamic linker When you start a program, the dynamic linker must ﬁnd and load the required
libraries. However, the program itself contains only a library’s name, not the name
of the ﬁle it is to be found in. To avoid having the dynamic linker search all of the
File/etc/ld.so.cache ﬁle system (or even only the /lib and /usr/lib directories), the File/etc/ld.so.cache
ﬁle contains an “index” of all known dynamic libraries. The dynamic linker ﬁnds
only those libraries that occur in this index.
ldconfig The index is created using the ldconfig program, which searches all directories
/etc/ld.so.conf listed in /etc/ld.so.conf as well as the two standard directories /lib and /usr/lib for
libraries. All found libraries are entered into the index. In addition, ldconfig takes
care of creating the “main version number” symbolic links for library ﬁles.

B /lib and /usr/lib are always searched for libraries, no matter what /etc/ld.
so.conf says.

index content You can inspect the current state of the index using
11.4 Dynamic Library Versioning 177

# ldconfig -p
909 libs found in cache `/etc/ld.so.cache'
libzvt.so.2 (libc6) => /opt/gnome/lib/libzvt.so.2
libz.so.1 (libc6) => /lib/libz.so.1
libz.so (libc6) => /usr/lib/libz.so

This does not generate a new index, but only outputs the content of ld.so.cache in
a readable format. If a library does not show up here, it cannot be found by the
linker, either.
To add a new dynamic library from the standard directories—besides /lib and New dynamic libraries
/usr/lib , this typically includes /usr/local/lib —, then a simple call to

# ldconfig

should suﬃce to update /etc/ld.so.cache . If your library is not located in one of
the directories mentioned in /etc/ld.so.conf , you have to add the directory to that
ﬁle ﬁrst.

B Usually the package installation tool of your distribution should take care of
invoking ldconfig when packages including dynamic libraries are installed.
Therefore, the previous paragraph applies mostly if you want to install li-
braries that you wrote yourself, or ones in software packages that you com-
piled from source code.

If you lack the administrator rights necessary to call ldconfig you can still use
your own dynamic libraries. On top of the cache in /etc/ld.so.cache , the dynamic
linker also searches the directories enumerated in the LD_LIBRARY_PATH environment LD_LIBRARY_PATH
variable. Its syntax corresponds to that of the PATH variable—directory names are
separated by colons:

$ export LD_LIBRARY_PATH=$HOME/lib:/opt/foo/lib

B The dynamic linker searches the content of the directories listed in LD_LIBRARY_PATH
(if set) ﬁrst, then the content of /etc/ld.so.cache , and then the content of
the /lib and /usr/lib directories (in that order—this is just a safety net in
case /etc/ld.so.cache does not exist). Libraries in a directory mentioned in
LD_LIBRARY_PATH whose names are the same as “oﬃcial” libraries are thus
found ﬁrst.

B The documentation of the dynamic linker (ld-linux.so (8)) contains many in-
teresting and thrilling options that allow you to do strange and wonderful
things.

Exercises
C 11.4 [!2] The directories in LD_LIBRARY_PATH are not considered when the pro-
gram to be started uses the Set-UID or Set-GID mechanism (see Section 3.5).
Why is that?

11.4 Dynamic Library Versioning
Users of Microsoft Windows know the problem called “DLL hell”2 : Many soft-
ware packages come with their own versions of important system libraries and do
2 DLLs, or “dynamically loadable libraries”, are the Windows equivalent to the dynamic libraries

of Linux.
178 11 Dynamic (AKA Shared) Libraries

not shy away from recklessly supplanting pre-existing versions of those libraries
with their own. If, of course, the supplanted version happens to be exactly the
one that another software package brought in, chaos is not far away.
third-party software Under Linux this problem is much less grave, since much less third-party
software is delivered in executable form to begin with. You receive binary soft-
ware mostly from your distributor, and other things are compiled from source
code (making use of pre-installed libraries as much as possible). Problems are
more likely to result from the fact that the Linux software landscape evolves fairly
quickly, which of course also applies to libraries.
version numbers For this reason, dynamic libraries are assigned version numbers that typically
contain up to three parts. For example, earlier on we talked about libcurl.4.1.0 .
These version numbers work as follows:
major version number • The ﬁrst number after the library name is the “major version number”. It
should be incremented whenever the programming interface oﬀered by the
library to other programs changes in an incompatible fashion. For example,
a function that used to accept two arguments might suddenly require three,
or an argument that used to be an integer might turn into a pointer to a
string.

B It goes without saying that a responsible developer should try to avoid
such changes whenever possible, as they imply that the programs us-
ing the library need to be adapted.

• The second number (after the second dot) is the “minor version number”.
It is incremented when the programming interface is extended without
changes to existing calls. For example, a completely new function might be
added. Older programs can still use the library because everything they
require is still available in an identical fashion; of course, newer programs
that do use the newly added calls will not work with older versions of the
library.
patch level • The third number (at the end) is the “patch level”. It is incremented when
there are changes to the library implementation that do not aﬀect the inter-
face. For example, a bug in the implementation of a function might be ﬁxed.
The fact that the library is dynamically linked makes it possible to simply
replace the defective library by the corrected one—all subsequently-started
programs automatically pick up the new version.

B Replacing a defective library by a new one has no impact on programs
daemons already running with the defective library—such as daemons. When in
doubt, such programs must be restarted manually to beneﬁt from the
improvements.

B The suﬀerers in this improvement process are, of course, programs that
rely on the presence of the implementation error. It sometimes happens
that knowledge of a bug gets passed around the developer commu-
nity, and programmers either exploit it (if it has some beneﬁcial side
eﬀect—which happens) or work around it in a way that does not take
into account the fact that at some point the error might no longer be
present3 . In such a situation, a correction amounts to an incompati-
ble change and should, in the worst case, cause even the major version
number to be incremented.
3 A long time ago, your author had to contend with a C compiler that would calculate the reciprocal

of the actual result when performing ﬂoating-point divisions. (Don’t ask.) In a software package writ-
ten by a colleague, there occured exactly one such division in a strategic place, where (uncommented
and long forgotten) the dividend and divisor had been swapped to accommodate this error. Ferreting
this out as the reason for a grisly sequence of other faults, after updating to a ﬁxed C compiler, took
us a few days.
11.4 Dynamic Library Versioning 179

Programs always require a certain major number of a library, such as libcurl.so.4 ,
which is mapped to a concrete ﬁle, such as /usr/lib/libcurl.so.4.1. , by means of
the symbolic link.
Hence it is quite feasible for a system to simultaneously contain programs re- multiple versions simultaneously
quiring multiple (major) versions of the same library. You can simply leave the
various library ﬁles installed and put the burden of ﬁnding the correct one for
each program on the dynamic linker.

B The LSB standard prescribes certain major versions of common libraries
and, in each case, standardises a programming interface for the library in
question. LSB-conforming distributions must oﬀer these libraries for the
beneﬁt of LSB executables, but these libraries do not need to be the ones
that all of the rest of the system is using. Hence it is quite possible to make
a distribution conform to LSB from the point of view of library support by
placing a set of the required libraries in a diﬀerent directory, and to foist
this on LSB executables instead of the usual system directories for libraries
through dynamic linker trickery (such as LD_LIBRARY_PATH ).

B The same strategy is also useful for “normal” third-party software that does
not rely on LSB. As long as the kernel-libc interface does not change incom-
patibly (which, fortunately, the kernel developers diligently try to avoid),
a software package can provide its own libc (plus, presumably, all sorts of
other libraries) and so try to become independent of the library support
oﬀered by the underlying Linux system. Whether this is always a good idea
in real life remains an open question—after all, as a software vendor one
may have to support independent bug ﬁxes to the libraries and distribute
them to all customers—, but it is a fact that on Linux there is no direct equiv-
alent to “DLL hell”. Score one for the penguin!

Commands in this Chapter
busybox A shell that already contains variants of many Unix tools
busybox (1) 174
file Guesses the type of a ﬁle’s content, according to rules file (1) 174
ldconfig Builds the dynamic library cache ldconfig (8) 176
ldd Displays the dynamic libraries used by a program ldd (1) 174
sash “Stand-Alone Shell” with built-in commands, for troubleshooting
sash (8) 174
strip Removes symbol tables from object ﬁles strip (1) 174

Summary
• Libraries oﬀer frequently-needed functionality in standardised form to sev-
eral programs.
• Dynamic libraries save memory and disk space and facilitate program main-
tenance.
• file lets you ﬁnd out whether a program is dynamically linked, and identify
shared libraries as such.
• ldd outputs the names of dynamic libraries that a program (or a dynamic
library) uses.
• Dynamic libraries are searched in /lib , /usr/lib , and the directories men-
tioned in /etc/ld.so.conf . The ldconfig command constructs an index.
• Several versions of the same library may be installed at the same time.
180 11 Dynamic (AKA Shared) Libraries

Bibliography
Pet08 Johan Petersson. “What is linux-gate.so.1?”, August 2008.
http://www.trilithium.com/johan/2005/08/linux- gate/
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

12
Software Package Management
Using Debian Tools

Contents
12.1 Overview. . . . . . . . . . . . . . . . . . . . . . . 182
12.2 The Basis: dpkg . . . . . . . . . . . . . . . . . . . . . 182
12.2.1 Debian Packages . . . . . . . . . . . . . . . . . . 182
12.2.2 Package Installation . . . . . . . . . . . . . . . . . 183
12.2.3 Deleting Packages . . . . . . . . . . . . . . . . . 184
12.2.4 Debian Packages and Source Code . . . . . . . . . . . 185
12.2.5 Package Information. . . . . . . . . . . . . . . . . 185
12.2.6 Package Veriﬁcation . . . . . . . . . . . . . . . . . 188
12.3 Debian Package Management: The Next Generation . . . . . . . 189
12.3.1 APT . . . . . . . . . . . . . . . . . . . . . . 189
12.3.2 Package Installation Using apt-get . . . . . . . . . . . . 189
12.3.3 Information About Packages . . . . . . . . . . . . . . 191
12.3.4 aptitude . . . . . . . . . . . . . . . . . . . . . 192
12.4 Debian Package Integrity . . . . . . . . . . . . . . . . . 194
12.5 The debconf Infrastructure . . . . . . . . . . . . . . . . 195
12.6 alien : Software From Diﬀerent Worlds . . . . . . . . . . . . 196

Goals
• Knowing the basics of Debian packaging tools
• Being able to use dpkg for package management
• Being able to use apt-get , apt-cache , and aptitude
• Being aware of the principles of Debian package integrity
• Knowing how to convert RPM packages to Debian packages using alien

Prerequisites
• Knowledge of Linux system administration
• Experience with Debian GNU/Linux or a Debian GNU/Linux derivative is
helpful

adm1-deb.tex (33e55eeadba676a3 )
182 12 Software Package Management Using Debian Tools

12.1 Overview
Software packages in Debian GNU/Linux and derived distributions such as
Ubuntu, Knoppix, Xandros, or Sidux are maintained using the dpkg tools. It
serves to install software packages, to manage dependencies, to catalog installed
software, to control updates to software packages, and to de-install packages that
are no longer required. Program such as aptitude serve as front-ends to dpkg , al-
lowing the convenient selection of software packages. The debconf infrastructure
is used to conﬁgure packages upon installation.

B The Debian and Red Hat package management systems were developed at
about the same time and have diﬀerent strengths and weaknesses. As usual
in the free software community, the religious wars around dpkg and rpm have
not led to one of the competitors carrying the day. With the increasing pop-
ularity of Debian derivatives—most notably Ubuntu—this remains unlikely
for the foreseeable future, too.

B The LSB standard for a basic Linux infrastructure that third-party vendors
can port their software to does prescribe a restricted version of RPM as its
package format. However, this does not imply that a LSB-compliant Linux
distribution must be RPM-based from the ground up, but only that it must
be able to install software packages from third-party vendors that conform
to the LSB ﬂavour of RPM.

For Debian GNU/Linux, this is of course a piece of cake. The reason
why Debian GNU/Linux is not oﬃcially touted as “LSB-compliant” is be-
cause LSB is run by an industry consortium of which Debian, being a non-
commercial project, is not a member. The description for the lsb package
on Debian GNU/Linux states:
The intent of this package is to provide a best current practice way
of installing and running LSB packages on Debian GNU/Linux.
Its presence does not imply that Debian fully complies with the
Linux Standard Base, and should not be construed as a statement
that Debian is LSB-compliant.

While its title talks about “Debian tools”, everything in this chapter also
applies to Ubuntu, since Ubuntu takes substantial parts of its infrastructure
from Debian GNU/Linux. We shall be pointing out signiﬁcant diﬀerences
that do exist.

12.2 The Basis: dpkg
12.2.1 Debian Packages
packages Within the Debian infrastructure, the software on the system is divided into pack-
package names ages. Packages have names that indicate the software contained within and their
version. The
hello_2.8-2_amd64.deb

ﬁle, for example, contains the hello program’s 2.8 version; in particular this is the
second release of this package within the distribution (for a future 2.9 package
the count would start over at 1). Packages like apt which have been speciﬁcally
developed for Debian do not include a “Debian release number”. The amd64 indi-
cates that the package contains architecture-speciﬁc parts for Intel and AMD x86
processors (and compatibles) in 64-bit mode—32-bit packages use i386 , and pack-
ages that contain only documentation or architecture-independent scripts use all
instead.
12.2 The Basis: dpkg 183

B A Debian package is an archive created using the ar program and generally package structure
contains three components:

$ ar t hello_2.8-2_amd64.deb
debian-binary
control.tar.gz
data.tar.gz

The debian- binary ﬁle contains the version number of the package format
(currently 2.0 ). In control.tar.gz there are Debian-speciﬁc scripts and control
ﬁles, and data.tar.gz contains the actual package ﬁles. During installation, installation
control.tar.gz is unpacked ﬁrst, so that a possible preinst script can be exe-
cuted prior to unpacking the actual package ﬁles. After this, data.tar.gz will
be unpacked, and the package will be conﬁgured if necessary by executing
the postinst script from control.tar.gz .

Exercises
C 12.1 [2] Obtain an arbitrary Debian package (such as hello ) and take it apart
using ar and tar . Can you ﬁnd the installation scripts? Which information
is contained in control.tar.gz , and which is in data.tar.gz ?

12.2.2 Package Installation
You can easily install a locally-available Debian package using the

# dpkg --install hello_2.8-2_amd64.deb

command, where --install can be abbreviated to -i . With --unpack and --configure
(-a ), the unpacking and conﬁguration steps can also be executed separately.

In real life, the short option names such as -i are convenient. However, if
you intend to pass the LPI-101 exam, you should be sure to learn the long
option names as well, since these, vexatingly, occur in the exam questions.
In the case of -i and --install , it is probably straightforward to come up with
the correspondence; with -a and --configure , this is already somewhat less
obvious.

B Options for dpkg can be given on the command line or else placed in the
/etc/dpkg/dpkg.cfg ﬁle. In this ﬁle, the dashes at the start of the option names dpkg.cfg
must be omitted.

If a package is installed using “dpkg --install ”, even though an earlier version Upgrade
already exists on the system, the older version is deinstalled before conﬁguring
the new one. If an error occurs during installation, the old version can be restored
in many cases.
There are various reasons that might prevent a successful package installation, installation problems
including:
• The package requires one or more other packages that either have not yet
been installed, or that are not included in the same installation operation.
The corresponding check can be disabled using the --force-depends option—
but this can severely mess up the system.
• An earlier version of the package is installed and set to hold (e. g., using
aptitude ). This prevents newer versions of the package from being installed.

• The package tries to unpack a ﬁle that already exists on the system and be-
longs to a diﬀerent package, unless the current package is explicitly labeled
as “replacing” that package, or the --force-overwrite option was speciﬁed.
184 12 Software Package Management Using Debian Tools

conflicts Some packages conﬂict with each other (see the possibilities for package depen-
dencies on page 187). For example, only one mail transport program may be in-
stalled at one time; if you want to install, e. g., Postﬁx, Exim (the Debian default
MTA) must be removed at the same time. dpkg manages this if certain conditions
are fulﬁlled.
Virtual packages Sometimes packages do not depend on a particular other package but on a
“virtual” package describing a feature that can, in principle, be provided by any
of several other packages, such as mail-transport-agent , which is provided by pack-
ages like postfix , exim , or sendmail . In this case it is possible to replace, say, Exim by
Postﬁx in spite of dependencies, as a program providing the “virtual” functional-
ity will be available at all times.

Exercises
C 12.2 [1] Download a Debian package—say, hello —from ftp.debian.org (or
any other Debian mirror) and install it using dpkg . (If you use anything else,
such as apt-get —see next section—, you’re cheating!) You can ﬁnd Debian
packages on the server reasonably conveniently given their names, by look-
ing in pool/main for a subdirectory corresponding to the ﬁrst character of
the name, and in there looking for a subdirectory whose name is that of the
package1 , in our case pool/main/h/hello . Exception: Since very many package
names start with lib , a package like libfoobar ends up in pool/main/libf .

C 12.3 [2] Locate the current list of virtual packages in Debian GNU/Linux.
Where did you ﬁnd it? When did the last update take place?

12.2.3 Deleting Packages
A package is removed using the

# dpkg --remove hello

command (“dpkg -r ”, for short). Its conﬁguration ﬁles (all the ﬁles listed in the
conffiles ﬁle within control.tar.gz ), though, are kept around in order to facilitate
a subsequent reinstallation of the package. The

# dpkg --purge hello

(or “dpkg -P ”) command removes the package including its conﬁguration ﬁles.

B The “conﬁguration ﬁles” of a Debian package are all ﬁles in the package
whose names occur in the conffiles ﬁle in control.tar.gz . (Look at /var/lib/
dpkg/info/ ⟨package name⟩.conffiles .)

Removal problems Package removal does not necessarily work, either. Possible obstacles include:
• The package is required by one or more other packages that are not about
to be removed as well.
• The package is marked “essential” (to system functionality). The shell, for
example, cannot simply be removed since some important scripts could no
longer be executed.
Here, too, the relevant checks can be disabled using suitable --force- … options (at
your own risk).

Exercises
C 12.4 [1] Remove the package you installed during Exercise 12.2. Make sure
that its conﬁguration ﬁles are removed as well.
1 The source code package, really, which may diﬀer. So don’t get too fancy.
12.2 The Basis: dpkg 185

12.2.4 Debian Packages and Source Code
When dealing with source code, a basic principle of the Debian project is to distin-
guish clearly between the original source code and any Debian-speciﬁc changes.
Accordingly, all changes are placed in a separate archive. In addition to the
Debian-speciﬁc control ﬁles, these include more or less extensive ﬁxes and cus-
tomisations to the software itself. For each package version, there is also a “source source control file
control ﬁle” (using the .dsc suﬃx) containing checksums for the original archive
and the changes ﬁle, which will be digitally signed by the Debian maintainer in
charge of the package:

$ ls hello*
-rw-r--r-- 1 anselm anselm 6540 Jun 7 13:18 hello_2.8-2.debian.tar.gz
-rw-r--r-- 1 anselm anselm 1287 Jun 7 13:18 hello_2.8-2.dsc
-rw-r--r-- 1 anselm anselm 697483 Mai 27 23:47 hello_2.8.orig.tar.gz

You can also see that the original source code archive does not change for all of the
2.8 version of the program (it does not contain a Debian release number). Every
new version of the Debian package of hello ’s 2.8 version comes with new .dsc
and .debian.tar.gz ﬁles. The latter contains all the changes relative to the original
archive (rather than the hello_2.8-2 package).

B In former times, the Debian project used a less complicated structure where
there was one single ﬁle (created with diff ) containing all Debian-speciﬁc
changes—in our example, hypothetically hello_2.8- 2.diff.gz . This approach
is still supported, and you may ﬁnd this structure with older packages that
have not been changed to use the new method instead. The new struc-
ture does have the advantage that diﬀerent changes—like the introduction
of the Debian-speciﬁc control ﬁles and any changes to the actual original
code—can be more cleanly separated, which greatly simpliﬁes maintaining
the package within the Debian project.

The dpkg-source command is used to reconstruct a package’s source code from dpkg-source
the original archive and the Debian changes such that you can recompile your
own version of the Debian package. To do so, it must be invoked with the name
of the source control ﬁle as an argument:

$ dpkg-source -x hello_2.8-2.dsc

The original archive and the .debian.tar.gz or .diff.gz ﬁle must reside in the same
directory as the source control ﬁle. dpkg-source also places the unpacked source
code there.

B dpkg-source is also used when generating source archives and Debian change Preparing Debian packages
ﬁles during Debian package preparation. However, this topic is beyond the
scope of the LPIC-1 certiﬁcation.

Exercises
C 12.5 [1] Obtain the source code for the package you installed during Exer-
cise 12.2, and unpack it. Take a look at the debian subdirectory of the result-
ing directory.

12.2.5 Package Information
You can obtain a list of installed packages using “dpkg --list ” (-l , for short): package list

$ dpkg --list
Desired=Unknown/Install/Remove/Purge/Hold
| Status=Not/Installed/Config-files/Unpacked/Failed-config/H
186 12 Software Package Management Using Debian Tools

|/ Err?=(none)/Hold/Reinst-required/X=both-problems (Status,
||/ Name Version Description
+++-==============-==============-==========================
ii a2ps 4.13b+cvs.2003 GNU a2ps - 'Anything to Po
ii aalib1 1.4p5-19 ascii art library
ii abcm2ps 4.0.7-1 Translates ABC music descr
ii abcmidi 20030521-1 A converter from ABC to MI

(truncated on the right for space reasons) This list can be narrowed down using
shell search patterns shell search patterns:

$ dpkg -l lib*-tcl
Desired=Unknown/Install/Remove/Purge/Hold
| Status=Not/Installed/Config-files/Unpacked/Failed-config/H
|/ Err?=(none)/Hold/Reinst-required/X=both-problems (Status,
||/ Name Version Description
+++-==============-==============-==========================
pn libdb3-tcl <none> (no description available)
un libdb4.0-tcl <none> (no description available)
un libdb4.1-tcl <none> (no description available)
un libmetakit-tcl <none> (no description available)
ii libsqlite-tcl 2.8.9-1 SQLite TCL bindings
rc libsqlite0-tcl 2.6.1-2 SQLite TCL bindings

The packages with version “<none> ” are part of the distribution but are either not
installed on the current system (status un ) or have been removed (status pn ).
package status You can ﬁnd out about an individual package’s status with the --status (-s )
option:
$ dpkg --status hello
Package: hello
Status: install ok installed
Priority: optional
Section: devel
Installed-Size: 553
Maintainer: Santiago Vila <sanvila@debian.org>
Architecture: amd64
Version: 2.8-2
Depends: libc6 (>= 2.4), dpkg (>= 1.15.4) | install-info
Description: The classic greeting, and a good example
The GNU hello program produces a familiar, friendly greeting. It
allows non-programmers to use a classic computer science tool which
would otherwise be unavailable to them.
.
Seriously, though: this is an example of how to do a Debian package.
It is the Debian version of the GNU Project's `hello world' program
(which is itself an example for the GNU Project).
Homepage: http://www.gnu.org/software/hello/

Besides the package name (Package: ), its output includes information about the
package’s status and priority (from required via important , standard and optional
down to extra ) and its approximate area of interest (Section: ). The Maintainer: is
the person who is in charge of the package on behalf of the Debian project.
Priorities B Packages of priority required are necessary for proper operation of the sys-
tem (usually because dpkg depends on them). The important priority encom-
passes packages one would expect to be available on a Unix-like system2 .
2 The deﬁnition is something like “A package is important if, should it be missing, experienced Unix

users would shake their heads and go “WTF?”.
12.2 The Basis: dpkg 187

standard adds those packages that make sense for a net but not overly re-
strictive system running in text mode—this priority describes what you get
if you install Debian GNU/Linux without selecting any additional pack-
ages. The optional priority applies to everything you might want to install
if you do not look too closely and have no particular requirements. This
includes things like the X11 GUI and a whole load of applications (such as
TEX). There should be no conﬂicts within optional . Finally, extra is used for
all packages that conﬂict with packages of other priorities, or that are only
useful for specialised applications.

B Packages may not depend on packages of lower priority. For this to hold in
all cases, the priorities of some packages have deliberately been tweaked.

An important area of information are the package dependencies, of which package dependencies
there are several types:
Depends The named packages must be conﬁgured for the package to be able to be
conﬁgured. As in the preceding example, speciﬁc versions of the packages
may be called for.
Pre-Depends The named packages must be completely installed before installation
of the package can even begin. This type of dependency is used if, for exam-
ple, the package’s installation scripts absolutely require software from the
other package.

RecommendsA non-absolute but very obvious dependency. You would nearly al-
ways install the named packages alongside this package, and only refrain
from doing so in very unusual circumstances.
Suggests The named packages are useful in connection with the package but not
required.
EnhancesLike Suggests , but in reverse—this package is useful for the named pack-
age (or packages).
ConflictsThis package cannot be installed at the same time as the named pack-
ages.

If a package isn’t installed locally at all, “dpkg --status ” only outputs an error
message:

# dpkg -s xyzzy
Package `xyzzy' is not installed and no info is available.
Use dpkg --info (= dpkg-deb --info) to examine archive files,
and dpkg --contents (= dpkg-deb --contents) to list their contents.

The --listfiles (-L ) option provides a list of ﬁles within the package: list of files

$ dpkg --listfiles hello
/.
/usr
/usr/share
/usr/share/doc
/usr/share/doc/hello
/usr/share/doc/hello/changelog.Debian.gz
/usr/share/doc/hello/copyright
/usr/share/doc/hello/NEWS
/usr/share/doc/hello/changelog.gz
/usr/share/info
/usr/share/info/hello.info.gz
/usr/share/man
188 12 Software Package Management Using Debian Tools

/usr/share/man/man1
/usr/share/man/man1/hello.1.gz
/usr/share/locale

package search Finally, you can use the --search (or -s ) option to ﬁnd out which package (if any)
claims a given ﬁle. Search patterns are allowed:

$ dpkg -S bin/m*fs
dosfstools: /sbin/mkdosfs
cramfsprogs: /usr/sbin/mkcramfs
util-linux: /sbin/mkfs.cramfs
smbfs: /sbin/mount.smbfs

The search may take some time, though.

B If you’re looking for the package for a ﬁle that is not on your system—for
example, if you plan to install that package afterwards—, you can use the
search form on http://www.debian.org/distrib/packages#search_contents . This
allows you to search any or all Debian distributions and architectures as
well as to search for exact ﬁle name matches and ﬁle names containing cer-
tain search terms.

Exercises
C 12.6 [3] How many packages whose names start with lib are installed on
your system? How many of those packages have priority required ?

12.2.6 Package Verification
integrity of an installed package The integrity of an installed package can be checked using the debsums program
(from the eponymous package):

$ debsums hello
/usr/share/doc/hello/changelog.Debian.gz OK
/usr/share/doc/hello/copyright OK
/usr/share/doc/hello/NEWS OK
/usr/share/doc/hello/changelog.gz OK
/usr/share/info/hello.info.gz OK

This compares the MD5 checksums of the individual ﬁles with the content of
the corresponding ﬁle in /var/lib/dpkg/info (here, hello.md5sums ). If an actual ﬁle’s
checksum does not match the set value, FAILED is displayed in place of OK .

B debsums can uncover “inadvertent” changes to a package’s ﬁles, but does not
protection from intruders provide protection from intruders who maliciously modify ﬁles. After all, a
cracker could place the checksum of a modiﬁed ﬁle in its package’s md5sums
list. Neither does this method help agains “Trojan” packages that hide ma-
licious code behind an innocuous facade. We shall be coming back to the
“integrity of packages” topic in Section 12.4.

Exercises
C 12.7 [!2] Change a ﬁle in an installed Debian package. (Look for a not-so-
important one, like that from Exercise 12.2.) You could, for example, append
a few lines of text to the README.Debian ﬁle (be root ). Check the integrity of the
package’s ﬁles using debsums .
12.3 Debian Package Management: The Next Generation 189

12.3 Debian Package Management: The Next Genera-
tion
12.3.1 APT
dpkg is a powerful tool, but still somewhat restricted in its potential. For example,
it is a bit aggravating that it will notice unﬁlled dependencies between packages,
but then just throw in the towel instead of contributing constructively to a solution
of the problem. Furthermore, while it is nice be able to install locally-available
packages, one would wish for convenient access to FTP or web servers oﬀering
packages.

B The dselect program, which in the early days of Debian served as an inter-
active front-end to package management, is oﬃcially deprecated today—its
inconvenience was proverbial, even though reading the manual did help as
a rule.

Quite early in the history of the Debian project (by today’s standards), the De-
bian community started developing APT, the “Advanced Packaging Tool”. This
project, in its signiﬁcance as in its ultimate pointlessness, is comparable to the
quest of the Knights of the Round Table for the Holy Grail, but, like the Grail
quest, APT development led to many noble deeds along the way. Although few
dragons were slain and damsels freed from distress, the APT developers produced
very important and powerful “partial solutions” whose convenience and feature
set remains unequalled (which is why some RPM-based distributions have begun
to ad-“apt” them for their purposes).

12.3.2 Package Installation Using apt-get
The ﬁrst of these tools is apt-get , which represents a sort of intelligent superstruc-
ture for dpkg . It does not oﬀer an interactive interface for package selection, but
could initially be used as a back-end for dselect , to install packages selected in
dselect . Today it is mostly useful on the command line. The most important prop-
erties of apt-get include the following:
• apt-get can manage a set of installation sources simultaneously. For exam- Several installation sources
ple, it is possible to use a “stable” Debian distribution on CD-ROM in par-
allel to a HTTP-based server containing security updates. Packages are nor-
mally installed from CD-ROM; only if the HTTP server oﬀers a more current
version of a package will it be fetched from the network. Certain packages
can be requested from certain sources; you can, for example, use a stable
Debian distribution for the most part but take some packages from a newer
“unstable” distribution.
• It is possible to update all of the distribution at once (using “apt-get dist-upgradeUpgrades
”),
with dependencies being resolved even in the face of package renamings
and removals.
• A multitude of auxiliary tools allows, e. g., setting up caching proxy servers auxiliary tools
for Debian packages (apt-proxy ), installing packages on systems that are not
connected to the Internet (apt-zip ), or retrieving a list of bugs for a pack-
age before actually installing it (apt-listbugs ). With apt-build , you can com-
pile packages with speciﬁc optimisations for your system and create a local
package repository containing such packages.
Package sources for apt-get are declared in /etc/apt/sources.list : package sources

deb http://ftp.de.debian.org/debian/ stable main
deb http://security.debian.org/ stable/updates main
deb-src http://ftp.de.debian.org/debian/stable main
190 12 Software Package Management Using Debian Tools

Binary packages will be fetched from http://ftp.de.debian.org/ , as will the corre-
sponding source code. In addition, the security.debian.org server is accessed, on
which the Debian project places updated package version that ﬁx security bugs.
operating procedure The standard operating procedure using apt-get is as follows: First you update
the local package availability database:

# apt-get update

This consults all package sources and integrates the results into a common pack-
age list. You can install packages using “apt-get install ”:

# apt-get install hello
Reading Package Lists... Done
Building Dependency Tree... Done
The following NEW packages will be installed:
hello
0 upgraded, 1 newly installed, 0 to remove and 0 not upgraded.
Need to get 68.7kB of archives.
After unpacking 566kB of additional disk space will be used.

This will also install or upgrade all packages mentioned in Depends: dependencies,
as well as any packages that these packages depend upon, and so on.
You may also install several packages at the same time:

# apt-get install hello python

or install some packages and install others simultaneously: The

# apt-get install hello- python python-django+

command would remove the hello package and install the python and python-django
packages (including their dependencies). The “+ ” is not mandatory but allowed.
With “apt-get remove ” you can remove packages directly.
simple update The “apt-get upgrade ” installs the newest available versions of all packages in-
stalled on the system. This will not remove installed packages nor install new
packages; packages that cannot be updated without such actions (because depen-
dencies have changed) remain at their present state.
“intelligent” update The “apt-get dist-upgrade ” command enables an “intelligent” conﬂict resolution
scheme which tries to resolve changed dependencies by judiciously removing and
installing packages. This prefers more important packages (according to their pri-
ority) over less important ones.
source code You can fetch a package’s source code using the “apt-get source ” command:

# apt-get source hello

This also works if the binary package is one of several that have been created from
a (diﬀerently named) source package.

B The apt programs are usually conﬁgured by means of the /etc/apt/apt.conf
ﬁle. This includes options for apt-get , apt-cache , and other commands from
the apt bunch.

Exercises
C 12.8 [!1] Use apt-get to install the hello package and then remove it again.

C 12.9 [1] Download the source code for the hello package using apt-get .
12.3 Debian Package Management: The Next Generation 191

12.3.3 Information About Packages
Another useful program is apt-cache , which searches apt-get ’s package sources: apt-cache

$ apt-cache search hello hello in name or description

grhino-data - othello/reversi boardgame - data-files
gtkboard - many board games in one program
hello - The classic greeting, and a good example
hello-debhelper - The classic greeting, and a good example
jester - board game similar to Othello

$ apt-cache show hello
Package: hello
Version: 2.8-2
Installed-Size: 553
Maintainer: Santiago Vila <sanvila@debian.org>
Architecture: amd64

The output of “apt-cache show ” mostly corresponds to that of “dpkg --status ”, except
that it works for all packages in a package source, no matter whether they are
installed locally, while dpkg only deals with packages that are actually installed.
There are also a few other interesting apt-cache subcommands: depends displays
all the dependencies of a package as well as the names of packages fulﬁlling that
dependency:

$ apt-cache depends hello
hello
Depends: libc6
|Depends: dpkg
Depends: install-info

B The vertical bar in the second dependency line indicates that the depen-
dency in this line or the one in the following line must be fulﬁlled. In this
example, the dpkg package or the install-info package must be installed (or
both).

Conversely, rdepends lists the names of all packages depending on the named pack-
age:

$ apt-cache rdepends python
python
Reverse Depends:
libboost-python1.4.1
mercurial-nested
mercurial-nested
python-apt
python-apt

B If a package occurs several times in the list, this is probably because the
original package speciﬁed it several times, typically with version numbers.
The python-apt package, for example, contains among other things

… python (>= 2.6.6-7~), python (<< 2.8), …

to signal that it will only work with particular versions of the Debian Python
package.
192 12 Software Package Management Using Debian Tools

stats provides an overview of the content of the package cache:

$ apt-cache stats
Total package names: 33365 (1335k) All packages in the cache
Normal packages: 25672 Packages that really exist
Pure virtual packages: 757 Placeholders for functionality
Single virtual packages: 1885 Just one implementation
Mixed virtual packages: 267 Several implementations
Missing: 4784 Packages in dependencies that no (longer?) exist
Total distinct versions: 28955 (1506k) Package versions in the cache
Total distinct descriptions: 28955 (695k)
Total dependencies: 182689 (5115k) Number of pairwise relationships
Total ver/file relations: 31273 (500k)
Total Desc/File relations: 28955 (463k)
Total Provides mappings: 5747 (115k)
Total globbed strings: 100 (1148)
Total dependency version space: 756k
Total slack space: 73.5k
Total space accounted for: 8646k

Exercises
C 12.10 [2] How can you ﬁnd all packages that must be installed for a particu-
lar package to work? (Compare the output of “apt-cache depends x11-apps ” to
that of “apt-cache depends libxt6 ”.)

12.3.4 aptitude
The program aptitude does package selection and management and has taken over
the old dselect ’s rôle in Debian GNU/Linux. On the console or inside a terminal
emulator, it features an interactive user interface with menus and dialogs, but also
provides command-line options that are roughly compatible to those of apt-get .
Since Debian 4.0 (popularly called “etch”), aptitude is the recommended program
for package installation and updates.

B Newer versions of aptitude include a GTK+-based user interface that can be
installed optionally.

improvements Compared to apt-get and dselect , aptitude oﬀers various improvements, includ-
ing:
• It does not necessarily need to be invoked as root , but asks for the root pass-
word before actions requiring administrator privileges are performed.
• It can remember which packages have been installed to fulﬁl dependencies,
and remove these automatically if all packages depending on them have
been removed. (In the meantime apt-get has learned to do that, too; see
apt-get (8), the autoremove command.)

• With aptitude , you have interactive access to all versions of a package avail-
able from various package sources, not just the most up-to-date one.
interactive UI The aptitude command invokes the interactive UI (Figure 12.1). Near the top
of the console (or terminal, as the case may be) you see a “menu bar”, below that
there is a line with a short help message and a line with the program’s version
number. The remainder of the screen is split in two parts: The upper half displays
an overview of the various types of package (updated, new, installed, and so on),
the lower half is used for explanatory messages.
With the ↑ and ↓ keys you can navigate in the package list. Lines starting
wiht --- represent the headings of “subtrees” of the package list, and ↩ can be
12.3 Debian Package Management: The Next Generation 193

Figure 12.1: The aptitude program

used to “open” the next level of such a subtree. (You can open all of the subtree
by typing ] .) / gives you a window that lets you enter a search term (or reg-
ular expression) for a package name. When scrolling through the package lists,
explanations for the packages encountered are displayed in the lower part of the
screen, and you can scroll these up or down using the a and z keys. The i key
lets you change from the explanatory text to a representation of the dependencies.
If the cursor bar sits on a package’s line, you can select it for installation or
updating using the + key, or mark it for deletion using - . If you want to remove
it completely (as in “dpkg --purge ”), use _ . = sets a package’s status to “hold”,
which means that it will no longer be automatically upgraded.
With u , you can update the package lists (like “apt-get update ”) and then check
the “Updated Packages” subtree to ﬁnd which packages aptitude would update.
Using U , you can mark all of these packages for actual updating. The “New
Packages” subtree displays those packages added since the last update; f empties
this list and places these packages among the “normal” lists. The Ctrl + t key
combination opens the menu bar, in which you can move using the arrow keys
and select a function using ↩ .
The g command starts the actual installation, update, or package removal. At
ﬁrst it shows an overview of the planned actions, which you may revise using the
usual commands. Another g starts the actual work: First all required new pack-
ages are fetched, then aptitude calls dpkg to atually install or remove the desired
packages.

B Contrary to popular perception, aptitude is not really a front-end to apt-get ,
but does by itself whatever apt-get would otherwise do.

If conﬂicts occur, aptitude oﬀers solution strategies by way of suitable proposals solution strategies
for installations, updates, or package removals, from which you can pick the one
that is most appropriate.

B In its default conﬁguration, aptitude automatically installs even those pack-
ages that a package marks Recommended: . This is not always what is wanted
and can be switched oﬀ from the “Options” menu.

You can install and use aptitude on Ubuntu, but it is not the recommended
program. For this reason, it does not agree 100% with the graphical tools
194 12 Software Package Management Using Debian Tools

proposed for package management by Ubuntu—so you should either do
everything like Ubuntu recommends, or else do everything using aptitude .

12.4 Debian Package Integrity
The debsums program is used to check the integrity of the ﬁles in a single package
(Section 12.2.6). This is nice but does not ensure that an attacker has not manip-
ulated both the ﬁles in the package and the .md5sums ﬁle containing the original
integrity of complete packages checksums. The question remains: How does the Debian project ensure the in-
tegrity of complete packages (and, based on this, the integrity of the whole distri-
bution)? This works as follows:
• Every Debian package is cryptographically signed by a Debian developer.
This means that the receipient of the package can use the developer’s public
key to verify that they received the package in the state it was in when the
developer released it.

B The Debian developer who signed the package must not necessar-
ily have been the person who assembled it. In principle, every De-
bian developer may sign and release any package in Debian (a “non-
maintainer upload”), and this is being done in practice for timely
updates ﬁxing critical security holes and to adopt “orphaned” pack-
ages. Furthermore, there are numerous people who help with Debian
GNU/Linux and, even though they are not formally Debian develop-
ers (or whose applications for developer status are pending), maintain
packages. These people cannot by themselves release packages, but
must do this via a “sponsor” who must be a Debian developer. The
sponsor assumes the responsibility that the package is reasonable.

A You should not overestimate the security gained through digital sig-
natures: A developer’s signature does not guarantee that there is no
malicious code in a package, but only that the developer signed the
package. Theoretically it is possible for a cracker to pass the Debian
developer accreditation procedure and be able to release oﬃcial pack-
ages into the distribution—whose control scripts most users will exe-
cute uncritically as root . Most other Linux distributions share the same
weaknesses.

• The Debian infrastructure only accepts packages for publication that have
been signed by a Debian developer.
• On the Debian server (and all servers mirroring Debian GNU/Linux) there
is a ﬁle (or several) called Packages.gz for each current Debian distribution.
This ﬁle contains the MD5 checksums of all packages in the distribution, as
they are on the server; since the server only accepts packages from accred-
ited developers, these are authentic.
• For each current Debian distribution on the server there is a ﬁle called
Release , which contains the MD5 checksums of the Packages.gz ﬁle(s) in-
volved. This ﬁle is cryptographically signed (the signature is in a ﬁle called
Release.gpg ).

With this chain of checksums and signatures, the integrity of packages in the dis-
tribution can be checked:

• A new package is downloaded and its MD5 checksum is determined.
• We check whether the signature of the Release ﬁle is correct, and, if so, read
the MD5 checksum of Packages.gz from that ﬁle.
12.5 The debconf Infrastructure 195

• With that checksum, we verify the integrity of the actual Packages.gz ﬁle.
• The MD5 checksum of the package given in Packages.gz must match that of
the downloaded ﬁle.
If the MD5 checksum of the downloaded ﬁle does not match the “nominal value”
from Packages.gz , the administrator is made aware of this fact and the package is
not installed (just yet, anyway).

B It is possible to conﬁgure a Debian system such that it only installs packages
that can be veriﬁed in this way. (Usually all you get is warnings which can be
overridden manually.) With this, you can construct an infrastructure where
only packages from a considered-safe-and-sensible “subdistribution” can
be installed. These packages may be from Debian GNU/Linux or else have
been made available locally.

B The APT infrastructure only trusts package sources for which a public
GnuPG key has been placed in the /etc/apt/trusted.gpg . The apt-key pro-
gram is used to maintain this ﬁle.

B The current public keys for Debian package sources are contained in the
debian-archive-keyring package and can be renewed by updating this ﬁle.
(Debian rotates the keys on a yearly basis).

You can ﬁnd out more about managing signed packages in Debian in [F+ 07,
chapter 7]. We explain GnuPG in the Linup Front training manual Linux Admin-
istration II.

12.5 The debconf Infrastructure
Sometimes questions come up during the installation of software packages. For
example, if you are installing a mail server package, it is important to know, in or-
der to generate an appropriate conﬁguration ﬁle, whether the computer in ques-
tion is connected directly to the Internet, whether it is part of a LAN with its own
dedicated mail server, or whether it uses a dial-up connection to access the net. It
is also necessary to know the domain the computer is to use for its messages and
so on.
The debconf mechanism is designed to collect this information and to store
it for future use. It is basically a database for system-wide conﬁguration set-
tings, which can, for example, be accessed by the installation scripts of a package.
To manipulate the database, debconf supports modular user interfaces covering
all tastes from very simple textual prompts to text-oriented dialogs and various
graphical desktop applications such as KDE and GNOME. There are also inter-
faces to popular programming languages like Python.
You can redo the initial debconf-based conﬁguration of a software package at
any time by giving a command like

# dpkg-reconfigure my-package

dpkg-reconfigurerepeats the questions asked during the original installation pro-
cess of the package, using the pre-set user interface.

B You can select another user interface on an ad-hoc basis by means of the
--frontend (or -f ) option. The possible names are given in debconf (7) if you
have installed the debconf-doc package. The default is dialog .

B To change the user interface temporarily if you are not calling dpkg-reconfigure
directly, use the DEBCONF_FRONTEND environment variable:

# DEBCONF_FRONTEND=noninteractive aptitude upgrade
196 12 Software Package Management Using Debian Tools

With dpkg-reconfigure , you can also control the level of detail of the questions
you will be asked. Use the --priority (or -p ) option followed by a priority. The
possible priorities are (in descending order):
critical Questions you absolutely must answer lest terrible things happen.

high Questions without a sensible default—your opinion counts.
medium Questions with a sensible default.
low Trivial questions with a default that works most of the time.

If you say something like

# dpkg-reconfigure --priority=medium my-package

you will be asked all priority critical , high , and medium questions; any priority low
questions will be skipped.

B For ad-hoc changes if debconf is called indirectly, there is also the DEBCONF_PRIORITY
environment variable.

The debconf infrastructure is fairly complex but useful. For example, it is pos-
sible to put the answers into an LDAP database that is accessible to all computers
on a network. You can thus install a large number of machines without manual
intervention. To explain this in detail would, however, be beyond the scope of this
manual.

Exercises
C 12.11 [1] How can you change the pre-set user interface for debconf on a
permanent basis?

12.6 alien : Software From Different Worlds
Many software packages are only available in the popular RPM format. Commer-
cial packages in particular are more likely to be oﬀered for the Red Hat or SUSE
distributions, even though nothing would prevent anyone from trying the soft-
ware on Debian GNU/Linux (serious use may be precluded by the loss of man-
ufacturer support if a non-approved platform is used). You cannot install RPM
packages on a Debian system directly, but the alien program makes it possible
to convert the package formats of various Linux distributions—besides RPM also
the Stampede and Slackware formats (not that these are desperately required)—to
the Debian package format (and vice-versa).
Important: While alien will let you convert packages from one format to another,
there is no guarantee whatsoever that the resulting package will be useful in any
way. On the one hand, the programs in the package may depend on libraries
that are not available (or not available in the appropriate version) in the target
distribution—since alien does not deal with dependencies, you must sort out any
problems of this type manually. On the other hand, it is quite possible that the
package integrates itself into the source distribution in a way that is impossible or
diﬃcult to replicate on the target distribution.
As a matter of principle, the farther “down” a package sits in the system the
smaller the probability that alien will do what you want. With packages that con-
sist of a few executable programs without bizarre library dependencies, the cor-
responding manual pages, and a few example ﬁles, chances are good for alien
to do the Right Thing. With system services that must integrate into the system
boot sequence, things may well look diﬀerent. And you should not even think of
replacing libc …
12.6 Bibliography 197

B alien is used, in particular, to convert LSB-compliant software packages for
installation on a Debian GNU/Linux system—LSB speciﬁes RPM as the
software distribution package format.

After these introductory remarks, we’ll show you quickly how to use alien to
convert a RPM package to a Debian package:

# alien --to-deb paket.rpm

(Where --to-deb represents the default case and may be left out.) The reverse is
possible using

# alien --to-rpm paket.deb

To assemble and disassemble RPM ﬁles, the rpm program must be installed (which
is available as a package for Debian GNU/Linux); to assemble deb packages you
need a few pertinent Debian packages which are listed in alien (1p). (As mentioned
in Section 12.2, you can take deb packages to bits on almost all Linux systems using
“on-board tools” such as tar , gzip , and ar .)

Commands in this Chapter
alien Converts various software packaging formats alien (1) 196
apt-get Powerful command-line tool for Debian GNU/Linux package manage-
ment apt-get (8) 189
aptitude Convenient package installation and maintenance tool (Debian)
aptitude (8) 192
dpkg Debian GNU/Linux package management tool dpkg (8) 182
dpkg-reconfigure Reconﬁgures an already-installed Debian package
dpkg-reconfigure (8) 195

Bibliography
F+ 07 Javier Fernández-Sanguino Peña, et al. “Securing Debian Manual”, 2007.
http://www.debian.org/doc/manuals/securing- debian- howto/
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

13
Package Management with RPM
and YUM

Contents
13.1 Introduction. . . . . . . . . . . . . . . . . . . . . . 200
13.2 Package Management Using rpm . . . . . . . . . . . . . . . 201
13.2.1 Installation and Update . . . . . . . . . . . . . . . 201
13.2.2 Deinstalling Packages . . . . . . . . . . . . . . . . 201
13.2.3 Database and Package Queries . . . . . . . . . . . . . 202
13.2.4 Package Veriﬁcation . . . . . . . . . . . . . . . . . 204
13.2.5 The rpm2cpio Program . . . . . . . . . . . . . . . . 204
13.3 YUM . . . . . . . . . . . . . . . . . . . . . . . . 205
13.3.1 Overview . . . . . . . . . . . . . . . . . . . . 205
13.3.2 Package Repositories . . . . . . . . . . . . . . . . 205
13.3.3 Installing and Removing Packages Using YUM . . . . . . . 206
13.3.4 Information About Packages . . . . . . . . . . . . . . 208
13.3.5 Downloading Packages. . . . . . . . . . . . . . . . 210

Goals
• Knowing the basics of RPM and related tools
• Being able to use rpm for package management
• Being able to use YUM

Prerequisites
• Knowledge of Linux system administration
• Experience with an RPM-based Linux distribution is helpful

adm1-rpm.tex (33e55eeadba676a3 )
200 13 Package Management with RPM and YUM

13.1 Introduction
The “Red Hat Package Manager” (RPM, for short) is a tool for managing software
packages. It supports the straightforward installation and deinstallation of pack-
ages while ensuring that diﬀerent packages do not conﬂict and that dependencies
between the packages are taken into account. In addition, RPM allows you to
specify package queries and to ensure the integrity of packages.
RPM’s core is a database. Software packages add themselves when they are
installed and remove themselves again when they are deinstalled. To allow this,
software packages must be provided in a speciﬁc format, i. e., as RPM packages.
The RPM package format is used used by many distributions (including those
by Red Hat, Novell/SUSE, TurboLinux, and Mandriva). An arbitrary RPM pack-
age, though, cannot generally be installed on any RPM-based distribution without
forethought: Since the RPM package contains compiled software, it must ﬁt the
processor architecture in use; since ﬁle system structure, the type of service con-
trol (init scripts, etc.), and the internal description of dependencies diﬀer between
distributions or even between diﬀerent versions of the same distribution, careless
installation across distribution may cause problems.

B RPM was originally developed by Red Hat and was accordingly called “Red
Hat Package Manager” at ﬁrst. Since various other distributions have taken
to using the program, it has been renamed to “RPM Package Manager”.

B At the moment there is a certain controversy as to who is in charge of further
development of this critical piece of infrastructure. After a long hiatus, dur-
ing which nobody really bothered to put out a canonical version, some Fe-
dora developers tried in 2006/7 to restart RPM development as an oﬃcially
distribution-neutral project (this project is now led by Panu Matilainen of
Red Hat, with developers aﬃliated with some other RPM-using distribu-
tions in support). Independently, Jeﬀ Johnson, the last oﬃcial RPM devel-
oper at Red Hat (who is no longer with the company), is putting work into
RPM and claims that his code represents “the oﬃcial code base”—although
no Linux distribution seems to pay attention.

file name An RPM package has a compound ﬁle name, for example

openssh-3.5p1-107.i586.rpm

which usually consists of the package name (openssh-3.5p1-107 ), the architecture
(i586 ) and the .rpm suﬃx. The package name is used to identify the package inter-
nally once it has been installed. It contains the name of the software (openssh ) and
the software version as assigned by its original developers (3.5p1 ) followed by a
release number (107 ) assigned by the package builder (the distributor).
The “RPM Package Manager” is invoked using the rpm command, followed by
basic mode a basic mode. The most important modes will be discussed presently, excepting
the modes for initialising the RPM database and constructing and signing RPM
packages, which are outside the scope of this course.
options There is a number of global options as well as supplementary, mode-speciﬁc
options. Since some modes and supplementary options are identical, the mode
must (unlike with tar ) be speciﬁed ﬁrst.
Global options include -v and -vv , which increase the “verbosity” of RPM’s
output.

B RPM’s conﬁguration is stored within the /usr/lib/rpm directory; local or in-
dividual customisations are made within the /etc/rpmrc or ~/.rpmrc ﬁles, but
should not be necessary for normal operations.
13.2 Package Management Using rpm 201

13.2 Package Management Using rpm
13.2.1 Installation and Update
An RPM package is installed in -i mode, followed by the package ﬁle’s path name,
such as
# rpm -i /tmp/openssh-3.5p1-107.i586.rpm

You may also specify the path as an HTTP or FTP URL in order to install package
ﬁles that reside on remote serers. It is permissible to specify several packages at
once, as in “rpm -i /tmp/*.rpm ”.
Additionally, there are the two related modes -U (“upgrade”) and -F (“freshen”).
The former removes any older versions of a package as well as installing the new
one, while the latter installs the package only if an earlier version is already
installed (which is subsequently removed).
All three modes support a number of options, which must absolutely be men-
tioned after the mode. Besides -h (“hash mark”, for a progress bar) there is --test ,
which prevents the actual installation and only checks for possible conﬂicts.
When a conﬂict occurs, the package in question is not installed. Conﬂicts arise
if
• an already-installed package is to be installed again,
• a package is to be installed even though it is already installed in a diﬀerent
version (mode -i ) or a more current version (mode -U ),
• the installation would overwrite a ﬁle belonging to a diﬀerent package,
• a package requires a diﬀerent package which is not already installed or
about to be installed.
If the installation fails for any of these reasons, you can force it to be performed
through options. For example, the --nodeps option disables the dependency check.
Further options can inﬂuence the installation itself (rather than just the security
checks). For example, you can move packages to diﬀerent directories on installa-
tion. This is the only way to install, e. g., Apache 1.3 and Apache 2.0 at the same
time, since usually both of them would claim /sbin/http for themselves: one of the
two must move to /usr/local .

13.2.2 Deinstalling Packages
Packages can be deinstalled using -e (“erase”) mode, e. g.,

# rpm -e openssh-3.5p1-107

Note that you need to specify the internal package name rather than the package
ﬁle path, since RPM does not remember the latter. (The next section will tell you
how to ﬁnd out the package name.) You can also abbreviate the package name as
long as it stays unique. If there is no other openssh package, you might also remove
it by

# rpm -e openssh

Again, RPM takes care not to remove packages that other packages depend upon.
The --test and --nodeps options have the same meaning as upon installation;
they must also appear after the mode.
When deinstalling, all of the package’s installed ﬁles will be removed unless
they are conﬁguration ﬁles that you have changed. These will not be removed but configuration files
merely renamed by appending the .rpmsave suﬃx. (The RPM package determines
which of its ﬁles will be considered conﬁguration ﬁles.)
202 13 Package Management with RPM and YUM

13.2.3 Database and Package Queries
The “RPM Package Manager” becomes even more useful if you do not just con-
sider it a package installation and removal tool, but also an information source.
The mode for this is -q (“query”), and you can specify in more detail what kind of
information you would like to obtain and from which package.

Specifying the Package Without further options, rpm expects an internal package
name, which may be abbreviated, and it outputs the full package name:

$ rpm -q openssh
openssh-3.5p1-107

This makes it easy to determine how current your system is. You can also ﬁnd the
package claiming a ﬁle, using the -f option:

$ rpm -qf /usr/bin/ssh
openssh-3.5p1-107

This lets you relate unknown ﬁles to a package. As a third possibility, you can
obtain a list of all installed packages with the -a option:

$ rpm -qa

This list may of course be processed further, as in the following example:1

$ rpm -qa | grep cups
cups-client-1.1.18-82
cups-libs-1.1.18-82
kdelibs3-cups-3.1.1-13
cups-drivers-1.1.18-34
cups-drivers-stp-1.1.18-34
cups-1.1.18-82

Finally, RPM allows you to query a non-installed package. Use -p followed by
the package ﬁle’s name:

$ rpm -qp /tmp/openssh-3.5p1-107.i586.rpm
openssh-3.5p1-107

This does not look too spectacular, since the internal package name was already
part of the package ﬁle name. But the ﬁle name might have been changed until it
no longer had anything to do with the actual package name, and secondly, there
are other questions that you might want to ask.

Specifying the Query If you are not just interested in the package name, you can
extend your query. Every extension may be combined with every way of specify-
ing a packet. Via

$ rpm -qi openssh

you can obtain detailed information (-i ) about the package; while -l provides a
list of all ﬁles belonging to the package, together with -v it forms an equivalent to
the ls -l command:
1 The naming and arrangement of packages is the package preparer’s concern; diﬀerences may occur

depending on the distribution and version.
13.2 Package Management Using rpm 203

$ rpm -qlf /bin/bash
/bin/bash
/bin/sh

$ rpm -qlvf /bin/bash
-rwxr-xr-x root root 491992 Mar 14 2003 /bin/bash
lrwxrwxrwx root root 4 Mar 14 2003 /bin/sh -> bash

It is important to note that the ﬁles listed for a package are only those that show
up in the RPM database, namely those that a package brought along when it was
installed. This does not include ﬁles that were generated during installation (by
the package’s installation scripts) or during system operation (log ﬁles, etc.).
We have already seen that RPM treats conﬁguration ﬁles specially (when de-
installing). The second class of special ﬁles are documentation ﬁles; these can be
omitted from installation. The -c and -d options of query mode behave like -l , but
they conﬁne themselves to conﬁguration and documentation ﬁles, respectively.

Advanced Queries The following details are not relevant for LPIC-1, but they
will improve your understanding of RPM’s concepts and database structure.
Dependencies between packages can belong to various types. For example, a dependencies
package may simply require a shell, i. e., /bin/sh . By means of

$ rpm -qf /bin/sh
bash-2.05b-105

it is straightforward to ﬁnd out that this ﬁle is provided by the bash package (the
same can be done for non-installed packages).
Things are diﬀerent, e. g., for the SAINT package, a security analysis tool which
requires a web browser. Every speciﬁc dependency on a particular web browser
would be unduly limiting. For this reason, RPM lets packages provide or depend
upon “capabilities”. In our example, SAINT requires the abstract capability “web
browser”. The ﬁles and capabilities that a package requires can be queried using
the --requires option:2

$ rpm -q --requires saint
web_browser
/bin/rm
/bin/sh
/usr/bin/perl

The packages providing these capabilities can be found using the --whatprovides
option:

$ rpm -q --whatprovides web_browser
w3m-0.3.2.2-53
mozilla-1.2.1-65
lynx-2.8.4-391

For SAINT, you need just one of these packages.
In the same manner, the --provides and --whatrequires options allow you to
query the services (or ﬁles, with the -l option) that a package oﬀers, and a ser-
vice’s consumers.
2 Here, again, the assignment and naming of capabilities is up to the package preparer; it may thus

diﬀer between distributions and versions.
204 13 Package Management with RPM and YUM

13.2.4 Package Verification
Pre-Installation Checks Two things may happen to a package which might pre-
clude its installation: It may have been damaged during the download, i. e., the
package is erroneous. Or the package is not what it pretends to be, i. e., it has
been falsiﬁed—for example, because some malicious person tries to pass a “Tro-
jan” package oﬀ as the original.
RPM safeguards you against both scenarios: with

$ rpm --checksig /tmp/openssh-3.5p1-107.i586.rpm
/tmp/openssh-3.5p1-107.i586.rpm: md5 gpg OK

an MD5 checksum of the package is compared to the checksum contained within
itself, which guarantees the proper transmission of the package. In addition, the
signature within the package, which was created using the private PGP or GPG
key of the package preparer, is checked using the package preparer’s public key.
This guarantees that the correct package has arrived.
Should the MD5 checksum be correct but not the signature, the output looks
correspondingly diﬀerent:

$ rpm --checksig /tmp/openssh-3.5p1-107.i586.rpm
/tmp/openssh-3.5p1-107.i586.rpm: md5 GPG NOT OK

Of course your distributor’s public key must be available on your system for
the signature checks.

Post-Installation Verification RPM lets you compare certain values within the
RPM database to the ﬁle system. This is done by means of the -V (“verify”) mode;
instead of one or more internal package names, this mode can use all speciﬁca-
tions made available for the query mode.

# rpm -V openssh
.......T c /etc/init.d/sshd
S.5....T c /etc/pam.d/sshd
S.5....T c /etc/ssh/ssh_config
SM5....T c /etc/ssh/sshd_config
.M...... /usr/bin/ssh

This output contains all ﬁles for which at least one “required” value from the
database diﬀers from the “actual” value within the ﬁle system: a “. ” signiﬁes
agreement, while a letter indicates a deviation. The following checks are per-
formed: access mode and ﬁle type (M ), owner and group (U , G ); for symbolic links,
the path of the referenced ﬁle (L ); for device ﬁles, major and minor device num-
bers (D ); for plain ﬁles the size (S ), modiﬁcation time (T ), and content (5 ). Since
conﬁguration ﬁles are unlikely to remain in their original state, they are labeled
with a c .
Even though the veriﬁcation of installed packages using RPM cannot replace
an “intrusion detection system” (why should an intruder not modify the RPM
database as well?), it can be useful to limit the damage, e. g., after a hard disk
crash.

13.2.5 The rpm2cpio Program
RPM packages are essentially cpio archives with a prepended “header”. You can
use this fact to extract individual ﬁles from an RPM package without having to
install the package ﬁrst. Simply convert the RPM package to a cpio archive using
the rpm2cpio program, and feed the archive into cpio . Since rpm2cpio works as a ﬁlter,
you can easily connect the two programs using a pipe:
13.3 YUM 205

$ rpm2cpio hello-2.4-1.fc10.i386.rpm \
> | cpio -idv ./usr/share/man/man1/hello.1.gz
./usr/share/man/man1/hello.1.gz
387 blocks
$ zcat usr/share/man/man1/hello.1.gz | head
.\" DO NOT MODIFY THIS FILE! It was generated by help2man 1.35.
.TH HELLO "1" "December 2008" "hello 2.4" "User Commands"
.SH NAME
hello \- friendly greeting program

Exercises
C 13.1 [2] Use rpm2cpio and cpio to display the list of ﬁles contained in an RPM
package.

13.3 YUM
13.3.1 Overview
The rpm program is useful but does have its limits. As a basic tool it can install pack-
ages that are available as ﬁles or URLs, but, for example, does not help with lo-
cating appropriate, possibly installable packages. Many RPM-based distributions
use YUM (short for “Yellow Dog Updater, Modiﬁed”, after the distribution for
which the program was originally developed) to enable access to package sources package sources
(repositories) available on the Internet or on CD-ROM.

B In RPM-based distributions, YUM takes up approximately the same “eco-
logical niche” occupied by apt-get in Debian GNU/Linux and its deriva-
tives.

B YUM is usually controlled via the command line, but the “yum shell ” com-
mand starts a “YUM shell” where you can enter multiple YUM commands
interactively.

13.3.2 Package Repositories
YUM introduces the concept of package repositories. A package repository is a set of
RPM packages that is available via the network and allows the installation of pack-
ages with YUM. The “yum repolist ” command outputs a list of conﬁgured package
repositories:

$ yum repolist
Loaded plugins: refresh-packagekit
repo id repo name status
fedora Fedora 10 - i386 enabled: 11416
updates Fedora 10 - i386 - Updates enabled: 3324
repolist: 14740

“yum repolist disabled ” yields a list of known but disabled repositories:

$ yum repolist disabled
Loaded plugins: refresh-packagekit
repo id repo name status
fedora-debuginfo Fedora 10 - i386 - Debug disabled
fedora-source Fedora 10 - Source disabled
206 13 Package Management with RPM and YUM

rawhide Fedora - Rawhide - Development disabled

To enable a repository, you need to give the --enablerepo= option, followed by the
“repo ID” from the list. This only works in connection with a “genuine” yum com-
mand; repolist is fairly innocuos:

$ yum --enablerepo=rawhide repolist
Loaded plugins: refresh-packagekit
rawhide | 3.4 kB 00:00
rawhide/primary_db | 7.2 MB 00:14
repo id repo name status
fedora Fedora 10 - i386 enabled: 11416
rawhide Fedora - Rawhide - Development enabled: 12410
updates Fedora 10 - i386 - Updates enabled: 3324
repolist: 27150

You can disable a repository using the --disablerepo option.

B Repositories are most conveniently made known to YUM by means of con-
ﬁguration ﬁles in the /etc/yum.repos.d directory. (You could also enter them
into /etc/yum.conf directly, but this is more inconvenient to manage.)

B YUM keeps itself current as far as the content of repositories is concerned.
There is no equivalent to the Debian tools’ “apt-get update ”.

13.3.3 Installing and Removing Packages Using YUM
To install a new package using YUM, you merely need to know its name. YUM
checks whether the active repositories contain an appropriately-named package,
resolves any dependencies the package may have, downloads the package and
possibly other packages that it depends upon, and installs all of them:

# yum install hello
Setting up Install Process
Parsing package install arguments
Resolving Dependencies
--> Running transaction check
---> Package hello.i386 0:2.4-1.fc10 set to be updated
--> Finished Dependency Resolution

Dependencies Resolved

=============================================================
Package Arch Version Repository Size
=============================================================
Installing:
hello i386 2.4-1.fc10 updates 68 k

Transaction Sum
=============================================================
Install 1 Package(s)
Update 0 Package(s)
Remove 0 Package(s)

Total download size: 68 k
Is this ok [y/N]: y
Downloading Packages:
hello-2.4-1.fc10.i386.rpm | 68 kB 00:00
13.3 YUM 207

====================== Entering rpm code =======================
Running rpm_check_debug
Running Transaction Test
Finished Transaction Test
Transaction Test Succeeded
Running Transaction
Installing : hello 1/1
====================== Leaving rpm code ========================

Installed:
hello.i386 0:2.4-1.fc10

Complete!

B YUM accepts not just simple package names, but also package names with
architecture speciﬁcations, version numbers, and release numbers. Check
yum (8) to ﬁnd the allowable formats.

Removing packages is just as simple:

# yum remove hello

This will also remove packages that this package depends upon—as long as these
are not required by another installed package, anyway.

B Instead of “yum remove ” you can also say “yum erase ”—the two are equivalent.
You can update packages using “yum update ”:

# yum update hello

checks whether a newer version of the package is available and installs that if
this is the case. YUM takes care that all dependencies are resolved. “yum update ”
without a package name attempts to update all installed packages.

B When the --obsoletes is speciﬁed (the default case), yum tries to handle the
case where one package has been replaced by another (of a diﬀerent name).
This makes full upgrades of the whole distribution easier to perform.

B “yum upgrade ” is the same as “yum update --obsoletes ”—but saves some typing
in the case that you have switched oﬀ the obsoletes option in the conﬁgura-
tion.

B YUM supports the idea of “package groups”, i. e., packages that together
are useful for a certain task. The available package groups can be displayed
using “yum grouplist ”:

$ yum grouplist
Loaded plugins: refresh-packagekit
Setting up Group Process
Installed Groups:
Administration Tools
Authoring and Publishing
Base
Dial-up Networking Support
Editors

B If you want to know which packages a group consists of, use “yum groupinfo ”:
208 13 Package Management with RPM and YUM

$ yum groupinfo 'Printing Support'
Loaded plugins: refresh-packagekit
Setting up Group Process

Group: Printing Support
Description: Install these tools to enable the system to
print or act as a print server.
Mandatory Packages:
cups
ghostscript
Default Packages:
a2ps
bluez-cups
enscript

A group is considered “installed” if all its “mandatory” packages are in-
stalled. Besides these there are “default packages” and “optional packages”.

B “yum groupinstall ” lets you install the packages of a group. The conﬁguration
option group_package_types determines which class package will actually be
installed—usually the “mandatory” and the “default packages”.

B “yum groupremove ” removes all packages of a group, without taking into ac-
count package classes (group_package_types is ignored). Note that packages
can belong to more than one group at the same time, so they may be miss-
ing from group 𝑋 after having been removed along with group 𝑌.

13.3.4 Information About Packages
The “yum list ” command is available to ﬁnd out which packages exist:

$ yum list gcc
Loaded plugins: refresh-packagekit
Installed Packages
gcc.i386 4.3.2-7 installed

You can also give a search pattern (it is best to put it inside quotes so the shell will
not mess with it):

$ yum list "gcc*"
Loaded plugins: refresh-packagekit
Installed Packages
gcc.i386 4.3.2-7 installed
gcc-c++.i386 4.3.2-7 installed
Availabe Packages
gcc-gfortran.i386 4.3.2-7 fedora
gcc-gnat.i386 4.3.2-7 fedora

The “installed packages” are installed on the local system, while the “available
packages” can be fetched from repositories. The repository oﬀering the package
is displayed on the far right.
To restrict the search to locally installed, or uninstalled but available, packages,
you can use “yum list installed ” or “yum list available ”:

$ yum list installed "gcc*"
Loaded plugins: refresh-packagekit
13.3 YUM 209

Installed Packages
gcc.i386 4.3.2-7 installed
gcc-c++.i386 4.3.2-7 installed
$ yum list available "gcc*"
Loaded plugins: refresh-packagekit
Available Packages
gcc-gfortran.i386 4.3.2-7 fedora
gcc-gnat.i386 4.3.2-7 fedora

B “yum list updates ” lists the packages that are installed and for which updates
are available, while “yum list recent ” lists the packages that have “recently”
arrived in a repository. “yum list extras ” points out packages that are in-
stalled locally but are not available from any repository.

To ﬁnd out more about a package, use “yum info ”:

$ yum info hello
Loaded plugins: refresh-packagekit
Installed Packages
Name : hello
Arch : i386
Version : 2.4
Release : 1.fc10
Size : 186 k
Repo : installed
Summary : Prints a Familiar, Friendly Greeting
URL : http://www.gnu.org/software/hello/
License : GPLv3+ and GFDL and BSD and Public Domain
Description: Hello prints a friendly greeting. It also serves as a
: sample GNU package, showing practices that may be
: useful for GNU projects.

The advantage over “rpm -qi ” is that “yum info ” also works for packages that are not
installed locally but are available from a repository.

B You con otherwise use “yum info ” like “yum list”—“yum info installed ”, for ex-
ample, displays detailed information about all installed packages.

Using “yum search ”, you can search for all packages in whose name or descrip-
tion a given string occurs:

$ yum search mysql
Loaded plugins: refresh-packagekit
============================ Matched: mysql ========================
dovecot-mysql.i386 : MySQL backend for dovecot
koffice-kexi-driver-mysql.i386 : Mysql-driver for kexi
libgda-mysql.i386 : MySQL provider for libgda

Unfortunately, the resulting list is unsorted and a little diﬃcult to read. yum uses
boldface to emphasise the places where the search string occurs.

B You can examine a package’s dependencies using “yum deplist ”:
$ yum deplist gcc
Loaded plugins: refresh-packagekit
Finding dependencies:
package: gcc.i386 4.3.2-7
210 13 Package Management with RPM and YUM

dependency: binutils >= 2.17.50.0.17-3
provider: binutils.i386 2.18.50.0.9-7.fc10
dependency: libc.so.6(GLIBC_2.3)
provider: glibc.i386 2.9-2

13.3.5 Downloading Packages
If you want to download a package from a repository but do not want to install it
outright, you can use the yumdownloader program. A command like

$ yumdownloader --destdir /tmp hello

searches the repositories for the hello package just like YUM would and down-
loads the corresponding ﬁle to the /tmp directory.
The --resolve option causes dependencies to be resolved and any other missing
packages to be downloaded as well—but only those that are not installed on the
local system.

B With the --urls option, nothing is downloaded at all, but yumdownloader out-
puts the URLs of the packages it would otherwise have downloaded.

B With the --source option, yumdownloader downloads source RPMs instead of
binary RPMs.

Commands in this Chapter
cpio File archive manager cpio (1) 204
rpm Package management tool used by various Linux distributions (Red Hat,
SUSE, …) rpm (8) 200
rpm2cpio Converts RPM packages to cpio archives rpm2cpio (1) 204
yum Convenient RPM package maintenance tool yum (8) 205

Summary
• RPM is a system for Linux software package management which is used by
various distributions such as those by Red Hat and Novell/SUSE.
• YUM is a front-end for rpm that gives access to package repositories over the
network.
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

A
Sample Solutions
This appendix contains sample solutions for selected exercises.

1.1 Access control applies to normal users but not the administrator. root may do
anything! The root account should only be used to execute commands that really
require root ’s privileges, e. g., to partition the disk, to create ﬁle systems, to add
user accounts, or to change system conﬁguration ﬁles. All other actions should be
performed from an unprivileged account. This includes invoking administration
commands to gather information (where possible) and unpacking tar archives.

1.2 As root you may do anything, therefore it is very easy to damage the system,
e. g., through inadvertently mistyped commands.

1.3 This question aims at a comparison to other operating systems. Depend-
ing on the system in question, there are no access controls at all (DOS, Win-
dows 95/98) or diﬀerent methods for access control (Windows NT/2000/XP or
Windows Vista). Accordingly, the former do not support administrator access (as
opposed to normal user access), while the latter even allow the creation of several
administrator accounts.

1.4 Basically you can log in as root , or create a UID 0 shell using su . The latter
method is better, e. g., because the change including the former UID is logged.

1.5 The shell prompt often looks diﬀerent. In addition, the id command may
help.

1.6 You can either log in directly or su . For frequent changes, it is a good idea to
log in on two consoles at the same time, and obtain a root shell using su on one.
Alternatively, you could open several terminal windows on a GUI.

1.7 You will generally ﬁnd the log entry in /var/log/messages .

1.10 The obvious advantage is that administration is possible from anywhere, if
necessary by using an internet-enabled cell phone on the beach, and without hav-
ing to have access to specialised hardware or software. The obvious disadvantage
is that you need to secure access to the administration tool very carefully, in order
to prevent unbidden guests to “misconﬁgure” your system (or worse). This may
imply that (the obvious advantage notwithstanding) you may be able to provide
the administration tool only from within the local network, or that you should
212 A Sample Solutions

secure access to it using strong cryptography (e. g., SSL with client certiﬁcates). If
you consider deploying Webmin in your company, you should discuss the possi-
bility of external access very carefully with the appropriate decision makers and/or
corporate data protection oﬃcers in order to avoid extremely dire consequences
that could hit you if problems appear. Consider yourself warned.

2.1 By their respective numerical UIDs and GIDs.

2.2 This works but is not necessarily a good idea. As far as the system is con-
cerned, the two are a single user, i. e., all ﬁles and processes with that UID belong
to both user names.

2.3 A pseudo-user’s UID is used by programs in order to obtain particular well-
deﬁned access rights.

2.4 Whoever is in group disk has block-level read and write permission to the
system’s disks. With knowledge of the ﬁle system structure it is easy to make
a copy of /bin/sh into a SUID root shell (section 3.5) by changing the ﬁle system
metadata directly on disk. Thus, group disk membership is tantamount to root
privileges; you should put nobody into the disk group whom you would not want
to tell the root password outright.

2.5 You will usually ﬁnd an “x ”. This is a hint that the password that would
usually be stored there is indeed stored in another ﬁle, namely /etc/shadow , which
unlike the former ﬁle is readable only for root .

2.6 There are basically two possibilities:
1. Nothing. In this case the system should turn you away after you entered
your password, since no user account corresponds to the all-uppercase user
name.
2. From now on, the system talks to you in uppercase letters only. In this case
your Linux system assumes that you are sitting in front of an absolutely
antediluvial terminal (1970s vintage or so) that does not support lowercase
letters, and kindly switches its processing of input and output data such that
uppercase letters in the input are interpreted as lowercase, and lowercase
letters in the output are displayed as uppercase. Today this is of limited
beneﬁt (except if you work in a computer museum), and you should log out
as quickly again as possible before your head explodes. Since this behaviour
is so atavistic, not every Linux distribution goes along with it, though.

2.7 Use getent , cut , and sort to generate lists of user names for the databases, and
comm to compare the two lists.

2.8 Use the passwd command if you’re logged in as user joe , or “passwd joe ” as
root .
In joe ’s entry in the /etc/shadow ﬁle there should be a diﬀerent value in the
second ﬁeld, and the date of the last password change (ﬁeld 3) should show the
current date (in what unit?)

2.9 As root , you set a new password for him using “passwd dumbo ”, as you cannot
retrieve his old one even though you are the administrator.

2.10 Use the command “passwd -n 7 -x 14 -w 2 joe ”. You can verify the settings
using “passwd -S joe ”.
A Sample Solutions 213

2.11 Use the useradd command to create the user, “usermod -u ” to modify the UID.
Instead of a user name, the ﬁles should display a UID as their owner, since no user
name is known for that UID …

2.12 For each of the three user accounts there should be one line in /etc/passwd
and one in /etc/shadow . To work with the accounts, you do not necessarily need a
password (you can use su as root ), but if you want to login you do. You can create
a ﬁle without a home directory by placing it in /tmp (in case you forgot—a home
directory for a new user would however be a good thing).

2.13 Use the userdel command to delete the account. To remove the ﬁles, use the
“find / -uid ⟨UID⟩ -delete ” command.

2.14 If you use “usermod -u ”, you must reassign the user’s ﬁle to the new UID,
for example by means of “find / -uid ⟨UID⟩ -exec chown test1 {} \; ” or (more eﬃ-
ciently) “chown -R --from= ⟨UID⟩ test1 / ”. In each case, ⟨UID⟩ is the (numerical) for-
mer UID.

2.15 You can either edit /etc/passwd using vipw or else call usermod .

2.16 Groups make it possible to give speciﬁc privileges to groups [sic!] of users.
You could, for example, add all HR employees to a single group and assign that
group a working directory on a ﬁle server. Besides, groups can help organise
access rights to certain peripherals (e. g., by means of the groups disk , audio , or
video ).

2.17 Use the “mkdir ⟨directory⟩” command to create the directory and “chgrp
⟨groupname⟩ ⟨directory⟩” to assign that directory to the group. You should also set
the SGID bit to ensure that newly created ﬁles belong to the group as well.

2.18 Use the following commands:

# groupadd test
# gpasswd -a test1 test
Adding user test1 to group test
# gpasswd -a test2 test
Adding user test2 to group test
# gpasswd test
Changing the password for group test
New Password:x9q.Rt/y
Re-enter new password:x9q.Rt/y

To change groups, use the “newgrp test ” command. You will be asked for the pass-
word only if you are not a member of the group in question.

3.1 A new ﬁle is assigned to your current primary group. You can’t assign a ﬁle
to a group that you are not a member of—unless you are root .

3.3 077 and ”‘u=rwx,go= ”’, respectively.

3.5 This is the SUID or SGID bit. The bits cause a process to assume the
UID/GID of the executable ﬁle rather than that of the executing user. You can
see the bits using “ls -l ”. Of course you may change all the permissions on your
own ﬁles. However, at least the SUID bit only makes sense on binary executable
ﬁles, not shell scripts and the like.
214 A Sample Solutions

3.6 One of the two following (equivalent) commands will serve:

$ umask 007
$ umask -S u=rwx,g=rwx

You may perhaps ask yourself why this umask contains x bits. They are indeed
irrelevant for ﬁles, as ﬁles are not created executable by default. However it might
be the case that subdirectories are desired in the project directory, and it makes
sense to endow these with permissions that allow them to be used reasonably.

3.7 The so-called “sticky bit” on a directory implies that only the owner of a ﬁle
(or the owner of the directory) may delete or rename it. You will ﬁnd it, e. g., on
the /tmp directory.

3.9 This doesn’t work with the bash shell (at least not without further trickery).
We can’t speak for other shells here.

3.11 You cannot do this with chattr alone, since various attributes can be dis-
played with lsattr but not set with chattr . Read up on the details in chattr (1).—In
addition, some attributes are only deﬁned for “plain” ﬁles while others are only
deﬁned for directories; you will, for example, ﬁnd it diﬃcult to make the D and
E attributes visible for the same “ﬁle system object” at the same time. (The E at-
tribute is to do with transparent compression, which cannot be used on directo-
ries, while D only applies to directories—write operations to such directories will
be performed synchronously.)

4.1 In the directory of a process below /proc there is a ﬁle called environ which
contains the environment variables of that process. You can output this ﬁle using
cat . The only blemish is that the variables in this ﬁle are separated using zero
bytes, which looks messy on the screen; for convenience, you might use something
like “tr "\0" "\n" </proc/4711/environ ” to display the environment.

4.2 Funnily enough, the limit is not documented in any obvious placees. In
/usr/include/linux/threads.h on a Linux 2.6 kernel, the constant PID_MAX_LIMIT is de-
ﬁned with a value of 32768; this is the lowest value that will by default not be
assigned to processes. You can query the actual value in /proc/sys/kernel/pid_max
(or even change it—the maximum for 32-bit platforms is actually 32768, while on
64-bit systems you may set an arbitrary value of up to 222 , which is approximately
4 million).
The PIDs assigned to processes rise monotonically at ﬁrst. When the above-
mentioned limit is reached, assignment starts again with lower PIDs, where PIDs
that are still being used by processes are of course not given again to others. Many
low PIDs are assigned to long-running daemons during the boot process, and for
this reason after the limit has been reached, the search for unused PIDs starts again
not at PID 1 but at PID 300. (The details are in the kernel/pid_namespace.c ﬁle within
the Linux source code.)

4.4 As we said, zombies arise when the parent process does not pick up the
return code of a child process. Thus, to create a zombie you must start a child
process and then prevent the parent process from picking up its return code, for
example by stopping it by means of a signal. Try something like

$ sh
$ echo $$ In the subshell
12345
$ sleep 20
In a diﬀerent window:
A Sample Solutions 215

$ kill -STOP 12345
Wait
$ ps u | grep sleep
joe 12346 0.0 0.0 3612 456 pts/2 Z 18:19 0:00 sleep 20

4.5 Consult ps (1).

4.6 Try

$ ps -o pid,ppid,state,cmd

4.7 Usually SIGCHLD (“child process ﬁnished”—sometimes called SIGCLD ), SIGURG
(urgent data was received on a network connection) and SIGWINCH (the size of the
window for a text-based program was changed). These three events are so inane
that the process should not be terminated on their account.

4.8 Something like

$ pgrep -u hugo

should suﬃce.

4.10 Use, e. g., the “renice -10 ⟨PID⟩” command. You can only specify negative
nice values as root .

6.2 sda1 , sda2 , sda5 , sda6 , and sdb1 , sdb5 , sdb6 , sdb7 .

7.2 Use tune2fs with the -c , -u and -m options.

7.3 mkreiserfs /dev/sdb5

7.6 /etc/fstab contains all frequently-used ﬁle systems and their mount points,
while /etc/mtab contains those ﬁle systems that are actually mounted at the mo-
ment.

8.1 The boot loader can be placed inside the MBR, in another (partition) boot
sector, or a ﬁle on disk. In the two latter cases, you will need another boot loader
that can “chain load” the Linux boot loader. Be that as it may, for a Linux system
you absolutely need a boot loader that can boot a Linux kernel, such as GRUB
(there are others).

8.3 Assign a password preventing the unauthorised entry of kernel parameters.
With GRUB Legacy, e. g., using

password --md5 ⟨encrypted keyword⟩

lock helps with the password request for a speciﬁc operating system.

9.3 You can display the previous and current runlevel using runlevel . If the pre-
vious runlevel is “N ” that means “none”—the system started into the current run-
level. To change, say “init 2 ”, then “runlevel ” again to check.
216 A Sample Solutions

9.4 A possible entry for the inittab ﬁle might be
aa:A:ondemand:/bin/date >/tmp/runlevel-a.txt

This entry should write the current time to the mentioned ﬁle if you activate it
using “telinit A ”. Don’t forget the “telinit q ” to make init reread its conﬁguration
ﬁle.

9.5 Call the syslog init script with the restart or reload parameters.

9.6 For example, by using “chkconfig -l ” (on a SUSE or Red Hat system).

9.7 It is tempting just to remove the symbolic links from the runlevel directory
in question. However, depending on the distribution, they may reappear after the
next automated change. So if your distribution uses a tool like chkconfig or insserv
you had better use that.

9.8 You should be prepared for the system asking for the root password.

9.10 Use the
# shutdown -h +15 'This is just a test'

command; everything that you pass to shutdown after the delay will be sent to your
users as a broadcast message. To cancel the shutdown, you can either interrupt
the program using the Ctrl + c key combination (if you started shutdown in the
foreground), or give the “shutdown -c ” command.

9.10 The ﬁle name will be sent as the message.

10.4 The unit ﬁle does not need to be modiﬁed in order to express dependencies.
This makes the automatic installation and, in particular, deinstallation of units
as part of software packages easier (e. g., in the context of a distribution-speciﬁc
package management tool) and allows the seamless updating of unit ﬁles by a
distribution.

10.10 There is no exact equivalent because systemd does not use the runlevel
concept. You can, however, display all currently active targets:
# systemctl list-units -t target

10.11 “systemctl kill ” guarantees that the signal will only be sent to processes
belonging to the unit in question. The other two commands send the signal to all
processes whose name happens to be example .

10.13 You can’t (“systemctl mask ” outputs an error message). You must deactivate
the service and then remove, move, or rename the unit ﬁle.

11.3 You can ﬁnd out using a shell script like
for f in /usr/bin/*
do
printf "%4u %s\n" $(ldd $f 2>/dev/null | wc -l) $f
done | sort -nr | head

(the 2>/dev/null suppresses the error message that ldd outputs if you feed it a shell
script rather than an executable program.) On the author’s system, mplayer is at
the top of the list, sporting an impressive 118 libraries.
A Sample Solutions 217

11.4 This would be a major security hole, since you could attempt to use a pri-
vate directory in LD_LIBRARY_PATH to sneak a “Trojan horse” version of a common
library into the program. For example, you can assume with near certainty that
a program such as passwd will call a C function like open() . If you should succeed
in arranging for this function to be called not from libc but your own library, you
would essentially be able to do whatever you wanted using root privileges. This
of course is not desirable (in the greater scheme of things, anyway).

12.11 Try

# dpkg-reconfigure debconf

13.1 This is most easily done using something like

$ rpm2cpio ⟨package⟩ | cpio -t
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

B
LPIC-1 Certification

B.1 Overview
The Linux Professional Institute (LPI) is a vendor-independent non-proﬁt organi-
zation dedicated to furthering the professional use of Linux. One aspect of the
LPI’s work concerns the creation and delivery of distribution-independent certi-
ﬁcation exams, for example for Linux professionals. These exams are available
world-wide and enjoy considerable respect among Linux professionals and em-
ployers.
Through LPIC-1 certiﬁcation you can demonstrate basic Linux skills, as re-
quired, e. g., for system administrators, developers, consultants, or user support
professionals. The certiﬁcation is targeted towards Linux users with 1 to 3 years
of experience and consists of two exams, LPI-101 and LPI-102. These are oﬀered
as computer-based multiple-choice and ﬁll-in-the-blanks tests in all Pearson VUE
and Thomson Prometric test centres. On its web pages at http://www.lpi.org/ , the
LPI publishes objectives outlining the content of the exams. objectives
This training manual is part of Linup Front GmbH’s curriculum for preparation
of the LPI-101 exam and covers part of the oﬃcial examination objectives. Refer
to the tables below for details. An important observation in this context is that
the LPIC-1 objectives are not suitable or intended to serve as a didactic outline for
an introductory course for Linux. For this reason, our curriculum is not strictly
geared towards the exams or objectives as in “Take classes 𝑥 and 𝑦, sit exam 𝑝,
then take classes 𝑎 and 𝑏 and sit exam 𝑞.” This approach leads many prospective
students to the assumption that, being complete Linux novices, they could book
𝑛 days of training and then be prepared for the LPIC-1 exams. Experience shows
that this does not work in practice, since the LPI exams are deviously constructed
such that intensive courses and exam-centred “swotting” do not really help.
Accordingly, our curriculum is meant to give you a solid basic knowledge of
Linux by means of a didactically reasonable course structure, and to enable you as
a participant to work independently with the system. LPIC-1 certiﬁcation is not a
primary goal or a goal in itself, but a natural consequence of your newly-obtained
knowledge and experience.

B.2 Exam LPI-101
The following table displays the objectives for the LPI-101 exam and the materials
covering these objectives. The numbers in the columns for the individual manuals
refer to the chapters containing the material in question.

adm1-objs-101.tex (33e55eeadba676a3 )
220 B LPIC-1 Certification

No Wt Title GRD1 ADM1
101.1 2 Determine and conﬁgure hardware settings – 5–6
101.2 3 Boot the system – 8–10
101.3 3 Change runlevels/boot targets and shutdown or reboot system – 9–10
102.1 2 Design hard disk layout – 6
102.2 2 Install a boot manager – 8
102.3 1 Manage shared libraries – 11
102.4 3 Use Debian package management – 12
102.5 3 Use RPM and YUM package management – 13
103.1 4 Work on the command line 3–4 –
103.2 3 Process text streams using ﬁlters 8 –
103.3 4 Perform basic ﬁle management 6, 11 7.3
103.4 4 Use streams, pipes and redirects 8 –
103.5 4 Create, monitor and kill processes – 4
103.6 2 Modify process execution priorities – 4
103.7 2 Search text ﬁles using regular expressions 7–8 –
103.8 3 Perform basic ﬁle editing operations using vi 5, 7 –
104.1 2 Create partitions and ﬁlesystems – 6–7
104.2 2 Maintain the integrity of ﬁlesystems – 7
104.3 3 Control mounting and unmounting of ﬁlesystems – 7
104.4 1 Manage disk quotas – 7.4
104.5 3 Manage ﬁle permissions and ownership – 3
104.6 2 Create and change hard and symbolic links 6 –
104.7 2 Find system ﬁles and place ﬁles in the correct location 6, 10 –

B.3 Exam LPI-102
The following table displays the objectives for the LPI-102 exam and the materials
covering these objectives. The numbers in the columns for the individual manuals
refer to the chapters containing the material in question.

No Wt Title ADM1 GRD2 ADM2
105.1 4 Customize and use the shell environment – 1–2 –
105.2 4 Customize or write simple scripts – 2–5 –
105.3 2 SQL data management – 8 –
106.1 2 Install and conﬁgure X11 – 11 –
106.2 1 Setup a display manager – 11 –
106.3 1 Accessibility – 12 –
107.1 5 Manage user and group accounts and related system ﬁles 2 – –
107.2 4 Automate system administration tasks by scheduling jobs – 9 –
107.3 3 Localisation and internationalisation – 10 –
108.1 3 Maintain system time – – 8
108.2 3 System logging – – 1–2
108.3 3 Mail Transfer Agent (MTA) basics – – 11
108.4 2 Manage printers and printing – – 9
109.1 4 Fundamentals of internet protocols – – 3–4
109.2 4 Basic network conﬁguration – – 4–5, 7
109.3 4 Basic network troubleshooting – – 4–5, 7
109.4 2 Conﬁgure client side DNS – – 4
110.1 3 Perform security administration tasks 2 – 4–5, 13
110.2 3 Setup host security 2 – 4, 6–7, 13
110.3 3 Securing data with encryption – – 10, 12
B LPIC-1 Certification 221

B.4 LPI Objectives In This Manual
101.1 Determine and configure hardware settings
Weight 2
Description Candidates should be able to determine and conﬁgure fundamen-
tal system hardware.
Key Knowledge Areas

• Enable and disable integrated peripherals
• Conﬁgure systems with or without external peripherals such as keyboards
• Diﬀerentiate between the various types of mass storage devices
• Know the diﬀerences between coldplug and hotplug devices
• Determine hardware resources for devices
• Tools and utilities to list various hardware information (e.g. lsusb , lspci , etc.)
• Tools and utilities to manipulate USB devices
• Conceptual understanding of sysfs, udev, dbus

The following is a partial list of the used ﬁles, terms and utilities:

• /sys/
• /proc/
• /dev/
• modprobe
• lsmod
• lspci
• lsusb

101.2 Boot the system
Weight 3
Description Candidates should be able to guide the system through the booting
process.
Key Knowledge Areas

• Provide common commands to the boot loader and options to the kernel at
boot time
• Demonstrate knowledge of the boot sequence from BIOS to boot completion
• Understanding of SysVinit and systemd
• Awareness of Upstart
• Check boot events in the log ﬁles

The following is a partial list of the used ﬁles, terms and utilities:

• dmesg
• BIOS
• bootloader
• kernel
• initramfs
• init
• SysVinit
• systemd

101.3 Change runlevels/boot targets and shutdown or reboot sys-
tem
Weight 3
222 B LPIC-1 Certification

Description Candidates should be able to manage the SysVinit runlevel or sys-
temd boot target of the system. This objective includes changing to single user
mode, shutdown or rebooting the system. Candidates should be able to alert users
before switching runlevels/boot targets and properly terminate processes. This
objective also includes setting the default SysVinit runlevel or systemd boot target.
It also includes awareness of Upstart as an alternative to SysVinit or systemd.
Key Knowledge Areas

• Set the default runlevel or boot target
• Change between runlevels/boot targets including single user mode
• Shutdown and reboot from the command line
• Alert users before switching runlevels/boot targets or other major system
events
• Properly terminate processes

The following is a partial list of the used ﬁles, terms and utilities:

• /etc/inittab
• shutdown
• init
• /etc/init.d/
• telinit
• systemd
• systemctl
• /etc/systemd/
• /usr/lib/systemd/
• wall

102.1 Design hard disk layout
Weight 2
Description Candidates should be able to design a disk partitioning scheme for
a Linux system.
Key Knowledge Areas

• Allocate ﬁlesystems and swap space to separate partitions or disks
• Tailor the design to the intended use of the system
• Ensure the /boot partition conforms to the hardware architecture require-
ments for booting
• Knowledge of basic features of LVM

The following is a partial list of the used ﬁles, terms and utilities:

• / (root) ﬁlesystem
• /var ﬁlesystem
• /home ﬁlesystem
• /boot ﬁlesystem
• swap space
• mount points
• partitions

102.2 Install a boot manager
Weight 2
Description Candidates should be able to select, install and conﬁgure a boot
manager.
Key Knowledge Areas

• Providing alternative boot locations and backup boot options
B LPIC-1 Certification 223

• Install and conﬁgure a boot loader such as GRUB Legacy
• Perform basic conﬁguration changes for GRUB 2
• Interact with the boot loader

The following is a partial list of the used ﬁles, terms and utilities:

• menu.lst , grub.cfg and grub.conf
• grub-install
• grub-mkconfig
• MBR

102.3 Manage shared libraries
Weight 1
Description Candidates should be able to determine the shared libraries that
executable programs depend on and install them when necessary.
Key Knowledge Areas

• Identify shared libraries
• Identify the typical locations of system libraries
• Load shared libraries

The following is a partial list of the used ﬁles, terms and utilities:

• ldd
• ldconfig
• /etc/ld.so.conf
• LD_LIBRARY_PATH

102.4 Use Debian package management
Weight 3
Description Candidates should be able to perform package management using
the Debian package tools.
Key Knowledge Areas

• Install, upgrade and uninstall Debian binary packages
• Find packages containing speciﬁc ﬁles or libraries which may or may not be
installed
• Obtain package information like version, content, dependencies, package
integrity and installation status (whether or not the package is installed)

The following is a partial list of the used ﬁles, terms and utilities:

• /etc/apt/sources.list
• dpkg
• dpkg-reconfigure
• apt-get
• apt-cache
• aptitude

102.5 Use RPM and YUM package management
Weight 3
Description Candidates should be able to perform package management using
RPM and YUM tools.
Key Knowledge Areas
224 B LPIC-1 Certification

• Install, re-install, upgrade and remove packages using RPM and YUM
• Obtain information on RPM packages such as version, status, dependencies,
integrity and signatures
• Determine what ﬁles a package provides, as well as ﬁnd which package a
speciﬁc ﬁle comes from

The following is a partial list of the used ﬁles, terms and utilities:

• rpm
• rpm2cpio
• /etc/yum.conf
• /etc/yum.repos.d/
• yum
• yumdownloader

103.3 Perform basic file management
Weight 4
Description Candidates should be able to use the basic Linux commands to
manage ﬁles and directories.
Key Knowledge Areas

• Copy, move and remove ﬁles and directories individually
• Copy multiple ﬁles and directories recursively
• Remove ﬁles and directories recursively
• Use simple and advanced wildcard speciﬁcations in commands
• Using ﬁnd to locate and act on ﬁles based on type, size, or time
• Usage of tar, cpio and dd

The following is a partial list of the used ﬁles, terms and utilities:

• cp
• find
• mkdir
• mv
• ls
• rm
• rmdir
• touch
• tar
• cpio
• dd
• file
• gzip
• gunzip
• bzip2
• xz
• ﬁle globbing

103.5 Create, monitor and kill processes
Weight 4
Description Candidates should be able to perform basic process management.
Key Knowledge Areas

• Run jobs in the foreground and background
• Signal a program to continue running after logout
• Monitor active processes
B LPIC-1 Certification 225

• Select and sort processes for display
• Send signals to processes
The following is a partial list of the used ﬁles, terms and utilities:
• &
• bg
• fg
• jobs
• kill
• nohup
• ps
• top
• free
• uptime
• pgrep
• pkill
• killall
• screen

103.6 Modify process execution priorities
Weight 2
Description Candidates should be able to manage process execution priorities.
Key Knowledge Areas
• Know the default priority of a job that is created
• Run a program with higher or lower priority than the default
• Change the priority of a running process
The following is a partial list of the used ﬁles, terms and utilities:
• nice
• ps
• renice
• top

104.1 Create partitions and filesystems
Weight 2
Description Candidates should be able to conﬁgure disk partitions and then
create ﬁlesystems on media such as hard disks. This includes the handling of
swap partitions.
Key Knowledge Areas
• Manage MBR partition tables
• Use various mkfs commands to create various ﬁlesystems such as:
– ext2/ext3/ext4
– XFS
– VFAT
• Awareness of ReiserFS and Btrfs
• Basic knowledge of gdisk and parted with GPT
The following is a partial list of the used ﬁles, terms and utilities:
• fdisk
• gdisk
• parted
• mkfs
• mkswap
226 B LPIC-1 Certification

104.2 Maintain the integrity of filesystems
Weight 2
Description Candidates should be able to maintain a standard ﬁlesystem, as
well as the extra data associated with a journaling ﬁlesystem.
Key Knowledge Areas

• Verify the integrity of ﬁlesystems
• Monitor free space and inodes
• Repair simple ﬁlesystem problems

The following is a partial list of the used ﬁles, terms and utilities:

• du
• df
• fsck
• e2fsck
• mke2fs
• debugfs
• dumpe2fs
• tune2fs
• XFS tools (such as xfs_metadump and xfs_info )

104.3 Control mounting and unmounting of filesystems
Weight 3
Description Candidates should be able to conﬁgure the mounting of a ﬁlesys-
tem.
Key Knowledge Areas

• Manually mount and unmount ﬁlesystems
• Conﬁgure ﬁlesystem mounting on bootup
• Conﬁgure user mountable removable ﬁlesystems

The following is a partial list of the used ﬁles, terms and utilities:

• /etc/fstab
• /media/
• mount
• umount

104.4 Manage disk quotas
Weight 1
Description Candidates should be able to manage disk quotas for users.
Key Knowledge Areas:

• Set up a disk quota for a ﬁlesystem
• Edit, check and generate user quota reports

The following is a partial list of the used ﬁles, terms and utilities:

• quota
• edquota
• repquota
• quotaon
B LPIC-1 Certification 227

104.5 Manage file permissions and ownership
Weight 3
Description Candidates should be able to control ﬁle access through the proper
use of permissions and ownerships.
Key Knowledge Areas

• Manage access permissions on regular and special ﬁles as well as directories
• Use access modes such as suid, sgid and the sticky bit to maintain security
• Know how to change the ﬁle creation mask
• Use the group ﬁeld to grant ﬁle access to group members

The following is a partial list of the used ﬁles, terms and utilities:

• chmod
• umask
• chown
• chgrp

107.1 Manage user and group accounts and related system files
Weight 5
Description Candidates should be able to add, remove, suspend and change
user accounts.
Key Knowledge Areas

• Add, modify and remove users and groups
• Manage user/group info in password/group databases
• Create and manage special purpose and limited accounts

The following is a partial list of the used ﬁles, terms and utilities:

• /etc/passwd
• /etc/shadow
• /etc/group
• /etc/skel/
• chage
• getent
• groupadd
• groupdel
• groupmod
• passwd
• useradd
• userdel
• usermod

110.1 Perform security administration tasks
Weight 3
Description Candidates should know how to review system conﬁguration to
ensure host security in accordance with local security policies.
Key Knowledge Areas

• Audit a system to ﬁnd ﬁles with the suid/sgid bit set
• Set or change user passwords and password aging information
• Being able to use nmap and netstat to discover open ports on a system
• Set up limits on user logins, processes and memory usage
228 B LPIC-1 Certification

• Determine which users have logged in to the system or are currently logged
in
• Basic sudo conﬁguration and usage

The following is a partial list of the used ﬁles, terms and utilities:

• find
• passwd
• fuser
• lsof
• nmap
• chage
• netstat
• sudo
• /etc/sudoers
• su
• usermod
• ulimit
• who , w , last

110.2 Setup host security
Weight 3
Description Candidates should know how to set up a basic level of host security.
Key Knowledge Areas

• Awareness of shadow passwords and how they work
• Turn oﬀ network services not in use
• Understand the role of TCP wrappers

The following is a partial list of the used ﬁles, terms and utilities:

• /etc/nologin
• /etc/passwd
• /etc/shadow
• /etc/xinetd.d/
• /etc/xinetd.conf
• /etc/inetd.d/
• /etc/inetd.conf
• /etc/inittab
• /etc/init.d/
• /etc/hosts.allow
• /etc/hosts.deny
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

C
Command Index
This appendix summarises all commands explained in the manual and points to
their documentation as well as the places in the text where the commands have
been introduced.

adduser Convenient command to create new user accounts (Debian)
adduser (8) 34
alien Converts various software packaging formats alien (1) 196
apt-get Powerful command-line tool for Debian GNU/Linux package manage-
ment apt-get (8) 189
aptitude Convenient package installation and maintenance tool (Debian)
aptitude (8) 192
blkid Locates and prints block device attributes blkid (8) 118
busybox A shell that already contains variants of many Unix tools
busybox (1) 174
cfdisk Character-screen based disk partitioner cfdisk (8) 93
chattr Sets ﬁle attributes for ext2 and ext3 ﬁle systems chattr (1) 51
chfn Allows users to change the GECOS ﬁeld in the user database
chfn (1) 27
chgrp Sets the assigned group of a ﬁle or directory chgrp (1) 44
chkconfig Starts or shuts down system services (SUSE, Red Hat)
chkconfig (8) 147
chmod Sets access modes for ﬁles and directories chmod (1) 43
chown Sets the owner and/or assigned group of a ﬁle or directory
chown (1) 44
cpio File archive manager cpio (1) 204
dd “Copy and convert”, copies ﬁles or ﬁle systems block by block and does
simple conversions dd (1) 120
debugfs File system debugger for ﬁxing badly damaged ﬁle systems. For gurus
only! debugfs (8) 108
dmesg Outputs the content of the kernel message buﬀer dmesg (8) 137
dpkg Debian GNU/Linux package management tool dpkg (8) 182
dpkg-reconfigure Reconﬁgures an already-installed Debian package
dpkg-reconfigure (8) 195
dumpe2fs Displays internal management data of the ext2 ﬁle system. For gurus
only! dumpe2fs (8) 108
dumpreiserfs Displays internal management data of the Reiser ﬁle system. For
gurus only! dumpreiserfs (8) 111
e2fsck Checks ext2 and ext3 ﬁle systems for consistency e2fsck (8) 107
e2label Changes the label on an ext2/3 ﬁle system e2label (8) 118
edquota Tool for entering and adjusting disk quotas edquota (8) 122
230 C Command Index

file Guesses the type of a ﬁle’s content, according to rules file (1) 174
fsck Organises ﬁle system consistency checks fsck (8) 101
gdisk Partitioning tool for GPT disks gdisk (8) 92
getent Gets entries from administrative databases getent (1) 32
getfacl Displays ACL data getfacl (1) 47
gpasswd Allows a group administrator to change a group’s membership and up-
date the group password gpasswd (1) 38
groupadd Adds user groups to the system group database groupadd (8) 37
groupdel Deletes groups from the system group database groupdel (8) 38
groupmod Changes group entries in the system group database groupmod (8) 37
groups Displays the groups that a user is a member of groups (1) 24
grub-md5-crypt Determines MD5-encrypted passwords for GRUB Legacy
grub-md5-crypt (8) 135
halt Halts the system halt (8) 151
id Displays a user’s UID and GIDs id (1) 24
initctl Supervisory tool for Upstart initctl (8) 150
insserv Activates or deactivates init scripts (SUSE) insserv (8) 147
kill Terminates a background process bash (1), kill (1) 58
killall Sends a signal to all processes matching the given name killall (1) 59
kpartx Creates block device maps from partition tables kpartx (8) 94
last List recently-logged-in users last (1) 24
ldconfig Builds the dynamic library cache ldconfig (8) 176
ldd Displays the dynamic libraries used by a program ldd (1) 174
losetup Creates and maintains loop devices losetup (8) 94
lsattr Displays ﬁle attributes on ext2 and ext3 ﬁle systems lsattr (1) 51
lsblk Lists available block devices lsblk (8) 119
lsmod Lists loaded kernel modules lsmod (8) 71
lspci Displays information about devices on the PCI bus lspci (8) 65
lsusb Lists all devices connected to the USB lsusb (8) 68
mkdosfs Creates FAT-formatted ﬁle systems mkfs.vfat (8) 114
mke2fs Creates ext2 or ext3 ﬁle systems mke2fs (8) 105
mkfs Manages ﬁle system creation mkfs (8) 100
mkfs.vfat Creates FAT-formatted ﬁle systems mkfs.vfat (8) 114
mkfs.xfs Creates XFS-formatted ﬁle systems mkfs.xfs (8) 111
mkreiserfs Creates Reiser ﬁle systems mkreiserfs (8) 111
mkswap Initialises a swap partition or ﬁle mkswap (8) 115
modprobe Loads kernel modules, taking dependencies into account
modprobe (8) 70
mount Includes a ﬁle system in the directory tree mount (8), mount (2) 116
nice Starts programs with a diﬀerent nice value nice (1) 61
nohup Starts a program such that it is immune to SIGHUP signals nohup (1) 61
pgrep Searches processes according to their name or other criteria
pgrep (1) 59
pkill Signals to processes according to their name or other criteria
pkill (1) 60
ps Outputs process status information ps (1) 56
pstree Outputs the process tree pstree (1) 57
quota Reports on a user’s quota status quota (1) 122
reboot Restarts the computer reboot (8) 151
reiserfsck Checks a Reiser ﬁle system for consistency reiserfsck (8) 111
renice Changes the nice value of running processes renice (8) 61
repquota Summarises ﬁlesystem usage and quota usage for many users
repquota (8) 122
resize_reiserfs Changes the size of a Reiser ﬁle system resize_reiserfs (8) 111
rpm Package management tool used by various Linux distributions (Red Hat,
SUSE, …) rpm (8) 200
rpm2cpio Converts RPM packages to cpio archives rpm2cpio (1) 204
runlevel Displays the previous and current run level runlevel (8) 145
C Command Index 231

sash “Stand-Alone Shell” with built-in commands, for troubleshooting
sash (8) 174
setfacl Enables ACL manipulation setfacl (1) 47
sfdisk Non-interactive hard disk partitioner sfdisk (8) 93
sgdisk Non-interactive hard disk partitioning tool for GPT disks sgdisk (8) 93
shutdown Shuts the system down or reboots it, with a delay and warnings for
logged-in users shutdown (8) 151
star POSIX-compatible tape archive with ACL support star (1) 47
strip Removes symbol tables from object ﬁles strip (1) 174
su Starts a shell using a diﬀerent user’s identity su (1) 16
sudo Allows normal users to execute certain commands with administrator
privileges sudo (8) 14
swapoff Deactivates a swap partition or ﬁle swapoff (8) 115
swapon Activates a swap partition or ﬁle swapon (8) 115
systemctl Main control utility for systemd systemctl (1) 157, 166
top Screen-oriented tool for process monitoring and control top (1) 61
tune2fs Adjusts ext2 and ext3 ﬁle system parameters tunefs (8) 108, 119
udevd Kernel uevent management daemon udevd (8) 73
update-rc.d Installs and removes System-V style init script links (Debian)
update-rc.d (8) 147
useradd Adds new user accounts useradd (8) 33
userdel Removes user accounts userdel (8) 36
usermod Modiﬁes the user database usermod (8) 36
vigr Allows editing /etc/group or /etc/gshadow with “ﬁle locking”, to avoid con-
ﬂicts vipw (8) 38
vol_id Determines ﬁle system types and reads labels and UUIDs
vol_id (8) 118
xfs_mdrestore Restores an XFS metadata dump to a ﬁlesystem image
xfs_mdrestore (8) 112
xfs_metadump Produces metadata dumps from XFS ﬁle systems
xfs_metadump (8) 112
yum Convenient RPM package maintenance tool yum (8) 205
$ echo tux
tux
$ ls
hallo.c
hallo.o
$ /bin/su -
Password:

Index
This index points to the most important key words in this document. Particu-
larly important places for the individual key words are emphasised by bold type.
Sorting takes place according to letters only; “~/.bashrc ” is therefore placed under
“B”.

/, 86 busybox , 174

access mode, 42 Cameron, Jamie, 18
adduser , 34 Card, Rémy, 102–103
administration tools, 14 cat , 31, 214
alien , 181, 196–197 cc , 120
--to-deb (option), 197 cd , 42
apt , 182, 190 cfdisk , 93
apt-cache , 181, 190–192 chage , 35
apt-get , 181, 184, 189–190, 192–193, chattr , 51, 214
205–206 -R (option), 51
dist-upgrade (option), 189–190 chfn , 27
install (option), 190 chgrp , 38, 44–45, 49
remove (option), 190 -R (option), 45
source (option), 190 chkconfig , 147, 216
upgrade (option), 190 chmod , 15, 43, 46, 48–49, 51
apt-key , 195 -R (option), 44
aptitude , 181–183, 192–193 --reference= ⟨name⟩ (option), 44
aquota.group , 122 chown , 36, 44–45
aquota.user , 122 -R (option), 45
ar , 182–183, 197 chsh , 27
at , 162 comm , 212
ATA, 78 cp , 116, 174
awk , 172 cpio , 130, 132, 204–205, 210, 230
cron , 147, 162
bash , 50, 56, 174, 214 cut , 212
bg , 54
/bin/sh , 212 D-Bus, 74
/bin/true , 27 dd , 88, 93–94, 110, 112, 116, 120–121, 138
blkid , 118–119 DEBCONF_FRONTEND (environment
/boot , 134 variable), 195
boot manager, 128 DEBCONF_PRIORITY (environment
boot script, 144 variable), 196
boot sector, 128 debsums , 188, 194
/boot/grub , 133 debugfs , 108
/boot/grub/custom.cfg , 134 -w (option), 108
/boot/grub/grub.cfg , 134 deﬁnitions, 12
/boot/grub/menu.lst , 132 demand paging, 50
Bottomley, James, 130 /dev , 73
btrfs , 114 /dev/block , 85
btrfs check /dev/mapper , 94
--repair (option), 114 /dev/null , 170
234 Index

/dev/psaux , 71 LD_LIBRARY_PATH , 177, 179, 216
/dev/scd0 , 107 PATH , 177
/dev/sda , 84, 88 VISUAL , 37
/dev/ttyS0 , 16 /etc , 17, 116
/dev/zero , 106 /etc/apt/apt.conf , 190
device , 73 /etc/apt/sources.list , 189
diff , 185 /etc/apt/trusted.gpg , 195
Dijkstra, Edsger, 136 /etc/dpkg/dpkg.cfg , 183
disk , 92 /etc/filesystems , 117–118
disk cache, 100 /etc/fstab , 85, 88, 101–102, 109,
dmesg , 137 116–117, 119, 121, 144, 157,
dpkg , 181–184, 186–189, 191, 193 215
-a (option), 183 /etc/group , 25, 27, 30–31, 33, 36–38
--configure (option), 183 /etc/grub.d , 134
--force-depends (option), 183 /etc/grub.d/40_custom , 134
--force-overwrite (option), 183 /etc/grub.inst , 133
-i (option), 183 /etc/gshadow , 31, 37–39, 231
--install (option), 183 /etc/inetd.conf , 157
-L (option), 187 /etc/init , 149
-l (option), 185 /etc/inittab , 142, 144–146, 151, 157, 160,
--list (option), 185 164–165
--listfiles (option), 187 /etc/ld.so.cache , 177
-P (option), 184 /etc/ld.so.conf , 176–177
--purge (option), 193 /etc/modprobe.conf , 71
-r (option), 184 /etc/modprobe.d , 71
-s (option), 186, 188 /etc/mtab , 120, 215
--search (option), 188 /etc/nologin , 151
--status (option), 186–187, 191 /etc/nsswitch.conf , 32
--unpack (option), 183 /etc/passwd , 25–28, 30–34, 36–37, 213
dpkg-reconfigure , 195 /etc/rpmrc , 200
-f (option), 195 /etc/securetty , 16
--frontend (option), 195 /etc/shadow , 26, 28–29, 31–33, 35–37, 39,
-p (option), 195 48, 212–213
--priority (option), 195 /etc/shells , 27
dpkg-source , 185 /etc/skel , 33
dselect , 189, 192 /etc/sysconfig , 18
dump , 50 /etc/udev , 73
dumpe2fs , 108 /etc/yum.conf , 206
dumpreiserfs , 111 /etc/yum.repos.d , 206

e2fsck , 107–108, 111 fdisk , 88–93
-B (option), 107 -l (option), 89
-b (option), 107–108 -u (option), 89
-c (option), 107 fg , 54
-f (option), 107 file , 174–175
-l (option), 107 ﬁle attributes, 50
-p (option), 107 finger , 27
-v (option), 107 fsck , 101–102, 107–109, 112, 139
e2label , 118 -A (option), 102
e4defrag , 109 -a (option), 102
EDITOR (environment variable), 37, 122 -f (option), 102
edquota , 122 -N (option), 102
-g (option), 122 -p (option), 102
-t (option), 122 -R (option), 102
egrep , 60 -s (option), 102
environment variable -t (option), 101, 112
DEBCONF_FRONTEND , 195 -V (option), 102
DEBCONF_PRIORITY , 196 -v (option), 102
EDITOR , 37, 122 fsck.ext2 , 107
Index 235

fsck.xfs , 112 initctl status , 150
initctl stop , 150
Garrett, Matthew, 130 insserv , 147, 216
gdisk , 92–93, 121 ISOLINUX , 128
getent , 31–32, 212
getfacl , 47 jobs ,
54
getty , 164 Johnson, Jeﬀ, 200
GNOME, 195 Journaling, 103
Gooch, Richard, 73
gpasswd , 38 KDE, 195
-A (option), 38 kill , 58–60, 147
-a (option), 38 killall , 58–60
-d (option), 38 -i (option), 59
grep , 31–32, 57, 59 -l (option), 59
group, 23 -w (option), 59
administrative, 31 Kok, Auke, 148
administrator, 38 konsole , 27
password, 31, 38 kpartx , 93–94, 96
groupadd , 37 -v (option), 94
-g (option), 37 Kroah-Hartman, Greg, 73
groupdel , 37–38
groupmod , 36–37 label, 118
-g (option), 37 last , 24–25
-n (option), 37 ld.so.cache , 177
groups, 15 LD_LIBRARY_PATH (environment variable),
groups , 24 177, 179, 216
GRUB, 128 ldconfig , 176–177
boot problems, 138 -p (option), 177
grub , 133 ldd , 174–176, 216
--device-map (option), 133 less , 31
lock (option), 215 /lib , 175–177
password (option), 135 /lib/modules , 70–71
grub-install , 133 login , 16, 27, 151
grub-md5-crypt , 135 losetup , 94
grub-mkconfig , 134–135 -a (option), 94
gzip , 197 -f (option), 94
lost+found , 108
halt ,151 ls , 26, 42–43, 51
hard disks -l (option), 26, 43, 51
geometry, 65 lsattr , 51, 214
SCSI, 79 -a (option), 51
hard quota, 121 -d (option), 51
hello , 182, 185 -R (option), 51
/home , 27–28, 86 LSB, 182
home directory, 23 lsblk , 119
/home/opt , 86 lsmod , 71
Homme, Kjetil Torgrim, 59 lspci , 65–68
-n (option), 67
id ,24, 26, 50, 211 -t (option), 66–67
-G (option), 24 -v (option), 66–67
-g (option), 24 lsusb , 68–69
-Gn (option), 24 -v (option), 69
-n (option), 24
-u (option), 24 mail , 27
init , 101, 135, 137, 144–146, 215 Mason, Chris, 100
init scripts, 146, 151 master boot record, 128
parameters, 146 Matilainen, Panu, 200
initctl , 150 mesg , 152
initctl start , 150 Minix, 102
236 Index

mkdosfs , 114–115 -t (option), 60
mke2fs , 100, 105–106, 109 -u (option), 60
-F (option), 106 pkill , 59–60, 168
mkfs , 100–101, 105–106, 113–114, 129 --signal (option), 60
-t (option), 100, 105–106, 114 Poettering, Lennart, 142, 156
mkfs.btrfs pre-emptive multitasking, 55
-d (option), 113 primary group, 26
-L (option), 119 priority, 60
mkfs.vfat , 114 /proc , 54, 56, 71, 214
mkfs.xfs , 111–112 /proc/filesystems , 117–118
-l (option), 112 /proc/pci , 66
mkreiserfs , 111 /proc/swaps , 115–116
mkswap , 115–116, 119 /proc/sys/kernel/pid_max , 214
/mnt , 106 process state, 55
modprobe , 70–71 ~/.profile , 122
-r (option), 71 ps , 47, 56–60
mount , 88, 109, 116–118 a (option), 56–57
grpquota (option), 122 ax (option), 57
-t (option), 117 -C (option), 57
usrquota (option), 121–122 --forest (option), 56, 58
mount point, 116 --help (option), 56
mv , 116 -l (option), 56
-o (option), 57
newgrp , 31 p (option), 60
nice , 61 r (option), 56
nohup , 61 T (option), 56
nohup.out , 61 U (option), 56
-u (option), 47
objectives, 219 x (option), 56–57
/opt , 86 pseudo-users, 25
pstree , 57–58
Packages.gz , 194–195
-G (option), 58
parted , 90–92
-p (option), 58
passwd , 26, 34–35, 37–38, 47–48, 212, 216 -u (option), 58
-l (option), 35
pwconv , 32
-S (option), 35
Python, 195
-u (option), 35
python , 172
passwd -n , 35
passwd -w , 35 quota , 122–123
passwd -x , 35 -g (option), 122
passwords, 23, 26, 28 -q (option), 122
changing, 34 quotacheck , 122
group —, 31, 38 quotaoff , 122
GRUB, 135 quotaon
setting up, 34 -g (option), 122
shadow –, 26
shadow —, 28 reboot , 151
PATH (environment variable), 177 Reiser, Hans, 110
PCI Express, 65 reiserfsck , 111
perl , 172 Release , 194
pgrep , 59–60 Release.gpg , 194
-a (option), 59 Remnant, Scott James, 142, 148
-d (option), 59 renice , 61
-f (option), 60 repquota , 122
-G (option), 60 resize_reiserfs , 111
-l (option), 59 restart , 216
-n (option), 60 return code, 55
-o (option), 60 Ritchie, Dennis, 48
-P (option), 60 rm , 42
Index 237

rpm ,
182, 197, 200, 202, 205, 209 swapon , 115–116
-a (option), 202 /sys , 73
-c (option), 203 /sys/block , 72
-d (option), 203 /sys/bus , 72
-e (option), 201 /sys/bus/scsi/devices , 85
-F (option), 201 /sys/class , 72
-f (option), 202 syslog , 147, 216
-h (option), 201 syslogd , 137, 147
-i (option), 200–202 systemctl , 157, 165–170, 216
-l (option), 202–203 --full (option), 167
--nodeps (option), 201 --kill-who (option), 167
-p (option), 202 -l (option), 167
--provides (option), 203 --lines (option), 167
-q (option), 201 -n (option), 167
-qi (option), 209 --now (option), 169
--requires (option), 203 -s (option), 167
--test (option), 201 --signal (option), 167
-U (option), 201 -t (option), 166–167
-V (option), 204 systemd , 168
-v (option), 200, 202 systemd-escape , 162
-vv (option), 200 -p (option), 162
--whatprovides (option), 203 -u (option), 162
--whatrequires (option), 203
rpm2cpio , 204–205 tar , 47, 130, 174, 183, 197, 200, 211
~/.rpmrc , 200 telinit , 144–146, 148
runlevel, 151 q (option), 144
changing —, 145 termination, 80
runlevel , 145, 168, 215 /tmp , 37, 49, 87, 213
runlevels, 142 top , 61
conﬁguring —, 147 touch , 37
meaning, 145 Ts’o, Theodore, 104
tune2fs , 107–109, 119, 215
sash , 174 -c (option), 215
/sbin/init , 142 -L (option), 119
SELinux, 14 -l (option), 107
setfacl , 47 -m (option), 215
sfdisk , 93, 121, 138 -u (option), 215
sgdisk , 93 Tweedie, Stephen, 103
shutdown , 15, 144, 151–152, 165–166
-c (option), 216 udevd , 73
-r (option), 151 udisksctl , 74
Sievers, Kay, 73, 142, 156 udisksd , 74
signals, 58 UID, 23
single-user mode, 147 umask , 46, 50
sleep , 60 -S (option), 46
soft quota, 121 umount , 116
sort , 212 uname , 24
/srv , 87 -r (option), 24
ssh , 24 update-grub , 134
sshd , 59 update-rc.d , 147
star , 47 usbview , 69
strip , 174 user accounts, 22
su , 16–17, 25, 211, 213 user database, 25, 28
sudo , 14, 17 stored elsewhere, 28
super user, 14 user name, 23
superblock, 100 useradd , 33–34, 36–37, 212
SuSEconfig , 18 userdel , 36–37
swap partition, 115 -r (option), 36
swapoff , 115 usermod , 36–37, 213
238 Index

/usr/lib , 175–177
/usr/lib/rpm , 200
/usr/local , 86, 201
/usr/local/lib , 177
UUID, 119

van de Ven, Arjan, 148
/var , 87
/var/lib/dpkg/info , 188
/var/lib/usbutils/usb.ids , 69
/var/log/messages , 17, 137, 211
/var/log/syslog , 137
/var/mail , 36, 121
vi , 19, 37
vigr , 37–38
-s (option), 38
vipw , 37–38, 213
-s (option), 37
VISUAL (environment variable), 37
vol_id , 118

w , 92
wall , 152–153
-n (option), 153
--nobanner (option), 153
Webmin, 18
write , 153

xclock , 56
xfs_copy , 112
xfs_info , 112
xfs_mdrestore , 112
xfs_metadump , 112
xfs_quota , 121
xfs_repair , 112
-n (option), 112
xfsdump , 112
xfsrestore , 112
xterm , 27

YUM, 205
yum , 205, 207–209
--disablerepo (option), 206
--enablerepo= (option), 205
--obsoletes (option), 207
yumdownloader , 210
--resolve (option), 210
--source (option), 210
--urls (option), 210

zombies, 55
zsh , 34