Principles of system administration Edition 2.1

Document technical information

Format pdf
Size 326.5 kB
First found Jun 9, 2017

Document content analysis

not defined
no text concepts found


Gaetano Donizetti
Gaetano Donizetti

wikipedia, lookup




Monday, July 26, 1999
Principles of system administration
Page: 1
Principles of system administration
Edition 2.1
Mark Burgess
Centre of Science and Technology
Faculty of Engineering, Oslo College
Copyright (C) 1998 Mark Burgess Permission is granted to make and distribute verbatim copies of this manual provided the copyright notice and this
permission notice are preserved on all copies.
For many years system administration has been passed on to new generations through manual pages, technical handbooks and by word of mouth. In all but
the most disciplined institutions this has been a haphazard affair with a disregard for theory and no common standard of practice. Put simply, no one has
seriously considered system administration to be a worthy academic pursuit. In recent years, the arrival of the internet has made this belief untenable
however. The level of complexity of networked operating systems together with the increasing problem of intrusion (`cracking') has properly elevated the
status of system administration to that of a discipline in its own right.
This text is an introduction to the aims and guiding principles of system administration. In this respect it is not like other books which focus mainly on
procedures. It needs to be used together with any and all information which you find elsewhere. It's main aim is to illuminate a sound and logical way
through the murky swamps of networked operating systems. It does not replace other books nor is it like any other. It is aimed at students in a College
environment, or apprentice administrators not at system administrators working in the field.
The course assumes a basic familiarity with Windows and Unix as well as a core knowledge of basic operating system concepts from the two preceding
courses Operating systems and Unix programming . It is intended to run intensively over one semester. You will be required to study an existing network,
then set up a GNU/Linux machine and integrate it into the network. You will learn system administration by trial and error. No one should underestimate the
value of making errors during the learning process. This is often the only way to really understand.
The practical work for the course is based on the Debian distribution of the GNU/Linux operating system. This choice has been made in order to maximize
the understanding of the important issues rather than the ease or simplicity of installation or of use. i.e. the aim is to gain experience, not to make things as
easy as possible.
Much of the section on security has been influenced by the thoughtful words exchanged at the USENIX/LISA conferences. Comments on this course,
observing suitable protocols and etiquette, to [email protected]
/M\, Oslo, Dec 1998
What is system administration about?
@hrule @vskip 0.3cm In this manual the word "host" is used to refer to a single computer system -- i.e. a single machine which has a name termed its
"hostname" See section Glossary , for other definitions.
What is system administration?
The job of a system administrator is complex. It a difficult job, which requires patience, understanding and a lot of knowledge and experience. It is like
working in the casualty ward of a hospital. You need to be a doctor, a psychologist, and--when your instruments fail--a mechanic. You need to work with the
limited resources you have. You need to be inventive in a crisis, you need to know a lot of facts and figures about the way computers work. You need to
recognize that the answers are not always written down for you to copy, that machines do not always behave the way you think they should. You need to do
all this--and, on top of everything, you need to learn a hundred new things a year.
Being a system administrator is as much a state of mind as it is about being knowledgeable. It is the sound of one hand tapping ...on the keyboard, while the
other is holding the phone, talking to a user and there is a queue of people waiting for help. You need to be ready for the unexpected, resigned to the
uncertain future, and you need to be able to plan for the future. It requires organization and the ability to be systematic. There is no right answer, but there is
often a wrong answer. It's about making something robust which works. In spite of stereotypes, today's system administrator is neither haphazard nor
messy. Today's computing systems require the very best of organizational skills and the most professional of attitudes.
It's also a matter of self-confidence. In the beginning a novice will be too nervous to do any real work, later, when more experienced, one learns to relax and
just do things. It's like dating or skate-boarding, there are three stages:
1. Under-confidence: you place your skate-board carefully onto a flat surface, well away from spectators, step shakily onto it and rock about
2. Optimal confidence: you glance about you, then throw your plank on the ground, jump onto it and skate off with admirable self-assurance.
3. Over-confidence: you don't look around you and throw your daemon-board onto the ground, running, into the path of an angry cyclist who then
breaks your nose for the inconvenience.
To get going, you need to know your stuff and have confidence in your abilities--but you also need to know your limitations.
Monday, July 26, 1999
Principles of system administration
Page: 2
If you have installed Windows, DOS or GNU/Linux on a PC, you might think that you know a lot about system administration, but in fact you know only
the very beginning. Networking is about cooperation and sharing in an environment with many users. Don't think that you know all the answers simply
because you know something about how your own machine works.
This book is an introduction to the concepts of system administration. As a system administrator, your first responsibility is to the greater network
community and then to the users of your system. An administrator's job is to make their lives bearable and to allow them to perform productive work..
A philosophy
How does one begin and how does one behave? What are bad habits and what are good habits? The value of an adequate theoretical understanding should
not be underplayed. The days when one could fly a system by the seat of one's pants are over. There is a strong need for practical skills, but nothing can
replace real understanding. Today, keeping up with the theory is as important as upgrading the software. One also needs to approach the problem correctly.
The following are ill-advised:
Insisting that there is a right or a wrong answer to every question.
Running to someone else whenever there is a problem.
Getting fraught and upset when things do not work the way one expects.
Expecting that every problem has a beginning, a middle and an end.
In contrast it is recommended to begin
Looking for answers in manuals and newsgroups.
Using trial and error (carefully!) to locate problems.
Practice being cheerful and patient with the users of your system and the machines.
Listening to people who tell you that there is a problem. It might be true, even if you can't see it yourself.
Writing down your experiences so that you will know how to solve the same problem again in the future.
Taking responsibility for one's own actions. (Be prepared for accidents. They are going to happen and they will be your fault. You will have to fix
Remembering the tedious jobs like vacuum cleaning the hardware once a year.
However much one might like one's own language, it is important to get used to the English language. English (actually American) is the language of the net.
System administrators need it to be able to read documentation, to be able to communicate with others and to ask questions on the Internet.
It is important to start by being systematic from the beginning. Every time you read something, or learn some isolated fact, think: how does this apply to
me? Do yourself a favour and buy an A-Z that you can use to write down your experiences. Start now. And try to find your own method for remembering
what you have learned.
What are the challenges of system administration?
System administration is not about installing operating systems. It is about planning and designing an efficient network of computers so that real users will
be able to get their jobs done. That means:
Designing a network which is logical and efficient.
Deploying large numbers of machines which can be easily upgraded later.
Deciding what services are needed.
Planning and implementing adequate security.
Providing a comfortable environment for users.
Developing ways of fixing errors and problems which occur.
Keeping track of and understanding how to use the enormous amount of knowledge which increases every year.
Some system administrators are responsible for both the hardware of the network and the computers which it connects, i.e. the cables as well as the
computers. Some are only responsible for the computers. Either way, an understanding of how data flow from machine to machine is essential as well as an
understanding of how each machine affects every other. A central axiom for network administration is:
A computer which is plugged into the network is not just yours. It is part of a society of machines which shares resources and communicates
with the whole. What your machine does affects other machines. What other machines do affects your machine.
The ethical issues associated with connection to the network are not trivial. Administrators are responsible for their organization's conduct to the entire rest of
the Internet. This is a great responsibility which must be borne wisely.
Is system administration a career?
The internet has grown considerably in the last ten years and operating systems have grown more and more complex. Unfortunately the number of qualified
system administrators has not grown in proportion, so there is a gap.
There is currently quite an appreciable market for consulting services in security and automation of system administrative tasks. Remember that most
companies have more money than time or expertise and that they need to be able to buy something in order to satisfy their board members and accountants.
For a business, doing something to address a problem means spending some money. That means that they are looking to pay someone to carry out a service.
Since the best system administration tools are free companies actively seek to pay consultants to set up and maintain administration tools for them. Not only
is system administration a very interesting line of work (it is extremely varied), it is also very satisfying.
NEVER-DOs in system administration
Some rules of thumb are useful to prevent embarrassing accidents.
The administrator or root account has unlimited privileges. Never log into the system as the root user. Use the su command to gain root privileges
when you need them and then quit at once. This is because one should never run ordinary programs with root privileges. If one does, it is possible
to allow the system to be invaded by viruses or to cause accidental damage to the system.
Monday, July 26, 1999
Principles of system administration
Page: 3
Never leave root shells on the console. It is possible to accidentally do something destructive without realizing that one has root privileges. (Put the
sledgehammer down when you are not using it, Eugene!)
Never leave root shells running so that others might gain access to them in an open room.
Never leave services running if they are not used for anything. They provide a possible back-door into the system for intruders.
Never give users physical access to a machine which stores important data. If users can touch the system, it's theirs.
Windows 95, Windows 98 and the MacIntosh are inherently insecure systems. They cannot be secure by virtue of their design (they have no access
control of any kind--all access is privileged). When setting up a network in a potentially hostile environment, use an operating system like Unix or
Know your network
System administration requires its operatives to know a lot of facts about hardware and software. The road to real knowledge is long and hard, so how does
one get started?
A top down approach is useful for understanding the network interrelationships. The best place to start is at the network level. In most daily situations, one
starts with a network already in place--i.e. one doesn't have to build one from scratch. It is important to know what hardware one has to work with and where
everything is to be found; how it is organized (or not) and so on. Here is a checklist:
How does the network fit together? (What is its topology?)
Which function does each host/machine have on the network?
Where are the key network services?
Having thought about the network as a whole, one can begin to think about individual hosts/machines. First there is a hardware list.
What kind of machines are on the network? What are their names and addresses and where are they? Do they have disks. How big? How much
memory do they have? If they are PC's, which screen cards do they have?
What operating systems are running on the network? MS-DOS, Novell, NT or Unix (if so which Unix? GNU/Linux, Solaris, HPUX?)
What kind of network cables are used? Is it thin/thick Ethernet? Is it a star net (hubs/twisted pair), or fiber optic FDDI net?
Where are hubs/repeaters/the router or other network control boxes located? Who is responsible for maintaining them?
What is the hierarchy of responsibility?
Then there is a software list:
How many different subnets does your network have?
What are their network addresses?
Find the router addresses (the default routes) on each segment.
What is the netmask?
What is the local timezone?
What broadcast address convention is used? 255 or the older 0?
Find the key servers on these networks. @itemize @bullet @item Where are the NFS network disks located? Which machine are they attached to?
@item The name services (DNS/NIS/NISPLUS) @item The WWW/http service. @end itemize
Finding and recording this information is an important learning process. The information will change as time goes by. Networks are not static; they grow
and evolve with time.
Frequent computer users are not usually aware of how they can be damaging their own health. Unlike cigarettes, computers do not have a government health
warning. Whether or not this is an issue for system administrators is open for discussion, but often the system administrator is the only person who thinks
about the users and the hardware they use. Certainly every administrator needs to look after his/her own health and, along the way, it is natural to think of the
health of others. Fortunately it is not difficult to avoid the worst problems.
Eyes should be protected, We only have one pair and they must last our entire lives. Ironically, users who wear glasses (not contact lenses) suffer less from
computer usage, because their eyes are partially protected from the radiation from the screen.
A computer screen works by shooting charged electrons at a phosphorescent surface. If one touches the screen one notices that it is charged with static
electricity. The effect of this is to charge dust particles and throw them out into users' faces. This can cause irritation to the eyes over long periods. Solution:
wear glasses or obtain an anti-static screen with an Earth-wire which counteracts this problem.
Another major cause of eye strain is through reflection. If there is a light source behind a user, it will reflect in the screen and the eyes will be distracted by
the reflection. The image on the screen lies on the screen surface, any reflected images lie behind the screen (as far behind the screen as the source is in front
of the screen). This confuses the eyes into focusing back and forth between the reflection and the image. The result is eye-strain. The solution is to (i)
eliminate all sharp light sources which can cause reflections, (ii) obtain an anti-reflective screen cover. This can be combined with an anti-static screen and it
is probably the best investment a user can make.
Prolonged eye strain can lead to problems reading and focusing. It can lead to headaches and neck ache from squinting.
Back. The back (spine) is one of the most complex and important parts of the body. It supports the upper body and head, and is attached to the brain (where
applicable). The upper body is held up by muscles in the stomach and lower back. If these muscles are relaxed by slouching for long periods unnecessary
strain in placed on muscles and bones which were not meant to bear the weight of the body.
To avoid back problems, users should (i) sit in a good chair, (ii) sit upright, using those all important flat-tummy muscles and lower back muscles to support
your upper body. Don't sit in a draft. Cold air blowing across the back and neck causes stiffness and tension.
Mouse strain Mouse strain is a strain in the tendons of the finger and forearm, which spreads to the shoulder and back and can be quite painful. It comes
from using the mouse too much. The symptoms can be lessened by making sure that users do not sit too far away from the desk where the mouse lies and
by having a support for the mouse forearm. The ultimate solution is simple: don't use the mouse. Use of the keyboard is far less hazardous. Learning
keyboard shortcuts is good for prolonged work.
Monday, July 26, 1999
Principles of system administration
Page: 4
Pregnancy and cancer. Some studies recommend that pregnant women wear protective aprons when sitting in front of computer screens.
Generally. Users should not sit for long periods without taking a break. Look away from the screen (to a far away object) at regular intervals relaxes the
eyes. Walking around exercises the back and relaxes the shoulders. Use of anti-static, anti-reflective screens is recommended.
Weather and environment affect computers.
strikes can destroy fragile equipment. No fuse will protect hardware from a lightning strike. Transistors and CMOS chips burn out much faster than
any fuse. Electronic spike protectors can help here.
failure can cause disk damage and loss of data. A UPS (Uninterruptible power supply) can help.
Heat. The blazing summer heat or a poorly placed heater oven can cause systems to overheat and suddenly black out. One should not let the ambient
temperature near a computer to rise much above about 25 degrees Centigrade. Heat can cause RAM to operate incorrectly and disks to
misread/miswrite. Good ventilation is essential for computers and screens for avoiding electrical faults.
Cold. Sudden changes from hot to cold are just as bad. They can cause unpredictable changes in electrical properties of chips and cause systems to crash.
In the long term, these changes could lead to cracks in the circuit boards and irreparable chip damage.
Dealing with users: etiquette
Although even the most stoical administrator's convictions might occasionally be called into question, system administration is a social service and it is
important to remain calm and reasonable. Users frequently believe that the system administrator has nothing better to do than to answer every question and
execute every whim and fancy. Dealing with users is not a small task.
Ethics and Responsibility
A system administrator wields great power. He or she has the means to read everyone's mail, change anyone's files, to start and kill anyone's processes. This
power can easily be abused and that temptation could be great.
Another danger is the temptation for an administrator to think that the system exists primarily for him or her and that the users are simply a nuisance to the
smooth running of things; if network service is interrupted, or if a silly mistake is made which leads to damage in the course of an administrator's work, that
is okay: the users should accept these mistakes because they were made trying to improve the system. When wielding such power there is always the chance
that such arrogance will build up. Some simple rules of thumb are useful.
The ethical integrity of a system administrator is clearly an important issue. Administrators for top secret government organizations and administrators for
small businesses have the same responsibilities towards their users and their organizations. One has only to look at the governing institutions around the
world to see that power corrupts. Few individuals, however good their intentions, are immune to the temptations of such power at one time or other. As with
governments it is perhaps a case of: those who wish the power are least suited to deal with it.
Administrators `watch over' backups, email, private communications and they have access to everyone's files. While it is almost never necessary to look at a
user's private files, it is possible at any time and users do not usually consider the fact that their files are available to other individuals in this way. Users need
to be able to trust the system and its administrator.
What kind of rules can you fairly impose on users?
What responsibilities do you have to the rest of the network community, i.e. the rest of the world?
Censoring of information or views
Restriction of personal freedom
Taking sides in personal disputes
Extreme views (some institutions have policies about this)
Unlawful behaviour
Objectivity of the administrator means avoiding taking sides in ethical, moral, religious or political debates, in the role of system administrator. Personal
views should be kept separate from professional views. However, the extent to which this is possible depends strongly on the individual and organizations
have to be aware of this. Some organizations dictate policy for their employees. This is also an issue to be cautious with: if a policy is too loose it can lead to
laziness and unprofessional behaviour; if it is too paranoid or restrictive it can lead to bad feelings in the organization. Historically, unhappy employees have
been responsible for the largest cases of computer crime. Other references:
Operating systems and programs are full of bugs. Learning to tolerate bugs is a matter of survival for system administrators. If one is lucky enough to be
using free software from the net, these bugs will usually be solved quickly and one can eliminate them by upgrading. If one uses commercial software it will
probably be necessary to wait a lot longer for a patch. Either way, one must be creative and work around these bugs. Bugs can be caused by many things.
They may come from
Shoddy software,
Little known bugs in the operating system,
Unfortunate collisions between incompatible software, i.e. one software package destroys the operation of another.
Totally unexplainable phenomena, cosmic rays and invasions by digital life-forms.
Operating systems
@hrule @vskip 0.3cm
Monday, July 26, 1999
Principles of system administration
Page: 5
For an operating system to be managed consistently it has to be possible to prevent its destruction by restricting the privileges of its users. Different
operating systems vary in their provisions for restricting privilege. In operating systems where any user can change any file, there is little or no possibility of
gaining true control over the system. Any accident or whim on the part of a user can make uncontrollable changes.
Insecure operating systems
The home computer revolution was an important development which spread cheap computing power to a large part of the world. As with all rapid
commercial developments, the focus in developing home operating systems was on immediate functionality not on planning for the future. The home
computer revolution preceded the network revolution by a number of years and home computer operating systems did not address security issues. Operating
systems developed during this period include Windows, MacIntosh, DOS, Amiga-DOS. All of these systems are completely insecure: they place no limits
on what a determined user can do.
Fortunately these systems will slowly be replaced by operating systems which were designed with resource sharing (including networking) in mind. Still,
there is a large number of insecure computers in use and many of them are now connected to the network. This should be a major concern for a system
administrator. In an age where one is forced to take security extremely seriously, leaving insecure systems where they can be accessed
by network
is a potentially dangerous situation. Such machines should not be allowed to hold important data and they should not be allowed any privileged access to
network services. We shall return to this issue in the chapter on security.
Administrator/root privilege
To distinguish them from insecure operating systems we shall refer to operating systems like Unix and NT as secure operating systems. This should not
give the impression that Unix and NT are completely secure by any stretch of the imagination: complete security is a fairy tale, a pipe dream, it will never
happen in any operating system. Nevertheless, these operating systems do have the mechanisms which make a basic level of security possible.
The most fundamental tenet of security is the ability to restrict access to certain parts of the system's resources. The main reason why DOS, Windows 9x
and MacIntosh are so susceptible to virus attacks is because any user can change the operating system's files. Properly configured and bug free Unix/NT
systems are theoretically immune to such attacks because ordinary users do not have the privileges required to change system files. Unfortunately the key
phrases properly configured and bug-free highlight the flaw in this theory.
In order to restrict access to the system we require a notion of ownership and permission . Ordinary users should not have access to the hardware devices
of a secure operating system's files, only their own files, and then they will not be able do anything to compromise the security of the system. System
administrators need access to the whole system in order to watch over it, make backups and keep it running. Secure operating systems thus need a privileged
account which can be used by the system administrator when he/she is required to make changes to the system. In Unix the privileged account is called
root . In NT, the Administrator account is similar to Unix's root, except that the administrator does not have automatic access to everything as does root.
Instead he must be first granted himself access to an object. However the administrator always has the right to grant him or herself access to a resource so in
practice this feature just adds an extra level of caution.
Administrator and root accounts should not be used for normal work, since they have far too much power. To use these an normal user accounts would be to
make the systems as insecure as the insecure systems we have mentioned above. In summary: It is desirable to restrict users' privileges for the greater
good of everyone on the network.
System identities
Whenever more than one computer is coupled together, there is a need for each computer to have a unique identity. As it turns out, this need has been
recognized many times in different contexts and the result is that todays computer systems can have many different names which identify them in different
contexts. The result is a confusion. For internet enabled machines, the IP address of the host is usually sufficient for most purposes.
Host ID
Circuit board identity number. Often used in software licensing.
Install name
Configured at install time. This is often compiled into the kernel, or placed in a file like `/etc/hostname'.
Application level name
Any name used by application software when talking to other hosts.
Local mapping
Originally the Unix `/etc/hosts' file was used to map IP addresses to names and vice versa.
Network Information Service
A local area network database service developed by Sun Microsystems.
Transport level address(es)
Each network interface can be configured with an IP address. This numerical converts into a text name through some naming service.
Network level address(es)
Each network interface (Ethernet/FDDI etc) has a hardware address burned into it at the factory. Some services (e.g. RARP) will turn this into a
name or an IP address through a secondary naming service like DNS.
DNS name(s)
The name returned by a Domain name server (BIND) based on an IP address key.
WINS name(s)
The name returned by a WINS server (Microsoft's name server) based on IP address.
In order to choose from all of these, it is necessary to understand why they are required (if at all). The host ID and network level addresses simply exist.
They are unique and nothing can be done about them, short of changing the hardware. For the most part they can be ignored. The network level address is
used by the network transport system for end-point data delivery, but this is not something which need concern most system administrators. The network
hardware takes care of itself.
At boot time, each host needs to obtain a unique identity. In today's networks that means a unique IP address and an associated name for convenience. The
only purpose for this name is to uniquely identify the host amongst all of the others on the world-wide network. Although every host has a unique Ethernet
Monday, July 26, 1999
Principles of system administration
Page: 6
address or token ring address, these addresses do not fall into a hardware-independent hierarchical structure. In other words Ethernet addresses cannot be
used to route messages from one side of the planet to the other in a simple way. In order to make that happen, a system like TCP/IP is required. At boot-time
then each host needs to obtain an identity. It has two choices
Ask for an address to be provided from a list of free addresses. (DHCP or BOOTP protocols)
Always use the same address, stored on its system configuration files. (Requires correct information on the disk)
The first of these possibilities is useful for terminal rooms containing large numbers of identical machines. In that case, the specific IP address is
unimportant as long as it is unique.
The second of these is the preferred choice for any host which is special in some way, particularly hosts which provide network services. Network services
should always be at a well-known location.
From the IP address a name can be automatically attached to the host through an internet naming service. There are several services which will perform this
conversion. DNS, NIS, WINS. In these remaining cases one needs to make a choice. As far as anyone on the network is concerned a host is its IP address.
Any names which are used internally by the kernel are quite irrelevant. The only difficulty with naming conflicts is that they can cause internal problems.
This is operating system dependent but it is always best not to risk it.
The only world-wide service is DNS (Domain Name Service) whose common implementation is called BIND (Berkeley Internet Name Domain); this
should always be used. This service binds a local network to the world-wide internet in several important ways. WINS (Windows Internet Name Service) is
a proprietary system built by Microsoft for NT. Any host can register data in this service; it is completely insecure and is therefore inadvisable in any
network which is trusted in some manner (either to provide a service, or to hold important data). DNS (which already existed prior to WINS) is a better
option regardless. WINS will be abandoned as of NT 5.0. NIS also provides a local service which has multiple database functions, not just hostname/IP
address conversion. It can also be avoided in favour of DNS for hostname mapping.
Under NT each system has a textual name which is chosen during the installation. A domain server will provide an SID (security ID) for the name which
helps prevent spoofing. When NT boots it broadcasts the name across the network to see whether it is already in use. If the name is in use, the user of the
workstation is prompted for a new name(!)
Although one should not forget that computers exist for users, it is also important to understand that users are often the biggest threat to the stability of their
computers. Two generations have now grown up with computers in their homes, but these computers were private machines which were not (until recently)
attached to a network. They were not part of an organization with many users--they were used by perhaps one or two family members. In short users have
grown up thinking that what they do with their computers is their business. That is not a good attitude in a network community.
In a shared environment, all users share the same machine resources. If one user is selfish that affects all of the other users. Given the opportunity users will
consume all of the disk space and all of the memory and CPU cycles somehow, whether through greed or simply through inexperience. Thus it is in the
interests of the user community to limit the ability of users to spoil things for other users.
One way of protecting operating systems from users and from faulty software is to place quotas on the amount of system resources which they are allowed.
Disk quotas
place fixed limits on the amount of disk space which can be used per user. The advantage of this is that the user cannot use more storage than this
limit; the disadvantage is that many software systems need to generate/cache large temporary files (e.g. compilers, or web browsers) and a fixed limit
means that these systems will fail to work as a user approaches his/her quota.
CPU time limit
Some faulty software packages leave processes running which consume valuable CPU cycles to no use. Users of multiuser computer systems
occasionally steal CPU time by running huge programs which make the system unusable for others. The C-shell limit cputime function can be
globally configured to help prevent accidents.
Policy decisions
Users collect garbage. To limit the amount of it, one can specify a system policy which includes items of the form: `Users may not have mp3, wav,
mpeg etc.. files on the system for more than one day'. To enforce such a policy, See section Controlling user resources.
Quotas have an unpleasant effect on system morale, since they restrict personal freedom. They should probably only be used as a last resort. There are other
ways of controlling the build up of garbage, See section Controlling user resources.
Internationalization support in computing involves three issue:
Choice of keyboard: American, British, German, Norwegian, Thai...etc.
Fonts: Roman, Cryllic, Greek, Arabic, far east...etc.
Translating program text messages into native language.
Inexperienced computer users usually want to be able to use computers in their own language. Experienced computer users often prefer the American
versions of keyboards and software.
Logging and Auditing
Operating system kernels share resources and offer services. They can be asked to keep lists of transactions which have taken place so that one can later go
back and see exactly what happened at a given time. This is called logging or auditing.
Full system auditing involves logging every single operation which the computer performs. This consumes vast amounts of disk space and CPU time and is
generally inadvisable unless you have a specific reason to audit the system. Part of auditing used to be called system accounting from the days when
computer accounts really were accounts for real money. In the mainframe days, users would pay for system time in dollars and thus accounting was
important since it showed who owed what.
Auditing has become an issue again in connection with security. Organizations have becomes afraid of break-ins from system crackers and want to be able
Monday, July 26, 1999
Principles of system administration
Page: 7
to trace the activities of the system in order to be able to look back and find out the identity of a cracker. The other side of the coin is that system accounting
is so resource consuming that the loss of performance might be more important to an organization than the threat of intrusion.
Executing jobs at regular times
The ability of a host to execute jobs at predetermined times is very important in automating the task of system administration. Unix has a time daemon called
cron: it's chronometer. Cron reads a configuration file called a `crontab' file which contains a list of shell-commands to execute at regular time intervals.
On modern UNIX systems, every user may create and edit a crontab file using the command
crontab -e
This command starts a text editor allowing the file to be edited. The contents of a user's crontab file may be listed at any time with the command crontab
-l. The format of a crontab file is a number of lines of the form
minutes 0-59
hours 0-23
day 1-31
month 1-12
weekday Mon-Sun
An asterisk or star `*' may be used as a wildcard, indicating `any'. For example:
# Run script every weekday morning Mon-Fri at 3:15 am:
15 3 * * Mon-Fri /usr/local/bin/script
A typical root crontab file looks like this:
# The root crontab
0 2 * * 0,4 /etc/cron.d/logchecker
5 4 * * 6
0 0 * * *
/usr/local/bin/cfwrap /usr/local/bin/cfdaily
30 * * * * /usr/local/bin/cfwrap /usr/local/bin/cfhourly
The first line is executed at 2a.m. on Sundays and Wednesdays, the second at 4:05 on Saturdays; the third is executed every night at 00:00 hours and the
final line is executed every half hour.
In old BSD 4.3 unix, it was only possible for the system administrator to edit the crontab file. In fact there was only a single crontab file for all users, called
`/usr/lib/crontab' or `/etc/crontab'. This contained an extra field, namely the username of under which the command was to be executed. This type
of crontab file is largely obsolete now, but may still be found on some older BSD 4.3 derivatives such as DEC's ULTRIX.
0,15,30,45 * * * * root /usr/lib/atrun
00 4 * * *
root /bin/sh /usr/local/sbin/daily
2>&1 | mail root
30 3 * * 6
root /bin/sh /usr/local/sbin/weekly 2>&1 | mail root
30 5 1 * *
root /bin/sh /usr/local/sbin/monthly 2>&1 | mail root
A version of cron has been ported to Windows.
File system planning
Most operating systems have hierarchical file systems with directories and subdirectories. This is a powerful tool for organizing data. Disks can also be
divided up into partitions. In many operating systems the largest supported partition size is 4GB since that is the maximum size which can be represented in
a 32 bit register. The point behind directories and partitions is to separate files so as not not mix together things which are logically separate. For example,
User home directories
Development work
Commerical software
Free software
Local scripts and databases
One of the challenges of system design is in finding an appropriate directory structure for all data which are not the operating system.
Data which are separate from the operating system should be kept in a separate directory tree, preferably on a separate disk partition. If they
are mixed with the operating system file-tree if makes re-installation or upgrade of the operating system unnecessarily difficult.
Apart from anything else, it makes no sense to mix logically separate file trees. Operating systems often have a special place for installed software.
Regrettably they often break the above rule and mix software with the operating system's file tree. In Unix machines the place for installed software is
traditionally `/usr/local'; fortunately in Unix separate disk partition can be placed anywhere in the file tree on a directory boundary, so this is not a
practical problem as long as everything lies under a common directory. In NT software is often installed in the same directory as the operating system itself;
also NT does not support partition mixing in the same way as Unix so the re-installation of NT means re-installation of all the software as well.
Data which are installed or created locally are not subject to any constraints however: they may be installed anywhere. One can therefore find a naming
scheme which gives the system logical clarity. This benefits users and management issues. Again we may use directories for this purpose. Operating
systems which descended from DOS also have the concept of drive numbers like A:, B:, C: etc. These are assigned to different disk partitions. Some Unix
operating systems have virtual file systems which allow one to add disks transparently without ever reaching a practical limit. Users never see partition
boundaries. This has both advantages and disadvantages since small partitions are a cheap way to contain groups of misbehaving users, without resorting to
disk quotas.
Each operating system has a model for laying out its files in a standard pattern, but user files are usually left unspecified. System administrators are usually
made responsible for both the safety and integrity of data on a network of computers. Choosing a sound layout for data can make the difference between an
Monday, July 26, 1999
Principles of system administration
Page: 8
incomprehensible chaos and a neat orderly structure. An orderly structure is useful not only for the users of the system, but also when making backups.
Some relevant issues are:
Disk partitions are associated with drives or directory trees when connected to operating systems. These need names.
Naming schemes for files and disks are operating system dependent.
The name of a partition should reflect its function or contents.
In a network the name of a partition ought to be a URL i.e. contain the name of the host.
It is good practice to consolidate file storage into a few special locations rather than spreading it out all over the network. Data kept on many
machines can be difficult to manage, compared to data collected on a few dedicated file servers. Also, insecure operating systems offer files on a
local disk no protection.
The URL model of file naming has several advantages. It means that one always knows the host-provider and function of a network resource. Also it fall
nicely into a hierarchical directory pattern. A simple but effective scheme is to use a three level mount-point for adding disks: each user disk is mapped onto
a directory with a name of the form
/site /host /purpose
This scheme is adequate even for large organizations and can be extended in obvious ways.
In DOS-derived operating systems one does not have the freedom to `mount' network filesystems into the structure of the local disk, network disks always
become a special `drive', like H: or I: etc. It is difficult to make a consistent view of the disk resources with this system, however it is rumoured that NT 5.0
will have this possibility and one can already use filesystems like the DFS on NT which do support this model.
Within an organization a URL structure provides a global naming scheme, like those used in true network filesystems like AFS and DFS. These use the
name of the host on which a resource is physically located to provide a point of reference. This is also an excellent way of labelling backups of partitions
since it is then immediately clear where the data belong. A few rules of thumb allow this naming scheme to live painlessly along side traditional Unix naming
When mounting a remote filesystem on a host, the client and server directories should always have exactly the same name. Anything else only
causes confusion and problems later.
The name of every filesystem mount-point should be unique and tell us something meaningful about where it is located and what its function is.
One can always make links from special unique names to more general names like `/usr/local'.
It doesn't matter whether software compiles in the path names of special directories into software as long as you follow the points above.
For example, the following scheme is used in Oslo at the University and at the College. The first link in the mount point is the part of the organization or
university faculty which the host belongs to, the second link is the name of the host to which the disk is physically connected, and the third and final link is a
name which reflects the contents of the partition. Some examples:
/site /hostname /purpose
The problem of drive names in NT and Windows is an awkward one if one is looking for Unix/NT interoperability. In this case Sun's PCNFS might be an
answer. In practice many networks based on NT and Windows will use Microsoft's model throughout, and while it might not gleam with elegance it does the
job. The problem of backups is confined to the domain servers, so the fact that Windows is not a fully distributed operating system restricts the problem to
manageable proportions.
@hrule @vskip 0.3cm
During the 1970's it was realized that expensive computer hardware could be used most cost-efficiently (by the maximum number of people) if it was
available remotely , i.e. if one could communicate with the computer from a distant location. Inter-system communication became possible through the use
of modems and UUCP and later wide area networks.
The large mainframe computers which served sometimes hundreds of users were painfully slow for interactive tasks, although they were efficient at
time-sharing. As hardware became cheaper many institutions moved towards a model of smaller computers coupled to file-servers and printers by a network.
This solution was relatively cheap but had problems of its own. At this time the demise of the mainframe was predicted. Today, however, mainframe
computers are very much alive for computationally intensive tasks, while the small networked workstation provides access to a world of resources via the
Dealing with networks is one of the most important aspects of system administration today. In order to be a system administrator it is necessary to
understand how and why networks are implemented, using a world-wide protocol: the internet protocol family.
Although it changes fairly frequently, the popular jargon of system administration is useful to know when reading the commercial press. Here we define a
few terms which can be found in the literature.
Monday, July 26, 1999
Principles of system administration
Page: 9
A company which sells hardware or software. A seller .
An Intel based personal computer, used by a single user.
A non-Intel based personal computer which might be used by several users. Workstations might be based on for example SPARC (Sun
Microsystems) or Alpha (Digital/Compaq) chip sets.
A small business network environment. Enterprise management is a popular concept today because NT has been aimed at this market. If nothing
else, Microsoft know about marketing and have highlighted this important market far more effectively than the Unix vendors have done before them.
Enterprise management typically involves running a web server, a database, a disk server and a group of workstations and common resources like
printers and so on. Many magazines think of enterprise management as the network model, but when people talk about Enterprise management they
are really thinking of small businesses with fairly uniform systems.
LISA Large Installation System Administration. This refers to environments with many (hundreds or thousands of) computers. The environments
typically consist of many different kinds of system from multiple vendors. These systems are usually owned by huge companies, organizations like
NASA or universities.
Consolidated system
A centralized mainframe type of solution for concentrating computing power in one place. This kind of solution makes sense for heavy calculations,
performed in engineering and for computer graphics.
Distributed system
A de-centralized solution, in which many workstations spread the computing power evenly throughout the network.
Open systems
is a concept promoted originally by Sun Microsystems for Unix. It is about software systems being compatible through the use of freely available
standards. Competitors are not prevented from knowing how to implement and include a technology in their products or from selling it under
Proprietary systems
is the opposite of open systems. These systems are secret and the details of their operation is not disclosed to competitors. The most infamous
example is Microsoft, but there are many others.
A domain is a logical group of hosts. This word is used with several different meanings in connection with different software systems. The most
common meaning is connected with DNS, the Domain Name Service. Here a domain refers to an Internet suffix, like `', or
`'. Internet domains denote organizations. Domain is also used in NT to refer to a group of hosts sharing the attributes of a common
file server. Try not to confuse Domain nameserver (DNS) server with NT Domain server.
Finally, it is important to distinguish between a user interface and an operating system. A window system is a graphical user interface (GUI), an operating
system shares resources and provides functionality. This issue has been confused by the arrival of operating systems called Windows and including a
graphical user interface. An operating system can be good or bad independently of whether its windowing system(s) is/are good or bad.
Common network sharing models
Different operating systems support different ideas about how networks should be used.
Unix is not tied to any specific model for utilizing network resources. Any Unix system can perform any function, as server, client or both. A Unix
network is fully distributed, there is no requirement about centralization of resources, but central models are commonly used. Unix is a multitasking,
multiuser system. It is is the most broadly configurable of the popular operating systems and many tools are freely available for administratrating it.
Unix contains tools for making many hosts work together and share resources, but each host can also be configured as a standalone system. Each
host either has a fixed IP address, or can be assigned one automatically at boot time by a service such as BOOTP or DHCP with ARP/RARP .
There are two types of NT system with separate software licenses: workstations and servers . NT revolves around a model in which programs are
run on a local workstation, but where network services are run on a centralized server. All systems must run NT or understand the proprietary
protocols which NT uses. IP addresses are fixed or may be assigned automatically by a network service such as BOOTP or DHCP . Several NT
servers can coexist. Each server serves a `domain' and clients subscribe to as many domains as they wish. NT is not a distributed system in the
sense that services are localized on server machines. NT supports two kinds of organizational groups: workgroups in which hosts share a simple
peer-to-peer network, perhaps with Windows 9x machines, and domains which have a central authority through a domain server. Domains provide
a common framework including user-id's (SID's in NT language), passwords and user profiles. Domains have a common user-database and a
common security policy. Any host which subscribes to a domain inherits the users and the security policy of the domain.
Novell The Novell Netware software is not an operating system. It is a network server for PCs which adds file storage, printing and other network services
on top of Windows, DOS, MacIntosh or GNU/Linux. The network protocol for local traffic is IPX which is faster than IP and is an
inter-networking protocol, but it is not a world wide protocol, thus Novell run PC's still need IP configurable interfaces. Each PC can also obtain an
IP address dynamically from a single server which replies to a BOOTP or DHCP broadcast request. Several Novell file servers can coexist. All
services run on these servers (which support a form of multitasking). The server is independent of the basic operating system run on the PCs it
supports, it will work with DOS or Windows or Windows 95, or NT workstation. Novell is not a distributed system like Unix. It requires a special
dedicated machine to perform a function as server.
Each MacIntosh is an independent system. Simple services like ftp can be run in a limited way from a normal machine. Macintosh uses its own
network protocol called Apple-talk which is incompatible with IP and IPX. Apple-talk servers allow networking and disk sharing. IP protocol disk
sharing available but does not mix well with the MacIntosh file system. System administration (actually everything) is by GUI only.
Recently, several companies (e.g. Auspex, Network Appliance) have begun producing solutions for integrating disk storage both for Unix and for Windows
worlds. IBM has traditionally produced software which works both on Unix and Microsoft platforms.
Another issue for networked systems is where configuration (preference) data should be stored. There are two possibilities here which correspond
approximately to the Unix approach and the approach used by all other systems.
Under Windows (9x/NT) and MacIntosh systems, each user is assumed to have his or her own personal workstation which will not normally be
used by other users. Configuration data or preferences which the user selects are thus stored locally on the system disk in a location provided by the
operating system. This location is common to all users. Only NT distinguishes between different users.
Under Unix, each user sets up personal preferences in his or her personal dot files which are stored in private user space. More general global
preferences are stored in a directory of the administrator's choice. Traditionally this has been the directory `/etc'.
Monday, July 26, 1999
Principles of system administration
Page: 10
The difficulties associated with the first of these approaches (having a fixed location for the configuration information which lies in the system files) are
several. In any single user operating system, one user can overwrite another users' preferences simply by changing them since the system is not capable of
telling the difference between users. This is a fundamental problem though which indicates that single user operating systems are basically unsuited to
networking. More pertinent to a networked world are the following point:
When the operating system is reinstalled, configuration information can easily be lost or overwritten if they are stored in an operating system
In a distributed environment, where users might not sit at the same physical workstation day after day, the user's personal configuration data will not
follow him or her from machine to machine.
NT partly solves these problems by maintaining user profiles which are stored on the domain server in a `\profiles' subdirectory of the system-root.
These data are copied into the local workstation when a user logs on to a domain server. On a Unix system, it is easy to specify the locations of
configuration files and these can then be kept separate from operating system files, e.g. on a different disk partition so that they are immune to accidental
deletion by system re-installation.
The primary difference between Unix and Windows/MacIntosh systems is that Unix is a multiuser system, i.e. any Unix machine can be used by an
arbitrary number of users from an arbitrary location on the network. Microsoft and Apple systems allow only a single user to use a workstation interactively,
from the console of the machine itself.
At the time of writing we are in the middle of a war of words between Unix and NT. In 1997, NT gained a lot of ground with newly started companies, eager
to get going on the Internet as quickly as possible. Microsoft efficient marketing and existing dominance of the PC world made NT an interesting option.
NT has suffered from setbacks as a result of bugs which affect security and stability, in the face of a vigorous marketing campaign. A major problem is the
need for compatibility with DOS, through Windows 9x to NT. Since both DOS and Windows 9x are insecurable systems, this has led to conflicts of
interest. Unix vendors have tried to keep step, in spite of poor public image of Unix (often the result of private dominance wars between different Unix
vendors) but the specially designed hardware platforms built by Unix vendors have had a hard time competing with inferior but cheaper technology from the
PC world.
In 1998 we see sales of both Unix and NT increasing but many users who have tried NT report that they are turning to cheap Unix solutions on Intel
platforms (GNU/Linux, FreeBSD etc) instead because of its proven reliability. Unix tends to work better in a distributed environment and delivers especially
good performance as a database server on 64-bit multiprocessor hardware. NT can run on 32-bit multiprocessor systems, but independent benchmarks show
that it does not scale as well as many Unix variants do. Also the issue of remote login is significant for some sites. Using standard Windows user interfaces
on NT, one cannot run graphical applications remotely as with X-windows on Unix. A basic X-windows release 6 is available for NT however, so one has
the option of replacing the Windows interface and choosing a more distributed user interface. Today some systems which are based on the Mach
micro-kernel can run native Unix and NT simultaneously. For example, Digital (now Compaq) Unix, GNU/Linux and NT can run on the Alpha processor.
User sociology
Most branches of computer science deal only with software systems and algorithms. System administration is made more difficult than this by the fact that it
deals with computer communities and is therefore strongly affected by what human beings do. In short, a large part of system administration is sociology .
A newly installed machine does not usually require attention until it is first used, but as soon as a user starts running programs and storing data, the
reliability and efficiency of the system are tested. This is where the challenge of system administration lies.
The load on computers and on networks is a social phenomenon: it peaks in response to patterns of human behaviour. For example, at universities and
colleges network traffic usually peaks during lunch breaks, when students rush to the terminal rooms to surf on the web or to read E-mail. In industry the
reverse can be true, as workers flee the slavery of their computers for a breath of fresh air (or carbonized air). In order to understand the behaviour of the
network, the load placed on servers and the availability of resources, we have to take into account the users' patterns of behaviour.
Models of system administration
There are several approaches to the management of computers in a network environment. Here are three broad categories which are used today.
Because the number of local networks has outgrown the number of experienced technicians, there are many administrators who are not skilled in the
systems they manage. A disturbing but common belief, which originated in the 1980's microcomputer era, is that problems with a computer can be
fixed by simply rebooting the operating system. Since home computer systems tend to crash with alarming regularity, this is a habit which has been
acquired from painful experience. The habit should be stifled however since more mature preemptive systems like Unix and NT should almost never
need to be rebooted(1), indeed it can be damaging to do so. Rebooting a multi-user system is extremely dangerous since users might be logged in
from remote locations and lose data. Moreover, the number of problems which can be fixed by rebooting a system is small compared to the number
of problems which eventually arise. For this reason we classify this point as a regrettable but real development in network management.
Another approach to system administration is the use of control systems such as Tivoli, HP OpenView and Sun Solstice. In this approach the
system administrator follows the state of the network on a graphical user interface. The administrator defines error conditions to look for and a
process on each host reports errors as they occur to the administrator. In this way the administrator has an overview of every problem on the
network from his/her single location and can either fix the problems by hand as they occur (if the system supports remote login), or distribute
scripts and antidotes which provide a partial automation of the process. The disadvantage with this system is that a human administrator usually has
to start the repair procedures by hand and thus this creates a bottleneck, since all the alarms go to one place to be dealt with serially. This means that
the amount of work required to run this system increased roughly linearly with the number of hosts on the network.
A relatively new approach to system management which is gaining growing popularity is the idea of equipping networked operating systems with a
simple immune system. By analogy with the human body, an immune system is an automatic system which every host possesses which attempts to
deal with emergencies. An immune system is the Fire, Police and Paramedic services as well as the garbage collection agencies. In an immune
system, every host is responsible for automatically repairing its own problems, without sending out warnings about what is going on. This avoids a
serial bottleneck created by a human administrator. The time spent on implementing this model is independent of the number of hosts on the
network. Unix administrators have run background scripts for many years. Such scripts (often called sanity checking scripts) run daily or hourly
and make sure that each system is properly configured, perform garbage cleaning and report any serious problems to an administrator. In an
immunological model, the aim is to minimize the involvement of a human being as far as possible. The tool Cfengine introduced technology which
promotes this approach.
Monday, July 26, 1999
Principles of system administration
Page: 11
These different approaches are easily implemented on Unix systems because of the large number of scripting languages and tools available. On Microsoft
systems the amount of freely available languages ported from Unix has now increased to the point where there are few problems there either. For MacIntosh
systems there seems to have been little development and one is locked into the commercial products available there. The main problem in implementing
monitoring software on insecure operating systems is that they are basically unstable. If a network administrator were informed every time a Windows 9x
machine crashed or was rebooted on a network of hundreds, there would be a considerable amount of irrelevant traffic. Since there is nothing an
administrator can do to change this fact, it has to be defined as a `feature' of these operating systems and tolerated.
Network details
This section offers a minimal introduction to some of the concepts required to discuss networking. You can refer to the bibliography if you would like to
read more, See section Recommended reading.
A network is a line of communications between two or more hosts. It is usually some kind of shared cable which is attached to several hosts simultaneously
by means of a network interface . Since it is impractical to have a private cable between every pair of hosts on a network (this would require N network
interfaces and cables per host and would be unmanageable) there are two main ways of connecting hosts together.
Bus/Ethernet approach
In the bus approach, every host is connected to a common cable or bus. Only one host can be using a given network cable at a given instant. It is like
a conference telephone call: what goes out onto a network reaches all hosts on that network (more or less) simultaneously, so everyone has to share
the line by waiting for a suitable moment to say something. You should not think data transmission over the network as being a little stream of bytes
rollin' down the track one behind the other, from origin to destination, like Thomas the Tank-Engine. Every bit, every 1 or 0, is a signal (a voltage or
light pulse) on a cable which fills the entire cable at a good fraction of the speed of light. It's like sending Morse code with a lighthouse. Everyone
sees the signal, but only the recipient bothers to read it.
Token ring/FDDI approach
In the token ring approach, each host has two network interfaces and the hosts are connected in a ring. If the signal arriving at one of the interfaces
is for the host itself then it is read. If it is not intended for the host itself, the signal is forwarded to the next host where the same applies. The most
common token ring in use today is the optical FDDI (Fiber distributed data interface).
Both of these approaches are susceptible to spying devices. If a host on the network wants to overhear a conversation between two others they have only to
listen. In some cases extra hardware switches can be used to isolate private connections (see below). In what follows an Ethernet type of network will be
assumed, since this is the commonest form in user-workstation environments.
Even with the bus approach, any host can be connected to several independent network segments. It must have a network interface for each network it is
attached to. Each network interface has a separate network address. Thus a host which is connected to several networks will have a different address on each
network. A host which is coupled to several networks and which forwards data from one network to another is called a router .
Network signals are carried by a variety of means. These days copper cables are being replaced by fiber-optic glass transmission for long distance
communication. In local area networks it is still copper cables which carry the signals. These cables are often called Ethernet cables, but in some cases the
name is used mainly for historical reasons since the Ethernet has been superseded by the twisted pair technology. Thick yellow cables (about 1cm thick)
carry thick Ethernet, thin black cables (0.5cm) with BNC connectors carry thin Ethernet. Fibre optic cables (FDDI) have varying appearances. Twisted pair
lines are sometimes referred to at 10baseT, 100baseT etc, or for short T1, T10, T100. The numbers indicate the capacity of the line, `base' indicates that the
cable is used in a baseband system and the `T' stands for twisted-pair. Twisted pair cables are very thin, so they are not tapped as are thin and thick
Ethernet cables. Instead each host has a single cable connecting it to a multi-way repeater or hub .
The cables of the network are joined together in segments by hardware which makes sure that messages are transmitted in the right direction to reach their
destinations. A host which is coupled to several network segments and which forwards data from one network to another is called a router . Routers not
only forward data but they prevent the spread of network messages which other network segments do not need to know about. The limits the number of
hosts which are sharing any given cable segment, and thus limits the traffic which any given host sees. A router knows which destination addresses lie on
which of the networks it is connected to and it does not let message traffic spread onto irrelevant cables.
A bridge is a hardware device which acts like a filter on busy networks. A bridge works like a `mini-router' and separates two segments of the same cable.
A bridge knows which parts of the cable do not contain a destination address and prevents traffic from spreading to this part of a cable. A bridge is use to
isolate traffic on busy sections of a network or conversely to splice networks together.
A repeater is an amplifier which strengthens the network signal over long stretches of cable. A multi-port repeater also called a hub does the same thing
and also splits one cable into N sub-cables for convenience. Hubs are common in twisted pair networks where it is necessary to fan a cable out into a star
pattern from the hub to send one cable to each host. A switch is a hub which can direct a message from one host cable directly to the intended host by
routing the signal directly. The advantage with this is that other machines do not have to see the traffic between two hosts. Each pair of hosts has an virtual
private cable. Switched networks are immune to spies, net-sniffing or network listening devices.
When learning about a new network you should obtain a plan of the physical setup. If you have done your homework, then you will know where all of these
boxes are on your network.
Information transactions take place by agree standards or protocols . Protocols exist to make sure that transmitted data are understood by the receiver in the
way that the sender intended. On a network, protocols are required to make sure that data are understood, not only by the receiver, but by all the network
hardware which carry them between source and destination. The data are wrapped up in envelope information which contains the address of the destination.
Each transmission layer in the protocol stack (protocol hierarchy) is prefixed with a some header information which contains the destination address and
other data which identify it. The Ethernet protocol also has a trailer.
---------------------------------------------------<- |ethernet |IP header| TCP
| data
<- | header |
| header| (ftp/telnet..)|trailer |
Monday, July 26, 1999
Principles of system administration
Page: 12
| Appl.
N e t w o r k
a l i n k
Wrapping data inside envelope information is called encapsulation and it is important to understand the basics of this mechanisms. Ten years ago network
administrators did not need to concern themselves with protocols and their like; today however network attacks make clever use of the features and flaws in
these protocols and system administrators need to understand them in order to protect systems from the attacks.
The Internet Family of protocols has been the basis of Unix networking for many years, since they were implemented as part of the Berkeley Software
Distribution (BSD) Unix. The hierarchy has the following form:
The transmission control protocol TCP is for reliable connection oriented transfer. The user datagram protocol UDP is a rather cheaper connectionless
service and the internet control message protocol (ICMP) is used to transmit error messages and routing information for TCP/IP. These protocols have an
address structure which is hierarchical and routable , that means that IP addresses can find their way from any host in the world to any other so long as they
are connected. The ethernet protocol does not know much more about the world than the cable it is attached to.
NT supports three network protocols
NETBIOS Extended User Interface, Microsoft's own network protocol. This was designed for small networks and is not routable.
Novell/Xerox's IPX/SPX protocol suite. Routable.
Standard Internet protocols. The default for NT 4.
Data formats
There are many problems which arise in networking when hardware and software from different manufacturers have to exist and work together. Some of the
largest computer companies have tried to use this to their advantage on many occasions in order to make customers buy only their products. An obvious
example is the choice of network protocols used for communication. Both Apple and Microsoft have tried to introduce their own proprietary networking
protocols. TCP/IP has won the contest because it was an inter -network protocol (i.e. capable of working on and joining together any hardware type) and
also because it is a freely open standard. Neither Appletalk nor Microsoft's protocols has either of these features.
This illustrates how networking demands standards. Another example of this is the way in which different operating systems represent numerical data.
Operating systems (actually the hardware they run on) fall into two categories known as big endian and little endian . The names refer to the byte-order
of numerical representations. The names indicate how large integers (which require say 32 bits or more) are stored in memory. Little endian systems store
the least significant byte first, while big endian systems store the most significant byte first. For example, the representation of the number 34,677,374 has
either of these forms.
| 34
126 |
----------------------------------126 |
| 17
Little |
Obviously if one is transferring data from one host to another, both hosts have to agree on the data representation otherwise there would be disastrous
consequences. This means that there has to be a common standard of network byte ordering . For example, Solaris (SPARC hardware) uses network byte
ordering (big endian), while GNU/Linux (Intel hardware) uses the opposite (little endian). This means that Intel systems have to convert the format every
time something is transmitted over the network.
IP addresses
Every network interface on the Internet needs to have a unique number which is called its address. At present the internet protocol is at version 4 and this
address consists of four bytes, or 32 bits. In the future this will be extended, in a new version of the internet protocol IPv6, to allow more IP addresses since
we are rapidly using up the available addresses. The addresses will also be structured differently. The form of an IP address in IPv4 is
Some IP addresses represent networks, whereas others represent hosts, or actually (if we are being 100 percent correct) they represent network interfaces.
The usual situation is that an IP address represents a host attached to a network. Such an address has a host part and a network part, but the format of the
address is not necessarily the same for all hosts and networks. There is some freedom to define local conventions.
Monday, July 26, 1999
Principles of system administration
Page: 13
In every IPv4 address there are 32 bits. We can choose to use these bits in different ways: we could use all 32 bits for host addresses and keep every host on
the same enormous cable, without any routers (this would be physically impossible in practice), or we could use all 32 bits for network addresses and have
only one host per network (i.e. a router for every host). Both these extremes are silly. After all, we are trying to save resources by sharing a cable between
convenient groups of hosts, but shielding other hosts from irrelevant traffic. What we want instead is to group hosts into clusters so as to restrict traffic to
localized areas.
Networks fall historically into three classes called class A, class B and class C networks. This distinction has proven to be mistake. Because of the
possibility of dividing into subnets , the distinction between these network types has to do with the historical development of the internet. The difference
between class A, B and C networks concerns which bits in the IP addresses on that network refer to the network itself and which bits refer to actual hosts.
IP addresses from to are class A networks, where the first byte is a network part and the last three bytes are the host address. This
allows @math{256^3} minus reserved addresses hosts on the network. Since this is a ludicrously large number, class A networks are no longer issued and
all the free addresses are wasted. Class A networks were intended for very large organizations and are only practical with the use of a netmask which divides
up the large network into manageable subnets.
IP addresses from to are class B networks. Here the first two bytes are the network part and the last two bytes are the host part.
This gives a maximum of @math{256^2} minus reserved host addresses. Class B networks are typically given to large institutions such as universities and
internet providers.
IP addresses from to are class C networks. Here the first three bytes are network addresses and the last byte is the host part.
This gives a maximum of 254 hosts per network.
Some IP addresses are reserved for a special purpose:
Default route
Loopback address
Network addresses
Broadcast addresses
Router or gateway (conventionally)
Multicast addresses
The default route is a default destination for outgoing packets on a subnet and is usually made equal to the router address. The loopback address is an
address which every host defines. It points straight back to the host itself. It is a kind of internal fake network which allows programs to use IP protocols to
address local services. The zeroth address of any network is reserved to mean the network itself, and the 255'th (or on older networks sometimes the zeroth)
is used for the broadcast address. Some internet addresses are reserved for a special purpose. These include network addresses (usually xxx.yyy.zzz.0),
broadcast addresses (usually xxx.yyy.zzz.255, but in older networks it was xxx.yyy.zzz.0) and multicast addresses (usually
In practice, a given `network' might consist of very many separate cable systems, coupled together by routers and bridges. One problem with very large
networks, like class B or class A networks is that broadcast messages (i.e. messages which are sent to every host) create traffic which can slow a busy
network. In most cases broadcast messages only need to be sent to a subset of hosts which have some logical or administrative relationship, but unless
something is done a broadcast message will by definition be transmitted to all hosts on the network. What is needed then is a method of assigning groups of
IP addresses to specific cables and limiting broadcasts to hosts belonging to the group. The purpose of subnets is to divide up networks into regions which
naturally belong together and to isolate regions which are independent. This reduces the propagation of useless traffic.
This logical partitioning can be achieved by dividing hosts up, through routers, into subnets. Each network can be divided into subnets by using a
netmask . Each address consists of two parts: a network address and a host address . A system variable called the netmask decides how IP addresses are
interpreted locally. The netmask decides the boundary between how many bits of the IP address will be kept for hosts and how many will be kept for the
network location name. There is thus a trade off between the number of allowed domains and the number of hosts which can be coupled to each subnet.
Subnets are usually separated by routers, so the question is how many machines do we want on one side of a router?
The netmask only has a meaning as a binary number. When you look at the netmask, you have to ask yourself -- which bits are ones and which are zeroes?
The bits which are ones decide which bits can be used to specify the domain and the subnets within the domain. The bits which are zeroes decide which are
hostnames on each subnet. The local network administrator decides how the netmask is to be used.
The host part of an IP address can be divided up into two parts by moving the boundary between network and host part. The netmask is a variable which
contains zeroes and ones. Every one represents a network bit and every zero represents a host bit. By changing the value of the netmask, we can trade many
hosts per network for many subnets with fewer hosts.
Class B address
----------------------------------net |
net | host | host |
----------------------------------Subnet mask
----------------------------------| net id | net id | subnet| host
----------------------------------Broadcast address
Monday, July 26, 1999
Principles of system administration
Page: 14
A subnet mask can be used to separate hosts which also lie on the same physical network, thereby forcing them to communicate through the router. This
might be useful for security or administrative purposes.
The IP address of a host is set in the network interface. The UNIX command ifconfig (interface-configuration) or the NT command ipconfig is used to
set this. Normally this is set at boot time by a shell script executed as part of the `rc' start-up files. These files are often constructed automatically during
the system installation procedure. The ifconfig command is also used to set the broadcast address and netmask for your subnet.
Each system interface has a name. You should find out what the name of the interface is on your system. A look at the manual page can help here. Here are
the network interface names commonly used by different UNIX types.
DEC ultrix
Look at the manual entry of your system for this command. Here is a typical ifconfig command which sets the internet address, netmask and broadcast
address on a SUN system which a Lance-ethernet interface.
ifconfig le0 up netmask broadcast
Normally you will not need to use this command directly, since it should be in the startup-files for the system, from the time you installed the system. But
you might be working in single-user mode or are trying to solve some special problem. You might discover that a system has been incorrectly configured.
Note that cfengine can also be used to set the netmask and broadcast address.
The internet protocol (IPv4)
The internet protocol consists of two user protocols (TCP/UDP) and a private control protocol (ICMP). The simplest protocol is the unreliable,
connectionless protocol UDP (User datagram protocol). UDP datagrams know their where they come from and where they are going, but have no idea about
how they relate to other datagrams or how what they are carrying fits into the big picture. UDP has no mechanism for determining whether datagrams arrive
at their destination or not.
TCP packets are reliable connection oriented data. They form streams or continuous data-flows with handshaking. This is accomplished by using a
three-way handshake based om so-called SYN (synchronize) and ACK (acknowledge) bits in the TCP header. Suppose host A wishes to set up a
connection with host B . Host A sends a TCP segment to host B with its SYN bit set and a sequence number X which will be used to keep track of the
order of the segments. Host B replies to this with its SYN and ACK bits set, with Acknowledgement=X+1 and a new sequence number Y. Host A then
replies to host B with the first data and the Acknowledge field=Y+1. The reason why each side acknowledges every segment with a sequence number which
is one greater than the previous number sent is that the Acknowledgement field actually determines the next sequence number expected. This sequence is
actually a weakness which network attackers have been able to exploit through different connections, in `sequence number guessing' attacks. Now many
implementations of TCP allow random initial sequence numbers.
The purpose of this circuit connection is to ensure that both hosts know about every packet which is sent from source to destination. Because TCP
guarantees delivery, it retransmits any segment for which is has not received an ACK after a certain period of time (the TCP timeout).
At the end of a transmission the sender sends a FIN (finished) bit, which is replied to with FIN/ACK. In fact closing connections is quite complex since
both sides must close their end of the connection reliably. See the reference literature for further details of this.
The next generation internet protocol (IPv6)
The current implementation of the internet protocol has a number of problems. It is straightforward to calculate that, because of the structure of the IP
addresses, divided into class A,B and C networks, something under two percent of the possible addresses can actually be used in practice. A recent survey
from Unix Review, March 1998 shows
Already allocated
82 74524
334 144201
75 182494
136 124154
2 31690
(Unknown Date)
Monday, July 26, 1999
Principles of system administration
Page: 15
210 115051
1575 687281
This means that, of the total numbers of addresses, these are already allocated:
Class A
Class B
Class C
Max possible
Percent allocated
Of course, this does not mean that all of the allocated addresses are in active use. After all, what organization has 65,535 hosts? In fact the survey showed
that under two percent of these addresses were actually in use. This is an enormous wastage of IP addresses. Amongst the class C networks, where smaller
companies would like address space, the available addresses are being used up quickly, but amongst the class A networks, the addresses will probably never
be used. A new addressing structure is therefore required to solve this problem. Other problems with IPv4 are that it is too easy to take control of a
connection by guessing sequence numbers. Moreover there is no native support for encryption or mobile computing.
In order to address these issues the IETF (Internet Engineering Task Force) has put together a workgroup to design a new 128-bit protocol which will be
called IPv6. Not only will the addressing structure be different but there will be a considerable number of extra addresses. Even with a certain inefficiency of
allocation, it is estimated that there will be enough IPv6 addresses to support a density of 10,000 IP addresses per square meter which ought to be enough
for every toaster and wristwatch on the planet and beyond.
Unless a host operates as a router in some capacity, it only requires a minimal routing configuration. Each host must define a default route which is a
destination to which outgoing packets will be sent for processing when they do not belong to the subnet. This is the address of the router or gateway on the
same network segment. It is set by a command like this:
route add default my-gateway-address
The syntax varies slightly between systems. On GNU/Linux systems you must write:
/sbin/route add default gw my-gateway-address
metric 1
The default route can be checked using the netstat -r command. The result should just be a few lines like this:
Kernel routing table
Ref Use
where my-gw is the address of your local gateway (usually subnet address 1).
If this default route is not set, a host will not know where to send packets and will therefore attempt to build a table of routes, using a different entry for every
outgoing address. This consumes memory rapidly and leads to great inefficiency. In the worst case the host might not have contact with anywhere outside its
subnet at all.
ARP is the (IP) address resolution protocol. ARP takes an IP address and turns it into an ethernet (hardware) address. The ARP service is mirrored by a
reverse ARP service (RARP). RARP takes a hardware address and turns it into an IP address.
Ethernet (or generally hardware) addresses are required when routing traffic from one device to another. While it is the IP addresses which contain the
structure of the internet and permit routing, it is the hardware address to which one must deliver packets in the final instance; this is the address which is
burned into the network interface.
The hardware addresses are cached by each host on the network so that repeated calls to the service ARP translation service are not required. Addresses are
checked later however, so that if an address from a host claiming to have a certain IP address originates from an incorrect hardware address (i.e. the packet
does not agree with the information in the cache) then this is detected and a warning can be issued to the effect that two devices are trying to use the same IP
address. ARP sends out packets on a local network asking the question `Who has IP address xxx.yyy.zzz.mmm?' The host concerned replies with its
hardware address.
For hosts which know their own IP address at boot-time these services only serve as confirmations of identity. Diskless clients (which have no place to store
their IP address) do not have this information when they are first switched on and need to ask for it. All they know originally is the unique hardware
(ethernet) address which is burned into their network interface. In order to bring up and configure an internet logical-interface they must first use RARP to
find out their IP addresses from a RARP server. Services like BOOTP or DHCP are used for this. Also the Unix file `/etc/ethers' and rarpd can be
The ability to read information about the performance of network hardware via the network itself is an attractive idea. Suppose one could look at a router on
the second floor of a building half a mile away and immediately see the load statistics, or number of rejected packets it has seen. That would be useful
diagnostic information. Similar information could be obtained about software systems on any host.
Monday, July 26, 1999
Principles of system administration
Page: 16
SNMP (Simple Network Management Protocol) is a protocol designed to do just this. SNMP supports two operations get and put , thus it can read and
perhaps modify the data stored on a host. SNMP access is mediated by a server process on each hardware node. Modern operating systems usually run
SNMP daemons or services. The services are protected by a rather weak password which is called the community string .
SNMP information is stored in data structures called MIBs (Management Information Bases). The MIBs are catalogued in RFC 1213 and give hardware
and software profiles. They are dynamically updated. An SNMP request specifies the information it wants to read/write by giving the name of a MIB. In
SNMP v2 there are standard MIBs for address translation tables, TCP/IP statistics and so on. There are 27 default parameters that can be altered by SNMP:
system name, location and human contact; interface state (up/down), hardware and IP address, IP state (forwarding gateway/not) IP TTL, IP next HOP
address, IP route age and mask, TCP state, neighbour state, SNMP trap enabling.
Operating systems define their own MIBs for system performance data. Some commercial network management systems like Tivoli and Hewlett Packard's
OpenView work by reading and writing MIBs using SNMP client-server technology. Most Unix variants now support SNMP. NT supports SNMP version
1. Its MIBs can be used to collect information from NT systems, such as the names of users who are logged on.
SNMP seems to be increasing in popularity, but like any public information database, it can be abused by network attackers. Some sites choose to disable
SNMP services altogether. All sites should filter SNMP packets to and from external networks to avoid illegal access of these services from intruders.
@node High level protocols
High level protocols
Internet networks use many high level protocols to provide the distributed services which most users take for granted.
HTTP The world wide web protocol for exchanging hypertext and multimedia data. All data are sent in clear text.
The secure world wide web protocol for exchanging hypertext and multimedia data. All data are encrypting using Netscape's secure socket layer
The File transfer protocol. Passwords are sent in clear text.
The secure shell. A replacement for the remote shell (rsh) Unix protocol. The secure shell provides full encryption and forwarding of X11 display
data through a secure pipe.
LDAP The Lightweight Directory Access Protocol is a generalized protocol for looking up data in simple databases. It is a lightweight version of the
service written for X.500 and is currently at Version 3. It can be used to register user information, passwords, telephone numbers etc and interfaces
through gateways to the NDS (Novell Directory Service), Microsoft's Exchange server and NIS (Sun's Network Information Service). The
advantage of LDAP will be a uniform protocol for accessing table lookups. Currently the spread of LDAP is hindered by few up-to-date
implementations of the protocol. Only Netscape's directory server supports the version 3 standard.
Security and reliability
@hrule @vskip 0.3cm
Security is about keeping the network of machines and their data safe. This includes protecting against
Malicious attacks.
Accidental erasure of data.
Disk crashes.
User ignorance.
There are two kinds of approach to maintaining system security: prevention and recovery.
Security is an increasingly important problem. Just in the last few years the number of attacks and break-ins to computer systems has risen to millions of
cases a year. Crackers(2) have found their way inside the computers of the pentagon, the world's security services, warships, fighter plane command
computers, banks and major services such as electrical power grids. With this kind of access the potential for causing damage is great. Computer warfare is
the next major battlefield we have to conquer. It is happening now, as you read these words. It is here, like it or not. Moreover, it is estimated that the banks
lose millions of dollars a year to computer crime.
If that is not enough to scare you into awareness, you will just have to learn the hard way.
Security is a huge subject, because modern computer systems are very complex and the connectivity of the internet means that millions of people can try to
break into networked systems. This chapter is not a comprehensive guide to security. There is not time in this introductory course to cover security in detail.
There are several reasons for this. One is the sheer size of the subject, the other is that security awareness requires a level of experience which we have not
had time to develop. A third reason is that one could easily spend every waking moment worrying about security, but in the end it comes down to a choice:
how much security do you need?
It is a good idea to have a security policy so that you know exactly what you mean by security. That way, when security is breached, you will know what to
do. Some sites which contain sensitive data require strict security and spend a lot of time enforcing it, others do not particularly care about their data and
would rather not waste their time on pointless measures to protect them.
Systems which do not implement security tend to attract only low-level crackers--and those who manage to break in, tend to use the systems only as a
springboard to go other places. The more security you implement, the more of a challenge it is for a cracker. So spending a lot of time on security might
only have the effect of asking for trouble.
When it comes down to it there is this: if you have extremely sensitive data, do not put them onto a network capable computer. If you do, live with the
User security
Password security
Monday, July 26, 1999
Principles of system administration
Page: 17
Password security is the first line of defense against intruders. Experience shows that many users have little or no idea about the importance of using a good
Consider some examples from a survey of passwords at a university. About 40 physicists had the password `Einstein', around 10 had `Newton' and several
had `Kepler'. Hundreds of users used their login-name as their password, some of them really went to town and added `123' to the end.
Passwords are not visible to ordinary users, but their encrypted form is often visible. There are many publicly available programs which can guess
passwords and compare them with the encrypted forms. No one with an easy password is safe. Passwords should never be any word in a dictionary or a
simple variation of such a word or name. It takes just a few seconds to guess these.
Some new operating systems like FreeBSD, NetBSD and Solaris and GNU/Linux have `shadow password files' which are not readable by normal users.
The regular password file contains an `x' instead of a password, and the encrypted password is kept in an unreadable file. This makes it much harder to scan
the password file for weak passwords.
Once a malicious user has gained access to an account, it is very much easier to exploit other weaknesses in security. Good passwords are the key to a safe
system. On the network, it is possible to fetch programs such as crack which are designed to break passwords. It is unusual not to find a few accounts with
trivial passwords in the space of a few seconds.
A useful way to choose a password is to use the PIN code from a little-used credit card as a part of your password. This means that you don't have to
remember too much--and it means that you do have secret numbers in your password. The worst passwords are names and words which are in any language
xhost access list
Although many users are not aware of it, it is often possible to download the screen image of another user who is using the X-windows system. It is equally
possible to `bug' the keyboard and listen to all the key-presses. The problem is an out-dated security mechanism which has long since been replaced, but
which is still used by very many users. The problem is the xhost program. This is used to grant other hosts permission to draw on your X server--in other
words, if you are remotely logged on to a host other than the one you are using as a display, you must grant the remote host access to write on your screen.
In the old X windows system, prior to release 5, one had to grant access to a particular host. One this was done, anyone on that host had access to your
server, not just you. This was later replaced by the xauth magic-cookie mechanism which works on a user basis. Some users still insist on using xhost
however, with a command like this:
xhost +
Any user writing this, opens their display to everyone in the world. The antidote, of course is the command xhost -. Users of the secure shell ssh, See
section Secure shell, can now have automatic X11 forwarding with authentication cookies. Everyone should therefore execute xhost - once and never use
the xhost mechnism again.
Secure shell
This is a secure replacement for the rsh commands. It protects against IP spoofing where a remote host pretends to be trusted host by faking IP datagrams;
DNS spoofing where an attacker forges name entries in the name-service; the interception of passwords in network packets and several other kinds of attack.
Note that the secure shell is NOT free software. It is only freely usable in academic contexts for Unix-like systems. Anything else you have to pay for.
To install this, collect it from an ftp site and perform the usual steps:
(get ftp file ssh-2.0.9.tar.gz)
host% tar zxf ssh-2.0.9.tar.gz
host% cd ssh-2.0.9
host% ./configure
host% make
host% su
Passwd: ********
host# make install
You then need to start the daemon by adding a command of the form
to the startup scripts on your system.
Accidental deletion of files
Once a file is deleted in UNIX, it is not possible to get it back. There is no way to undelete a file. Some system administrators like to protect ignorant users
by making an alias (in C shell)
alias rm
rm -i
which causes the rm command to prompt whether it should delete files before actually doing so. This is a simple idea and it is not foolproof. The only real
security against deletion is to keep extensive backups of user disks.
Monday, July 26, 1999
Principles of system administration
Page: 18
System security
Since the explosion of interest in the internet, the possibility of hosts being attacked from outside sources, has become a significant problem. With literally
millions of users on the net, the tiny percentage of malicious users becomes a large number.
Security Policy
The place to start in implementing security is to identify what it is you are trying to accomplish. First of all: what resources are you trying to protect ?
Some sites have secrets they wish to protect. They might be government or trade secrets or the solutions to a college exam.
Personell data
In your country there are probably rules about what you must do to safeguard sensitive personal information. This goes for any information about
employees, patients, customers or anyone else you deal with. Information about people is private.
CPU usage/System downtime
You might not have any data that you are afraid will fall into the wrong hands. It might simply be that your system is so important to you that you
cannot afford the loss of time incurred by having someone screw it up. If the system is down, everything stops. Perhaps you can't afford that.
Abuse of system
It might simply be that you do not want anyone using your system to do something for which they are not authorized, like breaking into other
Who are you trying to protect them from?
Competitors, who might gain an advantage by learning your secrets.
Malicious intruders. Note that people with malicious intent might come from inside or outside your organization. Do not think that the enemy is
simply everyone outside of your domain. Too many organizations think `inside/outside' instead of having a proper form of access control. If you
always ensure that systems and data are protected on a need-to-know basis, then there is no reason to discriminate between inside or outside of
an organization.
Old employees with a grudge. against your organization
Next: what will happen if the system is compromised?
Loss of money
Threat of legal action against you
Missed deadlines
Loss of reputation
How much work will you need to put into protecting the system? Who are the people trying to break in?
Sophisticated spies
Tourists, just poking around
Braggers, trying to impress
Finally: what risk is acceptable to you? If you have a secret which is worth 4 Lira, would you be interested in spending 5 Lira to secure it? Where does one
draw the line? How much is security worth?
The social term in the security' equation should never be forgotten. One can spend a hundred thousand dollars on the top of the range firewall to protect data,
but someone could walk into the building and look over an unsuspecting shoulder to obtain it instead, or use a receiver to collect the stray radiation from
your monitors. Are you leaving sensitive printouts lying around? Are you willing to place your entire building in a Faraday cage to avoid remote detection of
the radiation expelled by monitors? In the final instance someone could just point a gun at your head and ask you nicely for your secrets.
Some examples of security policies are here:
Site security handbook, RFC 1244
gopher:// A guide to developing computing security documents
Security through obscurity
There is a commonly held belief that, if one makes it difficult for intruders to find out information, they will not bother to try. This is completely wrong.
Often the reverse is true. If attackers can see that there is nothing worth finding they will leave systems alone. If everything is concealed they will assume that
there must be something interesting worth breaking in for.
Security through obscurity is a naive form of security which at best delays break-ins. Shadow passwords are an example of this. By making the encrypted
password list inaccessible to normal users one makes it harder for them to automate the search for poor passwords, but one does not prevent it! It is still
possible to guess passwords in exactly the same way as before, but it takes much longer. The NT password database is not in a readable format. Some
people have claimed that this makes it more secure than the Unix password file. Since then tools have been written which rewrite the NT password file in
Unix format with visible encrypted passwords. In other words, making it difficult for people to break in does not make it impossible.
Clearly there is no need to give away information to potential intruders. Information should be available to everyone on a need-to-know basis, whether they
be local users or people from outside the organization. But at the same time, obscurity is no real protection. Even the invisible man could get shot.
Time spent securing systems is better than time spent obscuring them. Obscurity might attract more attention than you want.
Security holes
One way that outside users can attack a system is by exploiting security holes in software. Classic examples usually involve setuid-root programs, which
give normal users temporary superuser access to the system. Typical examples are programs like sendmail and finger. These programs are constantly
Monday, July 26, 1999
Principles of system administration
Page: 19
being fixed, but even so, new security holes are found with alarming regularity. Faults in software leave back-doors open to intruders. The only effective way
of eliminating such attacks is to build a so-called firewall around your network, See section Firewalls.
The computer emergency response team (CERT) was established in the wake of the Internet Worm incident to monitor potential security threats. CERT
publish warnings to a mailing list about known security holes. This is also available on the newsgroup . Several other
organizations, often run by staff who work as security consultants, are now involved in computer security monitoring:
System administration and network security organization.
The computer emergency response team.
The computer incident center.
A source of all kinds of the latest security information.
System homogeneity
In a site with a lot of different kinds of platforms, perhaps several Unix variants, NT and Windows 9x, the job of closing security holes is much harder.
Inhomogeneity often provides the determined intruder with more possibilities for bug-finding. You might ask yourself whether you need so many different
kinds of platform. If you do, then perhaps a firewall solution would provide an extra level of protection, giving you a better chance of being able to upgrade
your systems before something serious happens.
Modem pools
Some companies expend considerable effort to secure their network connections, but forget that they have dial-in modems. Modem pools are a prime target
for attackers because they are often easy targets. There are many problems associated with modem pools. Sometimes they are quite unexpected. For
example, if one has network access to Windows systems using the same modem then those systems are automatically on a shared segment and can use one
another's resources, regardless of whether they have any logical relationship. Modems can also succumb to denial of service attacks by repetitive dialling.
Modems should never allow users to gain access to a part of the network which needs to be secure. Modems should never be back-doors into firewalled
Laptop computers are increasingly popular and they are popular targets for thieves. There have been cases of laptop computers being stolen containing
sensitive information, often enough to give crackers access to further systems, or simply to give competitors the information they wanted! One anecdote tells
of a high ranking officer in the US military whose laptop was stolen with sensitive national security information. It was later returned with a note from the
thief who did not want to be labelled a traitor.
If you make backups of important data (private data) then you must take steps to secure the backups also. If an intruder can steal your backups, then he/she
doesn't need to steal the originals.
A firewall is a network configuration which isolates some machines from the rest of the network. A firewall is a gate-keeper which limits access to and from
a network. You might think that this sounds like a strange idea: after all, what is the point of being connected to the internet if you just want to isolate
yourself from it! The point, of course, is not to cut oneself off from the network but to provide controls on what is allowed in and what is allowed out.
A firewall is a concept. It is not one specific thing; there is no single firewall solution. The name `firewall' is a collective description for a variety of methods
which restrict access to a network. They all involve placing restrictions on the way in which network packets are routed. A firewall might be a computer
which is programmed to act like a router, or it might be a dedicated router or a combination of routers and software systems. The idea with a firewall is to
keep important data behind a barrier which has some kind of passport-control and can examine and restrict network packets, allowing only `harmless'
packets to pass.
All traffic from inside to outside or vice versa must pass through the firewall.
Only authorized traffic is allowed to pass.
Potentially risky network services (like mail) are rendered safer using intermediary systems.
The firewall itself should be immune to attack.
A firewall cannot help with the following:
Badly configured hosts or misconfigured networks.
Data based attacks (where the attack involves sending some harmful information, like the code word which makes you take your own life, or an
E-mail which bolts a Trojan horse).
There are two firewall philosophies: block everything unless we make an explicit exception and Pass everything unless we make a specific exception . The
first of these is clearly the most secure or at least the most paranoid of the two.
Here's a few concepts which get bandied around in firewall-speak:
Screening router/Choke
A router which can be programmed to filter or reject packets directed at certain IP ports.
Monday, July 26, 1999
Principles of system administration
Page: 20
Bastion host
A computer, specially modified to be secure.
Dual-homed host
A computer with two net-cards, which can be used to link an isolated network to a larger network.
Application gateway
A filter, usually run on a bastion host, which has the ability to reject or forward packets at a high level (i.e. at the application level).
Screened subnet/DMZ
An isolated subnet, between the internet and the private network. Also called a DMZ (de-militarized zone). A DMZ is the bit between a screening
router and the firewall, bastion host. This is a good place for external WWW services.
The firewall philosophy builds on the idea that it is easier to secure one host (the bastion host) than it is to secure hundreds or thousands of hosts on a local
network. One focuses on a single machine and ensures that it is the only one effectively coupled directly to the network. One forces all network traffic to
stop at the bastion host, so if someone tries to attack the system by sending some kind of IP attack there can be little damage to the rest of the network
because the private network will never see the attack. This is of course a simplification. It is important to realize that installing a firewall does not give
absolute protection and it does not remove the importance of configuring and securing the hosts on the inside of the firewall.
Of course we do not want all traffic to stop, some things like E-mail and maybe HTTP should be able to pass through. To allow this, one uses a so-called
`proxy' service or a `gateway'. A common solution is to give the bastion host two network interfaces. One is connected to the unsafe part of the network and
the other is connected to the safe part. A service is said to be proxied if the bastion host forwards the packets from the unsafe network to the safe one. It only
does this for a packets which meet the requirements of your security policy. For instance, you might decide that the services you require to cross the firewall
are inbound/outbound telnet, inbound/outbound SMTP (mail), DNS, HTTP and FTP but no others.
Proxying requires some special software, often at the level of the kernel where the validity of connections can be established. For instance, packets with
forged addresses can be blocked. Data arriving at ports where there was no registered connection can be discarded. Connections can be discarded if they do
not relate to a known user-account.
A simple firewall configuration is shown below.
--------------- INTERNET ------------------- UNSAFE
External router
(pkt filter)
--------------------------------------------- DMZ
Bastion Host
acts as router
(pkt filter)
---------------------------------------------- SAFE
| | | | | |
Other hosts
In this example we have effectively two routers, a DMZ and a protected network. The first packet filtering router will route packets between the internet and
one of three hosts. FTP is routed directly to a special FTP server. The same applies to HTTP packets. These services are dealt with by separate hosts, so that
(if something should go wrong and the machines are broken into) it is no worse than having to restore these single hosts from backup. None of the servers
in the DMZ have user accounts, so there would be no help to crackers trying to crack password files there, if they managed to break in. The bastion host gets
all packets which are not for the other services. The bastion host forwards okay-looking packets to the internal router which is really just a further packet
filter (a backup in case of failure of the bastion host). The internal router accepts only packets passing between the safe network and the bastion host, all
others are rejected. The bastion host proxies all of the appropriate protocols including FTP and HTTP between the safe network and the DMZ. This is just
an example. In practice you might not have all the hardware you need to separate things as cleanly as shown here.
Although there is a public domain firewall toolkit,
most firewall software is commercial in nature because it needs to live in the kernel and make use of code which is proprietary.
Firewall management is a complex issue. One cannot set up a firewall and then forget about it. Firewalls need constant maintenance and they are susceptible
to bugs just like any other software. It is best to build up a firewall system slowly understanding each step. A good place to start is with packet filtering
routers to eliminate the most offensive or least secure service requests from outside your local network. These include NFS (RPC), IRC, ping, finger etc.
Access control with tcpd
Even without a firewall one can go half way by verifying the authenticity of packets coming into your network. This is a front line against `spoofing' -- i.e.
internet impersonation. The TCP wrappers package is a tool which can be used to this end.
The tcpd program is a `wrapper' for internet services which provides host-based access control. Instead of starting services directly from the
`/etc/inetd.conf' file, one starts the tcp-daemon with instructions to start a given service. The tcpd daemon checks where the request comes from and
only starts it if it comes from a trusted host. A trusted host is one which is listed in `/etc/hosts.allow'. Another file `/etc/hosts.deny' lists services
which are to be denied to non-trusted hosts.
Monday, July 26, 1999
Principles of system administration
Page: 21
finger stream tcp nowait nobody /usr/etc/in.fingerd in.fingerd
finger stream tcp nowait nobody /some/where/tcpd
pidentd authentication server
One of the problems with socket based communication is authentication. How do we determine the identity of the user making the connection? Normally the
only certain way is to require a password to be given, i.e. some kind of shared secret. The RFC documents RFC981 and RFC1413 specify a standard
protocol for identifying the username of a connecting user. In order for this to work, the a server which is being connected to has to contact the host which
is attempting to connect and check who owns the transmitting socket. This is done with the help of the identd daemon, or its implementation pidentd (the
portable identity daemon).
The identity or authentication service (port 113) is alas not installed as standard by most vendors. You need to get this daemon and install it yourself. This is
quite straightforward.
tar zxf pidentd-3.0.4.tar.gz
cd pidentd-3.0.4
configure --with-des=no
make install
The latest versions of the server are multithreaded and are most easily run outside of the inetd service. The daemon is started by
Although it can be started from inetd, you shuld never invoke this daemon with TCP wrappers, since TCP wrappers attempts to contact with daemon. This
will result in infinite loop. You might have to register the service in `/etc/services'
authentication tap ident
The identification service is a public service which you provide for your own and for others' benefit. It allows you and others to authenticate connections
from your hosts.
Honey-pots and sacrificial lambs
A Honey Pot is a host which is made to look attractive to attackers. It is usually placed on a network with the intention of catching an intruder or distracting
them from more important systems. A Sacrificial Lamb host is a host which is not considered to be particularly important to the domain. If it is
compromised by an attacker then that is an acceptable loss and no real harm is done.
Some network administrators believe that the use of such machines contributes to security. For example, WWW servers are often placed on Sacrificial Lamb
machines which are placed outside firewalls. If the machine is compromised then it can simply be reinstalled and the data reloaded from a secure backup.
This practice can seem rather dubious. There is certainly no evidence to support the idea that either Honey Pot machines or Sacrificial Lamb systems actually
improve security.
Trust relationships
A trust relationship lies at the basis of many software systems which grant access to services or resources. Trust relationships are important to grasp because
they can lead to security holes. Whenever one installs a new service which is available to more than one user it is appropriate to ask the questions:
Do I need this service?
Whom or what information do I have to trust in order to use this?
What will happen if someone abuses that trust?
For example, the rlogin feature of Unix has a file called `.rhosts' in which a user can add a list of trusted hosts. That user can log in to the host with the
`.rhosts' file from any one of those trusted hosts without giving a password. The user is clearly willing to trust this list of hosts. But that is not the only
trust relationship here. Unix uses DNS (the Domain Name Service) in order to verify the identity of connecting machines, so the rlogin service implicitly
trusts the DNS service. If someone could corrupt that service, there would be a potential security problem, See section DNS cache poisoning.
Another example is in software distribution, both for Unix and NT. In order to distribute software from a central server to many clients, the clients have to
trust the information being sent to them from the server. They have to give the server permission to install unknown files which might be security hazards.
SNMP control systems accept information from a controller, based only on a fairly weak password (community string). The password has a default value of
`public' which many sites forget to change (a potentially huge security risk). This information can be used to change control functions of key network
components and is even used for performing remote system administration in certain products.
Cfengine places all of its trust in the correctness of its input file, it does not accept input from the network at all. In software distribution it will trust files
from a software server of its own choosing, but arbitrary servers cannot send data to it uninvited.
Monday, July 26, 1999
Principles of system administration
Page: 22
There are many ways to attack a networked computer in order to gain access to it, or simply disable it. Some well-known examples are listed below. The
actual attack mechanisms used by attackers are often intricate and ingenious, but the common theme in all of them is to exploit naive limitations in the way
network services are implemented. Time and again one sees crackers make use of software systems which were written in good faith, by forcing them into
unnatural situations where the software fails through inadequate checking.
Ping attacks
The RFC 791 specifies that internet datagrams shall not exceed 64kB. Some implementations of the protocol can send packets which are larger than this, but
not all implementations can receive them.
ping -s 65510 targethost
Some older network interfaces can be made to crash certain operating systems by sending them a `ping' request like this with a very large packet size. Most
modern operating systems are now immune to this problem (e.g. NT 3.51 is vulnerable, but NT 4 is not). If not, it can be combatted with a packet filtering
router. See .
Denial of service (DoS) attacks
Another type of attack is to overload a system with so many service requests that it grinds to a halt. One example is mail spamming(3), in which an attacker
sends large numbers of repetitive e-mail messages, filling up the server's disk and causing the sendmail daemon to spawn rapidly and slow the system to a
Newer versions of Berkeley sendmail have built in anti-spamming mechanisms to help protect from this problem. Vendors' sendmails are less advanced.
Denial of service attacks are almost impossible to protect against. It is the responsibility of local administrators to prevent their users from initiating such
attacks wherever possible.
Denial of service attacks on NT can be embarrassingly simple
%myhost telnet nt-host 1028
Trying ???.???.???.???...
Connected to nt-host
Escape character is '^]'.
Hello there.
This sends NT4 inetinfo server into a loop which can stop all services. Service Pack 2 fixes this. There are many other examples. Starting full system
auditing can also bring a respectable Pentium system to a virtual standstill.
TCP/IP spoofing
Most network resources are protected on the basis of the host IP addresses of those resources. Access is granted to by a server to a client if the IP address is
contained in an access control list (ACL). Since the operating system kernel itself declares its own identity when packets are sent, it has not been common to
verify whether packets actually do arrive from the hosts which they claim to arrive from. Ordinary users have not traditionally had access to privileges which
allow them to alter network protocols. Today everyone can run a PC with privileged access to the networking hardware.
Normally an IP datagram passing from host A to host B has a destination address `host B' and source address `host A'. IP spoofing is the act of forging IP
datagrams in such a way that they appear to come from a third party host. i.e. an attacker at host A creates a packet with destination address `host B' and
source address `host C'. The reasons for this a varied. Sometimes an attacker wants to appear to be host C in order to gain access to a special resource
which host C has privileged access to. Another reason might be to attack host C , as part of a more elaborate attack. Usually it is not quite this simple
however, since the forgery is quickly detected.
--------------> HOST B
The TCP handshake is such that host A sends a packet to host B and then replies to the source address with a sequence number which has to match the
next number of an agreed sequence. If another packet is not received with an agreed sequence number the connection will be reset and abandoned. Indeed, if
host C received the confirmation reply for a message which it never sent, it would send a reset signal back immediately, saying effectively `I know nothing
about this'. To prevent this from happening it is common to take out host C first by attacking it with some kind of Denial of Service method, or simply
choosing an address which is not used by any host. This prevents it from sending a reset message. The advantage of choosing a real host C is that the
blame for the attack is placed on host C .
SYN flooding
IP spoofing can also be use>
Monday, July 26, 1999
Principles of system administration
Page: 23
Transfer interrupted!
hoosing an address for host C which is not in use so that it cannot reply with a reset, host A can send SYN packets (new connections) on the same and
other ports repeatedly. The RECV queue quickly fills up and cannot be emptied since the connections cannot be completed. Because the queues are filled
the services are effectively cut off.
These attacks could be prevented if routers were configured so as to disallow packets with forged source addresses.
TCP sequence guessing
Random sequences
This attack allows an attacker to make a TCP connection to a host by guessing the initial TCP sequence number used by the other end of the connection.
This is a form of IP spoofing. The attack was first described in the references below. It was made famous by the break in into Tsutomo Shinomrua's
computers which led to the arrest of Kevin Mitnick. This attack is used to impersonate other hosts for trusted access.
1. R.T. Morris, "A weakness in the 4.2 BSD Unix TCP/IP software", Computer Science Technical Report #117,
2. S.M. Bellovin, "Security problems in the TCP/IP protocol suite", Computer Communications Review 19:2, April 1989, pp. 32-48.
4. Jonathan Littman, The Fugitive Game ,
UDP fragmentation (Teardrop)
The Teardrop attack was responsible for the now famous twelve hour attack which `blue-screened' thousands of NT machines all over the world. This attack
uses the idea of datagram fragmentation. Fragmentation is something which happens as a datagram passes through a router from one network to another
network where the transmission rate is lower. Large packets can be split up into smaller packets for more efficient network performance. In the Teardrop
attack, the attacker forges two UDP datagrams which appear to be fragments of a larger packet, but with data offsets which overlap.
When fragmentation occurs it is always the end host which reassembles the packets. In order to allocate memory for the data, the kernel calculates the
difference between the end of the datagram and the offset at which the datagram fragment started. In a normal situation that would look like this:
------------------UDP frag #1
size = 100 - 0
-----------------UDP frag #2
size = 200 - 100
In a Teardrop attack the packets are forged so that they overlap like this:
--------------------------------UDP frag #1
---------------------------------------------| UDP frag #2 |
size = 120 - 0
size = 90
- 120
The assumption that the next fragment would follow on from the previous one leads to a negative number for the size of the fragment. As the kernel tries to
allocate memory for this it calls malloc(size ) where the size is now a negative number. The kernel panics and the system crashes on implementations
which did not properly check the bounds.
ICMP flooding (Smurf)
ICMP flooding is another denial of service attack. The ICMP protocol is the part of TCP/IP which is used to transmit error messages and control
information between hosts. Well known services like ping and echo use ICMP. Normally all hosts respond to ping and echo requests without question,
since they are useful for debugging. In an ICMP flooding attack, the attacker sends a spoofed ICMP packet to the broadcast address of a large network. The
source address of the packet is forged so that it appears to come from the host which the attacker wishes to attack. Every host on the large network receives
the ping/echo request and replies to the same host simultaneously. The host is then flooded with requests. The requests consume all the system resources.
DNS cache poisoning
This attack is an example of the exploitation of a trusted service in order to gain access to a foreign host. Again it uses a common theme, that of forging a
network service request. This time however the idea is to ask a server to cache some information which is incorrect so that future look-ups will result in
incorrect information being given instead of the correct information.
DNS is a hierarchical service which attempts to answer queries about IP names and addresses locally. If a local server does not have the information
requested it asks an authoritative server for that information. Having received the information from the authoritative server it caches it locally to avoid having
to contact the other server again, after all, since the information was required once, it is likely that the same information will be required again soon. The
information is thus placed in the cache for a period of time called the TTL (Time To Live ). After that time has expired it has to be obtained again from the
authoritative server.
In a cache poisoning attack, the aim is to insert incorrect information into the cache of a server. Once it is there it will be there for the TTL period. In order to
arrange this an attacker does the following.
Monday, July 26, 1999
Principles of system administration
Page: 24
1. The attacker launches his/her attack from the authoritative nameserver for his/her network. This gives him/her the chance to send information to
another nameserver which will be trusted.
2. The attacker sends a query for the IP address of the victim host to the victim's default DNS server in order to obtain a DNS query ID. This provides
a point of reference for guessing i.e. forging the next few query IDs from that server.
3. The attacker then sends a query asking for the address of a host which the victim machine trusts, i.e. the host which the attacker would like to
4. The attacker hopes that the victim host will soon need to look up the IP address of the host it trusts; he/she sends a fake `reply' to such a DNS
lookup request, forged with the query ID to look as though it comes from a lookup of the trusted host's address. The answer for the IP address of
the trusted host is altered so that it is the IP address of the attacker's host.
5. Later when the victim host actually sends such a DNS request it finds that it has already received a UDP reply to that request (this is the nature of
UDP) and it ignores the real reply because it arrives later. Now the victim's DNS cache has been poisoned.
6. The attacker now attempts to connect directly to the victim host, posing as the trusted host. The victim host tries to verify the IP address of the host
by looking up the address in its DNS server. This now responds from its cache with the forged address.
7. The attacker's system is accepted.
This kind of attack requires the notion of external login based on trust, e.g. with Unix `.rhosts' files. This doesn't help with NT because NT doesn't have
trusted hosts in the same way. On the other hand, NT is much easier to gain access to through NULL sessions.
Protecting against attacks
Look out for users with weak passwords. This is the easiest way for attacker to enter the system.
Do not give trusted access to other hosts unless absolutely necessary. Make sure there are no NIS wildcards `+' in `/etc/hosts.equiv'. Avoid
using `.rhosts' files altogether.
Disable unused services in `/etc/inetd.conf' which might contain security leaks, like UUCP, TFTP.
Make sure the router filters all unnecessary traffic. Usually there is no reason to permit RPC traffic outside of the local domain for instance.
Make sure that the latest security patches are installed on all systems.
Monitor connections using netstat -a to show all listening connections. Use tcpd logging.
Monitor processes running on your system. How many copies of important processes are running? How many should be running. Often it is
possible to see that one is under attack by looking at what processes are running and who is running them. For instance an attempt at port sniffing
or spamming might be seen with a bunch of processes like this:
inetd is a multiplexer which starts internet services on many ports. Normally it is only root who runs this. The above indicates that a user is trying
to use the well-known account nobody to start services, or to overload the system with requests.
Check filesystems for suspicious looking hidden files. i.e. files with names like `.. .' These are often used to hide dangerous programs or shells
which users can use to gain root privileges. Cfengine performs this task automatically when it examines filesystems.
Make sure that `.' is not in root's path. It is possible to inadvertently execute a Trojan horse program.
Make sure that files like `/var/adm/utmp' are not world writable allowing crackers to cover their tracks.
Password sniffing
Many communication protocols (telnet, ftp etc) were introduced before security was a concern amongst those on the internet. So many of these protocols are
very insecure. Passwords are often sent over the network as plain text. This means that a sophisticated cracker could find out passwords simply by listening
to everything happening on the network and waiting for passwords to go by. If a cracker has privileged access to at least one machine with a network
interface on the same subnet he/she use tcpdump to capture all network traffic. Normal users do not have this privilege for precisely this reason. These days
however, anyone with a laptop, an ethernet card and a GNU/Linux installation could do this. Switched networks are immune to this problem since traffic is
routed directly from host to host.
Programs which dump all network traffic include tcpdump, etherfind, snoop. Here is a sample of the output from Solaris' snoop program showing the
ethernet traffic on segment of cable. Snoop recognizes common high level protocols (SMTP/FTP/ARP etc) and lists them explicitly. Unknown protocol
types (in this case IPX) are simply listed as ETHER. In the right hand column is the information which an intruder would try to use to sniff passwords.
Using device /dev/hme (promiscuous mode) -> nexus
SMTP C port=4552 oJyhnJycoZyhnKCcnGCc -> nexus
SMTP C port=54621 AGoHRPVU9VT3
nexus -> SMTP R port=54621 -> nexus
FTP C port=1093
nexus -> pc111-75 FTP R port=1093 226 Transfer complet
nexus -> SMTP R port=4552 -> nexus
nexus -> SMTP R port=4546 -> nexus
SMTP C port=4546 H2AQcBBwH1AfYBAQH1Af -> nexus
SMTP C port=11424 O3Jw+XF7cMFCCwe\r\nEQ/
nexus ->
SMTP R port=11424 -> nexus
SMTP C port=4552 \niYmJgomChomChoaChoK
nexus ->
nexus -> (broadcast)
nexus ->
? -> *
? -> *
? -> *
SMTP R port=4546
ARP C Who is, takpeh ?
SMTP R port=4552
ETHER Type=0000 (LLC/802.3), size = 86 bytes
ETHER Type=0000 (LLC/802.3), size = 128 bytes
ETHER Type=0000 (LLC/802.3), size = 80 bytes
Monday, July 26, 1999
Principles of system administration
Page: 25
One way to avoid the problem of password sniffing is to use fullt encrypted links such as ssh and SSL (Secure Socket Layer) enabled services which
replace the standard services. Another is to use a system of one-time passwords. One time passwords are designed to eliminate the need for users to send
their passwords over the network at all. Instead of typing an actual password, one types the remote password for a host into a program on a local machine, in
order to generate a sequence of throw-away passwords which can be used in place of the actual remote password. The passwords are used only once so,
even if someone gets to overhear them, it will already be too late: the password will have expired. Also the system is ingeniously designed so that the actual
remote password (which is used to generate the one-time passwords) never gets sent over the network at all. S/KEY is such a system. Here is an example of
how it works:
You want to make a connection from host A to host B.
You have earlier set a password on host B.
You telnet to host B from host A.
Host B prompts you with a code string: 659 ta55095 and asks for your user name. you type your user name and host B asks you for the one-time
5. You now need to find the one-time password by running a local program on host A with the code string as an argument:
key 659 ta55095
passwd: *******
The key program prompts you for your secret password on host B. When you type this it does not go across the network. The key program returns
a clear text, one-time password valid for one session: "EASE FREY WRY NUN ANTE POT"
6. You type "EASE FREY WRY NUN ANTE POT" on host B (sent over the network) and the password is accepted.
7. Next time you follow the same procedure and get a different password.
Port scanning
In order to find back-doors into vulnerable systems many network attackers scan ports on network hosts in order to find out which services are running on
them. Programs for performing such scans can be obtained freely from the network, as can many other intrusion tools, so crackers require little or no
intelligence in order to carry out these simple attacks these days.
In a poorly configured system a cracker might find active services which even the system owner did not realize were running. UUCP and TFTP are typical
examples. These services can often be exploited to install files in illegal places. Known faults in services can be exploited if one knows about the services
which are running.
Primitive port scanning is detectable if one follows network activity closely. Strings of attempted `connect' requests to one port after the other are easily
spotted. Recently however the trend has expanded to include `stealth scanning' in which scans are performed at random over long periods of time to avoid
attracting attention. Port scanning is only dangerous if there are poorly configured hosts on the network.
Social engineering
Network attackers i.e. system crackers are people. We are often so consumed by the world of the network, that we forget that people can just walk into a
building and steal something in the real world . If one can avoid complex technical expertise in order to break in to a system, then why not do it? There is
more than one way to crack a system.
The only secure computer is a computer which is locked into a room, not connected to a network, shielded from all electromagnetic radiation. In social
studies of large companies, it has been demonstrated that--in spite of expensive firewall software and sophisticated anti-cracking technology--all most
crackers had to do to break into the system was to make a phone call to an unwary employee of the company and ask for their username and password.
Some crackers posed as system administrators trying to fix a bug, others simply questioned them as in a marketing survey until they gave away information
which allowed the crackers to guess their passwords. Some crackers will go one step further and visit the building they are trying to break into, going
through the garbage/refuse to look for documents which would give clues about security. Most people do not understand the lengths that people will go to to
break into systems if they really want to.
Another social phenomenon which motivates break-ins is bragging. Crackers who have broken into the system like to tell people that they have been there
and done that. They sometimes try to scare administrators by telling them how much damage they have caused. Here the trick is not to panic and do
something hasty, but to try to verify what the crackers claim. In many cases it is nonsense, empty words.
There is a few things which can be done to counteract social threats.
Never disclose information over the telephone, especially through a voice-mail system. Phone calls can be spoofed.
Examine system logs, check the system regularly, run cfengine to ensure consistency.
Make hard-copies of messages sent with all the headers printed out. Most people don't know how to hide their true identity on the internet.
Make proper backups of the system regularly.
Inform the whole system of attacks so that anyone who knows something can help you.
Do not assume that what crackers tell you is true. Make a judgment as to whether you want to act or ignore the event.
Have a clear security policy. Death threats, serious or not, should probably be reported to the company responsible for sending the message, and
perhaps even the police, but not the person sending the message.
Current attack profiles
The popular operating systems to attack are
Windows 9x
Windows NT
Irix (Silicon graphics)
Monday, July 26, 1999
Principles of system administration
Page: 26
In the Wheelgroup and NetSolve Security Survey, which includes an analysis of a half-million security alarms, it was found that serious attacks to
companies take place at a rate of between 0.5-5.0 per month. Electronic commerce sites (sites which deal with money) are those most at risk. The attack
methods follow standard, well-known bugs
FTP (Bounce)
Intrusion detection
In the last year or two the reality of network intrusion has led to several attempts to build systems which can detect break-ins, either while they are in
progress or afterwards. In order to detect an intrusion in progress, programs like Network Flight Recorder (NFR) and Big Brother (Bro) attempt to
examine every packet on the network in order to look for tell-tale signatures of network break-in activity. This is an extremely resource consuming task and
it is best with a number of problems. One problem is that of fragmentation . Fragmentation is something which occurs to IP datagrams which pass between
networks with different transmission rates. Larger packets can be broken up into smaller packets in order to optimize transmission. These fragments are
reassembled at the final destination. This presents a problem for intrusion detection systems because the fragmented packets might not contain enough data
to identify them as hostile. This would allow them to get past the detection system. An intruder might be able to generate packets which were fragmented in
such as way as to confound the attempts at detection. In spite of the difficulties network intrusion detection is a hot research topic.
Network forensics is what one does after an intrusion. The idea is to examine logs and system audits in order to name the intruder and determine the
damage. Network forensics is perhaps most important for the purpose of possible legal action against intruders. The cost of keeping the necessary logs and
audits is very great and the work required after a break-in is far from trivial.
User administration
@hrule @vskip 0.3cm
User registration
One of the most important issues for a system administrator is to issue new accounts for users. Surprisingly this is an area where vendors provide virtually
no help. The tools provided by operating systems for this task are at best primitive and almost never suitable for use without some modification.
Many systems provide shell scripts or user interfaces for installing new users, but most of these scripts will be useless to you, because they will have
different ideas about how the system should work than you. Also, you will probably want to use some system for centralizing passwords, so that each user
has the same password on every host on your network.
To add a new user to your local system one needs to
Find a unique username, user-id (uid number) and password for the new user.
Update the system database of user accounts, e.g. add a line to the file `/etc/passwd' in Unix (or on the centralized password server of a network)
for the new user.
Create a login directory (home directory) for the user.
Choose a shell for the user (if appropriate).
Copy some configuration files like `.cshrc' or `.profile' into the new user's directory, or update the system registry.
Because every site is different, user registration requires different tools and techniques in almost every case. For example: where should users' home
directories be located? GNU/Linux has an `adduser' script which assumes that the user will be installed on the local machine under `/home/user', but
many users belong to a network and their disk space lies physically on a different host which is mounted by NFS. In NT all users begin in the root directory
by default. It is customary to create a `\users' directory for home directories. If one creates an account on one host, what about the other hundred hosts at
the site? Should users be allowed to log on to all of them? Should they have the same files on every host, or different ones? All of these questions need to be
Perl scripts are excellent ways of making user installation scripts. Interactive programs are almost useless since users are seldom installed one by one. At
universities hundreds of students are registered at the same time. No system administrator would type in all the names by hand. More likely they would be
input from some administrative list generated by the admissions department. The format of that list is not a universal standard, so no off-the-shelf software
package is going to help here.
Local and network accounts
Both Unix and NT support the creation of accounts both locally on a single host, or `globally' within a network domain. With a local account, a user has
permission to use only the local host. With a network account, the user can use any host which belongs to the network `domain'.
Local accounts are configured on the local host itself. Unix reisters local users by added them to the files `/etc/passwd' and `/etc/shadow'. In NT the
Security Accounts Manager (SAM) is used to add local accounts to a given workstation.
For network accounts, Unix uses NIS (Network Information Service, formerly called Yellow Pages or simply YP). NT uses its model of domain servers,
which is like a NIS server only much more. A user in the SAM of a primary domain controller is registered within that domain and has an account on any
host which subscribes to that domain.
An NT domain server involves not only shared databases but also shared administrative policies and shared security models. A host can subscribe to one or
more domains and one domain can be associated with one another by a trust relationship. When one NT domain `trusts' another then accounts and groups
defined in the trusted domain can be used in the trusting domain.
The DCE (Distributed Computing Environment) can also be used to provide a seamless world-wide distributed network domain. The DCE has been ported
to both Unix and NT and requires a special login after normal login to Unix/NT.
Monday, July 26, 1999
Principles of system administration
Page: 27
Login environment
When a new user logs in for the first time, he or she expects the new account to work straight away. Printing should work, programs should work and there
should be no strange error messages about files not being found or programs not existing. Most users want to start up a window environment.
Unix is about the ability to customize. Everything in UNIX is configurable and advanced users like to play around and many create their own setups, but
there will also be a lot of users who do not want to do this. As system administrator, it is your job to make sure that everything works properly with
acceptable defaults, right from the start. Here is a checklist for configuring a user environment
If the default shell for users is a C shell or derivative, then you need to supply a default `read commands' file for this shell. This should set a path
and a terminal type and any environment variables which your local system requires.
If the default shell is a Bourne-again shell like bash or ksh, then you will need to supply this file to set a PATH variable and terminal type and any
environment variables which your system requires.
This file specifies what windows and window manager will be used when the X-windows system is created. It is a shell script which should begin
by setting up applications in the background (with a `&' symbol after them) and end up exec-ing a window manager in the foreground. If the
window manager is called as a background process, the script will be able to exit immediately and users will be logged out immediately. Some older
systems use an outdated file called `.xinitrc' but this file is officially obsolete. The official way to start the X11 window system is through the
xdm program, which provides a login prompt window. For some reason, many distributions of GNU/Linux seem to encourage the use of the
obsolete command startx which starts the X windows system from a tty-shell. The older startx system used the `.xinitrc' file, whereas xdm
uses `.xsession'. Most GNU/Linuxes hack this so that one only needs a `.xsession' file.
This file configures the default menus etc for the mwm window manager.
This file customizes the behaviour of the fvwm window manager.
This file customizes the behaviour of the fvwm2 window manager.
This file customizes the behaviour of the fvwm95 window manager. This is a mock windows-95 interface.
If you are running the TkGoodStuff toolbar as a part of fvwm2 then you might like to create a default toolbar setup for users.
Controlling user resources
Disks fill up at an alarming rate. Users almost never throw away files unless they have to. If one is lucky enough to have only very experienced and
extremely friendly users on the system, then one can try asking them nicely to tidy up their files. Most administrators do not have this luxury however. Most
users never think about the trouble they might cause others by keeping lots of junk around. After all, multiuser systems and network servers are designed to
give every user the impression that they have their own private machine.
To keep hosts working it is necessary to remove files, not just add them. Quotas limit the amount of disk-space users can have access to, but this does not
solve the real problem. The real problem is that in the course of using a computer many files are created as temporary data but are never deleted afterwards.
The solution is to delete them.
Some files are temporary by definition.
For example the byproducts of compilation, `*.o' files, files which can easily be regenerated from source like TeX `*.dvi' files, cache files in
`.netscape/' loaded in by Netscape's browser program, etc.
Some files can be defined as temporary as a matter of policy.
Files which users collect for personal pleasure like `*.mp3', video formats and pornography.
When a Unix program crashes, the kernel dumps its image to disk in a file called `core'. These files crop up all over the place and have no useful purpose.
To most users they are just fluff on the upholstery and should be removed. A lot of free disk space can be claimed by deleting these files. Many users will
not delete them themselves however, because they do not even understand why they are there.
Disk quotas mean that users have a hard limit to the number of bytes they are allowed to use on the disk. They are an example of a more general concept
known as system accounting whereby you can control the resources used by any user, whether they be number of printed pages sent to the printer or the
number of bytes written to the disk. Disk quotas have advantages and disadvantages.
The advantage is that users really cannot exceed their limits. There is no way around this.
Disk quotas are very restrictive and when a user exceeds their limit they do not often understand what has happened. Usually users do not even get a
message unless they are logging in. Quotas also prevent users from creating large temporary files which can be a problem when compiling
programs. They carry with them a system overhead, which makes everything run a little slower.
Deciding whether to delete files automatically is a policy decision. It might be deemed totalitarian to delete files without asking. On the other hand, this is
often the only way to ever clear anything up. Many users will be happy if they do not have to think about the problem themselves, however one has to be
careful never to delete file which cannot be regenerated or re-acquired if necessary.
A useful strategy is to delete files one is not sure about only if they have not be accessed for a certain period of time, say a week. This allows users to use
files freely as long as they need to, but prevents them from keeping the files around for ever. Cfengine can be used to perform this task, See section Using
cfengine. For example:
actionsequence = ( tidy )
Monday, July 26, 1999
Principles of system administration
Page: 28
# 3 days minimum, remember weekends
home/.netscape/cache pattern=*
recurse=inf age=14
recurse=inf age=0
Killing old processes
Cfengine can also be used to clean up old processes. Processes sometimes do not get terminated when they should. There are several reasons for this.
Sometimes users forget to log out, sometimes poorly written terminal software does not properly kill its processes when a user logs out. Sometimes
background programs simply crash or go into loops from which they never return. One way to clean up processes in a work environment is to look for user
processes which have run for more than a day. (Note that the assumption here is that everyone is supposed to log out each day and then log in again the next
day--that is not always the case.) Cfengine's processes commands are used to match processes in the process table (which can be seen by running ps ax).
Here is an example:
actionsequence = ( processes )
This rule works by noticing that, when processes are more than a day old, the date (i.e. the name of the month) appears in the process listing. Thus to find
processes which are more than a day old we only need to search for any line containing the name of a month and then pick out a subset of those which
contain the strings `tcsh', `xterm' etc. This is an extremely useful way of cleaning up old processes which poor terminal software leave behind, or which
forgetful users leave behind when they forget to log out. In the latter case, user security is also protected by this clean up operation.
Network orientation
@hrule @vskip 0.3cm
This chapter assumes that you have a reasonably functional network already. It is about how to understand how a network works. We begin from the
perspective of someone who has just moved to a new job where the network is unfamiliar.
The first step in familiarizing oneself with a computer network is to find out about the local domain, all the hosts connected to it and who is responsible for
what. Then one needs an overview of where the cables run and what operating systems are running.
Knowing the domain
The familiarization process begins by analysing the network and all of its inter-relationships. It is especially important to know who is responsible for
maintaining different parts of the network. It might be you or it might be someone else. Who runs the DNS domain above yours? Who do you call if the
internet connection is broken? What service contracts might exist on hardware, what upgrade possibilities are there on software? What system is in use for
making backups? How does one obtain a backup should the need arise? In short, it is essential to know where to begin in solving any problem which might
arises, and who to call if the responsibility for a problem lies with someone else.
After these organizational points, one needs an overview of the hosts. A host list can probably be obtained from the DNS using nslookup. If there are Unix
systems, one can start by logging onto each machine and using the uname command to find out what OS is being used:
nexus% uname -a
SunOS nexus 5.5 Generic sun4m
Monday, July 26, 1999
Principles of system administration
Page: 29
borg% uname -a
Linux borg 1.3.62 #2 Mon Feb 12 11:06:19 MET 1996 i586
This tells us that host nexus is a SunOS kernel version 5.5 (colloquially known as Solaris 2.5) system with a sun4m series processor, and that host borg is
a GNU/Linux system kernel version 1.3.62.
If the uname command doesn't exist, you know that your operating system is an old dinosaur from BSD 4.3 days and you will have to find out what it is by
different means. Try the following commands: arch and mach.
How much memory does a host have? (Most systems print this when they boot. Sometimes the information can be coaxed out of the system in
other ways.) What disks and other devices are in use?
Use locate and find and which and whereis to find important directories and software. How is the software layed out?
What binary directories exist? `/usr', `/usr/bin'?
Where are the important files? For example, is your filesystem table called `/etc/fstab', `/etc/vfstab', `/etc/checklist' etc.. or something
Do Unix systems have a C compiler and a C++ compiler installed? This is often needed for installing software.
Using nslookup
nslookup is a program for querying the Domain Name Service (DNS). The name service provides a mapping or relationship between internet numbers and
internet names, and contains useful information about domains: both your own and others. The first thing you need to know is your domain name. This is
the suffix part of the internet names for your network. For instance, our domain at Oslo College, Faculty of Engineering has the domainname
`'. Hosts in this domain have names like `'.
If you don't know your DNS domain name, you can probably find out by looking at the file `/etc/resolv.conf'. For instance:
borg% more /etc/resolv.conf
Most UNIX systems have a command called domainname. This prints the name of the local Network Information Service (NIS) domain which is not the
same thing as the DNS domainname (though, in practice, most people would use the same name for both). Do not confuse the output of this command with
the DNS domain name.
Once you know your domainname, you can find out the hosts which are registered in your domain by running the program nslookup, or the name service
lookup program. Type the command and you should see something like this:
borg% nslookup
Default Server:
nslookup always prints the name and the address of the server from which it obtains its information. Then you get a new prompt `>' at which you can type
commands. Typing help provides you with a list of commands which nslookup supports.
hostname/IP lookup
Type the name of a host or internet (IP) address and nslookup returns the equivalent translation. Example:
dax% nslookup
Default Server:
In this example we look up the internet address of the host called and the name of the host which has internet address In
both cases the default server is the name server which has internet address
Note that the default server is the first server listed in the file `/etc/resolv.conf' which answers for queries when you start nslookup.
Special information
Monday, July 26, 1999
Principles of system administration
Page: 30
The Domain name service identified certain special hosts which perform services like the name service itself and mail-handlers (called mail exchangers).
These servers are identified by special records so that people outside of a given domain can find out about them. After all, the mail service in one domain
needs to know how to send mail to a neighbouring domain. It also needs to know how to find out the names and addresses of hosts for which it does not
keep information personally.
We can use nslookup to extract this information by setting the `query type' of a request. For instance, to find out about the mail exchangers in a domain we
> set q=mx
> domain name
For example
> set q=mx
Non-authoritative answer:
preference = 0, mail exchanger =
Authoritative answers can be found from:
nameserver =
nameserver =
internet address = internet address =
Here we see that the only mail server for is
Another example, is to obtain information about the nameservers in a domain. This will allow us to find out information about hosts which is not contained
in our local database See section Listing hosts belonging to a domain. To get this, we set the query-type to `ns'.
> set q=ns
Non-authoritative answer:
nameserver =
nameserver =
Authoritative answers can be found from: internet address =
internet address =
Here we see that there are two authoritative nameservers for this domain called and
Finally, if we set the query type to `any', we get a summary of all this information.
Listing hosts belonging to a domain
To list every registered internet address and hostname for a given domain you use the ls command inside nslookup. For instance
> ls
server =
server =
server =
First the nameservers are listed and then the host names and corresponding IP addresses are listed.
If you try to look up hosts in a domain for which your default name server has no information, you will get an error message. For example, suppose we try
to list the names of the hosts in the domain over ours:
> ls
*** Can't list domain Query refused
This does not mean that you cannot find out information about other domains, only that you cannot find out information about other domains from your
default server See section Changing to a different server.
Changing to a different server
Monday, July 26, 1999
Principles of system administration
Page: 31
If you know the name of a server which contains authoritative information for a domain, you can tell nslookup to use that server instead. That way you can
list the hosts in a remote domain and find out detailed information about that domain, See section Listing hosts belonging to a domain. You always get
more information from an authoritative server than you do from a remote server. To change the server you simply type
> server new-server
> lserver new-server
There is a subtle difference between these two commands. If you use the first command to change the server to another host which is not running a named
daemon (the DNS daemon), you will find yourself in a situation where you can no longer lookup hostnames or IP addresses because the host you have
specified is not a server. In this case, you can use the second form which always uses the default (the first) server to look up the names you use. This is only
relevant if you use internet names on the command line and not IP addresses (which are always recognized).
We can use this now to list all of the data for a remote domain. First we change server; then once this is done we use ls to list the names.
> server
Default Server:
> ls
(listing ..)
Another advantage with using the server which is directly responsible for the DNS data, is that we obtain extra information about the domain, namely a
contact address for the person responsible for administrating the domain. For example:
> server
Default Server:
nameserver =
preference = 0, mail exchanger =
nameserver =
origin =
mail addr =
serial = 1996120503
refresh = 3600 (1 hour)
= 900 (15 mins)
expire = 604800 (7 days)
minimum ttl = 86400 (1 day)
nameserver =
nameserver = internet address =
internet address =
This is probably more information than you are interested in, but it does tell you that you can address queries and problems concerning this domain to
[email protected] (Note that DNS does not use the @ symbol for `at' in these data.)
Contacting other domains
Sometimes you will need to contact other domains, perhaps because you believe there is a problem with their system, or perhaps because an unpleasant user
from another domain is being a nuisance and you want to ask the administrators there to put that person to a long and painful death. You now know how to
obtain one contact address using nslookup. Another good bet is to mail the one address which every domain must have: [email protected] . Any domain
which does not define this mail address deserves to have its wires cut(4).
Various unofficial standards also encourage sites to have the following mail addresses which you might try:
Apart from these sources, there is little one can do to determine who is responsible for a domain. A limited number of domains in the USA register with
another network database service called the whois service. In some cases it is possible to obtain information this way. For example:
Financial Connections, Inc (MONEYWORLD-DOM)
2508 5th Ave, #104
Seattle, WA 98121
Monday, July 26, 1999
Principles of system administration
Page: 32
Administrative Contact, Technical Contact, Zone Contact:
Williams, Bob (BW747) [email protected]
206 269 0846
Record last updated on 13-Oct-96.
Record created on 26-Oct-95.
Domain servers in listed order:
The InterNIC Registration Services Host contains ONLY Internet Info
(Networks, ASN's, Domains, and POC's).
Please use the whois server at for MILNET Information.
NVRAM settings
Some hosts come with a basic `monitor' panel which is a ROM based program which can be used to set the configuration of Non Volatile RAM variables
even before the system has booted up a true operating system. You should probably familiarize yourself with the kinds of variables which can be set in
NVRAM for your system, if any. These could control basic choices about how your machine works, like which network interface is to be used: thick
ethernet or twisted pair, etc. On a PC system running GNU/Linux, this corresponds to the BIOS settings.
On solaris hosts there is a program called eeprom which can be used to set values in NVRAM.
Hardware awareness
To be a system administrator one does not usually need to know very much about hardware, but it is very useful to have a basic appreciation of how to install
hardware and how to treat it. If you do not feel comfortable handling hardware, then don't do it!
Read instructions
When dealing with hardware, always look for and read instructions in a manual. Do not assume that you know what you are doing. Instructions
are there for a reason.
Interfaces and connectors
Hardware is often connected to an interface or connector. Make sure that you have the correct kind of cable. Modem cables in particular can damage
your computer or modem if they are incorrectly wired. Some computers supply power to certain pins which can damage equipment which does not
expect to find a power supply coming across the cable. Network interfaces are often built in. Some interfaces are for thin ethernet, some for thick
ethernet and others are for twisted pair connectors. If your computer has the wrong type of interface, you will need to buy a transceiver which
converts the signal and connection to the right type.
Handling components
Modern day CMOS chips work at low voltages (typically 5 volts). Standing on the floor with insulating shoes, you can pick up a static electric
charge of several thousand volts. Such a charge can instantly destroy computer chips. Before touching any computer components, earth yourself by
touching the metal casing of the computer. If you are installing equipment inside a computer, wear a conductive wrist strap.
Disks The most common disk types are IDE (integrated drive electronics) and SCSI (small computer software interface). IDE disks are usually cheaper
than SCSI disks, but SCSI disks are more efficient at handling multiple accesses, and are therefore better in multitasking systems. SCSI comes in
several varieties, SCSI 1, SCSI 2, wide SCSI, fast-wide etc etc. The difference has to do with the width of the data-bus and the number of disks
which can be attached to each controller. Each disk has its own address (or number) which must be set by changing a setting on the disk-cabinet or
by changing jumper settings inside the cabinet. Disk chains must be terminated with a proper terminating connector.
Memory chips are sold on small circuit boards called SIMMs. These SIMMs are sold in different sizes and with different speeds. Your computer
has a number of slots where SIMMs can be installed. When buying and installing RAM, remember @itemize @bullet @item The physical size of
SIMMs is important. Most have 72 pins and some older SIMMs have 30 pins. @item SIMMs are sold in 1MB, 4MB, 16MB, 64MB sizes etc.
Find out what size you can use in your system. In most cases you are not allowed to mix different sizes. @item Do not buy slower RAM than that
which is recommended for your computer, or it will not work. @item On some computers you need fill up RAM slots in a particular order,
otherwise the system will not be able to find them. @end itemize
Physical network issues
There are two common styles of network:
Thin ethernet
In the thin ethernet scheme, hosts are chained together by stretches of cable. Each host taps into the main cable at suitable locations. The ends of the
cable must be terminated by a 50-ohm resistance. Any break in the cable disables the cable, potentially for all machines. Finding cable breaks
requires special tools. Most signal losses are caused through ignorance by users who break the cable integrity. Thin ethernet cable lengths are
limited to 30m per segment (between machines) but signal amplifiers (called repeaters) or signal splitters (called multi-port repeaters) can be used to
increase this length. A maximum of three repeaters can be used. A total maximum length of 300m cannot be exceeded. To make a longer network,
one needs to join to separate networks together by a router or repeater.
Star net/twisted pair (xBaseT)
This scheme is far less susceptible to cable problems. Each twisted pair (ISDN) telephone style socket is connected by an independent cable to a
central HUB. A twisted pair network uses a cable directly from computer to HUB, so the cable is automatically terminated. A connection board is
used to wire each computer into a HUB which acts as an exchange, routing packets internally between the legs of the star.
Other network devices include bridges which are small-scale routers which separate two segments of a network. Bridges do no forward packets which do
not need to cross the bridge, so they may be used to isolate busy segments of a network, thereby reducing traffic in another segment.
See the pictures at
Monday, July 26, 1999
Principles of system administration
Page: 33
Using cfengine
@hrule @vskip 0.3cm
Many of the routine maintenance jobs you are required to do to keep your system running smoothly can be automated by a configuration robot called
cfengine. Cfengine, or the configuration-engine, can set up your system from scratch and can watch over the system as time goes on, reporting about and
even fixing errors.
What is cfengine?
System maintenance involves a lot of jobs which are repetitive and menial. There are half a dozen languages and tools for writing programs which will
automatically check the state of your system and perform a limited amount of routine maintenance automatically. Cfengine is one such tool which has
acquired wide acceptance on the net. It is a very high level language (much higher level than shell or Perl) and a robot for interpreting your programs and
implementing them. Cfengine is a general tool for structuring, organizing and maintaining information system on a network. Because it is general, it does not
try to solve every little problem you might come across, instead it provides you with a framework for solving all problems in a consistent and organized way.
Cfengine's strength is that it encourages organization and consistency of practice--also that it may easily be combined with other languages.
Cfengine is about (i) defining the way you want all hosts on your network to be set up (configured), (ii) writing this in a single `program' which is read by
every host on the network, (iii) running this program on every host in order to check and possibly fix the setup of the host. Cfengine programs make it easy
to specify general rules for large groups of host and special rules for exceptional hosts. Here is a summary of cfengine's capabilities.
Check and configure the network interface on network hosts.
Edit textfiles for the system or for all users.
Make and maintain symbolic links, including multiple links from a single command.
Check and set the permissions and ownership of files.
Tidy (delete) junk files which clutter the system.
Systematic, automated (static) mounting of NFS filesystems.
Checking for the presence or absence of important files and filesystems.
Controlled execution of user scripts and shell commands.
Process management.
By automating these procedures, you will save a lot of time and irritation, and make yourself available to do more interesting work.
A cfengine program is probably not like other programming languages you are used to. It is more like a Makefile. Instead of using low-level logic, it uses
high-level classes to make decisions. Actions to be carried out are not written in the order in which they are to be carried out, but listed in bulk. The order in
which commands are executed is specified in a special list called the action-sequence . A cfengine program is a free-format text file, usually called
`cfengine.conf' and consisting of declarations of the form.
list of actions
The action type tells cfengine what the commands which follow do. The action type can be from the following list.
You may run cfengine scripts/programs as often as you like. Each time you run a script, the engine determines whether anything needs to be done -- if
nothing needs to be done, nothing is done! If you use it to monitor and configure your entire network from a central file-base, then the natural thing is to run
cfengine daily with the help of cron.
Cfengine configurations can save you an enormous amount of time by freeing you from repetitive tasks. Finally you run your system chauffeur driven with
your own programmable dog. Totally excellent.
Monday, July 26, 1999
Principles of system administration
Page: 34
The simplest way to use cfengine
The simplest cfengine configuration you can have consists of a control section and a shellcommands section, in which you collect together scripts and
programs which should run on different hosts or host-types. Cfengine allows you to collect them all together in one file and label them in such a way that the
right programs will be run on the right machines.
domain = ( mydomain
actionsequence = ( shellcommands )
# All GNU/Linux machines
# Just one host
"/bin/echo Hi there"
While this script does not make use of cfengine's special features, it shows you how you can control many machines from a single file. Cfengine reads the
same file on every host and picks out only the commands which apply.
A simple file for one host
Although cfengine is designed to organize all hosts on a network, you can also use it just on a single stand-alone host. In this case you don't need to know
about classifying commands.
Let's write a simple file for checking the setup of your system. Here are some key points:
Every cfengine must have a control: section with an actionsequence list, which tells it what to do, and in which order.
You need to declare basic information about the way your system is set up. Try to keep this simple.
#!/usr/local/gnu/bin/cfengine -f
# Simple cfengine configuration file
actionsequence = ( checktimezone netconfig resolve files shellcommands )
= ( )
= ( )
= ( MET )
# Add these name servers to the /etc/resolv.conf file
# nexus
/etc/passwd mode=644 owner=root action=fixall
Monday, July 26, 1999
Principles of system administration
Page: 35
A file for multiple hosts
If you want to have just a single file which describes all the hosts on your network, then you need to tell cfengine which commands are intended for which
hosts. Having to mention every host explicitly would be a tedious business. Usually though, we are trying to make hosts on a network basically the same as
one another so we can make generic rules which cover many hosts at a time. Nonetheless there will still be a few obvious differences which need to be
accounted for.
For example, the Solaris operating system is quite different from the GNU/Linux operating system, so some rules will apply to all hosts which run Solaris,
whereas others will only apply to GNU/Linux. Cfengine uses classes like solaris:: and linux:: to label commands which apply only to these systems.
We might also want to make other differences, based not on operating system differences but on groups of hosts belonging to certain people, or with a
special significance. We can therefore create classes using groups of hosts.
The idea of classes is central to the operation of cfengine. Saying that cfengine is `class oriented' means that it doesn't make decisions using if...then...else
constructions the way other languages do, but only carries out an action if the host running the program is in the same class as the action itself. To
understand what this means, imagine sorting through a list of all the hosts at your site. Imagine also that you are looking for the class of hosts which
belong to the computing department, which run GNU/Linux operating system and which have yellow spots! To figure out whether a particular host satisfies
all of these criteria you first delete all of the hosts which are not GNU/Linux, then you delete all of the remaining ones which don't belong to the computing
department, then you delete all the remaining ones which don't have yellow spots. If you are on the remaining list, then you are in the class of all
computer-science-Linux-yellow-spotted hosts and you can carry out the action.
Cfengine works in this way, narrowing things down by asking if a host is in several classes at the same time. Although some information (like the kind of
operating system you are running) can be obtained directly, clearly, to make this work we need to have lists of which hosts belong to the computer
department and which ones have yellow spots.
So how does this work in a cfengine program? A program or configuration script consists of a set of declarations for what we refer to as actions which are
to be carried out only for certain classes of host. Any host can execute a particular program, but only certain action are extracted -- namely those which refer
to that particular host. This happens automatically because cfengine builds up a list of the classes to which it belongs as it goes along, so it avoids having to
make many decisions over and over again.
By defining classes which classify the hosts on your network in some easy to understand way, you can make a single action apply to many hosts in one go
-- i.e. just the hosts you need. You can make generic rules for specific type of operating system, you can group together clusters of workstations according
to who will be using them and you can paint yellow spots on them -- what ever works for you.
A cfengine action looks like this:
action-type :
compound-class ::
A single class can be one of several things:
The name of an operating system architecture e.g. ultrix, sun4 etc. This is referred to henceforth as a hard class .
The (unqualified) name of a particular host. If your system returns a fully qualified domain name for your host, cfengine truncates it so as to
unqualify the name.
The name of a user-defined group of hosts.
A day of the week (in the form Monday Tuesday Wednesday..).
An hour of the day (in the form Hr00, Hr01 ... Hr23).
Minutes in the hour (in the form Min00, Min17 ... Min45).
A five minute interval in the hour (in the form Min00_05, Min05_10 ... Min55_00)
A day of the month (in the form Day1 ... Day31).
A month (in the form January, February, ... December).
A year (in the form Yr1997, Yr2001).
An arbitrary user-defined string.
A compound class is a sequence of simple classes connected by dots or `pipe' symbols (vertical bars). For example:
A compound class evaluates to `true' if all of the individual classes are separately true, thus in the above example the actions which follow
compound_class:: are only carried out if the host concerned is in myclass, is of type sun4 and the day is Monday! In the second example, the host
parsing the file must be either of type sun4 or ultrix or osf. In other words, compound classes support two operators: AND and OR, written `.' and
`|' respectively. Cfengine doesn't care how many of these operators you use (since it skips over blank class names), so you could write either
Monday, July 26, 1999
Principles of system administration
Page: 36
depending on your taste. On the other hand, the order in which cfengine evaluates AND and OR operations does matter, and the rule is that AND takes
priority over OR, so that `.' binds classes together tightly and all AND operations are evaluated before ORing the final results together. This is the usual
behaviour in programming languages. You can use round parentheses in cfengine classes to override these preferences.
Cfengine allows you to define switch on and off dummy classes so that you can use them to select certain subsets of action. In particular, note that by
defining your own classes, using them to make compound rules of this type, and then switching them on and off, you can also switch on and off the
corresponding actions in a controlled way. The command line options -D and -N can be used for this purpose.
A logical NOT operator has been added to allow you to exclude certain specific hosts in a more flexible way. The logical NOT operator is (as in C and
C++) `!'. For instance, the following example would allow all hosts except for myhost:
action :
and similarly, so allow all hosts in a user-defined group mygroup, except for specialhost, you would write
action :
which reads `mygroup AND NOT myhost'. The NOT operator can also be combined with OR. For instance
class1 |!class2
would select hosts which were either in class 1, or those which were not in class 2.
Finally, there is a number of reserved classes. The following are hard classes for various operating system architectures. They do not need to be defined
because each host knows what operating system it is running. Thus the appropriate one of these will always be defined on each host. Similarly the day of the
week is clearly not open to definition, unless you are running cfengine from outer space. The reserved classes are:
ultrix, sun4, sun3, hpux, hpux10, aix, solaris, osf, irix4, irix, irix64
freebsd, netbsd, openbsd, bsd4_3, newsos, solarisx86, aos,
nextstep, bsdos, linux, debian, cray, unix_sv, GnU
If these classes are not sufficient to distinguish the hosts on your network, cfengine provides more specific classes which contain the name and release of the
operating system. To find out what these look like for your systems you can run cfengine in `parse-only-verbose' mode:
cfengine -p -v
and these will be displayed. For example, solaris 2.4 systems generate the additional classes sunos_5_4 and sunos_sun4m, sunos_sun4m_5_4.
Cfengine uses both the unqualified and fully host names as classes. Some sites and operating systems use fully qualified names for their hosts. i.e. uname
-n returns to full domain qualified hostname. This spoils the class matching algorithms for cfengine, so cfengine automatically truncates names which
contain a dot `.' at the first `.' it encounters. If your hostnames contain dots (which do not refer to a domain name, then cfengine will be confused. The moral
is: don't have dots in your host names! NOTE: in order to ensure that the fully qualified name of the host becomes a class you must define the domain
variable. The dots in this string will be replaced by underscores.
In summary, the operator ordering in cfengine classes is as follows:
Parentheses override everything.
The NOT operator binds tightest.
The AND operator binds more tightly than OR.
OR is the weakest operator.
We may now label actions by these classes to restrict their scope:
{ /etc/motd
PrependIfNoSuchLine "Plan 9 was a better movie and a better OS!"
Monday, July 26, 1999
Principles of system administration
Page: 37
{ /etc/motd
AppendIfNoSuchLine "Your rpc.spray is so last month"
{ /etc/motd
AppendIfNoSuchLine "Please keep your finger commands to yourself!"
Actions or commands which work under a class operator like solaris:: are only executed on hosts which belong to the given class. This is the way one
makes decisions in cfengine: by class assignment rather than by if..then..else clauses.
Using cfengine as a front end for cron
One of cfengine's strengths is its use of classes to identify systems from a single file or set of files. Many administrators think that it would be nice if the
cron daemon also worked in this way. One possible way of setting up cron from a global configuration would be to use the cfengine editfiles facility to
edit each cron file separately. A much better way is to use cfengine's time classes to work like a user interface for cron. This allows you to have a single,
central cfengine file which contains all the cron jobs on your system without losing any of the fine control which cron affords you. All of the usual
advantages apply:
It is easier to keep track of what cron jobs are running on the system when you have everything in one place.
You can use all of your carefully crafted groups and user-defined classes to identify which host should run which programs.
The central idea behind this scheme is to set up a regular cron job on every system which executes cfengine at frequent intervals. Each time cfengine is
started, it evaluates time classes and executes the shell commands defined in its configuration file. In this way we use cfengine as a wrapper for the cron
scripts, so that we can use cfengine's classes to control jobs for multiple hosts. Cfengine's time classes are at least as powerful as cron's time specification
possibilities, so this does not restrict you in any way. The only price is the overhead of parsing the cfengine configuration file.
To be more concrete, imagine installing the following `crontab' file onto every host on your network:
# Global Cron file
0,15,30,45 * * * * /usr/local/cfengine/inputs/run-cfengine
This file contains just a single cron job, namely a script which calls cfengine. Here we are assuming that you will not want to execute any cron script more
often than every fifteen minutes. If this is too restrictive, the above can be changed. We refer to the time interval between runs of the script `run-cfengine'
as the `scheduling interval' and discuss its implications in more detail below.
The script `run-cfengine' would replace any `cfdaily' or `cfhourly' scripts which you might have, and can as simple as this
# Script run-cfengine
export CFINPUTS=/usr/local/cfengine/inputs
# Should we pipe mail to a special user?
or it could be more fancy. You could also use the `cfwrap' script, if you have perl on all your systems, to pipe mail to the mail address described in the
cfengine file.
# Global Cron file
0,15,30,45 * * * * path /cfwrap path /run-cfengine
You might not want to run your entire system configuration `cfengine.conf' every time cron fires up cfengine. An alternative would be to keep a separate
file for cron jobs called, say, `cf.cron'. You would then replace the `run-cfengine' file by
# Script run-cfengine
export CFINPUTS=/usr/local/cfengine/inputs
/usr/local/gnu/bin/cfengine -f cf.cron
# Should we pipe mail to a special user?
Monday, July 26, 1999
Principles of system administration
Page: 38
There is no particular advantage to doing this unless you are running cfengine on some very slow hardware. A better way to approach the problem is to think
of the `cf.cron' file as a module which can be imported into the main configuration file. This gives you the maximum amount of flexibility, since it allows
you to decide exactly what you want to happen any any given time from the central file.
Building flexible time classes
Each time cfengine is run, it reads the system clock and defines the following classes based on the time and date:
Yrxx ::
The current year, e.g. `Yr1997', `Yr2001'. This class is probably not useful very often, but it might help you to turn on the new-year lights, or
shine up your systems for the new millennium!
Month ::
The current month can be used for defining very long term variations in the system configuration, e.g. `January', `February'. These classes
could be used to determine when students have their summer vacation, for instance, in order to perform extra tidying, or to specially maintain some
administrative policy for the duration of a conference.
Day :: The day of the week may be used as a class, e.g. `Monday', `Sunday'.
Dayxx ::
A day in the month (date) may be used to single out by date, e.g. the first day of each month defines `Day1', the 21st `Day21' etc.
Hrxx ::
An hour of the day, in 24-hour clock notation: `Hr00'...`Hr23'.
Minxx ::
The precise minute a which cfengine was started: `Min0' ... `Min59'. This is probably not useful alone, but these values may be combined to define
arbitrary intervals of time.
Minxx _xx ::
The five-minute interval in the hour at which cfengine was executed, in the form `Min0_5', `Min5_10' .. `Min55_0'.
Time classes based on the precise minute at which cfengine started are unlikely to be useful, since it is improbable that you will want to ask cron to run
cfengine every single minute of every day: there would be no time for anything to complete before it was started again. Moreover, many things could
conspire to delay the precise time at which cfengine were started. The real purpose in being able to detect the precise start time is to define composite classes
which refer to arbitrary intervals of time. To do this, we use the group or classes action to create an alias for a group of time values. Here are some creative
# synonym groups:
LunchAndTeaBreaks = ( Hr12 Hr10 Hr15 )
= ( Hr22 Hr23 Hr00 Hr01 Hr02 Hr03 Hr04 Hr05 Hr06 )
= ( Day26 Day27 Day29 Day30 )
= ( Min00 Min15 Min30 Min45 )
= ( Min01 Min02 Min03 Min33 Min34 Min35)
In these examples, the left hand sides of the assignments are effectively the OR-ed result of the right hand side. This if any classes in the parentheses are
defined, the left hand side class will become defined. This provides an excellent and readable way of pinpointing intervals of time within a program, without
having to use `|' and `.' operators everywhere.
Choosing a scheduling interval
How often should you call your global cron script? There are several things to think about:
How much fine control do you need? Running cron jobs once each hour is usually enough for most tasks, but you might need to exercise finer
control for a few special tasks.
Are you going to run the entire cfengine configuration file or a special light-weight file?
System latency. How long will it take to load, parse and run the cfengine script?
Cfengine has an intelligent locking and timeout policy which should be sufficient to handle hanging shell commands from previous crons so that no overlap
can take place.
General guidelines for using cfengine
Setting up a network configuration takes a long time. Even with a tool like cfengine, you will be changing things all the time. The main strength of cfengine
is that it allows you to make changes without losing sight of what you have done. Here are some tips to bear in mind while putting together your
Get hold of the cfengine manual and read it. See
Start writing your configuration from the earliest stages of setting up your network. A configuration file takes a long time to develop and it is best to
do a little at a time.
Add to your file a little at a time so that you always understand and remember what is in your configuration.
Begin with network services and configuration, then add file checking and symbolic links. Use tidy and disable only when you know that
cfengine is working the way it should.
Avoid mounting filesystems from host A onto host B and filesystems from host B onto host A. This can lead to NFS deadlocks.
If you are thinking of linking files like the `/etc/passwd' file to another disk where you keep all of your site data -- don't! When a host boots, it
normally has access only to the root file system (no other disks), so if you link a critical file to disk which isn't available booting will fail and you
might have to reinstall the system! Do not expect all disks to be mounted during the whole boot process.
We do not have the space in this introduction to go into cfengine in great detail. You should look at the manual yourself if you find yourself in charge of a
Monday, July 26, 1999
Principles of system administration
Page: 39
Key systems and services
@hrule @vskip 0.3cm
Summoning daemons
Network services are run by daemons. Once you have done the deed of configuring a network service, See section Installing a new service, by editing some
textfiles and chanting some dark rites and maybe sacrificing a few doughnuts, you reach the point where you have to actually start the daemon in order to see
the fruits of those labours. There are two ways to start network daemons:
When the system boots, by adding an appropriate shellcommand to one of the system's startup scripts. When you use this method, the daemon
hangs around in the background all the time waiting for connections.
On demand: that is, only when a network request arrives. You use the inetd daemon to monitor requests for your new service and it starts the
daemon to handle requests on a one-off basis. Not all services should be started in this way. You should normally follow the guidelines in the
documentation for the service that you are running
The behaviour of unix systems at boot-time is very far from being standard. Older unix systems use a series of scripts called `/etc/rc*' (short for `read
commands'). On such system one normally finds a file called `/etc/rc.local' where it is possible to add your own commands. Newer unix varieties use
a program called init and a series of run-levels to control what happens when the machine boots, See section Booting unix.
System 5 init
The SVR4 version of the init program is attaining some popularity and is used by several GNU/Linux distributions, such as Debian and Redhat. The idea
with this program is to start the system in one of a number of run-levels. Runs levels decide how many services will be started when the system boots. The
minimum level of operation is single user mode, or run level `s'. Full operation is usually run level 2 or 3, depending on the system you are using. (NB: be
sure to check this!) When entering a run level, init looks in a directory called `/etc/rc?.d' and executes scripts in this directory. For instance, if we are
entering run-level 2, init would look in the directory `/etc/rc2.d' and execute scripts lying there in order to start necessary services for this run-level. All
you have to do to add a new service is to make a new file here which conforms to init's simple rules. The files in these directories are usually labelled
according to the following pattern:
Snumber -function
Knumber -function
Files beginning with S are for starting services and files beginning with K are for killing them again when the system is halted. The number is used to
determine the order in which the scripts are read. It does not matter if two scripts have the same number. Finally the function tells us what the script does.
Each script is supposed to accept a single argument, the word `start' or the word `stop'. Let's consider an example of how we might start the httpd daemon
using init. Here is a checklist:
1. Determine the correct run-level for the service. Let us suppose that it is run level 2.
2. Choose an unused filename, say `S99http'.
3. Create a script accepting a single argument:
case $1 in
start) /usr/local/bin/httpd -d /usr/local/lib/httpd ;;
kill `cat /usr/local/lib/httpd/logs/` ;;
*) echo Syntax error starting http
The advantage of this system is that software packages can be added and removed transparently just by adding or removing a file. No special editing is
required as is the case for BSD unix.
BSD init
The BSD style init program is rather simple. It starts executing a shell script called `/etc/rc' which then generally calls other child-scripts. These scripts
start important daemons and configure the system.
To add your own local modifications, you have to edit the file `/etc/rc.local'. This is a Bourne shell script.
The BSD approach has a simpler structure than the system 5 inittab directories, but it is harder to manipulate package-wise.
inetd configuration
The internet daemon is a service demultiplexer . In English, that means that it is a daemon which listens on the network for messages to several services
simultaneously. When it receives a message intended for a specific port, it starts the relevant daemon to handle the request just long enough to handle one
request. inetd saves the system some resources by starting daemons only when they are required, rather than having the clutter up the process table all the
The format this file can differ slightly on older systems. The best way to gleaning its format is to look at the entries which are already there. Here is a
Monday, July 26, 1999
Principles of system administration
Page: 40
common example for the format.
# Service|type|protocol|wait|user|daemon-file|command line
# NB wu-ftpd needs -a now
stream tcp nowait root
in.ftpd -a
telnet stream tcp nowait root
finger stream tcp nowait nobody /local/etc/in.fingerd in.fingerd
cfinger stream tcp nowait nobody /local/etc/in.cfingerd in.cfingerd
The first column is the name of the service from `/etc/services'. The next column is the type of connection (stream or dgram or tli), then comes the
protocol type (tcp/udp etc). The wait column indicates whether the service is to be single or multithreaded, i.e. whether new requests should wait for an
existing request to complete or whether a new daemon should be started in parallel. The last two columns contain the location of the program which should
handle the request and the actual command line (including options) which should be executed.
To add a new service, you have to edit the file `/etc/inetd.conf' and then send the inetd process the HUP signal. To do this, you find the process id:
ps aux | grep inetd
Then you type:
kill -HUP process-id
TCP wrapper security
One of the problems with inetd is that it accepts connections from any host and passes them to to services registered in its configuration file without
question. In todays network climate this is a dangerous step and it is usually desirable to limit the availability of certain services. For instance, services which
are purely local (like RPC) should never be left open so that outside users could try to exploit them. In short, services should only be made available to those
who need them. If they are left open to those who do not need them, you invite attacks on your system.
TCP wrappers is a solution to this problem. In short it gives you the possibility of adding Access Control Lists (ACLs) to your network services.
TCP wrapper expects to find daemons in a special directory.
# hosts.allow
in.fingerd: LOCAL
in.cfingerd: LOCAL
sendmail: ALL
cfd: LOCAL
in.rlogin: ALL
in.telnetd: ALL
sshd: ALL
All other services are denied access:
Planning for an efficient network
The efficiency of your network can be improved greatly by planning carefully how you run networks services particularly NFS, NIS and DNS. You should
think about:
Which hosts should have disks and which host(s) should be the NFS disk server.
Separate users' home directories over several disks and keep problem disk-users on a partition for themselves, away from honest users.
Which services will you provide. Which hosts should be servers and for which services?
You will need a binary server (for software) for each operating system type you maintain, since a Sun machine cannot run software compiled on a
GNU/Linux host etc.
Do you want to centralize your services on one host, or distribute them across many? The first possibility is easier to administrate, but the second
might be more efficient and less prone to host crashes.
Try not to create disk-deadlocks whereby host A needs to mount host B and host B needs to mount host A.
Will one server be enough? Do you need a backup server?
Normally you want to use your most powerful system as the server, but perhaps you want to reserve this for running some especially heavy
software, and use a less powerful host as server.
Setting up a nameserver
File structure
The Domain Name Service (DNS) is that internet service which converts fully qualified hostnames (FQHN) like into IP addresses
like and vice versa. A FQHN always includes the full name of the host and the domain in which is it registered. The service consists of a
database for your local domain together with a daemon named or in.named which handles lookups in the database.
Recently, the BIND software has been rewritten to solve a number of pressing problems. The resulting version is called BIND 8. Most vendor releases do
not incorporate this new BIND as of 1998, but you would do well to get the bind software and install it yourself, if you intend running a nameserver.
Since the mapping of (even fully qualified) hostnames to IP addresses is not one-to-one (a host can have several aliases and network interfaces), the DNS
Monday, July 26, 1999
Principles of system administration
Page: 41
database needs information about conversion both from FQHN to IP address and the other way around. To set up a primary name server, you will need to
complete the following checklist.
Make a directory in your site-dependent files (don't mix it up with your OS files) called `dns' or `named' and change to this directory.
Make subdirectories `pz' and `sz' for primary and secondary data. We shall only be using the primary data at this stage. Secondary data are
cached files from a different primary nameserver, which act as mirror for backup purposes.
Find out your domain name (let's supposed it's called ) and create a file `pz/ '. We shall worry about its
contents shortly. This file will contain data for converting names into addresses.
Find out your network numbers and create files for them. The domain, for instance, has four networks:,, and DNS does not use the trailing zero for the network addresses, so create files `pz/128.39.89',
`pz/128.39.73' etc., one for each network. These files will contain data for converting addresses into `canonical' names, or official hostnames (as
opposed to aliases). We shall call the network files `pz/network '
Create a file for the loopback network (everyone needs one of these) `pz/127.0.0'
Create a cache file `named.cache' which will contain the names of the Internet's primary (root) name servers.
(Old BIND v 4.x) Create a boot configuration file for the name-service daemon `named.boot'. We shall later link this file to `/etc/named.boot'
where the daemon expects to find it. We place it here, however, so that it doesn't get lost or destroyed if we should choose to upgrade the operating
(New BIND v 8.x) Create a configuration file `named.conf'. We shall later link this file to `/etc/named.conf' where the daemon expects to
find it. We place it here, however, so that it doesn't get lost or destroyed if we should choose to upgrade the operating system.
Link the boot/conf file to the `/etc' directory, so that it appears to be at the location `/etc/named.boot'. Start the name-service daemon by
typing in.named.
At this stage you should have the following directory structure in your site dependent files.
named/named.boot (old BIND)
named/named.conf (new BIND)
The DNS tables also tell E-mail services how to deliver mail. You will need to have a so-called `mail-exchanger' record in the DNS tables in order to tell
E-mail which host handles E-mail for the domain. An entry of the form
MX priority
tells E-mail services that mail sent to name @domain-name should be routed to the host mailhost . For instance, MX 10 nexus
tells our server that mail addresses of the form name should be handled by host nexus. The priority number 10 is chosen at random.
Several records can be added with backup servers if the first server does not respond.
Mail records are also possible on a per-host basis. If you want mail sent to host XXX handled by host YYY, you would add a record,
This would mean that mail sent to name @XXX.domain-name would be handled by YYY. For instance, name would be handled by
`Sample named.boot' for BIND v 4.x
The boot file tells the name-service daemon which files provide information for which networks. The syntax of this file is somewhat bizarre and has its roots
in history. It begins with the name of the directory in which you have chosen to store your data. Next it contains a line telling the daemon the name of the
cache file. Finally we list all primary and secondary domains and networks. In our case we have only primary data, no mirrored data from other servers.
Suffice it to say that the network addresses have to be written backwards for reverse lookup (i.e. IP address to FQHN resolution) and the string is appended(5). Here is an example file from the primary server at
; Boot file for primary name server
; Note that there should be one primary entry for each SOA record.
; type
source file or host
; running directory for named
; cache (mandatory). Needed to "prime" name server with startup info so it
; can reach the root name servers.
; Primary and secondary name/address zone files follow.
Monday, July 26, 1999
Principles of system administration
Note that comments are written after a semi-colon in DNS files.
Sample `named.conf' for BIND v 8.x
If you are going to use the new BIND software (recommended if you run a nameserver) then you need to replace the `named.boot' file with a new format
file called `named.conf'. The information contained in this file is the same as that for `named.boot', but many more options can be set in the new file,
particularly in connection with logging. Here is a translation of the above `named.boot' file into the new format:
directory "/local/iu/dns";
check-names master ignore;
check-names response ignore;
check-names slave warn;
named-xfer "/local/iu/bind/bin"; /* Location of daemon */
fake-iquery no;
/* security */
notify yes;
zone "."
type hint;
file "named.cache";
// Primary and secondary name/address zone files follow.
zone ""
type master;
file "pz/127.0.0";
zone ""
type master;
file "pz/128.39.73";
zone ""
type master;
file "pz/128.39.74";
zone ""
type master;
file "pz/128.39.75";
zone ""
type master;
file "pz/128.39.89";
zone ""
type master;
file "pz/";
// server options
transfer-format many-answers;
channel admin_stuff
file "/local/iu/logs/admin" versions 7;
Page: 42
Monday, July 26, 1999
Principles of system administration
severity debug;
print-time yes;
print-category yes;
print-severity yes;
channel xfers
file "/local/iu/logs/xfer" versions 7;
severity debug;
print-time yes;
print-category yes;
print-severity yes;
channel updates
file "/local/iu/logs/updates" versions 10;
severity debug;
print-time yes;
print-category yes;
print-severity yes;
channel security
file "/local/iu/logs/security" versions 7;
severity debug;
print-time yes;
print-category yes;
print-severity yes;
category config
category parser
category update
category load
category security
category xfer-in
category xfer-out
category db
category lame-servers
category cname
Sample `named.cache'
Page: 43
Monday, July 26, 1999
Principles of system administration
Page: 44
The data in the cache file can be retrieved by anonymous ftp from the most illustrious repository of internet information The list of internet
root servers (which bind together all internet domains) are listed in a file called `netinfo/root-servers.txt' The retrieved data have to be written in the
following form
; From's netinfo/root-servers.txt, as of December 96 (mark)
Sample `pz/ '
The main domain file contains data identifying the IP addresses of hosts in your domain, it defines possible aliases for those names and it also identifies
special servers such as mail-exchangers which mail-relay programs use to learn where to send electronic mail to your domain.
Each host has a canonical name or CNAME which is its official name. One may then define any number of aliases to this canonical name. For instance, it
is common to create aliases to hosts which provide well known services, like and so that no one needs to
remember a special host name in order to access these services in your domain. Here is an abbreviated example file. There are several kind of records here
Indicates authority for this domain (referred to as `@').
Lists a nameserver for this domain.
Lists a mail exchanger for this domain (with priority).
Create an A record, i.e. define the canonical name of a host with a given IP address.
CNAME Associate an alias with a canonical name.
HINFO Host information.
; @ is an alias for this
1996111300 ; Serialnumber
; Refresh, 1 hr
; Retry, 10 min
604800 ; Expire 1 week
; Minimum TTL, 1 day
Solaris 2
; aliases
; Name servers:
; Mail exchangers:
; Domain data:
; Router
; 89 net
Monday, July 26, 1999
Principles of system administration
; 73 net
Sample `pz/network '
The network files are responsible for producing a fully qualified domain name given an IP address. This is accomplished with so-called PTR records. The
reverse lookup-file looks like this:
1996111300 ; Serialnumber
; Refresh, 1 hr
; Retry, 10 min
604800 ; Expire 1 week
86400 ) ; Minimum TTL, 1 day
; Name servers:
; Glue records:
; Mail exchangers:
; Domain data:
Note carefully how the names end with a full-stop. If you forget this, the name server appends the domain name to the end, resulting in something like
Sample `pz/127.0.0'
; Zone file for "localhost" entry.
1995070300 ; Serialnumber
; Refresh
; Retry
3600000 ; Expire
14400 ) ; Minimum
; Glue
; (none needed, no name servers in this domain)
; Domain data
localhost. IN NS IN NS IN PTR localhost.
Page: 45
Monday, July 26, 1999
Principles of system administration
Page: 46
Getting and compiling BIND v 8.x
The new BIND software can be collected from the Internet Software Consortium `'. This software contains everything you need to build
replacement name-daemon software and nslookup. It also allows you to build resolver libraries for system lookup. Unless you really know what you are
doing, I don't recommend replacing the resolver libraries on systems with shared-libraries (Solaris, for instance). If you are running Solaris 2.6 or better, you
can make do without these. The most important element is the name server daemon named.
Here is a brief checklist for building and installing the nameserver:
Make a new directory somewhere, called `bind'. Collect the gzipped tar file from the WWW site above. The tar file unpacks into a sub-directory
called `src', since it is part of a larger tree of code, so make sure that you unpack in a fresh directory.
Decide on a directory where the compiled code will reside. For instance `/site/host/bind'.
You will need to be root in order to follow the build procedure. (Silly!)
Change directory to the sources and follow the instructions there:
cd src
make DST=/site/host/bind SRC=`pwd` links
cd /site/host/bind
make clean
make depend
The installation requires that you have byacc installed on your system. If you do not have this, you need to edit one of the Makefiles and replace it
with yacc or bison. Note that, if you use bison, you should use the command bison -y -d for compatibility. Once the sources are build, you can
move them to a permanent location. For instance
host# mv bin/named/named /local/sbin/named
and so on for the other programs. Remember to set appropriate permissions on the files. None of them need to be setuid root! Remember to set the
correct path to the compiled xfer-daemon in the file `named.conf'. Remove any startup code you have on the system for older named versions.
Add code to start up the new daemon.
Setting up a WWW server
The World Wide Web (or W3) service is mediated by the http-daemon. There are several publicly available daemons which mediate the WWW service.
This text is based on the freely available Apache daemon which is widely regarded as the best and most up-to-date. you can get it from
At the practical level, the web service is just another daemon which you have to start on some server. To start a WWW service you need some html-files
containing web data and a server which is configured with the filenames and locations of these data. You then need to set some configuration files which tell
the daemon where to find the web pages it will be publishing, and to tell it what you do not want it to tell the outside world. The security of your system
depends on which files and directories outsiders should have access to. It is up to you to decide this. Here is your checklist:
Create a directory in your site-dependent files called `www', where you will keep your web pages. In particular you will need your master file
`www/index.html' which is the root of your web site.
Create a directory in your site dependent files called `httpd' where the web server program will be installed. This is collected as a package, so you
should keep the files together in this directory. Unpack the files into this directory.
Edit the file `httpd/conf/access.conf' with your favourite text editor. Set up the directories to point to your data, See section Sample
`access.conf' file.
Edit the file `httpd/conf/srm.conf' and set the paths to point to your data, See section `srm.conf' file. This can also be used to specify any
special error handling, such as customized error pages.
The sample files for in which the root files for WWW are attached to a special user called `www'. The daemon will run with the access rights of this user for
all operations. The UserDir directive tells the daemon that all users' personal WWW pages will be kept in a subdirectory of their home directories called
`www'. Thus using this arrangement (user www, with subdir www) the DocumentRoot for the main WWW pages becomes
which, using the file scheme /site/host/contents becomes
Having installed the WWW server, you might want to provide a connection with a database engine, such as an MSQL database. This is covered briefly in the
last section.
Sample `access.conf' file
# access.conf: Global access configuration
# Online docs at
<Directory /local/site/httpd/cgi-bin>
Options Indexes FollowSymLinks
Monday, July 26, 1999
Principles of system administration
<Directory /site/server/home/www/www>
Options Indexes FollowSymLinks Includes
<Directory /site/server/home/www/www/cgi-bin-public>
Options Indexes FollowSymLinks Includes
AllowOverride All
<Limit GET>
order allow,deny
allow from all
`srm.conf' file
## srm.conf -- Apache HTTP server configuration file
# With this document, you define the name space that users see of your http
# server. This file also defines server settings which affect how requests are
# serviced, and how results should be formatted.
# See the tutorials at for
# more information.
# Originally by Rob McCool; Adapted for Apache
# DocumentRoot: The directory out of which you will serve your
# documents. By default, all requests are taken from this directory, but
# symbolic links and aliases may be used to point to other locations.
DocumentRoot /iu/nexus/ua/www/www
# UserDir: The name of the directory which is appended onto a user's home
# directory if a ~user request is recieved.
UserDir www
# DirectoryIndex: Name of the file or files to use as a pre-written HTML
# directory index. Separate multiple entries with spaces.
DirectoryIndex index.html
# FancyIndexing is whether you want fancy directory indexing or standard
FancyIndexing on
# AddIcon tells the server which icon to show for different files or filename
# extensions
AddIconByEncoding (CMP,/icons/compressed.gif) x-compress x-gzip
(TXT,/icons/text.gif) text/*
(IMG,/icons/image2.gif) image/*
(SND,/icons/sound2.gif) audio/*
(VID,/icons/movie.gif) video/*
/icons/binary.gif .bin .exe
/icons/binhex.gif .hqx
/icons/tar.gif .tar
/icons/world2.gif .wrl .wrl.gz .vrml .vrm .iv
/icons/compressed.gif .Z .z .tgz .gz .zip
/icons/a.gif .ps .ai .eps
/icons/layout.gif .html .shtml .htm .pdf
/icons/text.gif .txt
/icons/c.gif .c
/icons/p.gif .pl .py
/icons/f.gif .for
/icons/dvi.gif .dvi
/icons/uuencoded.gif .uu
/icons/script.gif .conf .sh .shar .csh .ksh .tcl
/icons/tex.gif .tex
/icons/bomb.gif core
/icons/back.gif ..
/icons/hand.right.gif README
/icons/folder.gif ^^DIRECTORY^^
/icons/blank.gif ^^BLANKICON^^
# DefaultIcon is which icon to show for files which do not have an icon
# explicitly set.
Page: 47
Monday, July 26, 1999
Principles of system administration
DefaultIcon /icons/unknown.gif
# AddDescription allows you to place a short description after a file in
# server-generated indexes.
# Format: AddDescription "description" filename
ReadmeName is the name of the README file the server will look for by
default. Format: ReadmeName name
The server will first look for name.html, include it if found, and it will
then look for name and include it as plaintext if found.
HeaderName is the name of a file which should be prepended to
directory indexes.
ReadmeName README
HeaderName HEADER
# IndexIgnore is a set of filenames which directory indexing should ignore
# Format: IndexIgnore name1 name2...
IndexIgnore .??* *~ *# HEADER* README* RCS
# AccessFileName: The name of the file to look for in each directory
# for access control information.
AccessFileName .htaccess
# DefaultType is the default MIME type for documents which the server
# cannot find the type of from filename extensions.
DefaultType text/plain
# AddEncoding allows you to have certain browsers (Mosaic/X 2.1+) uncompress
# information on the fly. Note: Not all browsers support this.
AddEncoding x-compress Z
AddEncoding x-gzip gz
AddLanguage allows you to specify the language of a document. You can
then use content negotiation to give a browser a file in a language
it can understand. Note that the suffix does not have to be the same
as the language keyword -- those with documents in Polish (whose
net-standard language code is pl) may wish to use "AddLanguage pl .po"
to avoid the ambiguity with the common suffix for perl scripts.
# LanguagePriority allows you to give precedence to some languages
# in case of a tie during content negotiation.
# Just list the languages in decreasing order of preference.
LanguagePriority en fr de
Redirect allows you to tell clients about documents which used to exist in
your server's namespace, but do not anymore. This allows you to tell the
clients where to look for the relocated document.
Format: Redirect fakename url
# Aliases: Add here as many aliases as you need (with no limit). The format is
# Alias fakename realname
# Note that if you include a trailing / on fakename then the server will
# require it to be present in the URL. So "/icons" isn't aliased in this
# example.
Alias /icons/ /local/iu/share/apache/icons/
# ScriptAlias: This controls which directories contain server scripts.
# Format: ScriptAlias fakename realname
ScriptAlias /cgi-bin/ /site/server/home/www/www/cgi-bin-public/
# If you want to use server side includes, or CGI outside
# ScriptAliased directories, uncomment the following lines.
# AddType allows you to tweak mime.types without actually editing it, or to
# make certain files to be certain types.
# Format: AddType type/subtype ext1
# For example, the PHP3 module (not part of the Apache distribution)
# will typically use:
Page: 48
Monday, July 26, 1999
Principles of system administration
AddType application/x-httpd-php3 .phtml
AddType application/x-httpd-php3-source .phps
AddType application/x-httpd-php3 .php3 .php
AddHandler allows you to map certain file extensions to "handlers",
actions unrelated to filetype. These can be either built into the server
or added with the Action command (see below)
Format: AddHandler action-name ext1
# To use CGI scripts:
#AddHandler cgi-script .cgi
# To use server-parsed HTML files
AddType text/html .shtml
AddHandler server-parsed .shtml
# Uncomment the following line to enable Apache's send-asis HTTP file
# feature
#AddHandler send-as-is asis
# If you wish to use server-parsed imagemap files, use
AddHandler imap-file map
# To enable type maps, you might want to use
#AddHandler type-map var
Action lets you define media types that will execute a script whenever
a matching file is called. This eliminates the need for repeated URL
pathnames for oft-used CGI file processors.
Format: Action media/type /cgi-script/location
Format: Action handler-name /cgi-script/location
# MetaDir: specifies the name of the directory in which Apache can find
# meta information files. These files contain additional HTTP headers
# to include when sending the document
#MetaDir .web
# MetaSuffix: specifies the file name suffix for the file containing the
# meta information.
#MetaSuffix .meta
# Customizable error response (Apache style)
# these come in three flavors
1) plain text
#ErrorDocument 500 "The server made a boo boo.
# n.b. the (") marks it as text, it does not get output
2) local redirects
#ErrorDocument 404 /missing.html
# to redirect to local url /missing.html
#ErrorDocument 404 /cgi-bin/
# n.b. can redirect to a script or a document using server-side-includes.
3) external redirects
#ErrorDocument 402
ErrorDocument 404 /400.html
ErrorDocument 401 /400.html
ErrorDocument 500 /500.html
# mod_mime_magic allows the server to use various hints from the file itself
# to determine its type.
#MimeMagicFile /local/iu/etc/apache/magic
The following directives disable keepalives and HTTP header flushes.
The first directive disables it for Netscape 2.x and browsers which
spoof it. There are known problems with these.
The second directive is for Microsoft Internet Explorer 4.0b2
which has a broken HTTP/1.1 implementation and does not properly
support keepalive when it is used on 301 or 302 (redirect) responses.
BrowserMatch "Mozilla/2" nokeepalive
BrowserMatch "MSIE 4.0b2;" nokeepalive downgrade-1.0 force-response-1.0
# The following directive disables HTTP/1.1 responses to browsers which
# are in violation of the HTTP/1.0 spec by not being able to grok a
# basic 1.1 response.
BrowserMatch "RealPlayer 4.0" force-response-1.0
BrowserMatch "Java/1.0" force-response-1.0
BrowserMatch "JDK/1.0" force-response-1.0
Page: 49
Monday, July 26, 1999
Principles of system administration
Page: 50
ScriptAlias /cgi-bin-public /iu/nexus/ua/www/www/cgi-bin-public
# More script aliases here...
Notice that this file can be used to create customized error messages, tailored to your own special environment, perhaps with personal logo etc. The
ErrorDocument directive is used for this. It traps error numbers and maps them to special pages. For example, to map to a standard local file in the root of
the server's pages one would add
ErrorDocument 500 /errorpage.html
ErrorDocument 401 /errorpage.html
ErrorDocument 402 /errorpage.html
Perl script for generating script aliases
A convenient way of generating script aliases for all users is to write a short Perl script which rewrites the `srm.conf' file by looking through the password
file and adding an entry for every user. In addition, a general cgi-bin directory is often desirable, where it is possible to place scripts which anyone can use.
In the example below we call this alias cgi-bin-public. Each user has a script alias called cgi-bin-username .
# Build script aliases from password file
# Path to the srm.conf file
$srmconf = "/local/httpd/conf/srm.conf";
$srmbase = "/local/iu/lib/apache/srm.conf";
open (OUT,">$srmconf") || die;
open (BASE,"$srm") || die;
while (<BASE>)
print OUT $_;
# Copy base file to output
close (BASE);
while (($name,$pw,$uid,$gid,$qu,$com,$full,$dir) = getpwent)
# SKip system accounts
next if ($uid < 100);
print OUT "ScriptAlias /cgi-bin-$name $dir/www/cgi-bin\n";
last if ($name eq "");
close OUT;
`mime.types' file
This file tells the server how to respond to file requests containing special data. It consists of a list of protocol names followed by a list of file extensions.
Unrecognized files are displayed in your browser as text/ascii files. If you see graphics files (like vrml files) displayed as text, then you need to add a line
here to inform the server about the existence of such files. Here is a brief excerpt:
mpeg mpg mpe
qt mov
Private directories
In some cases you will want certain information to be made available to local users of your domain but not to general outside users. This can be
accomplished by using a `.htaccess' file to override the default access rights set in the server configuration files. The assumes that you have set
AllowOverride All.
Creating a directory that is only available from the local domain is a simple matter of creating the directory and creating a `.htaccess' file owned by the
`www' user (i.e. the user running the daemon) with read permission for `www', containing the lines:
order deny,allow
deny from all
allow from .local.domain
Monday, July 26, 1999
Principles of system administration
Page: 51
For many sites the possibility to combine a database of information with its web pages is a powerful one. The language PHP is designed for this purpose. In
order to use PHP it has to be coupled to a local WWW server and to a specific database. The order in which these three components is configured is
important. Here is an example using MySQL, PHP and the Apache web server. This is a powerful and popular combination of free components.
For some of its graphical functions, PHP uses a library called gd. This can be obtained from and added with a configuration
option as secribed below.
The first step is to compile the database engine. Note that the source tree has to be remain on the system, so it should be unpacked in its final location. The
other software packages expect to find the msql server in the directory `/usr/local/mysql'. You should make a special user for the mysql daemon so that
it does not need to run as root.
cd /usr/local
tar zxf mysql-xxx.tar.gz
cd /local/mysql-xxx
configure --with-pthread --with-unix-socket-path=/home/mysql.socket
make install
--with-mysqld-user=mysql --prefix=/usr/local/mysql
# After check that the daemon really is started as mysql user!!!
Next it is necessary to make a temporary build of the WWW server. After this is done, PHP can be compiled and installed. Then one must go back and
make a proper build of the WWW server. These can be kept in `/usr/local' or in site dependent files `/local/site', as you see fit. Again, the source
trees have to remain on the system.
cd /local/site
tar zxf apache_1.3.3.tar.gz
cd /local/site/apache_1.3.3
./configure --prefix=/local/site
cd /local/site
tar zxf php3.tar.gz
cd /local/site/php3
./configure --prefix=/local/site --with-apache=/local/site/apache_1.3.3 --with-mysql=/local/mysql --with-zlib --with-gd=/usr/local/lib
make install
cd /local/site/apache_1.3.3
./configure --prefix=/local/site --activate-module=src/modules/php3/libphp3.a
make install
There are lots of things which can go wrong with options here. These options work for building a php interpreter into the apache server. You might also want
to use php as a CGI scripting language. In that case you need to build a binary separately. To do this you configure as follows:
cd /local/site/php3
./configure --prefix=/local/site --with-mysql=/local/mysql --with-zlib --with-gd=/usr/local/lib --enable-discard-path
make install
E-mail configuration
Nightmares on ELM street!
Electronic mail configuration is something which you do not want to change very often. Once you have a working system, it is well worth leaving it alone.
Configuration of E-mail is one of the most complex issues for the system administrator.
Why should it be so complex? Part of the trouble is that, in the past, there were many different kinds of networks, many different ways of connecting up to
different hosts. This made it quite a complex issue to relay messages all over the world. Today things are much simpler: most sites use the internet protocols
and mail configuration can be simplified quite a lot.
If you are lucky, and you are using an operating system like GNU/Linux, then a program will automatically help you set up E-mail, but you will still need
some information about the way mail is handled at your site. There are two main models for handling electronic mail at a domain. One is that every host
receives mail independently. This method is a bit old-fashioned since, usually, all users have the same password and account on all of the hosts on a network.
Hence the second method.
The second approach is to have a mail `hub', or central mail processor. In this model, all incoming mail is diverted to the hub and all outgoing mail is sent via
the hub. With this approach, you put all of your hard work into fixing mail on the hub, and all other machines have a simple configuration to pass all mail
onto the hub.
In order for mail to be diverted to a hub, one has to arrange for the mail exchanger data in DNS to point to the hub, for every system. i.e. for every host in
DNS one should have an MX record accompanying the A record:
Monday, July 26, 1999
Principles of system administration
Page: 52
Without such an MX record, mail which is addressed to [email protected] will be sent directly to hostname. With such a record the mail for hostname
is sent to mailhub.domain instead. It can later be forwarded to hostname if desired using a mailertable. This has several advantages: first of all it means that
mail configuration can be centralized, spam filtering can be performed even for dumb hosts and aliases can be expanded here without the need for a
distributed alias database like NIS. The second advantage concerns security. If all mail is forced to pass through this hub then a secure set-up here will
prevent SNMP attacks on weaker hosts, thus this simplifies the security administration of mail also. One may concentrate most of one's effort on securing
this hub, knowing that nothing very bad will happen. A further precaution is then to configure the site router to accept SNMP traffic only for the mailhub
since it is supposed to go there anyway. That way, if one forgets an MX record in DNS there will be no back-doors for would-be attackers.
In this section we shall look only at the mail agent called Berkeley sendmail . This is the most up to date version of sendmail. If you are using non-free
software from a vendor, you should strongly consider collecting Berkeley sendmail from the network and compiling it in place of your vendors version.
Sendmail is very susceptible to attack from the network, and only the Berkeley version is well enough equipped with deal with this threat.
Collecting sendmail
You can find out about sendmail and also collect the latest version from the web site.
Compiling and installing sendmail
Once you have unpacked the distribution, you need to run make to compile it. Before doing this, you should make sure that you have all of the libraries you
need to compile. Sendmail uses BIND and TCP-wrappers libraries. You should consider searching for the latest versions of these libraries on the internet
before compiling. BIND is the resolver library. The official place to get BIND is `'. This also contains a library
lib44bsd.a which you should make available. You can pick up the latest version of TCP wrappers from
`'. Many of the database lookup features require the Berkeley `db' package. You should get this
from `' and compile it.
Here is an example for sendmail-8.9.1:
host# cd sendmail-8.9.1
host# ls
cf.tar mail.local praliases src
KNOWNBUGS RELEASE_NOTES contrib mailstats rmail
Makefile cf
host# cd src
host# sh makesendmail
Configuration: os=SunOS, rel=5.5.1, rbase=5, rroot=5.5,
arch=sun4, sfx=
Making in obj.SunOS.5.5.1.sun4
The script `makesendmail' selects your operating system type and compiles the program. In the process it creates a directory for the compilation. In the
example above, it creates `obj.SunOS.5.5.1.sun4'.
You might still have to edit the Makefile in this new directory, so do a CTRL-C to stop the compilation and edit the file which corresponds to
`obj.SunOS.5.5.1.sun4/Makefile' in the example above. You will have to choose the exact Makefile based on your own operating system.
In the Makefile, you can switch on several features. The first thing you should switch off is NIS alias lookups. The best way to do alias lookups is to use
only the Berkeley db database. That means editing out the line beginning DBMDEF= as in the example below. Use of NIS, NIS+ or other databases is not
recommended for several reasons. The main reason is that it not required as long as mail passes through a central hub, which is a good approach to mail
configuration for small to medium sized sites.
You must also decide where you want your distribution to be placed. A good place is `/usr/local/mail' or `/usr/local/lib/mail'. Create this
directory now and make a subdirectory `bin' where you will keep your executable file.
Here is an example Makefile:
This Makefile is designed to work on the old "make" program. It does
not use the obj subdirectory. It also does not install documentation
automatically -- think of it as a quick start for sites that have the
old make program (I recommend that you get and port the new make if you
are going to be doing any significant work on sendmail).
This has been tested on Solaris 2.5.
8.10 (Berkeley) 4/13/97
# use O=-O (usual) or O=-g (debugging)
# warning: do not use -O with versions of gcc prior to 2.6
Monday, July 26, 1999
Principles of system administration
Page: 53
DESTDIR = /local/mail
# define the database mechanism used for alias lookups:
-DNDBM -- use new DBM
-DNEWDB -- use new Berkeley DB
-DNIS -- include NIS support
# The really old (V7) DBM library is no longer supported.
# See READ_ME for a description of how these flags interact.
# environment definitions (e.g., -D_AIX3)
# see also conf.h for additional compilation flags
# include directories
# library directories
LIBDIRS=-L/usr/sww/lib -L/local/lib
# libraries required on your system
# delete -l44bsd if you are not running BIND 4.9.x
# add -ldb if you add -DNEWDB above (in DBMDEF)
#LIBS= -lresolv -l44bsd -lsocket -lnsl -lkstat
-lwrap -lresolv -l44bsd -lsocket -lnsl -lkstat -ldb
# location of sendmail binary (usually /usr/sbin or /usr/lib)
# location of file (usually /var/log or /usr/lib)
# location of sendmail.hf file (usually /usr/share/misc or /usr/lib)
When you have compiled the program successfully, the finished executable files must be installed. Store these executables in your local files, by copying
them like this:
cp obj.SunOS.5.5.1.sun4/sendmail /usr/local/mail/bin/sendmail
cp obj.SunOS.5.5.1.sun4/makemap /usr/local/mail/bin/makemap
Your operating system most likely expects to find the sendmail executable file in either the `/usr/lib/' directory, or the `/usr/sbin' directory on newer
systems. You should replace the old executable in these directories by making a link to the new executable. For example:
mv /usr/lib/sendmail /usr/lib/
ln -s /usr/local/mail/bin/sendmail /usr/lib/sendmail
Configuring sendmail
To finish of the installation, you need to create configuration files for your mail domain. Begin by going back to the sendmail distribution and copying the
`cf' directory to your own directory, like this:
cp -r sendmail-8.9.1/cf /usr/local/mail
Next make a `lib' directory.
mkdir /usr/local/mail/lib
To create a `' file, you need to create a so-called macro file `/usr/local/mail/lib/'. Here is an example file for domain
`'. You should only need to change the domain name and the OS name of your mailhost in the first trhee lines. Using this file you will be
able to build the sendmail configuration more or less automatically. This example is for sendmail 8.9.1 for a mail hub:
VERSIONID(`$Id:,v 1.1 1997/04/08 08:52:28 mroot Exp mroot $')
Monday, July 26, 1999
Principles of system administration
FEATURE(domaintable, `hash -o /local/iu/mail/lib/domaintable')
FEATURE(mailertable, `hash -o /local/iu/mail/lib/mailertable')
FEATURE(access_db, `hash -o /local/iu/mail/lib/access_db')
FEATURE(genericstable, `hash -o /local/iu/mail/lib/genericstable')
FEATURE(virtusertable, `hash -o /local/iu/mail/lib/virtusertable')
define(`ALIAS_FILE', /local/iu/mail/lib/aliases)dnl
define(`HELP_FILE', /local/iu/mail/lib/sendmail.hf)dnl
define(`STATUS_FILE', /local/iu/mail/etc/
define(`QUEUE_DIR', /var/spool/mqueue)
define(`LOCAL_MAILER_CHARSET', iso-8859-1)dnl
define(`SMTP_MAIL_CHARSET', iso-8859-1)dnl
define(`confMAX_MESSAGE_SIZE', `20000000')
define(`confHOST_STATUS_DIRECTORY', `.hoststat')
define(`confPRIVACY_FLAGS', `authwarnings,noexpn,novrfy')
define(`confME_TOO', `True')
define(`confMIME_FORMAT_ERRORS', `False')
define(`confTIME_ZONE', `MET-1METDST')
define(`confDEF_CHAR_SET', `iso-8859-1')
define(`confEIGHT_BIT_HANDLING', `m')
define(`confCW_FILE', `/local/iu/mail/lib/')
define(`confCT_FILE', `/local/iu/mail/lib/sendmail.ct')
define(`confUSERDB_SPEC', `/local/iu/mail/lib/userdb.db')
define(`confSMTP_MAILER', `esmtp')
FEATURE(rbl) dnl vixie's black hole database for spammers
Next create `/usr/local/mail/Makefile' which will build the configuration for you:
cf/ $(ALIASES).db $(USERDB).db .restart
$(MAKEMAP) btree $(USERDB) < $(USERDB)
$(MAKEMAP) hash $(ACCESSDB) < $(ACCESSDB) $(MCFILE) cf/hack/rewrite_from.m4 $(MCFILE)
m4 -D_CF_DIR_=$(CF_DIR) cf/m4/cf.m4 $(MCFILE) >
.restart: lib/
kill -1 `head -1 $(PIDFILE)
touch .restart
This is a shorter example for a system attached to a mail hub, whose only function is to send the mail to the hub for processing.
Page: 54
Monday, July 26, 1999
Principles of system administration
Page: 55
Typing make in the `/usr/local/mail' directory should now result in a configuration file /usr/local/mail/ You should wait until you
have read the next section before doing this.
You will need to create a file `lib/' which contains a list of possible machines or domains for which the sendmail program will accept mail. It
is, amongst other things, this file which allows you to send mail of the form [email protected], i.e. to an entire domain, without specifying a particular
machine. This file should contain a list of all the valid addresses, like this:
Finally, you will need to make the key files readable for normal users. There is no harm in giving everyone read access to all the files and directories.
Rewriting outgoing addresses
The `lib/userdb' file and `lib/aliases' file are used by sendmail for resolving aliases. It is common for sites to create mail aliases for all their users,
by taking the full name of each user and joining the pieces with dots. For example, user `mark' with fullname `Mark Burgess' would map to an alias
Aliases of this kind look nice and are user friendly to outsiders, but they do not work unless you set up the system yourself. To do this you need to create
files `lib/aliases' and `lib/userdb'
For each user, the `userdb' file should contain a line of the form.
[email protected]:maildrop [email protected]
which tells sendmail where to deliver incoming mail which is sent to the user alias `Mark.Burgess' at the mail hub.
Outgoing messages are processed if you have a line of the form:
mark:mailname [email protected]
This means that mail messages which originate from `mark' will be rewritten so that they look as though they orginate from `[email protected]'.
This form usually only applies to mail which originates from the local mail host, since that is the only case where your outgoing name is just your short
user-name `mark'. Mail which is sent from other hosts to the mail-hub for outgoing processing generally produces from-information in the form
`[email protected]'. In order for this to be rewritten, you need a line of the form as well.
[email protected]:mailname [email protected]
The `aliases' file is set up as in the earlier section, See section Mail aliases.
The sendmail remote shell is a security measure to prevent system crackers from executing an arbitrary program on your system. The smrsh program is
contained in the sendmail distribution and is configured by using the FEATURE
Spam and junk mail
Spam mail is E-mail which is sent repetetively as a would-be denial of service attack. The word comes from the Monty Python spam song sketch. Junk mail
is unwanted mail, often advertisements about financial opportunities or pornography, sometimes hoaxes. Often these two kinds of unwanted mail are lumped
together and called collectively `spam'.
Spam has become a major problem since it is very easy to send E-mail and very hard to pick out what it useful from what is useless. There are two
approaches to the filtering of spam, both of which are needed together:
Site rules for rejecting mail (ACLs).
Private user-rules for rejecting mail.
The reason why both of these is needed is that what one user wants to reject, another user might be glad to receive. Users prospecting for finanical
opportunities or collecting the latest `artwork' might live for the messages which most of us get annoyed with.
Sendmail has rules for filtering mail at the site level. These include the ability to deny access to connecting mailers from certain domains.
At the user level, users of procmail can use junkfilter to create their own rules for rejecting spam. See for
Policy decisions
Monday, July 26, 1999
Principles of system administration
Page: 56
In order to protect your site from E-mail attacks, even ones made in innocence you might want to restrict mail by other criteria. For example, multimedia
attachments can now allow users to sent huge files by email. This is a very inefficient way of sending large amounts of data and it causes problems for
mailbox storage space. A possibility is to limit the size of mail messages handled by sendmail so that mail which is too large will be rejected with an error
message. For example, the following rules limit E-mail to approximately 20MB. Even with such a large reject size a handful of messages per month are
rejected on the basis of this rule.
define(`confMAX_MESSAGE_SIZE', `20000000')
Again, this must be a policy decision like garbage collection of users' files. It is never desirable to restrict the personal freedom of users, but it becomes a
matter of survival. If one provides an opening, it will be exploited either through ignorance or malice.
Overview of Unix
@hrule @vskip 0.3cm
Useful Unix commands
Typed commands are infinitely more flexible than graphical (GUI) based programs. You can tell the system what you want to do, rather than having to
search through the menus to find out whether or not you are allowed to do what you want. As a system administrator you will find most GUI programs
useless for any real tasks.
Here is a list of commands which you will find useful during your stint as a system administrator. Always check the manual page on your local system
before trying these commands. Versions, optional and even names differ, especially on older systems.
Who am I?
whoami Prints your user name.
who am i
Prints your real and effective user id, and terminal.
GNU program which prints all your user ids and groups.
Remote logins
The telnet command is the most reliable way of logging onto a remote unix host. The rlogin or rsh commands can be used to this effect, but they will
sometimes hang without reason, where telnet works without problem.
The rlogin command can be used to login without a password using the `.rhosts' authority file for trusted hosts and users.
Monitoring disk usage
Display the usage of all mounted disk partitions if no argument is given. If a directory is named, the state of the disk partition on which the given
directory resides is displayed. On SVR4 systems the output of this command is hard to understand unless the `-k' option is used.
Show disk usage on a per-file basis. The file sizes are either in kilobytes or in 512 byte blocks. The `-k' option forces output to be in kilobytes.
The `-s' option prevents du from outputting information about every file and yields a summary of the named directory instead.
swap -s
System 5 program to show swap space.
BSD program to show swap space.
Disk backups
Raw dump of a disk partition to a file or to tape.
rdump Same as dump, but this can be done over the network, remotely without need for physical contact with the host.
Solaris/SVR4 replaces dump with this command.
cp -r
Restores a disk partition from a filesystem dump.
Copy a directory and all files recursively to a new location. This does not preserve symbolic links but makes multiple copies of the file instead. See
tar below.
A simple way to copy an entire filesystem, preserving symbolic links is to do the following:
cd source-dir ; tar cf - . | (cd destination-dir ; tar xf - )
This pipes the output directly to the new directory using the streams interface for standard IO.
Mounting filesystems
mount Mount a local or remote disk.
umount Unmount a local or remote disk. Note the peculiar spelling.
Show all hosts who are mounting filesystems from this server.
Monday, July 26, 1999
Principles of system administration
Page: 57
Packing and unpacking archives
tar cf tarfile .tar source-dir
Packs all the files and sub-directories in the directory source-dir into a single `tape-archive' file. If the `-f' argument is missing, tar expects to be
able to write data to a default tape-streamer device and will complain with an error message.
tar zcf tarfile .tar.gz source-dir
Same as above, but piped through gzip to compress the data. This only works with GNU tar.
tar xf tarfile .tar
Unpacks the contents of a tar-file into the current directory.
tar zxf tarfile .tar.gz
Same as above, but pipes through gzip to uncompress data. This only works with GNU tar.
Shared libraries
Display the shared libraries used by a compiled executable file (SunOS only?).
Some systems require this command to be run after installing or upgrading shared libraries. It updates symbolic links to the latest version of the
library and in some cases generates a cache file of library names. Esp. GNU/Linux and SunOS prior to Solaris.
Handling binaries
This command lists all of the strings in a binary file. It is useful for finding out information which is compiled into software.
Prints the type of data a file contains.
Remove debugging information from a compiled program. This can reduce the size of the program substantially.
Files and databases
locate GNU fast-find command, part of the GNU find package. Locates the names of files matching the argument string in part, by reading from a
database. See `updatedb' below.
Locate by searching through every directory. Slow but powerful search facilities.
which Locate an executable file by searching through directories in the PATH or path variable lists.
whatis Gives a one-line summary of a command from the manual page (see catman).
catman -M
This program builds the apropos or man -k `whatis' databases.
This shell script updates the locate fast-find database.
Process management
ps aux Show all processes on the system (BSD).
ps -ef Show all processes on the system (SysV).
Send a signal to the named process (pid), not necessarily to kill it. The process ID is the one listed by the ps command. Typical options are `-HUP'
to send the hangup signal. This is used by many system daemons like inetd and cron as a signal which tells them to reread their configuration
files. Another option is `-9' which is a non-ignorable kill instruction.
Run a program with a non-default scheduling priority. This exists both as a shell command and as a C-shell builtin. The two versions use different
syntax. Normal users can only reduce the priority of their processes (make them `nicer'). Only the superuser can increase the priority of a process.
The priority values differ between BSD and SysV systems. Under BSD, the nice values run from -20 (highest priority) to 19 (lowest priority) with
0 being the default. Under SysV, priorities run from 0 to 39, with 20 being the default. The C-shell builtin priorities are always from -20 to 20 for
renice new-priority
-p pid
Resets the scheduling priority of a process to a new value. The priority values used by the system (not C shell) apply here.
Modern releases of Unix use the crontab command to schedule commands or scripts which are to be run at a specified time, or at regular intervals.
The crontab -l command lists currently registered jobs. The crontab -e command is used to edit the crontab file. Each user has his or her own
crontab file on every host. On older BSD systems, only root could alter the crontab file, which was typically a single file `/etc/crontab' or
`/usr/lib/crontab' containing usernames and jobs to be performed.
Mail management
Sometimes mail gets stuck and cannot be delivered for some reason. This might be because the receiving mailhost is down, or because there is insufficient
disk space to transfer the message, or many other reasons. In that case, incoming and outgoing mail gets placed in a queue which usually lies under the unix
directory `/var/spool/mail', `/var/mail' or one of these with `/var' replaced by `/usr'.
mailq Display any messages waiting in the mail queue. Same as sendmail -bp.
sendmail -q -v
Manually process the mail queue in verbose mode.
Disk management
format Sun's interactive disk formatting and repair tool.
The filesystem check program. A disk doctor. This checks the consistency of the filesystem (superblock consistency etc) and repairs simple
newfs Creates a new filesystem on a disk partition, erasing any previous data. This is analogous to formatting a diskette.
swapon This command causes the system to begin using a disk partition or swap file for system swapping/paging. swapon -a starts swapping on all
devices registered in the filesystem table `/etc/fstab' or equivalent.
mkfile Creates a special file for swapping inside a filesystem. The file has a fixed size, it cannot grow or shrink, or be edited directly. Normally swapping
should be to a raw partition. Swapping to this kind of file is inefficient, but is used by (for instance) diskless clients.
Monday, July 26, 1999
Principles of system administration
Page: 58
Name service lookups
An interactive query program for reading domain data from the Domain Name Service (DNS/BIND)
A non-interactive query program for reading domain data from the Domain Name Service (DNS/BIND)
Displays information about who is responsible for a limited number of domains in the US. For example, the highly irritating domain can be found with whois
System statistics
iostat Displays I/O summary from the disks at an interval of time-in-seconds .
vmstat Displays virtual-memory summary info at an interval of time-in-seconds .
Show all current network socket connections.
netstat -i
Show statistics from all network interfaces.
netstat -r
Show the static routing table.
Show NFS statistics. The -c option shows client-side data, while the -s option shows server-side data, where appropriate.
Send a sonar `ping' to see if a host is alive. The -s option sends multiple pings on some types of UNIX.
Show the route, passing through all gateways to the named host. This command normally has to be made setuid-root in order to open the network
kernel structures. Here is an example:
traceroute to (, 30 hops max,
40 byte packets
1 ca30-gw ( 3 ms 1 ms 2 ms
2 ( 5 ms 4 ms 5 ms
3 ( 15 ms 15 ms 19 ms
4 ( 43 ms 34 ms 32 ms
5 ( 40 ms 31 ms 38 ms
6 ( 37 ms 21 ms 29 ms
7 icm-uk-1-H1/ ( 58 ms 57 ms
8 icm-pen-1-H2/ ( 162 ms 136 ms
9 icm-pen-10-P4/ ( 198 ms 134
10 ( 146 ms 297 ms
11 * ( 144 ms 120 ms
12 ( 116 ms 116 ms 123
13 ( 131 ms 136 ms
14 ( 133 ms 124 ms
15 ( 138 ms 129
16 ( 128 ms
17 ( 129 ms 170 ms 143 ms
18 B24-RTR-FDDI.MIT.EDU ( 129 ms 147 ms 148
19 ( 149 ms * 130 ms
20 ( 134 ms 129 ms 134 ms
21 * * *
22 * * *
<--- routing problem here
Dump ethernet packet activity to console, showing traffic etc, SunOS.
snoop Newer version of etherfind in Solaris.
Configure or summarize the setup of the a network interface. e.g. ifconfig -a shows all interfaces. Used to set the broadcast address, netmask
and internet address of the host.
route Make an entry in the static routing table. Hosts which do not act as routers need only a default route. e.g. route add default .1
1 or in GNU/Linux route add default gw .1.
Internet searches
These days the simple way to search for software is to use the World Wide Web and a suitable search engine. For locating anonymous ftp data try the URL .
Before this, the program archie was used to search the archives.
Strengths and weaknesses of Unix
The main strength of Unix in a network environment is its stability and experience. It uses open standards which means that bugs are more quickly found. It
scales well under load.
Amongst the disadvantages: GNU/Linux is an attractive target for hackers. Unix has expensive commercial software. Courses and books on Unix tend to be
ten years out of date.
Monday, July 26, 1999
Principles of system administration
Page: 59
Installation procedure
@hrule @vskip 0.3cm
Information is the beginning of the system administrators wisdom, but you need to know how to find new information yourself by looking at manuals and
by asking others. You should get used to the fact that you will almost never find exactly what you are looking for, because everybody's needs are different.
Let us state another rule for the record:
We begin therefore by describing some useful information for beginners and certain key procedures. These are things you will be doing quite often so wake
up and pay attention!
Installing a new unix host
Installing a new machine is a simple affair these days. The operating system comes on some removable medium (like a CD). You put it in the player and run
a program, usually called install. Alternatively, you boot a machine directly from the CD and the program is run automatically. Then you simply answer a
few questions and the installation is done for you.
Operating systems are getting big so they are split up into packages. You are expected to choose whether you want to install everything or whether you just
want to install certain packages. Most operating systems provide a fancy package installation program which helps you with this. In most cases these
programs are quite stupid, they don't tell you that something will break if you don't install certain packages. For that reason it is strongly recommended that
you always install the complete operating system: every single package. Whether you know it or not, you almost certainly need the whole thing--and the
stuff you don't need probably doesn't take up much space anyway. Disk is cheap, but time spent trying to find out what went wrong with an installation is
In order to answer the questions about installing a new host, you need to collect some information and make some choices:
You must decide a name for your machine.
You will need an unused internet address.
You need to decide how much swap space to allocate. A good rule of thumb is at least twice the amount of RAM you have installed.
You need to know the local netmask.
You need to know the local timezone.
You need to know the name of your local domain.
You need to know whether you are using the Network Information Service (NIS) and, if so, what the name of the NIS server is.
When you have this information, you are ready to begin. As an example, you can look at the installation instructions for Debian GNU/Linux at
Once you have installed your operating system, remember:
When installing a large group of client machines, the easiest way might be to install one client by hand, and make the other clients a copy of this "client
master". This saves time in selection, installation and configuration of software packages. Here is an example from our college, which uses the Debian
GNU/Linux operating system.
When all software packages have been installed, and all necessary configuration has been done, make a backup of the machine with something like this
tar --exclude /iu --exclude /proc --exclude /lib/ \
--exclude /etc/hostname --exclude /etc/hosts -c -v
-f /iu/nexus/ud/ds/dump/my-machine.tar /
Note that several files must be excluded from the backup. In this case, all our NFS disks reside in /iu, and we don't want them in our backup. The file
/lib/ is the C library, if we try to write this file back from backup, the destination computer will crash immediately. /etc/hostname and /etc/hosts
contains definitions of the hostname of the destination computer, and must be left unchanged.
On the destination clients, make a minimal GNU/Linux installation. Make sure that you install NFS and network support, as the backup will be installed
from an NFS disk. When your new client is finished, mount the NFS disk where the backup resides, and restore the tar archive, for example by this
(cd / ; tar xvfp /mnt/ds/dump/my-machine.tar; lilo)
It is essential that the lilo command is run immediately after the tar archive has been restored. If you forget it, your system will not be able to reboot!
Then, remember to check that the file `/etc/init.d/network' is correct. You also have to create your NFS directories by hand, for example:
Check that the correct swap partition is defined in `/etc/fstab'. You might also have to change `/etc/X11/Xserver' and `/etc/X11/XF86Config'.
Monday, July 26, 1999
Principles of system administration
Page: 60
Finally, reboot your system.
Marrying unix and NT
If you are installing a PC, you will probably be interested in the idea of being able to choose between operating systems. You might want DOS to play
games, Unix for serious work, NT or Windows for its well-known applications and perhaps OS/2. You will want each of these operating systems to live on a
different partition of your hard disk(s). The question is: how do we marry these operating systems in such a way that they do not try to kill each other?
A word of warning: because of the wide range of system cracking tools available to users, it can be risky to install dual-boot operating systems on any
important host. NT is particularly vulnerable to password editing on dual boot systems.
The installation programs for NT and GNU/Linux do not always respect each other's independence. Experience says: install NT first, then the OS/2 boot
manager, then unix. The OS/2 boot manager is a wonderful program which allows you to choose operating system at boot-time. All you need is about 2
megabytes of free space on your bootable hard disk to install it. You get hold of the diskettes for OS/2/Warp and run the installation program: after a couple
of messages you will end up in the fdisk partition manager. If you select an area of free space, you can choose install-boot-manager from the menu and the
boot manager install itself. Once you have come through the fdisk partition program, you do not have to proceed with the rest of the OS/2 installation
unless you feel that you want to lend moral support to this historical and much-underrated operating system.
You can then install GNU/Linux or your favourite free-unix, being careful not to choose to install a `master boot record' as you move through the
installation menus.
If you do not have access to the OS2 boot manager, you can use NT's own bootmanager if you trick it by copying the boot blocks from the unix filesystem
to a DOS filesystem. The command
dd if=/dev/hdalinux-partition
of=/dev/hdaDOS /bootsect.linux bs=512 count=1
copies the bootsector to `c:\bootsect.linux'. It can now be used by the bootmanager, by editing `C:\boot.ini'
[boot loader]
default=multi(0)disk(0)rdisk(0)partition(NT partition number )\WINNT
[operating systems]
multi(0)disk(0)rdisk(0)partition(NT Partition )\Winnt="Windows Nt"
An alternative method is to copy the program loadlin and kernel to the DOS drive so that UNIX can be started from a windows command (which you
might like to put in a batch file)
loadlin vmlinuz root="/dev/hdaLinux-partition "
Note -- it might not be a good idea to use this method from a multitasking Windows NT machine since it does not give the system a chance to shut down
properly and damage to the NT partition might result. This method was meant for DOS PCs.
Problems have been known to occur with partitions using the GNU/Linux disk formatting tool. Sometimes this tool fails to acknowledge a partition or free
space with a message `Bad primary partition'. If this occurs, try to use a different formatter to create a partition which uses up the free space and try again.
This seems to make the problem go away in the cases I have seen.
Another problem is that the lilo disk boot information might not get written to the GNU/Linux partition. To fix this problem, boot with the GNU/Linux boot
diskette and edit the file `/etc/lilo.conf'. Make sure that the correct partition is bootable (use fdisk) if you need to see a list of partitions. When you
are done you must type the command lilo and press ENTER. If the lilo boot information is not correctly stored, the boot manager might not recognize the
partitions and claim that the GNU/Linux partition is not formatted.
Installing a diskless client
If you plan to run diskless unix workstations, i.e. workstations which have no physical disk attached but use a network server for all disk operations, then
you need to remember a few things.
Generally speaking, diskless workstations are not a good idea. Disks are cheap these days and diskless workstations are expensive in terms of network
bandwidth, time wasted because they run slowly and--not least--time spent fixing problems with them.
Most vendors will supply a script for creating a diskless workstation. You should use this script or you will just invite trouble. Here is a checklist of things
to remember:
A diskless workstation needs its own root filesystem and its own swap space, but it can share system files under `/usr'. You need to create disk
areas for these. Call them something like `/export/swap/hostname ' and `/export/root/hostname '. These areas need to be exported to the
clients with root privileges granted.
`/etc/ethers' on the server must contain the ethernet addresses of the clients.
When a diskless system is switched on for the first time, it has no files and knows nothing about itself except the ethernet address on its network
card. It proceeds by sending a RARP (Reverse address resolution protocol) out onto the subnet in the hope that a RARP server (in.rarpd) will
respond by telling it its internet address.
A subsequent call to the bootparamd daemon transfers data about where the diskless station can find its server, and what its swap-area and root
directory are called in the file tree of this server.
Diskless workstations swap to files. The command mkfile is used to create a fixed-size file for swapping.
Cloning systems
We are almost never interested in installing every machine separately. A real system administrator usually has to install ten, twenty or even a hundred
machines at a time. She would also like them to be as far as possible the same, so that users will always know what to expect. This might sound like a
Monday, July 26, 1999
Principles of system administration
Page: 61
straightforward problem, but it is not. There are several approaches.
Some operating systems provide a solution to this using package templates so that the installation procedure becomes standardized.
You copy the hard-disks of one machine physically and then change the hostname and address afterwards.
You can put all your software on one machine and share the disk via NFS.
Each of these approaches has its attractions. The NFS approach is without doubt the least amount of work, since it involves installing the software only once,
but it is also the slowest in operation for users.
As an example of the first, here is how Debian GNU/Linux tackles this problem using the Debian package system:
Install one system
dpkg --get-selections > file
On the remaining machines you type
cat file | dpkg --set-selections
before you call dselect After that, just install.
Booting a machine
You should read the whole of this section before attempting to boot or reboot a system. You can cause damage to files and disks by your careless actions.
Booting unix
Normally it is sufficient to switch on the power to boot a UNIX system. Sometimes you might have to type `boot' or `b' to get it going. Unix systems can
boot in several different modes or run levels . The most common modes are called multi-user mode and single-user mode. On different kinds of unix, these
might translate into run-levels with different numbers, but there is no consensus. In single-user mode no external logins are permitted. The purpose of
single-user mode is to allow the system administrator access to the system without fear of interference from other users. It is used for installing disks or
when repairing filesystems, where the presence of other users on the system would cause problems.
In the olden days, the unix boot procedure was controlled entirely by a file called `/etc/rc' meaning `run commands' and subsidiary files like
`rc.local'. These files were no more than shell scripts. Newer versions of unix have made the boot process insufferably complex by introducing a
program called init. init reads a configuration file called `/etc/inittab' and a directory called `/etc/rc.d'. `/etc/inittab' defines a number of
run-levels, and starts scripts depending on what run-level you choose. The idea behind `inittab' is to make unix installable in packages, where each
package can be started or configured by a separate script. Which packages get started depends on the run-level you choose.
The default form for booting is to boot in multi-user mode. You should find out how to boot in single-user mode on your system, in case you need to repair
a disk at some point. Here are some examples.
Under the SunOS and Solaris versions of unix, one interrupts the normal booting process by typing stop a, where stop represents the `stop key' on the left
hand side of the keyboard. If you do this, you should alway give the `sync' command to synchronize disk caches and minimize filesystem damage.
Stop a
ok? sync
ok? boot -s
If the system does not boot right away, you might see the line
type b) boot, c) continue or n) new command
In this case, you should type
b -s
in order to boot in single-user mode.
Under the GNU/Linux operating system, using the LILO boot system, you need to interrupt the normal boot sequence by pressing the SHIFT key. You
should then see the prompt
To boot, you must normally specify the name of a kernel file which is `linux'. To boot in single-user mode, you should type
Boot: linux single
Or at the LILO prompt, you can type ? in order to see a list of kernels. There appears to be a bug in some versions of GNU/Linux so that this does not have
the desired effect. In some cases you will be prompted for a run-level. The correct run-level should be determined from the file `/etc/inittab'. It is
normally called S or 1 or even 1S.
Once you are in single-user mode, you can return to multi-user mode just by exiting the single-user login.
Shutting down unix
Anyone can start a unix system, but you have to be the superuser to shut one down correctly. Of course, you could just pull the plug, but if you are a
well-informed system administrator, then you will know that this can ruin the disk filesystem. Even when no users are touching a keyboard anywhere, a unix
Monday, July 26, 1999
Principles of system administration
Page: 62
system can be writing something to the disk--if you pull the plug, you may interrupt a crucial write-operation which destroys the disk contents.
The correct way to shutdown a unix system is to run one of the following programs.
Stops the system immediately and without warning. All processes are killed with the TERMinate signal 15 and disks are synchronized.
reboot As halt, but the system reboots in the default manner immediately.
This program is the recommended way of shutting down the system. It is just a friendly user-interface to the other programs, but it warns the users
of the system about the shutdown and allows them to finish what they are doing before the system goes down.
Here are some examples of the shutdown command. The first is from BSD unix:
shutdown -h +3 "System halting in three minutes, please log out"
shutdown -r +4 "System rebooting in four minutes"
The option -h implies that the system will halt and not reboot automatically. The option -r implies that the system will reboot automatically. The times are
specified in minutes.
System V unix R4 (e.g. solaris) has a different syntax which is based on its confusing system of run-levels. The shutdown command allows one to switch
run-levels in a very general way. One of the run-levels is the `not running' or `halt' run-level. To halt the system, we have to call this.
shutdown -i 5 -g 120 "Powering down os...."
The `-i 5' option tells SVR4 to go to run-level 5, which is the power-off state. Run level 0 would also suffice here. The `-g 120' option tells shutdown to
wait for a grace-period of 120 seconds before shutting down.
WARNING! Never assume that the run levels on one system are the same as those on another. This would be a big mistake.
Mounting NFS disks
The sharing of disks over the network is the province of NFS (Network File System). Unix disks on one host may be accessed across the network by other
UNIX hosts, or by PC's running PC-NFS. A disk attached physically to a host called a server is said to be mounted on a client host. In order to maintain
a certain level of security, the server must give other hosts permission to mount disks. This is called exporting or sharing disks.
Server side exporting
In order to mount a disk on a server you must export the disk to the client (this is done on the server) and you must tell the client to mount the disk.
Permission to mount disks is given on the server in a file which is called `/etc/exports' or on recent SVR4 hosts `/etc/dfs/dfstab'. The format for
information in these files differs from system to system so you should always begin by looking at the manual page for these files. Here are two examples.
The first is from GNU/Linux
# See exports(5) for a description.
# This file contains a list of all dirs exported to other computers.
# It is used by rpc.nfsd and rpc.mountd.
/iu/borg/local daystrom(rw) worf(rw) nanite(rw)
In this example, a file system called `/iu/borg/local' is exported read-write explicitly to the client hosts daystrom, worf, and nanite. It is also exported
read-only to any host in the domain This last feature is not available on most types of UNIX.
On some brands of unix (such as SunOS 4.1.*) one must run a command after editing this file in order to register the changes. The command is exportfs
-a to export all filesystems. The command exportfs alone shows which filesystems are currently exported and to whom.
Our second example is from Solaris (SVR4). The file is called `/etc/dfs/dfstab'. Under Solaris, one can use the share command to export filesystems
manually from the shell, using a command line of the form
share -F nfs -o rw=hostname
The `/etc/dfs/dfstab' file is in fact a shell script which simply executes such a command for each filesystem of interest. This has several advantages
over traditional export files, since one may define variables, as in the example below.
place share(1M) commands here for automatic execution
on entering init state 3.
share [-F fstype] [ -o options] [-d "<text>"] <pathname> [resource]
share -F nfs -o rw=engineering -d "home dirs" /export/home2\\\
rw=$hostlist /iu/nexus/local
rw=$hostlist,root=dax /iu/nexus/u1
rw=$hostlist,root=dax /iu/nexus/u2
rw=$hostlist,root=dax /iu/nexus/u3
rw=$hostlist,root=dax /iu/nexus/u4
rw=$hostlist /var/mail
Monday, July 26, 1999
Principles of system administration
Page: 63
This script exports the six named filesystems, read-write to the entire list of hosts named in the variable hostlist. The command shareall runs this script,
or it can be run manually by typing sh /etc/dfs/dfstab. The command share without arguments shows the currently exported filesystems. Notice that
the hostname pc-torejo is repeated, once unqualified and again with a fully qualified hostname. This is sometimes necessary in order to make the entry
recognized. The mount daemon is not particularly intelligent when it verifies hostnames. Some systems send the fully qualified name to verify and others
send the unqualified name. If in doubt, list both like this.
Client side mounting
Clients may mount any subdirectory of the exported directory onto any local directory by becoming root and either executing a shell command of the form
mount server :remote-directory
or by adding a line to the filesystem table file, usually called `/etc/fstab'. On some brands of unix, this file has been renamed as `/etc/checklist' or
`/etc/filesystems'. On Solaris systems it is called `/etc/vfstab'. The advantage of writing the disks in the filesystem table is that the mount
commands will not be lost when you reboot your system. The filesystems in the filesystem table file are mounted automatically when the system is booted.
You can also mount all file systems in this file with the simple command mount -a.
You should begin by looking at the manual page on the appropriate file for your system, or better still looking at examples which are already in the file. The
form of a typical filesystem table is as below(6) .
This example is from GNU/Linux. Notice the left hand column. These are disks which are to be mounted. The first disks which begin with `/dev' are local
disks, physically attached to the host concerned. Those which begin with a hostname followed by a colon (in this case host nexus) are NFS filesystems
which lie physically on the named host. The second column in this table is the name of a directory on which the disk or remote filesystem is to be mounted
on ---i.e. where the files are to appear in the local host's file-tree. The remaining columns are options and filesystem types: rw means mount for read and
write access, bg means `background' which tells mount to continue trying to mount a filesystem in the background if it fails on a first attempt.
Editing the `/etc/fstab' (or equivalent) file is a process which can be automated very nicely with the help of the system administration tool cfengine. We
shall discuss this in the next chapter.
If you get a message telling you `Permission denied' when you try to mount a remote filesystem, you may like to check the following:
Did you remember to add the name of the client to the `export' or `dfstab' file on the server?
Some systems require a fully qualified hostname (i.e. hostname with domainname appended) in the export file. Try using this.
Did you mis-spell the name of the client or the server?
Are the correct network daemons running which support nfs? On the server side, you must be running mountd or rpc.mountd. This is an
authentication daemon. The actual transfer of data is performed by nfsd or rpc.nfsd. On older systems there should be at least four of these
daemons running to handle multiple requests. Modern systems use a multi-threaded version of the program, so that only one daemon is required.
On the client side, some systems use the binary input/output daemon to make transfers more efficient. This is not strictly necessary to get NFS
working. This daemon is called biod on older systems and nfsiod on newer systems like FreeBSD. Solaris no longer makes use of this daemon,
its activities are now integrated into a kernel thread.
Installing a new disk
Adding a new disk or device to a unix machine generally involves a little planning. The first thing you should know about your system is what kind of
hard-disks you use already, and what kind you intend to use in the future. If you do not know much about hard-disks, it might be a good idea to talk to
someone who can advise you.
The first concern is what type hard-disk. There are several types of disk interface used for communicating with hard-disks. Some are cheap and cheerful
(IDE), while others are more expensive and reliable (SCSI).
You can use a maximum of 2 disks and the size of the disks may be limited to less than a megabyte, but these disks are cheap and cheerful.
EIDE An extended version of the IDE interface. Allows 4 disks.
SCSI (Small computer systems interface)
Most small Unix systems use SCSI disks. This interface can be used for devices other than disks too. It is better than IDE at multitasking. The
original SCSI interface was limited to 7 devices in total per interface. Wide SCSI can deal with 14 disks. SCSI interfaces are developing rapidly and
can be found in SCSI I and SCSI II flavours, with fast, wide, differential. Talk to a dealer about the possibilities.
On some PC's IDE disks boot before SCSI disks because of the way the BIOS is configured. You can get problems with a PC with a SCSI disk and IDE if
you want to boot from the SCSI device.
Here is a typical checklist for adding a SCSI disk to a Unix system.
Connect disk and terminate SCSI chain with proper terminator.
Set the SCSI id of the disk so that it does not coincide with any other disks.
On SUN machines you could now use the ROM command probe-scsi to check that the system is able to find all your disks.
Partition and label the disk. Update the defect list.
Edit the `/etc/fstab' filesystem table, or equivalent to mount the disk. See also next section.
Monday, July 26, 1999
Principles of system administration
Page: 64
`mount' and `umount'
To make a disk partition appear as part of the file tree it has to be "mounted". We say that a particular filesystem is mounted on a directory. The command
`mount' mounts filesystems and disks defined in the filesystem table file. This is a file which holds data for mount to read.
The filesystem table has different names on different implementations of UNIX.
Solaris 1 (SunOS)
Solaris 2
IRIX /etc/fstab
OSF1 /etc/fstab
These files also have different syntax on different machines. The eventual standard which most systems comply with (SunOS, HPUX, OSF1) is
# SunOS 4* / Solaris 1
4.2 rw
4.2 rw
1 1
1 2
rw,suid 0
rw,suid 0
rw,suid 0
rw,suid 0
0 1
0 2
0 0
The syntax of the command is
mount filesystem directory type (options)
There are two main types of filesystem -- a disk filesystem (ufs, hfs) (which means a physical disk) and the NFS network filesystem. If we mount a 4.2
filesystem it means that it is, by definition, a local disk on our system and is described by some logical device name like `/dev/something'. If we mount an
NFS filesystem, we must specify the name of the filesystem and the name of the host to which the physical disk is attached.
Here are some examples, using the SunOS filesystem list above:
-at nfs
-at 4.2
all in fstab
all in fstab which are type nfs
all in fstab which are type 4.2
only this fs with options given in fstab
(The `-t' option does not work on all UNIX implementations.) Of course, we can type the commands manually too, if there is no entry in the filesystem
table. For example, to mount an nfs filesystem on machine `gandalf' called `/site/gandalf/local' so that it appears in our filesystem at
`/mounted/gandalf', we would write
mount gandalf:/site/gandalf/local /mounted/gandalf
The directory `/mounted/gandalf' must exist for this to work. If it contains files, then these files will no longer be visible when the filesystem is mounted
on top of it, but they are not destroyed. Indeed, if we then unmount using
umount /mounted/gandalf
(the spelling is correct) then the files will reappear again. Some implementations of NFS allow filesystems to be merged at the same mount point, so that the
user sees a mixture of all the filesystems mounted at the same point.
Disk device names
The convention for naming disk devices in BSD and system 5 unix differs. Let us take SCSI disks as an example. Under BSD, the SCSI disks have names
according to the following scheme
Monday, July 26, 1999
Principles of system administration
First partition of disk 0 of the standard disk controller. This is normally th root file system `/'.
Second partition of disk 0 on the standard disk controller. This is normally used for the swap area.
Third partition of disk 1 on the standard disk controller. Note that this partition is usually reserved to span the entire disk, as a a reminder of how
large the disk is.
System 5 unix employs a more complex, but also more general naming scheme. Here is an example from solaris 2:
Disk controller 0, target (disk) 3, device 0, segment (partition) 0
Disk controller 1, target (disk) 1, device 0, segment (partition) 4
Not all systems distinguish between target and device. On many systems you will find only t or d but not both.
Registering a printer
The way in which you register a printer depends (i) on what kind of UNIX you are using, and (ii) on whether you are running any special network printer
software. The main difference is between BSD-like systems and System V.
There are also two types of printer which you might want to register: (i) a printer which is connected physically to a UNIX host by a cable attached to its
printer port, and (ii) a stand-alone printer which is simply coupled to the network with its own IP address.
In order to use a printer, you need to do the following:
Think of a name for your printer.
Decide whether it is going to be connected directly to a host or stand alone on the network.
Make a `spool' directory for its queue files. This should probably lie under `var/spool' or `/usr/spool'.
mkdir /var/spool/printer-name
Register the printer with the printing system so that the daemons which provide the print service know how to talk to it. You will need to decide
whether you will be sending all data to a common central server, or whether you will be attaching printers locally to several machines, each running
their own print-server.
Most unix systems assume the existence of a default printer which is referred to by the name `lp'. If you do not specify a particular printer when printing,
your data are sent to the default printer. It is up to you to name or alias one of your printers `lp'. Each printer may have several names or aliases.
BSD printer with lpd
The file `/etc/printcap' is used to register a printer with a BSD system. If you are lucky, there might be a script or a user-interface for helping you to
register a printer, if not you will have to follow the recipe below.
The format of the `/etc/printcap' file can be quite simple in most cases. The manual page for printcap contains a description of the file format. This
file consist of a list of entries (each on a single line, or split over several lines using the continuation character `\'). A simple template entry looks like this:
printer-name-1 |printer-alias-1 |printer-alias-2 :\
:sd=spool-directory :\
:rm=remote machine or IP address of printer :\
:rp=name of remote printer on remote machine :
This file should be installed on all hosts which need to access the printer, regardless of whether the printer is physically attached to them or not. Here is an
example which registers two printers. The first is called `myprinter' and is connected physically to the remote host nexus. The second is a stand-alone
printer which we have named `diff-engine' and has IP address
# /etc/printcap
myprinter|lp|default|SPARCprinter, a Sun SPARCprinter printer:\
diff-engine|HP laser stand-alone:\
Note that the rp field exists in case a given printer has a different name on the remote host to the one you have given it locally on your machine. On a
stand-alone printer this name is irrelevant.
Sys V
Page: 65
Monday, July 26, 1999
Principles of system administration
Page: 66
The lpadmin command is used to install printers under system 5. This commmand is complex and has many command line options to add and remove
printers. If you have system 5 systems, consult the manual page on your system.
Sun Microsystems provide a script front end to help simplify this procedure called AdminTool.
A recent and welcome addition to the printer debate is the Next Generation LPR package by Patrick Powell. LPRng is a drop in replacement for both BSD
and system 5 print systems. It is configured quite simply in a manner very similar to (but not identical to) the Berkeley printcap system. LPRng can be
obtained from @uref{} and it is a real god send if you have system 5 (e.g. Solaris) hosts to grapple with. The
software uses a printcap file and two other optional files called `lpd.conf' and `lpd.perms'. The printcap file is like a regular printcap file but without the
backslash continuation characters.
LPRng provides effectively both `lpr', `lpd', `lpq' and `lprm' commands from Berkeley and `lp', `lpstat' and `cancel' commands from system
5. The daemon reads the three configuration files an handles spooling. The configuration is challenging but straightforward and there is extensive
documentation. Here is a simple example for a network printer (with its own IP address) which allows logged on users to start and delete their own
# /etc/printcap (lprng)
# LF/CR filter
The IP address of the printer is xxx.yyy.zzz.mmm and it must be written in numerical form. The percent symbol marks the standard port 9100. The
`lpd.conf' file is slightly mysterious but has a number of useful options. Most, if not all of these can be set in the printcap file also, but options set here
apply for all printers. One nice feature for instance is the ability to reject printouts of binary (non-printable) files. This can save a few rain forests if someone
is kind enough to dump `/bin/ls' to the printer.
# lpd.conf
# Purpose: name of accounting file (see also la, ar)
# Purpose: accounting at start (see also af, la, ar)
as=jobstart $H $n $P $k $b $t
# Purpose: check for nonprintable file
# Purpose: default printer
# Purpose: error log file (servers, filters and prefilters)
# Purpose: lpd lock file
# Purpose: lpd log file
# Purpose: /etc/printcap files
# Purpose: suppress headers and/or banner page
The `lpd.perms' file sets limits on who can access the printers and from where, unlike the traditional services which are open to everyone.
# lpd.perms
# allow root on server to control jobs
# allow anybody to get status
# reject all others, including lpc commands permitted by user_lpc
# allow same user on originating host to remove a job
# allow root on server to remove a job
Monday, July 26, 1999
Principles of system administration
Page: 67
# All other operations disallowed
Environment variable PRINTER
The BSD print command and some application programs read the environment variable PRINTER to determine which printer destination to send data to. The
System V print command lp does not.
Setting up an X-terminal
An X-terminal is a computer without a CPU. It cannot execute programs, but it has just enough power to display an X-windows graphical desktop.
X-terminals have no disk, but they need to run software called an X-server. They need to download this program from the network. Each X-terminal
manufacturer provides software for its X-terminals. As a system administrator you must decide on a boot-server for X-terminals: i.e. a system which will
transfer X-terminals' software to them when they boot and tell them what their internet addresses are.
Most X-terminals boot using a protocol called bootp in which the X-terminal sends a broadcast signal to the subnet asking for a boot server. A unix
machine which runs the bootpd daemon bootpd daemon looks at the ethernet address of the X-terminal sending the request and uses a configurable table to
determine an internet address for it. It returns this information to the X-terminal, together with the filename of the software it needs to download. Most
X-terminals then download their software using the tftp (trivial file transfer program) protocol, which is a way of downloading files without a password
from a restricted area of the server's disk.
The server keep all the X-terminals files in a directory which is usually called `/tftpboot'(8). In order to make the chosen server work for the X-terminal,
it must be configured to run the tftp and bootp services. This is done as follows:
Edit the file `/etc/services' on the server and add the protocol lines for these services if they do not already exist, See section Installing a new
# boot program server
Check to see whether you have an up to date bootpd. The version supplied by some vendors is old. You can collect a modern version by
anonymous ftp from the internet, See section Internet searches. Compile and install this program.
The bootp configuration table is called `/etc/bootptab'. Remember that keeping configuration files in a safe place is always a good idea. Store
this file in your site-dependent files and make a symbolic link to `/etc/bootptab' so that you don't lose the file if you re-install the OS. The
format of this file is like this:
# Legend:
bf -#
bs -#
cs -#
df -#
dn -#
ds -#
ef -#
gw -#
ha -#
hd -#
hn -#
ht -#
im -#
ip -#
lg -#
lp -#
ns -#
ra -#
rl -#
rp -#
sa -#
sm -#
sw -#
tc -#
td -#
to -#
ts -#
vm -#
Tn --
(see bootptab.5)
field -- hostname (not indented)
bootfile size in 512-octet blocks
cookie servers
dump file name
domain name
domain name servers
extension file
hardware address
home directory for boot-files
host name set for client
hardware type
impress servers
host IP address
log servers
LPR servers
IEN-116 name servers
reply address
resource location protocol servers
root path
boot server address
subnet mask
swap server
template host (points to similar host entry)
TFTP directory
time offset (seconds)
time servers
vendor magic number
generic option tag n
# Define two variables/macros
#name :
:sm= :tc=.default:
Monday, July 26, 1999
Principles of system administration
Page: 68
Finally you must edit the file `/etc/inetd.conf' which is used to start the daemons for the bootp and tftp services. Make sure that you have lines
of the following forms (BSD style)
tftp dgram udp wait root /usr/sbin/in.tftpd in.tftpd -s /path/tftpboot
bootp dgram udp wait root /local/bin/bootpd bootpd -i -d
Note that the path names above should be altered to fit in with your local configuration. You should use the true path-names here, not use symbolic
links. Security issues prevent tftpd from reading through symbolic links. When you are finished editing the file you must restart the inetd
daemon, or send it the hang-up signal using kill -HUP pid .
If you have trouble setting up X-terminals, here is a troubleshooting list.
X-terminal does not find a bootp server: check that the configuration in `/etc/inetd.conf' is correct for the bootp-daemon. If your bootp setup
is working, you should be able to type telnet server bootp and obtain a response.
Error message permission denied on X-terminal while downloading software or fonts: Could be that the pathname to `tftpboot' directory was
incorrect or was only a symbolic link to the real directory. You must use the real directory name here. Check that the software files are readable to
PPP internet over serial lines
PPP stands for the Point to Point Protocol. It is a way of using a computer whose only link to the outside world is a serial line (a modem coupled to the
telephone, for instance). Other protocols exist for this purpose: SLIP, for instance or Serial Line Internet Protocol, but PPP is generally regarded as better.
To run PPP, say from your portable or home computer using a modem, you need a program which talks PPP. On unix systems this program is called pppd.
This daemon provides the backbone of the PPP service. It opens a logical network interface, analogous to the ethernet interface eth0, called ppp0. When
PPP is configured, you can see this interface just like a standard network interface by typing ifconfig -a. The pppd program is an ungraceful workhorse
which requires an enormously long command line. It requires the assistance of another program called chat to communicate with the modem. Both pppd
and chat need to be available and in your command path.
PPP setup depends on several things: the kind of unix you are starting from, the kind of system you are logging on to and the kind of modem that you have.
Obviously, you do not want PPP to be using your telephone continuously, so you need to write a script which will switch the service on when needed and off
when not needed. Let's take a look at such a script.
# set path to pppd and chat if necessary
pppd connect 'chat -v ABORT BUSY ABORT "NO CARRIER" \
"" ATDTteleph-num CONNECT "" ">" username ">" passwd ' \
-detach debug defaultroute netmask noipdefault \
: ipcp-accept-local ipcp-accept-remote -vj \
proxyarp /dev/ttyS0 38400
The above script is supposed to be typed in on a single line. The backslash characters at the ends of the lines above are shell continuation characters which
mean that the following lines are to be treated as one.
This script is just an example, it is unlikely that you will be able to use it directly. Instead, you should obtain instructions from the local service you
are dialing.
In the chat part, double quoted strings such as ">" are text which the modem can expect to receive and should be waited for. These are generally
prompts. On other systems they might be strings like `Username:' and `Password:'.
The username, password and telephone number and netmask values relate to the remote system.
The device `/dev/ttyS0' is the serial device. On some systems this is called `/dev/modem', on others it will be called something else. You will
need to find out.
The final number is the modem speed at which the connection is to be tried. There is no harm in setting an optimistic transfer rate here, since
modems negotiate their own transfer rates according to what is possible at the time of connection. This is the value they start at, and if it does not
succeed the modems will try successively slower rates until a working connection is established.
Maintenance procedures
@hrule @vskip 0.3cm
One of your main jobs as a system administrator is to maintain the system. This includes adding new information to databases, upgrading and installing
software, adding users and tidying up when disks get name just a few things. In the next chapter we'll look at a tool which will help you to automate
some of these procedures: here we describe what you need to know before you can do that. Remember that, when ever you wield the power of the system
administrator, you can cause great damage. Work carefully and test your changes: do not just assume that they will work. Even small changes are often
followed by hundreds of mail messages from angry users because of something you broke while doing your job.
Configuring DNS/BIND
The name service must be configured in order for a system to be able to look up hostnames and internet addresses. The most important file in this
Monday, July 26, 1999
Principles of system administration
connection is `/etc/resolv.conf'. Ancient IRIX systems seem to have placed this file in `/usr/etc/resolv.conf'. This old location is extremely
obsolete. Without the resolver configuration file, a host will probably stop dead whilst trying hopelessly to look up internet addresses. The most important
features of this file are the definition of the domain-name and a list of nameservers which can perform the address translation service. These nameservers
must be listed as IP numerical addresses (since DNS can't look up any names until it knows the name of a server to look them up on, and that's what we're
trying to do, nez pah?). The format of the file is as shown in the box below.
As you should know, DNS has several competing services. A mapping of hostnames to IP addresses is also performed by the `/etc/hosts' database.
Moreover, network version of this file can also be shared using NIS or NISPLUS. As a result, most Unix variants allow you to choose the order in which
these competing services are given priority when looking up hostname data. Unfortunately there is no standard way of configuring this. GNU/Linux and
public domain resolver packages for old SunOS (resolv+), use a file called `/etc/hosts.conf'. The format of this file is
order hosts,bind,nis
multi on
This example tells the lookup routines to look in the `/etc/hosts' file first, then to query DNS/BIND and then finally look at NIS. The resolver routines
quit after the first match they find, they do not query all three databases every time.
Solaris uses a file called `/etc/nsswitch.conf' which is a general configuration for all database services, not just the hostname service.
An example file that could be copied over to /etc/nsswitch.conf; it
does not use any naming service.
"hosts:" and "services:" in this file are used only if the
/etc/netconfig file has a "-" for nametoaddr_libs of "inet" transports.
files dns
The print queue
BSD print queue
`lpr -p printer
Send file to named print queue.
`lpq' Show the printer queue for the default printer, or the printer specified in the environment variable PRINTER if this is set. This lists the queue-ids.
`lprm queue-id '
Remove a job from the print queue. Get the queue id using lpq.
Start the print service. (Must be killed to stop again)
An incredibly stupid user interface for print administration. This program tells lies.
SysV print queue
`lp -d printer
file '
Send a file to the named print queue.
`lpstat -o all'
Show the printer queue for the default printer. This lists the queue-ids.
`lpstat -a'
Tells lies about when the print service was started.
Start the print service
Stop the print service.
`cancel queue-id '
Remove a job from the print queue. Get the queue id using lpstat.
The solaris operating system has an optional printing system called Newsprint in addition to the SVR4 printing commands.
Building a kernel
In the past it was necessary to recompile the operating system kernel for every new device you wanted to add to a unix system The kernel was an enormous
statically linked program.
Page: 69
Monday, July 26, 1999
Principles of system administration
Page: 70
In modern unix systems, the kernel is divided up into modules which can be loaded and reloaded as needed. Under system 5 (SVR4) unix, you can
reconfigure some aspects of the kernel online without having to stop the system.
Solaris (SVR4)
Edit the file `/etc/system' to change kernel parameters. Reboot if necessary with the command
reboot -- -r
which reconfigures the system automatically.
The linux kernel is subject to more frequent revision than other systems. It must be recompiled when new changes are to be included, or when an
optimized kernel is required. Change directory to `/usr/src/linux'. The command make config can be used to set kernel parameters. A more
user friendly windows based program make xconfig is also available.
These days it is often unnecessary to build custom kernels. The default kernels supplied with many OSes are good enough. Still, performance enhancements
are obtainable, particularly on busy servers.
Installing software
Unlike many other systems, UNIX has grown up around people who write their own software rather than relying on off-the-shelf products. The internet
contains gigabytes of software for UNIX systems which is completely free. Large companies like the oil industry and newspapers can afford off-the-shelf
software for UNIX but most people can't.
There are therefore two kinds of software installation: the installation of commercial software and the installation of freeware. Commercial software is
usually installed from a CD by running a simple script and by following the instructions carefully; the only decision you need to make is where you want to
install the software. Freeware usually comes in an uncompiled form and must therefore be compiled. UNIX programmers have gone to great lengths to make
this process as simple as possible for system administrators.
Structuring software
The first step in installing software is to decide where you want to keep it. You could put the software anywhere you like, but you should bear in mind the
You should keep installed software separate from the operating system installed files, so that the OS can be reinstalled or upgraded without ruining
your software installation.
Compiled software should be grouped together, with a `bin' directory and a `lib' directory so that binaries and libraries conform to the usual
UNIX conventions. This makes the system more consistent and easier to understand and it also makes it easier to administer the PATH for
user-commands is programs are collected in only a few places.
You should try to keep files and programs which are special to your site separate from files which could be used anywhere.
The directory traditionally chosen for installed software is called `/usr/local'. One then makes sub-directories `/usr/local/bin' and
`/usr/local/lib' and so on.
UNIX has a de-facto naming standard for directories which you should try to stick to, so that others will understand how your system is built up.
`bin' Binaries or executables for normal user programs.
`sbin' Binaries or executables for programs which only system administrators require. Those files in `/sbin' are often statically linked to avoid problems
with libraries which lie on umounted disks during system booting.
`lib' Libraries and support files for special software.
`etc' Configuration files.
Files which might be shared by several programs or hosts. For instance, databases or help-information; other common resources.
Here is one suggestion for structuring installed software.
We divide these into three categories: regular installed software, GNU software and site-software. The reason for this is as follows:
`/usr/local' is the traditional place for software which does not belong to the OS. You could keep everything here, but you will end up installing
a lot of software after a while, so you might like to create two other sub-categories.
GNU software, which is written by the Free Software Foundation, forms a self-contained set of tools which replace many of the older UNIX
equivalents, like ls and cp. GNU software has its own system of installation and set of standards. GNU will also eventually become an operating
system in its own right, and should therefore be kept separate.
Site specific software includes programs and data which you build locally to replace the software or data which follows with your operating system.
It also includes special data like the database of `aliases' for e-mail and the DNS tables for your site. Since it is special to your site, you should
keep it separate so that you can back it up separately and you always know where to find site-specific stuff.
Monday, July 26, 1999
Principles of system administration
Page: 71
When you are installing software, you are expected to give the name of a prefix for installing the package. The prefix in the above cases is `/usr/local'
for ordinary software, `/usr/local/gnu' for GNU software and `/usr/local/site' for site specific software. Most software installation scripts place
files under `bin' and `lib' automatically.
To begin compiling software, you should always start by looking for a file called `README' or `INSTALL'. This will tell you what you have to do to compile
and install the software. In most cases, you will only have to type a couple of commands, like in the following example.
GNU software example
Let us now illustrate the GNU method of installing software which most others have since copied. The steps are as follows
Collect the software package by ftp from a site like or Use a program like ncftp for painless anonymous login.
Unpack the file using tar zxf software .tar.gz.
Enter the directory which is unpacked, cd software .
Type: configure --prefix=/usr/local/gnu
Type: make
If all goes well, type make install. This should be enough to install the software.
Some installation scripts leave files with the wrong permissions so that ordinary users cannot access the files. You might have to check that the files
have a mode like 755 so that normal users can access them.
This procedure should be more or less the same for just about any software pick up. Older software packages sometimes provide only Makefiles which you
must customize yourself. Some X11 based windowing software requires you to use the xmkmf X-make-makefiles command instead of configure. You
should always look at the README file.
Installing shared libraries
Systems which use shared libraries or shared objects sometimes need to be reconfigured when new libraries are added to the system. This is because the
names of the libraries are cached to fast access. The system will not look for a library if it is not in the cache file.
SunOS (not solaris)
After adding a new library, you must run the command ldconfig lib-directory . The file `/etc/' is updated.
You can add new library directories to the file `/etc/'. Then run the command ldconfig. The file `/etc/' is updated.
How often should I upgrade?
Some software (especially free software) gets updated very often. You could easily spend your entire life just chasing the latest versions of your favourite
software packages. Don't!
It is a waste of your time.
Sometimes new versions contain more bugs than the old one, and an even-newer-version is just around the corner.
Users will not thank you for changing things all the time. Stability is a virtue. Everyone likes time to get used to the system before you change it!
Installing a new service
Sooner or later you will find yourself installing software which requires it's own network service to be set up. You need to configure the system to accept a
new service by editing the file `/etc/services'. This file contains the names of services and their protocol types and port numbers.
The format of entries is like this:
There are two ways in which a service can run under unix: one is that a daemon runs all the time in the background, handling connections. This method is
used for services which are used often. Another way is to start the daemon only when an outside connection wishes to use it; this method is used for less
frequently used services. In the second case, a master `internet' daemon is used, which listens for connections for several services at once and starts the
correct daemon only long enough to handle one connection. The aim is to save the overhead of running many daemons.
If you want to run a daemon all the time, then you just need to make sure that you start the daemon in the appropriate `rc' start-up files for the system. To
add the service to the internet daemon, on the other hand, you need to add a line of the following form the the configuration file `/etc/inetd.conf'.
# service type protocol threading user-id server-program server-command
stream tcp nowait root /local/etc/pop3d
cfengine stream tcp nowait root /local/iu/bin/cfpush
cfpush -x -c
The software installation instructions will tell you what you should add to this file.
Once you have configured a new service, you must start it by running the appropriate daemon, See section Summoning daemons.
Mail queue
When mail cannot be delivered immediately it is placed in the mail queue. To see what mail is waiting in the mail-queue, you must log into the mailserver (i.e.
Monday, July 26, 1999
Principles of system administration
Page: 72
the host which handles outgoing mail) on your network. You can see what is in the queue by typing
nexus% mailq
nexus% sendmail -bp
These two forms are equivalent. You can force sendmail to process the mail queue by typing:
nexus% sendmail -q -v
Mail aliases
One of the first things you should locate on your system is the sendmail alias file. This is a file which contains e-mail aliases for users and system services.
Common locations for this file are `/etc/aliases' and `/etc/mail/aliases'. On some systems, the mail aliases are in the NIS network database.
If this file actually lies in the `/etc' directory, or some other place amongst the system files, then you should move it to your special area for site-dependent
files and make a symbolic link to `/etc/aliases' instead. Your mail aliases are valuable and you want to make sure that nothing happens to them if you
reinstall the OS.
The format of the mail aliases file is as follows:
# Alias for mailer daemon; returned messages from our MAILER-DAEMON
# should be routed to our local Postmaster.
postmaster: mark, toreo
MAILER-DAEMON: postmaster
nobody: /dev/null
# alias: list of addresses
drift:[email protected],[email protected]
root:[email protected],[email protected]
# Alias for distribution list, members specified elsewhere:
# alias : :include:file of names
lsk: :include:/iu/nexus/u2/wiikl/www/lsk/maillist/maillist
# Dump mail to a file
ace: /iu/nexus/local/iu/archive/ACEmailinglist
Setting clocks
You need to keep your system clock synchronized. A time-server is used for this purpose. The network time protocol daemon ntpd is used to synchronize
clocks from a reliable time server.
Another option for BSD-like systems is the rdate command, which sets the local clock according to the clock of another UNIX host. Some systems (like
GNU/Linux) do not have this command, but it can be emulated with a script like this one, provided remote-shell access is permitted by the server.
# Fake rdate script for linux - requires rsh access on server
echo Trying time server
DATE=`/bin/su -c '/usr/bin/rsh time-server
date' remote-user `
echo Setting date string...
/bin/date --set="$DATE"
Another more reliable way of keeping clocks synchronized in the long run is the use the NTP protocol, or network time protocol. The daemon xntpd can be
used to synchronize hosts against a master host. Two configuration files are needed to set up this service: `/etc/ntp.conf' and `/etc/ntp.drift'.
`/etc/ntp.conf' looks something like this, where the IP address is that of the master time server, whose clock you trust.
driftfile /etc/ntp.drift
authdelay 0.000047
Monday, July 26, 1999
Principles of system administration
Page: 73
The `/etc/ntp.drift' file must exist, but its contents are undetermined. you should touch this file.
Posix ACLs
ACLs, or access control lists are as modern replacement for file modes and permissions. With access control lists you can specify precisely the access rights
to files for each user individually. Earlier in Unix systems, it was necessary to make a group if you wanted to open a file for several users, but not for
ACLs were introduced in the DOMAIN OS by Apollo, and were then copied by Novell, HP and other vendors. A POSIX standard for ACLs has been
drafted, but as of today there is no standard for ACLs and each vendor has a different set of incompatible commands and datastructures. Sun Microsystem's
solaris (NFS3) implementation is based on the POSIX draft. We shall follow Solaris ACLs in this section. Not all Unix systems have ACLs. GNU/Linux
does not have them. If you grant access to a file which is shared on the network to a machine which doesn't support them, they will be ignored.
ACLs are literally lists of access rights. Each file has a list of data structures with pairs of names and permissions:
You specify an ACL by saying what permissions you would like to grant and which user or group of users the permissions apply to. An ACL can grant
access or deny access to a specific user. Because of the amount of time required to check all the permissions in an ACL, ACLs slow down file search
Under solaris, the commands to read and write ACLs have the horrible names
getfacl file
Examine the ACLs for a file
setfacl file
-s permission
Set ACL entries for a file, replacing the entire list.
setfacl file
-m permission
Set ACL entries for a file, adding to an existing list.
For example. If we create a new file, it ends up with a default ACL which is based upon the unix umask value. Suppose umask is 077, giving minimal rights
to others.
dax% touch testfile
dax% getfacl testfile
# file: testfile
# owner: mark
# group: iu
This tells us that a new file is created with read/write permission for the user (owner) of the file, and no other rights are granted. To open the file for a
specific user, one writes
dax% setfacl -m user:ds:rw- testfile
dax% getfacl testfile
# file: testfile
# owner: mark
# group: iu
To open a file for reading by a group iu, except for one user called robot, one would write:
dax% setfacl -m group:iu:r--,user:robot:--- testfile
dax% getfacl testfile
# file: testfile
# owner: mark
# group: iu
Monday, July 26, 1999
Principles of system administration
Page: 74
Notice that this is accomplished by saying that the group has read permission whilst the specific user should have no permissions.
NT Overview
@hrule @vskip 0.3cm
NT is a multitasking operating system from Microsoft which allows one user at a time to log in to a console. The consoles may be joined together in a
network with common resources shared by a `domain'. An NT host is either a network domain server or a personal workstation. NT is quite new, at least in
terms of age, and Microsoft has reinvented many well known systems rather than use tried and tested solutions. This has led to a history of bugs and
security issues and some disillusionment in the popular press. Together with GNU/Linux and IRIX, NT is the most popular target for network attacks. NT
has a long way to go before it becomes mature, so significant changes can be expected in the next few years. The pace of development is already rapid and
NT is getting steadily closer to Unix. Here we present a provisional overview as a point of reference and comparison.
The basic NT distribution contains few tools which can be used for network administration. The NT Resource Kit is an extra package which, for a tidy sum
of money, provides many essential tools. Other tools can be obtained free of charge.
NT can be both easier and harder to administrate than Unix. It can be easier because the centralized model of having one domain server running the network
services means that all configuration information can be left in one place (on the server), and that each workstation can be made (at least to a limited degree)
to configure itself from the server's files.
It is harder to administrate because the only tools provided for system administration tasks work by the GUI (graphical user interface ) and this is not a
suitable tool for addressing the issues of hundreds of hosts. The Resource Kit and free tools go some way to helping with this problem, but the Resource kit
adds a substantial sum to the cost of running NT. It provides tools analogous to cron and script languages for automating tasks.
NT does not have a remote shell login feature like Unix, so the administrative job cannot be handled remotely, unless the PC is tied specifically to a central
server for all of its maintenance. Fortunately the free Perl Win32 package and related tools provides tools for solving a number of problems with NT from a
script viewpoint.
Logging in
In order to log on to NT the user must type CTRL-ALT-DEL. This is a `feature' which is intended to serve a security purpose. Any login screen which one
might see prior to pressing this key combination is likely to be a Trojan horse program. Windows NT is not a multiuser operating system, only one user
may be logged in at any time. The superuser account is called the Administrator account. In fact any account which is a member of the group Administrators
has the same privileges. These accounts have the power to do anything with the system, just as in Unix. However it is possible to remove access rights to the
Administrator as a precautionary measure; the Administrator then has to grant him/herself access before continuing.
System layout
The first thing to learn about a new operating system is where all the important files live on the disk, where new software should be installed and where new
users should have home directories.
The layout of the NT filesystem has changed through the different versions, in an effort to improve the structure. This description relates to NT 4.0. The
system root is usually in `C:\WinNT' and is referred to in the system environment variable %SystemRoot% .
`I386' This directory contains binary code and data for the NT operating system. This should normally be left alone.
`Program Files'
This is NT's official location for new software. Program packages which you buy should install themselves in subdirectories of this directory. More
often than not they choose their own locations however, often with a distressing lack of discipline.
`Temp' Temporary scratch space, like Unix's `/tmp'.
This is the root directory for the NT system. This is mainly for operating system files, so you should not place new files under this directory
yourself unless you really know what you are doing. Some software packages might install themselves here.
Configuration information for programs. These are generally binary files so the contents of NT configuration files is not very interesting.
This is the so-called system root. This is where most system applications and datafiles are kept.
The registry
In addition to these files and directories, there is a set of files which is normally invisible called the system registry . The system registry is a place where
configuration information, preferences, boot info and options are kept. These files contain the analogue of Unix dot-files and `/etc' setup files, as well as
the kind of information which dmesg provides. They replace the `.INI' files of earlier windows systems. The hive
There are advantages and disadvantages to the registry approach to host configuration. A nice feature is that the database lies in one well-defined place. A
disadvantage is that it lies in the system workspace--meaning that it is not a distributed database. If users want special preferences their registry data have to
be installed manually on each host. This can be contrasted with Unix, say, where configuration files exist per-host and per-user. Thus what one gains in
Monday, July 26, 1999
Principles of system administration
Page: 75
tidiness, is lost in convenience.
The registry has a hierarchical structure made from `keys'. Keys are analogous to directories: they may hold values or sub-keys. The values of keys are
protected by ACLs. The registry files are called hives and lie in the directory `C:\WINNT\system32\config'. The default ACLs on the registry allow
everyone to read them.
These files can only be viewed by starting the registry editor program `regedt32'. There are six base keys which cannot be changed
Contains file associations and OLE (Object Linking and Embedding) data (Link to HKEY_LOCAL_MACHINE\SOFTWARE\Classes)
Contains data about the user who is currently logged in on the console. It is linked to HKEY_USERS.
Contains configuration data for the local machine. This contains information about installed software, analogous to GNU/Linux's `Packages.gz'
file, so that new software can be downloaded or upgraded from a server. It also contains setup information concerning running services, hardware
and applications.
This subtree contains user ID's (security ID's) for users who are known to the machine. This is analogous to the `/etc/passwd' and
`/etc/shadow' files on unix. It does not contain info about the SAM (security access manager).
Data about the local machine configuration which might change in the future.
Sub tree containing dynamically created performance data. For NT's private use only.
The information in the registry is updated mainly by application programs, but can also be edited using the Registry editor mentioned above. Perl for
WIN32 can also manipulate the registry data, and as of service pack 4 and NT 5.0, the registry can be managed remotely.
It is hazardous to change the registry without considerable care since ordinary users can lose control of the machine if the ACLs are changed.
User login directories
NT has no specified place for user directories, so you may place them where-ever you like. A naming scheme, such as the one described earlier for Unix
disk partitions can still be used even though NT cannot mount disks at arbitrary places in a file system.
Starting and stopping under NT
Like Unix, processes under NT can live in the foreground or in the background. A background process can be started with
start /B
In order to kill the process it is necessary to purchase the Resource kit which contains a kill command. A background process detaches itself from a login
session and can continue to run even when the user is logged out.
The shutdown of the whole system must be performed from the Windows menu. Any logged on user can shut down the a host. This is not a major problem
since only one user can use the system at a time. However, background processes die when this happens, including other users' background processes. A
shutdown command also exists for shutting down local or remote systems.
Mounting disks
Network disks and local disks must be mounted in NT just as in UNIX, but this occurs more transparently. Network disks become visible when you join an
appropriate NT domain. This is a nice feature, but there is a price for the simplification. It is not possible to add a remote (network) disk to your file tree at
an arbitrary point, as one can with Unix file systems. Each new device is assigned a drive number, a hang over from the old DOS ways. Disks may be
assigned A:, B:, C:, D: etc. Moreover, you don't have any control over what designation a device gets. Two machines might assign different drive names to
the same device, so be careful. Users home directories are attached to H: if possible. Although only a few years old NT has nearly as many historical quirks
as Unix.
It is rumoured that NT 5.0 will have support form transparent mounting at an arbitrary point within a directory structure as is the case with Unix file
Windows NT uses a system of network file sharing based on their own SMB (Server message block) protocol. Samba is a unix daemon based service which
makes unix disks visible to Windows NT.
Samba maps usernames, so you need an account with the same name on the NT server and the unix server. It maps username textually, without any fancy
Samba configuration is in true unix style, by editing the text-file `/etc/smb.conf'. Here is an example file.
Note carefully the `hosts allow' line which restricts access to disks to specific IP addresses. Without this option, anyone could fake packets and mess with
your disks.
printing = bsd
printcap name =
load printers =
guest account =
invalid users =
Monday, July 26, 1999
Principles of system administration
workgroup = UNIX
hosts allow = 128.39.
comment = Home Directories
browseable = no
read only = no
create mode = 0644
comment = All Printers
browseable = no
path = /tmp
printable = yes
public = no
writable = no
create mode = 0644
ACLs and ACEs
Access control lists, or Access control entries are set and checked with either the `Explorer' program (File/Properties/Security/Permissions menu) or the
cacls command. This command works in more or less the same way as the POSIX setfacl command, but with different switches. Also the lists apply to
usernames not groups. The switches are
Grant access to user.
Edit ACE instead of replacing.
Act on all files and subdirectories
Revoke (remove) access rights to a user.
Deny access rights to a given user.
The access rights are not read, write, execute as in Unix, but N (no rights, R (read), C (change/write), F (full).
hybrid> CACLS testfile
C:\home\mark\testfile BUILTIN\Administrators:F
hybrid> CACLS testfile /G ds:F
wait for 30 seconds..
Are you sure(Y/N)?
hybrid> CACLS testfile
C:\home\mark\testfile HYBRID\ds:F
In this example the original ACL consisted of three entries. We then replace it with a single entry for user ds on the local machine HYBRID, granting full
rights. The result is shown in the last line. If, instead of replacing the ACE, we want to supplement it, we write
hybrid> CACLS testfile /E /G mark:R
wait for 30 seconds
Are you sure(Y/N)?
hybrid> CACLS testfile
C:\home\mark\testfile HYBRID\ds:F
Unix/NT correspondence
As NT improves it has been steadily converging on Unix. It is not only interesting, but useful to compare the Unix and NT software interfaces.
Strangely it does not seem to be possible to rename directories under NT. Instead one has to cut and paste them using the explorer program! If you have
collected the Unix shell programs, you can use the usual mv command.
Here is a list of some Unix commands and their NT equivalents.
alias doskey Create a shell alias.
The at and soon commands in the resource pack for scheduling background batch tasks
type to print a file or copy a+b+c to concatenate
cd Change directory
chgrp None: files don't have group attributes.
chmod cacls Change access rights on a file.
compact/expand Compress and uncompress files. Also gzip
copy Copy a file
cp -r xcopy Copy directories recursively
cp -p scopy Copy preserving attributes
at Periodic task scheduler
Show disk usage. Must be simulated with script or free software additions.
Page: 76
Monday, July 26, 1999
Principles of system administration
Page: 77
net share Make disks available to other hosts via network.
chkdsk Check disk filesystem integrity.
ipconfig /all Ethernet interface configuration.
A kill command is in the Resource Kit.
ln -s None. Short cuts are similar to symbolic links.
mkdir mkdir Make a new directory.
mount net use Add a remote disk to the local host.
netstat Show network connections and statistics.
nslookup Query the DNS/BIND service.
passwd net user Change password.
pstat, pview in the Resource Kit show active processes.
A version of rsh is provided for login to Unix systems.
GNU tar, and pax available for packing/unpacking tape archives.
tracert Trace route through TCP/IP network.
uptime net statistics Show system statistics.
UNIX tools on NT
@hrule @vskip 0.3cm
GNU tools for NT
The Unix GNU tools from Cygnus software are obtainable free of charge. To obtain them, log on as Administrator.
Make directory `C:\usr\local' where the unix commands will be installed. (New software should really go into the `Program Files' directory,
according to the NT setup, but this package follows Unix formalities and uses `/usr/local'.)
Start the internet explorer or netscape if you have it. Go to a GNU ftp site, for instance `'. Go to `pub/win-nt/gcc' folder
and get everything.
If you can find gunzip or uncompress get them, but otherwise you will need to download the files to a Unix machine first and uncompress them
there with gunzip You can then use the explorer or the Windows NT ftp program to collect the unpacked tar file.
Move the tar file and `tar.exe' file to the `C:\' directory and start an NT/DOS command prompt. Type
tar xf gnu-bin.tar
and so on. The files unpack themselves.
To start a shell window which recognizes the unix commands, you will need to set up a `short-cut' to NT's CMD.EXE program and tell the command
window to execute a setup script. (NT command windows do not seem to execute a setup script analogous to `.cshrc' automatically.) Click on the
right (menu) mouse button over the Window interface background to bring up the icon menu. Select New from this menu and Shortcut from the
sub-menu which pops up. As the command line, type
C:\\winNT\system32\cmd.exe /K C:\usr\local\congruent\setenv C:
and choose a name for the short cut e.g. Unix-shell. A little icon pops up and when you click on this, it should now start a shell where you can use
all of the installed unix commands including the ntemacs editor.
The command shell you just started is still not very intelligent, although you can set up on-line editing and command history, it is not possible to use
filename or command completion. ntemacs can do this however. To get such a shell, type ntemacs and then ESC x shell.
Tcsh for NT
If you want a real Unix shell with completion and internationalization, there is a port of the popular tcsh program for windows NT. This could be a useful
environment for performing system tasks. You can collect a ready compiled binary from `ftp:/'. This shell supports
all of the usual unix syntax and job control goodies, with the exception of certain signal handlers which NT does not have.
David Korn of AT&T (the author of ksh, the Korn shell) has written an beautiful package of code for NT which allows you to run many standard unix
services for free! There is a development package for porting Unix programs to NT.
Whether you like unix or not, this package is invaluable since it gives you access to telnet services, rsh and many other goodies. Many of the GNU tools
are bundled with this package too. In other words, you can now remote log onto an NT machine and do all of the things you are used to doing on unix
except for running window based software.
The Uwin package belongs to AT&T, it is not free software in the sense of having access to source code or having the right to redistribute the package. But
this is mainly a formality and anyone can download the libraries. You can collect everything from
Monday, July 26, 1999
Principles of system administration
Page: 78
Recommended reading
Computer security , D. Gollmann, Wiley.
DNS and BIND , Paul Albitz and Cricket Liu, O'Reilly & Assoc.
TCP/IP Network administration , Craig Hunt, O'Reilly & Assoc.
Practical UNIX security , Simson Garfinkel and Gene Spafford, O'Reilly & Assoc.
Building Internet Firewalls , D.B. Chapman and E.D.Zwicky, O'Reilly & Assoc.
Windows NT: User administration , A.J. Meggitt and T.D. Ritchey, O'Reilly & Assoc.
Computer networks, a systems approach , L.L. Peterson and B.S. Davie, Morgan Kaufman.
Security reference ,
Bootstrapping a machine. This comes from the expression `to lift yourself by your bootstraps', which is supposed to reflect the way computers are
able to start running their programs from scratch.
Berkeley Internet Name Domain. The library part of DNS, the routines which perform nameservice lookups.
Complementary Metal Oxide Semiconductor. p-n back-to-back transistor technology, low dissipation.
A centralized mainframe type of solution for concentrating computing power in one place. This kind of solution makes sense for heavy calculations,
performed in engineering of computer graphics.
William Gibson's name for the virtual world of the net. From his novel Neuromancer.
A de-centralized solution, in which many workstations spread the computing power evenly throughout the network.
DNS The Domain Name Service, which converts internet names into IP addresses and vice versa.
A small business network environment. Enterprise management is a popular concept today because NT has been aimed at this market. If nothing
else, Microsoft know about marketing and have highlighted this important market far more effectively than the Unix vendors have done before them.
Enterprise management typically involves running a web server, a database, a disk server and a group of workstations and common resources like
printers and so on. Many magazines think of enterprise management as the network model, but when people talk about Enterprise management they
are really thinking of small businesses with fairly uniform systems.
Free software
This is a concept promoted by the Free Software foundation. Free means freedom to share with your friend, rather than gratis. Free software might
cost money (usually it does not), but it comes with a license which gives everyone the right to copy, distribute and alter the program for whatever
purpose they wish. As Richard Stallman (who founded the FSF) puts it, think `free speech' not `free beer'. Several companies make money out of
supporting and marketing Free Software, e.g. Cygnus and Red Hat.
Graphical user interface.
Internetworking protocol
A protocol which can send messages across quite different physical networks, binding them together into a unified communications base.
IP address
Internet address. Something like
LISA Large Installation System Administration. This refers to environments with many (hundreds or thousands of) computers. The environments
typically consist of many different kinds of system from multiple vendors. These systems are usually owned by huge companies, organizations like
NASA or universities.
Open Source
This is like free software. It is a kind of registration stamp for software for which the source code is available to the user. Mostly users are not
allowed to modify the code and resell it however. See
Open systems
is a concept promoted originally by Sun Microsystems for Unix, which is about all systems being compatible through the use of freely available
standards. While competitors might have to pay to implement a feature, they are not prevented from knowing how to include that feature or from
selling it later.
An Intel based personal computer, used by a single user.
Process identity number.
Proprietary systems
is the opposite of open systems. These systems are secret and the details of their operation is not disclosed to competitors. The most infamous
example is Microsoft, but there are many others.
RAID Redundant array of inexpensive (ha-ha) disks. A disk array with automatic redundancy and error correction. Can tolerate a failure of one disk in the
array without loss of data.
SCSI Small Computer Systems Interface. Used mainly for disks on multiuser systems and musical instruments.
Security identity number (NT)
SIMM Memory chip arrays.
Impersonation, faking, posing as a false identity.
A way of spreading data over several disk controllers to increase through out. Striping can be dangerous, since files are stored over several disks,
meaning that if one disk fails, all data are lost.
SVR4 System 5 release 4 unix. AT&T's code release.
Time to live/Transistor-Transistor Logic
User identity number (Unix)
URL Uniform resource locator. A network `filename' including the name of the host on which the resource resides and the network service (port number)
which provides it.
A company which sells hardware or software. A seller .
Monday, July 26, 1999
Principles of system administration
A non-Intel based personal computer which might be used by several users. Workstations might be based on for example SPARC (Sun
Microsystems) or Alpha (Digital/Compaq) chip sets.
The Unix windows system.
`.cshrc', `.cshrc'
`.mwmrc', `.mwmrc'
`.profile', `.profile'
`/etc/ethers', `/etc/ethers'
`/etc/fstab', `/etc/fstab'
`/etc/inetd.conf', `/etc/inetd.conf'
`/etc/rc' files
`/etc/resolv.conf', `/etc/resolv.conf'
`/usr/etc/resolv.conf' on IRIX
A record
Access control for services
Access control lists
ACEs in NT
ACLs, network services
Aliases in mail
Alive, checking a host
Anonymous ftp databases
arch program
Background process, NT
Bad primary partition error
Page: 79
Monday, July 26, 1999
Principles of system administration
Big endian
BIND version 8
BIND, setting up
Boot scripts
Booting unix
bootp protocol
BOOTP protocol
Broadcast address, Broadcast address
BSD 4.3
Byte order
Cache file, DNS
Cache poisoning
CACLS command
Canonical name
Canonical names
catman command
chat program
Checking the mode of installed software
Checking whether host is alive
Class A,B,C networks
Classes, compound
Classes, defining and undefining
Clock synchronization
Community string
Community strings
Components, handling
Compound classes
Contact with the outside world
cp command
cron, cron
Cron jobs, controlling with cfengine
crontab command
Daemons and services
David Korn
Day of the week
Death to the users
Default nameserver
Default printer, Default printer
Default route, Default route, Default route
Denial of service attack
df command
Disk backups
Disk doctor
Disk partition names
Disk quotas
Disk repair
Disk statistics
DNS cache file
DNS, BIND setup
DNS, mail records
Domain name
Domain name service
Domain name, definition
Domain OS
Domain, listing hosts in
Page: 80
Monday, July 26, 1999
Principles of system administration
DoS attack
Dots in hostnames
Down, checking a host
du command
dump command
etherfind command
exportfs command
Exporting on GNU/Linux
External hosts do not seem to exist
File type problem in WWW
Filesystem table
find command
Finding a mail server
Finding domain information
Finding the name server for other domains
format program, Sun
Formatting a filesystem
Fragmentation of UDP
Free software foundation
fsck program.
ftp databases
Fully qualified names
fvwm window manager
fvwm2 window manager
fvwm95 window manager
GNU software
Grouping time values
groups and time intervals
groups in cfengine
Handling components
Hangup signal
Home directory
Host name gets truncated
Host name lookup
Hostname lookup
ifconfig command, ifconfig command
inetd master-daemon
Interface configuration, Interface configuration
Internet domain
iostat command
IP address
IP address lookup
IP address, setting
IP addresses, IP addresses
Page: 81
Monday, July 26, 1999
Principles of system administration
kill command
kill, NT process
Korn Shell
ldconfig command
ldd command
Linux, exports
Little endian
locate command
Logical NOT
Login directory
Looking up name/domain information
Lookup hosts in a domain
Loopback address
Loopback network in DNS
lp, lp
lp default printer
lpr, lpr
lpstat -a
lpstat -o all
mach program
Mail address of administrator, Mail address of administrator
Mail aliases
Mail exchangers
Mail queue, Mail queue
Mail records in DNS
Mail spool directory
Mail, finding the server
Management information base
mkfile command
mount -a
mount command
Mounting filesystems
Mounting filesystems.
Mounting problems
Multi-port repeater, Multi-port repeater
multi-user mode
Multicast address
mwm window manager
MX records
Name service lookups
Nameserver for other domains
Nameserver list
Netmask, Netmask
netstat -r and routing table
netstat -r command
Page: 82
Monday, July 26, 1999
Principles of system administration
netstat command
Network address
Network byte order
Network information service
Network interface, Network interface
Network interfaces
Network numbers
Network, transmission method
newfs command
NFS client/server statistics
nfsstat command
No contact with outside world
NOT operator
Novell, Novell
nslookup, nslookup
NT GNU tar file
NT installation
NT, tcsh for
One time passwords, One time passwords
Operator ordering
OS/2 boot manager
Password sniffing
Permissions on installed software
Ping attacks
ping command, ping command
Point to point protocol
Port sniffing
Print spool area
Print-queue listing, Print-queue listing
Print-queue, remove job, Print-queue, remove job
Print-queue, start, Print-queue, start
Print-queue, stop, Print-queue, stop
Printer registration
Printer, choosing a default
Probe SCSI
ps command
PTR records
q=any, nslookup
q=mx, nslookup
q=ns, nslookup
`rc' files
rdump command
Registering a printer
Registry, NT, Registry, NT
renice command
Repairing a damaged disk
Repeater, Repeater
Resolver, setting up
Page: 83
Monday, July 26, 1999
Principles of system administration
Restarting daemons
restore command
Reverse lookup, DNS
rlogin command
rm -i command
Root partition
route command
Routing information
Routing table, Routing table
rsh command
Rsh for NT
Rule 2
Rule 3
Rule 8
`run-cfengine' file.
Running jobs at specified times
Scheduling priority
SCSI probe on SunOS
Security holes
Sequence guessing
Server message block
Service configuration
Services and daemons
Setuid programs
share, share
Simple Network Management Protocol
Single user mode
single-user mode
Site specific data
SMB protocol
Smurf attack
snoop command
Socket connections
Start up files for unix
Statistics, disks
Statistics, NFS
Statistics, virtual memory
Swap partition.
Swap space
swapon command
Swapping, switching on
Switched networks
SYN flooding
System accounting
System type, System type
tar command, tar command
TCP wrappers
TCP/IP spoofing
tcsh for NT
telnet command
Telnet for NT
Time classes
Time service
Time, executing jobs at specified
Page: 84
Monday, July 26, 1999
Principles of system administration
traceroute command
Twisted pair
ufsdump command
Undeleting files
Unix dot-files
Up, checking a host
updatedb script
Virtual memory statistics
vmstat command
whatis command
which command
whois command
X server
X-terminal troubleshooting
xhost access control
xhost command
This document was generated on 25 Febuary 1999 using the texi2html translator version 1.50.
Page: 85

Similar documents


Report this document